thoughts

find a research article for each strength (Utilize the ASU Library, this cannot be a website, or a newspaper/magazine article but an actual study) that supports how that strength (or a topic similar to your strength) contributes to overall happiness and provide the following:

How that strength contributes to happiness (your fact)

Don't use plagiarized sources. Get Your Custom Essay on

thoughts

Just from $13/Page

Order Essay

A paragraph (7 – 10 sentences minimum) on each thoughts on the study and how it contributes to your life

Citation…..APA or MLA format, just be consistent

6pts total – 1pt on how each strength contributes to your happiness and 1pt on the supplemental research article that supports what you are saying (2pts per strength X 3 strengths/research articles = 6pts)

Remember, websites will NOT be accepted as a credible source. Also, please be sure you cite all of your references (APA or MLA format). If you are going to use direct quotes, be sure to include quotation marks as well as PAGE NUMBERS of where you found that information. If you don’t, it could be treated as plagiarism since they are not considered your own words or ideas.

first thought : Leadership answer it from this book with citation and quotation of page number.

second thought : self control answer it from this book with citation of page number and book

third thought: Humor and playfulness , answer it from this book with citation and quotation with the page number.

Chapter 1
Introduction: Jokes, humor, and taste

The importance of a shared sense of humor is made obvious by its absence.
It is almost impossible to build a relationship with someone who never
makes you laugh, who never laughs at jokes you make or even worse: who
tries really hard to be funny but insists on telling the wrong jokes. Very few
things are more painful than an attempt at humor that is not appreciated by
those listening.

To tell the right joke at the right time requires considerable cultural
knowledge. Someone who doesn’t laugh when others do or who laughs
when the rest are silent, exposes himself as an outsider: he reveals his lack
of awareness of codes, habits, and rules. He doesn’t belong. In social rela-
tionships, humor has the role of measuring mutual understanding and sig-
naling good intentions. When a joke fails, listeners usually feel like reject-
ing the joke teller, and often do.

What counts as “good” humor differs from group to group, from person
to person, and from moment to moment. The extent to which people differ
in their opinions of what is funny is sharply illustrated in a form of humor
prominent in day-to-day interactions: in joke telling. The joke is a humor-
ous genre about which opinions are extremely mixed. There are true joke
lovers – proverbial uncles at parties producing enough jokes, one after the
other, to last the whole evening – but also self-declared adversaries. Certain
groups welcome jokes with great enthusiasm while others reject the telling
of a joke with demonstrative silence; the latter group sees joke telling as
tasteless and vulgar.

Judging jokes goes further than expressing personal style or taste. Sense
of humor is connected to social milieu and background. Not only are there
individual differences in how humor is appreciated but there are also dif-
ferences between men and women, between people with different educa-
tional advantages, between old and young, and of course differences be-
tween people from different cultures and countries. What people think is
funny – or not funny – is strongly determined by how they were brought up
and the company they keep.

This book has to do with the relationship between sense of humor and
social background. As the starting point for understanding and plotting
these social differences in sense of humor, I will be looking at how people

Kuipers, G. (2006). Good humor, bad taste : A sociology of the joke. Retrieved from http://ebookcentral.proquest.com
Created from asulib-ebooks on 2020-02-27 15:44:40.

C
o
p
yr

ig
h
t
©

2
0
0
6
.
D

e
G

ru
yt

e
r,

I
n
c.

.
A

ll
ri
g
h
ts

r
e
se

rv
e
d
.

2 Introduction: Jokes, humor, and taste

think about one specific humorous genre: the joke. More precisely: I will
be looking at the standardized or “canned” joke: a short humorous story,
ending in a punch line, which the teller usually does not claim to have in-
vented himself. Dutch, the language in which most of this research was
originally carried out, has a specific word for this genre: mop. In English,
“joke” can refer both to this specific genre and more generally to something
said or done to amuse people. However, even though there is no separate
word in English for the specific genre, the joke was clearly recognized as
being separate by the Americans I interviewed.

It may perhaps surprise the reader that I chose this, of all genres. The
subject “humor” is capable of suggesting something profound and of
prompting people to contemplate human nature, the importance of creativ-
ity, or the connection between suffering, humor, and detachment, but the
joke evokes many fewer grandiloquent associations. Jokes are amusement
more than anything else, without many pretensions or profound purposes:
they are meant to make people laugh and no more.

The joke is perhaps not a particularly chic genre but it does evoke em-
phatic reactions. As I did my research, I interviewed people who consid-
ered telling jokes to be “the acme of humor”, “just part of any good night’s
fun”, and even: “the essence of togetherness and pleasure, a reason to laugh
till you cry”. But I also talked to people who denigrated jokes as “a form of
spiritual poverty”, “tiresome things, enormously disrupting to conversa-
tion”. One of my informants stated very decidedly: “Jokes that’s not hu-
mor”. In saying these things, people are also stating implicitly what humor
means to them: what they consider funny, hilarious, corny, far-fetched,
vulgar, or banal. They are verbalizing a decision about good and bad hu-
mor. But just behind the scenes, ideas are lurking about what a good con-
versation entails, what an enjoyable evening looks like, what being sociable
means, and more generally: how people are supposed to interact with each
other. Judgments about humor are directly connected with ideas about what
constitutes pleasant and unpleasant communication.

Researching jokes

This variance of opinion itself makes the genre of the joke a suitable start-
ing point for research into differences in how humor is appreciated. Re-
search into preference for and aversion to jokes leads to more general ques-
tions about humor: why do some people love certain forms of humor while
others can’t stand them? Which subjects are preferred joke material and

Kuipers, G. (2006). Good humor, bad taste : A sociology of the joke. Retrieved from http://ebookcentral.proquest.com
Created from asulib-ebooks on 2020-02-27 15:44:40.
C
o
p
yr
ig
h
t
©
2
0
0
6
.
D
e
G
ru
yt
e
r,
I
n
c.
.
A
ll
ri
g
h
ts
r
e
se
rv
e
d
.

Researching jokes 3

which are not? Why do people consider something funny, amusing, hilari-
ous or, rather, corny, feeble, or vulgar? What do people mean by “sense of
humor”? How do people differ in their opinions on this? And what are the
consequences of such differences in humor style?

In order to answer these questions I conducted interviews, I did a survey
of humor styles, and I collected a large number of jokes. The research on
which this book is based was carried out mostly in the Netherlands. How-
ever, the final chapter presents the results of a similar, though smaller,
study in the United States. In the Netherlands, I spoke extensively to sev-
enty Dutch people about jokes and humor. First, I talked to thirty-four joke
lovers, people who knew and told a lot of jokes. These were acquired
through newspaper advertisements and through the grapevine, but primarily
gleaned from people participating in the selection for the Dutch television
program Moppentoppers, a program aired by RTL4, a large commercial TV
channel. Moppentoppers (the name is a contamination of toppers, which
means approximately the same in English, and moppentapper, a rather
jocular word for joke teller) was a highly popular joke-telling contest for
amateur joke tellers.

I also interviewed four editors of joke books. After that, I interviewed
thirty-two “ordinary people” about their sense of humor: men and women,
young and old, of different educational and professional backgrounds, joke
lovers and joke haters. All these interviewees, under fictitious names, will
be cited frequently in this book. The group of thirty-two was a sample
taken from a group of 340 Dutch people who had filled in a questionnaire
about jokes and humor in 1997/1998. In addition to this, I collected many
thousands of jokes: I found them on the Internet, in joke books and maga-
zines, in archives, and they were told to me by friends, acquaintances, and
people whom I interviewed.

Differences in the appreciation of jokes touch upon three of the most
important social distinctions in the Netherlands: gender, age, and particu-
larly class. In the Netherlands, like in other Western countries, it is often
said that class no longer plays a role of any importance. Classical distinc-
tions between high and low culture are said to be fading; people can freely
choose from a great diversity of “lifestyles”. That educational level and
social milieu played such a huge role in appreciating humor surprised me
too. My questions about jokes, humorists, and humorous television pro-
grams seemed to lead automatically to the subjects of vulgarity and good
taste, high and low culture, common and elitist humor. The discourse about
humor in the Netherlands turned out to be imbued with references to class.

4 Introduction: Jokes, humor, and taste

Five years after I did these interviews in the Netherlands, I carried out a
similar, though much smaller, study in the United States: I interviewed
twenty-eight people and 143 people filled in a questionnaire similar to the
one used in the Netherlands. Not only did this enable me to compare Dutch
patterns of humor styles and social background with American humor
styles, it also gave me an opportunity to include national differences in the
comparison. Moreover, the American study functioned as a cross-cultural
validation of the approach to humor and social background I had developed
in the Dutch study: it turned out that, in a different cultural context, the
approach and the concepts still worked, even though actual social distinc-
tions in the US were markedly different.

Jokes and humor

Jokes are – as all humor is – meant to amuse, to make people laugh. Ever
since antiquity, many superior and inferior thinkers have reflected on hu-
mor, and there is but one thing upon which they all agree: humor is a pleas-
ant experience, often (but by no means always) accompanied by laughter.
Humor is not solely amusement; it can bring people closer to each other,
embarrass, ridicule, cause to reflect, relieve tension, or put into perspective
serious affairs. However, if people do not like the joke, humor cannot fulfill
these other functions competently. Humor can fulfill a great number of
functions, but the first goal of the joke is to provoke mirth, amusement, and
preferable laughter.

The joke is but one form of humor. In addition to jokes, many other
humorous genres and styles exist, varying from slapstick to doggerel to
cabaret critical of society. To investigate social differences in sense of hu-
mor, it seemed most sensible to me to focus on one genre. It is simply not
possible to allow all styles and genres sufficient space in a single book. It is
also questionable whether one can make meaningful generalizations about
such divergent genres as jokes, cabaret, revue, stand-up comedy, clowns,
TV satire, sitcoms, humorous talk shows, let alone cartoons, regular col-
umns, trick cigars, fake turds, clowns’ noses, or the humor in advertise-
ments, on signboards, or carved into toilet doors. And then these represent
only standardized humor: spontaneous jokes like those made every day fall
outside these categories. Therefore, I sought to limit my scope.

After serious consideration, I chose the joke, one of the most widely
distributed and most recognizable humorous genres in the Netherlands and
throughout the world. It is basically a short humorous text with at its end an

Jokes and humor 5

unexpected turn or denouement, the punch line. Usually j o k e s have the
form o f a story, but riddles are also considered to be jokes. These have, as
j o k e s do, a clear punch line and are based frequently on the same themes.
What happens in a j o k e generally follows a standard pattern: things often
take place three times. There are standard formulas for the telling too: ” A
man walks into a bar…”; ” A Dutchman, a German and a Belgian…” or their
American counterparts, ” A Polack, an Irishman and a . . . ” Themes, set-
tings, and personages are largely standard as well: a dumb blonde, a woman
at the doctor’s, a man in a bar, a f l y in the soup, three persons on the Eiffel
Tower, in an airplane, or on a desert island – all personages and situations
the good listener immediately associates with jokes. Some o f these charac-
ters have a more national flavor: in the Netherlands the dumb character is
usually Belgian, the little boy outwitting adults is called Jantje (Johnnie),
and Jewish stock characters Sam and Moos (short for Samuel and Moses)
will be out walking in the Kalverstraat, Amsterdam’s main shopping street.
However, the same jokes, featuring different characters, can be found in
other countries around the world, from America to India or Chile.

The strong standardizing o f j o k e s has to do with the fact that jokes are
orally transmitted. Jokes are written down, for instance in j o k e books, but
the majority of the jokes contained in these derives from the oral culture o f
storytelling. Such oral genres often have standard formulas and themes: this
makes them easy to remember and repeat to others. In addition, new punch
lines can be built into an existing pattern. Everyone w h o tells a j o k e is dip-
ping into an enormous, pre-existing, repertoire of jokes. A j o k e teller will
then never – at least almost never, and then very seldom rightfully – claim
that he thought up the joke himself. Not only the j o k e itself is at issue in
joke telling, but also the art of telling it: whether or not someone knows
how to “present” it.

This verbal transmission means that jokes grant researchers a v i e w of
the role, so difficult o f access, that humor plays in normal, day-to-day in-
teractions. Jokes are not the domain o f professional humor producers, but
are mainly told in everyday situations by “ordinary people”. Jokes differ in
this from other standardized humor usually taking place on paper or on a
podium. Comedy often is a rather one-sided form of communication: the
role o f the audience is limited to laughing or not, laughing right out loud or
less enthusiastically – where in humor on paper, radio, or television the
humorist does not even get to hear whether this happens. However, the
teller makes direct contact with his audience in telling jokes. A l s o , the divi-
sion of roles is not standardized: different people gathered together can tell

6 Introduction: Jokes, humor, and taste

jokes in turn. Jokes thus provide the opportunity to look at humor as a
mode of communication.

The joke is a preeminently social phenomenon. Jokes belong to every-
one: they are not thought up by any one person, but are told again and again
and continuously redesigned in the interaction. A joke is a joke only if it is
repeated: only at the moment of repetition does a joke become a joke, a
“social event” instead of an individual creation. This too is one of the rea-
sons why I chose the joke: research into cabaret or other humor whose au-
thor can be clearly indicated would quickly have become a history of im-
portant names and conflicting artistic movements. In research of this na-
ture, I would also have run the risk of placing the emphasis on creation
instead of communication. Jokes are what Emile Dürkheim ([1895] 1964)
has referred to as social facts: phenomena that cannot be reduced to the
level of individual decisions and motivations.

The consideration that finally determined my choice for the joke is how
people think about jokes. As I have already mentioned: the joke has very
definite advocates and opponents. In general, however, the joke’s status is
low. This means that the joke forms a good basis for this research: it is
precisely the “low” and controversial genres that evoke explicit reactions
and thus make visible social distinctions. Explicit judgments about the joke
have a lot to do with the fact that joke appreciation is often couched in
terms of good and bad taste. “Taste” does not usually point to matters of
life and death but rather to mundane things like preferences for interior
decoration, clothing, or television series. And yet, in judgments having to
do with taste, preference or aversion is often highly present and deeply felt.
Social boundaries are sharply delineated by what seem to be trivial matters,
in which “tastes differ”.

Given that this is mostly research into my own society, throughout the
research I was very much aware of these opinions on good and bad taste. I
have never been particularly tempted to tell jokes (in spite of my prolonged
contact with them, I have never become a virtuoso joke teller) and in my
social milieu I seldom hear them. This “anthropological impulse” also
spurred me on to choose the joke as my research subject: researching some-
thing you don’t know well often produces more insight, even into what is
very familiar and trusted, than researching something with which you are
intimately involved. In choosing jokes, I was not choosing an unknown
subject but a “strange” one, nevertheless, to those in my own circles.

Humor as a social phenomenon 7

Humor as a social phenomenon

Thinking about humor has always been predominantly the domain of phi-
losophers and psychologists, and more recently of linguists. While humor is
preeminently a social phenomenon, social scientists have only dealt with
humor research sporadically. The scientific background of the majority of
humor researchers has influenced the questions they have asked about hu-
mor. Philosophers see humor to a large extent as something intrinsically
present in a text or an event. The question then quickly is focused on what
“the substance” or “the essence” of humor is: what are the distinguishing
characteristics of “humor”, “the laugh”, or “the comic”? (Morreall 1983,
1987). Within psychological research, the emphasis is strongly placed on
humor as an individual matter: the confrontation between an individual,
with specific moods, distinguishing personal characteristics, aspects of
character and interests, and a joke (e.g. Martin 1998; Ruch 1998). Lin-
guists, finally, have tended to focus on the formal characteristics of the
humorous text: what distinguishes jokes and other funny texts from serious
ones? (Raskin 1985; Attardo 1994, 2001)

In this book, I want to look at humor primarily as a social phenomenon:
a form of communication that is embedded in social relationships. For this
reason I have also chosen a working definition of humor in which the social
aspect is prominent: I see humor as “the successful exchange of joking and
laughter”. Humor in this definition is viewed as an exchange involving a
number of people. This communication can be more or less successful;
there is only question of humor if the joke “succeeds”. An unsuccessful
exchange does still contain an aspiring joke – an attempt to make people
laugh – but this is not successful humor: no one laughs, smiles, or other-
wise acknowledges the joke.

While humor also can be unintentional, I will be looking primarily at
conscious attempts to make people laugh: jokes, performances of comedi-
ans, television comedy – all of these socially stylized invitations to laugh-
ter. The ideal reaction to such a joke is always a laugh. Everyone who tells
a joke hopes that it will be laughed at. The joke tellers I spoke to said with-
out exception that the attractiveness of telling jokes lies in the fact that
people laugh at them, “that people fall off their chairs laughing”. “That the
canteen resounds with laughter. The more people laugh, the more fun you
have telling them.” “What’s fun about telling jokes? If I go to a bar, the
moment I come in, everybody starts laughing. That’s the nice thing about
telling jokes.”

8 Introduction: Jokes, humor, and taste

The explicit mention of laughter in this definition may be something of
a provocation in current humor research, where humor and laughter are
generally considered separate and partly unconnected phenomena. Not
everyone who appreciates a joke expresses that by laughing, and there are
many forms of laughter that are not responses to humor (Douglas 1975;
Pro vine 2000). Still, everyone who makes a joke hopes for laughter as the
result; and everyone who hears a laugh assumes that something funny has
happened (and will also want to know “what’s so funny?”).

As the sociologist Rose Coser wrote: “To laugh, or to occasion laughter
through humor and wit, is to invite those present to come close.” (Coser
1959:172) Laughter signals the acceptance of this invitation. As so often
happens with invitations, acceptance of the invitation is often interpreted as
an acceptance of the inviter at the same time. Thus, humor and its counter-
part and reward, laughter, are among the strongest signals of social solidar-
ity and togetherness.1

Therefore the laugh is a fundamental part of the way people perceive
humorous communication. It is the expected, intended, and coveted reac-
tion of any joke teller. Laughter is the idealtypical expression of the emo-
tion of amusement. As such, it cannot be ignored as a social phenomenon
and a form of communication in any study of humor. But there are of
course other possible reactions: these range from smiling and grinning to
complicated (but culturally coded) reactions such as half-exhausted sighing
at a corny joke, or verbal expressions of appreciation. When I was living in
the United States, I was rather puzzled at the prevalence of verbal acknowl-
edgements of jokes: “That’s so funny” in addition to, or even instead of,
laughing.

The description employed bypasses the crucial question about humor:
what is it that makes people laugh? If people try to get other people to
laugh by using a joke, how do they do that? Ever since Plato and Aristotle,
people have asked themselves these questions but it is very difficult, if not
impossible, to answer them conclusively and definitively. From the per-
spective chosen here, this is not necessary either. Here we are concerned
not with the essence of humor, but with its social functions and meanings.
Thus, even though the question how humor works, and what mechanisms
are central to it, will emerge several times in the course of this book, most
extensively in Chapters 7 and 8, this book is not an attempt to construct a
theory of the workings and mechanisms of humor.

Humor has peculiar contradictory meanings: a joke can be an invitation,
as Coser states, but it can also put people off and exclude them (Bergson
[1900] 1999; Billig 2004). Humor brings people together but it can also

Humor as a social phenomenon 9

emphasize and augment differences in status. Humor can shock, insult,
hurt, and consecutively be used as an excuse (“it was just a joke”) but nev-
ertheless a sense of humor counts as a positive feature. This multiplicity of
contradictory functions has a lot to do with the fact that humor is “not seri-
ous”. Something that is said in the guise of a joke should not be taken liter-
ally (Bateson 1972; Mulkay 1988). Because of this, the same joke can have
different functions and meanings at one and the same time. Signals like
laughter and smiling, as well as verbal cues of humorous intent (“Have you
heard the one about…”) separate humor from ordinary, serious communi-
cation. This so-called framing separates playing behavior from serious be-
havior; it separates what actors do on stage from what is done in the “real
world”; and it separates humor from seriousness.

The different functions of humor often intersect and can hardly be sepa-
rated in concrete situations (Palmer 1994; Ziv 1984). A regularly occurring
event such as telling an ethnic joke – for instance about the Turkish minor-
ity in the Netherlands – can be an attempt simply to amuse, as well as an
expression of a shared negative attitude regarding a specific ethnic group.
Perhaps a joke of this type is also an attempt to acquire status or to bring up
in conversation the sensitive subject of migrants. If there are Turks present,
it may be an attempt to shock, insult, or exclude them. Among Turks and
Dutch people who know each other well it can, instead, be a way of show-
ing that they are above such sentiments. One and the same phenomenon
can therefore have a diversity of functions for different persons or in differ-
ent situations; for separate persons it can, moreover, have another function
than for the group or society as a whole.

The polysemy of a joke makes it impossible to say with certainty which
function it fulfills or what the joke teller meant: humor is by definition an
ambivalent form of communication. “The” function of the joke or humor-
ous genre can thus not be firmly established. What’s more: even “the”
function of a single joke about Turks in one, specific, social setting gener-
ally cannot be established firmly. Quite probably even the person telling a
joke does not know for sure why he’s doing it, let alone why he chose that
joke. The only thing he will probably know for sure is that he wants to
make people laugh.

If humor is seen as a social phenomenon, in addition to that communica-
tive aspect two other aspects are of special importance. Firstly: differences
in appreciation of humor are for a large part socially and culturally deter-
mined. What people think is funny varies from culture to culture and from
group to group: even within one culture there are differences in taste. Even

10 Introduction: Jokes, humor, and taste

though many scholars have commented on the cultural variability of hu-
mor, comparative research on humor is almost absent.2

Secondly, humor often touches upon social and moral boundaries. Jokes
often deal with taboos or “painful subjects”; this means that social and
moral boundaries are often transgressed to some extent (Douglas 1966,
1975). Humor, however, also marks social boundaries: it is a powerful
means of pulling people together and, in doing this, automatically shutting
other people out. Sometimes this takes place directly, by laughing at peo-
ple, but it can also be indirect: taking place through shared standards of
what is funny and what is not, or because the joke includes a reference that
not everyone understands. The laugh makes group boundaries clearly visi-
ble and palpable: he who laughs belongs, he who does not laugh is ex-
cluded.

Humor is a form of communication, a question of taste, a marking of
social boundaries. These three aspects determine the social functions and
meanings of humor, and these aspects will serve here as guidelines for an
exploration of the sociology of the joke. These subjects lie at the cutting-
edge of humor research and sociology: within social theory, taste, commu-
nication, and social boundaries are important themes. Two of these, the
connection between humor and social boundaries, and the role of humor as
a form of communication, are important themes in the existing research in
the social sciences into humor. The third – the connection between humor
and taste – has hardly been investigated; thus in this book I have tried to
establish a connection between humor research and theories about taste and
taste difference in the social sciences.

Humor and taste

Anthropological research into humor also shows a clear connection be-
tween humor and culture: the documented humor from cultures far distant
from our own in time and place often seems coarse, strange, absurd, or
simply unintelligible to us (Apte 1985). The impossibility of understanding
someone else’s humor has much to do with cultural knowledge: people do
not understand each other’s jokes because they fail to understand crucial
references. Additionally, cultural background affects cultural boundaries,
and that also affects sense of humor: something that is funny in one culture
can be shocking, embarrassing, or even repugnant in another.

But mostly, cultural differences in sense of humor are also connected
with culturally determined differences in style. For instance, the difference

Humor and taste 11

between British and American humor has been summarized as the contrast
between understatement and overstatement. This difference lies not in the
themes or subjects of the joke but rather in the general tone or attitude: it is
possible to conceive two completely different jokes about similar subjects,
and with similar techniques, but with a completely different presentation,
tone, and purport. Such differences, which reflect more general notions of
what good and bad humor are, are matters of style. Style differences of this
sort are difficult to verbalize; they are more subtle but more crucial than the
more obvious differences in subject choice or language use.

Not only among cultures but also within any given culture, huge differ-
ences exist in what people find funny. Within the Netherlands there is, for
instance, a cultural difference between city and countryside, between the
college and non-college educated and between young and old. These cul-
tural differences are reflected in humor: the highly provocative stand-up
comic Hans Teeuwen is thus for the young while the late Toon Hermans,
who often seemed more a clown than a comic, and whose humor ranged
from the absurd to the sentimental, is preferred by those who are older.
Andre van Duin, a comic from the music-hall tradition, known for his im-
personations of rather bizarre characters, is the popular humorist while
Freek de Jonge, a critical comedian from the slightly moralist Dutch caba-
ret tradition, with his dense, poetic style, and fast and chaotic presentation,
appeals more to intellectuals. (Short descriptions of Dutch humorists and
television programs mentioned in this book are provided in Appendix 2.)

The differences in humor between different groups within Dutch soci-
ety, just as the differences between cultures, can to some extent be clarified
by differences in cultural knowledge and sensitivity to certain boundaries.
Most cultural differences within the Netherlands are, however, also a ques-
tion of style. This is, for instance, the case for Andre van Duin, the most
prominent popular comedian in the Netherlands. Probably very few people
in the Netherlands have difficulty understanding him, he does not antago-
nize people by transgressing boundaries, and no one doubts his professional
skill. And yet there are many people who don’t like him. Differences in the
appreciation of van Duin are questions of style, of taste: his humor is a
close or less close match with what people expect of good humor.

“Style” and “taste” are connected concepts: both have to do with es-
thetic preference for and aversion to something. Each concept can be used
both in the evaluative and the descriptive sense: someone who “has style”,
has good style; someone “with taste”, has good taste. There are, however,
differences: taste primarily has to do with judgment and appreciation while
style implies a more active role: it can be linked to both appreciation and

12 Introduction: Jokes, humor, and taste

creation. Someone with a certain taste likes a certain style. Taste has to do
with “what people see in something” while style has to do with “how peo-
ple do it”.

The social formation and determination of taste is one of the classical
themes in sociology (e.g. Bourdieu [1979] 1984; Gans [1974] 1999; Vehlen
[1899] 2001). Sociologists tend to speak of “taste” rather than “style” when
describing how social background relates to esthetic appreciation. This
reflects and reproduces a distinction between consumption and production
of esthetic preferences, which does not make much sense when discussing
joke telling and humor, where appreciation and creation are intimately con-
nected. In this book, I will therefore often speak of “humor style” rather
than “humor taste”. However, the main theoretical inspiration of this book
comes from sociologists studying taste, most notably Pierre Bourdieu.

The central proposition of sociological studies of taste is that they can
be used to demarcate social boundaries (Lamont and Fournier 1992; La-
mont 1992, 2000; Lamont and Molnar 2002). It is preferences for mundane
things such as furniture, books, paintings, clothing, television programs,
and hobbies that come to have exceptionally strong symbolic meaning.
Choices of this nature can and do evoke vehement and emphatic reactions.
Taste, just as humor, is felt to be something extremely personal and sponta-
neous, but it also serves as a way of establishing whether or not people are
on the same “wavelength”. In this way, these taste differences function as a
direct social delimitation, not just between cultures, but also between
groups within one culture.

Cultural differences within one society are not completely independent
of one another. Although different groups within Dutch society have to
some extent their own values, norms, ideas, and symbols, differing from
those of other Dutch subcultures, taste and style are defined and formed in
relationship, and even in contrast, to tastes and styles of other groups (see
Kuipers 2006a). Youth culture, for instance, often includes a more or less
conscious rejection of the adult culture (Hebdige 1979; Willis et al. 1990).
Class culture as well is often partially based on rejecting the taste, behavior
and lifestyle of other classes. The higher middle class has always tried to
distinguish itself from the culture of the lower classes. “Taste” is one of the
weapons used by the higher classes in doing this: the power to define the
preferences of other people as less valuable than one’s own.

That taste is a way of distinguishing oneself – a means of distinction – is
the central proposition of La Distinction (1984). In this book, Pierre
Bourdieu shows that esthetical preference, therefore taste, is strongly con-
nected to class. Not only the art form people appreciate, but also what they

Humor and taste 13

prefer to eat, how they furnish their houses, what they prefer to photograph,
is connected with their class position. Taste is not just a way to distinguish
oneself from others, it also reveals one’s status. The preferences of the col-
lege educated are socially more valued than those of the non-college edu-
cated. Their taste has more status because they have more status, and vice
versa: they have more status because of their taste. Taste of the higher cir-
cles very quickly becomes the legitimate taste. Taste is then, in Bourdieu’s
words, a form of symbolic capital: the status of certain groups is automati-
cally equated with the status of their esthetical preferences.

In La Distinction, Bourdieu distinguishes two different sorts of capital:
cultural and economic. Class, and thus taste, is connected with possessions
and wealth: economic capital. Beside this, there is the influence of cultural
capital: the education one has had and, connected to this, one’s knowledge
of “high culture”. These two forms of capital do not completely coincide.
There are people with a great deal of cultural but very little economic capi-
tal – penniless but very cultivated intellectuals and artistic persons – and
the nouveau riches with a great deal of money but the “wrong” cars,
clothes, hairdos, and holiday destinations. The esthetical preferences of
these two groups do not completely coincide: people with more economic
capital are often somewhat more conservative; people with more cultural
capital are often somewhat avant-garde. These are all “legitimate” tastes:
they are not shared, however, in all high circles.

This relationship between taste and status is also the motor behind
the trickle-down effect (Fallers 1954; Elias [1939] 1978): the mechanism
by which many tastes, styles, and preferences diffuse through a society.
People attempt to distinguish themselves from others in a positive sense by
acquiring tastes and styles with higher status. The effect of this is that this
taste or style diffuses further and further downward in the society, after
which those in the upper layer seek something else to distinguish them-
selves. The drive to distinguish oneself is thus an important source of cul-
tural change.

Since Bourdieu’s theory of cultural and economic capital was devel-
oped, in the France of the 1960s, class systems have eroded considerably in
most Western countries. Social hierarchies have become less strict, social
mobility has increased, and other social distinctions, such as age or ethnic-
ity, have become more influential. As a result of this, relationships between
taste cultures are less clear-cut than Bourdieu’s theory might suggest (cf.
Holt 1997, Peterson 1997). Rather than “trickle down”, processes of emula-
tion and distinction may lead to elements of working-class and youth cul-
ture “moving up” (e.g. blue jeans, pop music), or to taste groups combining

14 Introduction: Jokes, humor, and taste

high and popular tastes (Peterson and Kern 1996). Additionally, processes
of distinction and emulation may not be limited to one society, leading to
incorporation of foreign, cultural elements into taste cultures.

That taste is a means of distinction, does not however, mean that people
necessarily allow themselves to be led by the need to be different or better
than others in their preferences for humor, art, or television programs. Nei-
ther is it so that assuming the taste of people with higher status, the legiti-
mate taste, solely arises from the drive to imitate. Taste is more than an
imposed attempt to belong somewhere; taste happens automatically. “Each
taste is experienced as natural – and it is practically that, as habitus – which
amounts to rejecting others as unnatural and consequently bad.” (Bourdieu
1984: 56) Variations in taste and style, according to Bourdieu, are the em-
bodiment of social distinctions, they are “symbolic re-creations of actual
differences”. Esthetic judgments are inevitably intertwined with social dis-
tinctions.

Taste is an integral part of what Bourdieu calls the habitus·, embodied
culture, culture that has become part of people’s most automatic reactions
and preferences. Habitus is culture worn in, as it were, to the body. This
includes not only esthetical preference or aversion, but also how someone
speaks, how someone moves, how someone sees himself and others. All
these things are formed by culture and social position. What someone finds
humorous is also part of habitus: the reaction to humor is almost a reflex –
you either laugh or you don’t. But at the same time the reactions – as will
become apparent from this book – are strongly connected to social position:
different people laugh at different things.

And yet “sense of humor” is not only linked to status arising from cul-
tural and economic capital. Bourdieu and other sociologists of taste de-
scribed primarily how taste differences mark class boundaries. Next to
these, there are other, different, important social distinctions such as age,
gender, religion, regional or ethnic background – all these distinctions are
generally connected with taste and lifestyles, and thus: humor style.

Making the appropriate jokes and knowing which jokes to laugh at are
always skills held in high esteem. A good sense of humor also is a way to
distinguish oneself within one’s own social group. The ability to tell jokes
well is also a form of capital; the only thing is that it does not have the
same value in every situation. Having “a sense of humor” always goes to-
gether with social status, but not by definition with economic or cultural
capital.

A good sense of humor, the ability to make, share, and appreciate the
right jokes is always symbolic capital. But still, the humor of the groups

The context of Dutch humor 15

with the highest status is not everyone’s idea of the funniest humor. I will
try in this book to classify not the lifestyles but the humor styles: the differ-
ences in “humorous habitus”.

The context of Dutch humor

A study of a country’s humor inevitably becomes something of an ethnog-
raphy of that country. This book, although primarily a book about humor,
will probably reveal much about the Netherlands. The final chapter, while
intended mainly as a cross-cultural validation of the approach developed in
the first nine chapters, will also shed light on American culture.

T o approach a country through its humor is to produce a rather unusual
image of that country. This study highlights the backstage areas of private
interactions with family and friends, the minutiae of everyday communica-
tion, and the subtle distinctions and judgments that people make about oth-
ers. It is also so that the jokes I collected and discussed with my informants
give a view of the Netherlands that may differ from the official “front-
stage” image of the Netherlands. Like j o k e s around the world, Dutch jokes
are mostly concerned with the sensitive areas in Dutch culture and society.

This book not only looks at Dutch society through the lens of humor, it
looks at differences in humor. This focus on social differences highlights
cultural rifts and distinctions within Dutch society. It w a s n ‘ t until I did a
similar project in the U S that I realized there was something particularly
Dutch about Dutch respondents, despite their great differences.

This Dutchness, and commonality within Dutch society, may need a
short introduction here. Given the b o o k ‘ s focus on class differences, proba-
bly it is important to note that the Netherlands is a fairly egalitarian society,
even by Western European standards. Traditionally, the Netherlands had
hardly any nobility, and n o upper class to speak of. Ever since the seven-
teenth century, Dutch society has been solidly middle class, with the elite
consisting of merchants and bankers rather than counts and princes. The
country did not become a monarchy until the nineteenth century, after the
defeat of Napoleon. The Royal Family, though very prominent in jokes, has
never had m u c h real power and has always worked hard to cultivate a nor-
mal, accessible, and middle-class image. Social mobility is relatively high
and traditional indicators of class inequality such as elite universities and a
very uneven distribution of income are not present in the Netherlands
(Dronkers and Ultee 1995). The class difference in sense of humor found in
this study is by no means a reflection of the Dutch self-image. Like Ameri-

16 Introduction: Jokes, humor, and taste

cans, Dutch tend to think of themselves as part of a classless society, mak-
ing references to social class slightly taboo.

Currently, the Netherlands is one of the most secularized countries in
the world. Until recently, religion significantly influenced all domains of
life; witness the characteristic “pillarization”: Catholics and the various
Protestant denominations each had their own churches, political parties, and
leisure activities (Lijphart 1968). A remnant of this system can still be
found in public television: each of the old “pillars” has its own broadcast-
ing corporation. In general, the various cultural revolutions of the past fifty
years seem to have hit the Netherlands harder than other countries: the
Netherlands has not only become one of the world’s most secularized, and
one of the world’s most permissive countries (Kennedy 1995). This shows
in Dutch humor, which has changed very quickly since the 1950s, becom-
ing more transgressive and explicit than in other countries.

Another characteristic of the Netherlands that may be important to keep
in mind is its ethnic composition. The Netherlands had a large Jewish mi-
nority until the Second World War, when a large percentage of Dutch Jews
was deported. Apart from the Jewish community, the country was ethni-
cally very homogeneous. It was only after 1970 that the country gradually
became more mixed, with the arrival of migrants from the (former) colo-
nies of Surinam and the Netherlands Antilles, and guest workers, mainly
from Turkey and Morocco. However, at the time of this research, even
though migrants had become the target of many jokes, for the most part
they did not take part in Dutch joke culture (this is changing quickly, see
Meder 2001). The former guest workers in particular did not speak Dutch
well as a rule. Thus, the Dutch research addresses class, age, and gender
differences, but not ethnic differences. The sample simply did not include
enough members of ethnic minority groups to say much about ethnicity.

A final aspect of Dutch society that may need some introduction is its
humorous tradition. Although the research dealt primarily with jokes, to
create a larger perspective for the appreciation of jokes, informants also
were asked about comedians, humorists, comic television shows, and other
standardized forms of humor. This was done both in the interviews and in
the surveys, both in the Netherlands and in the United States.

The Dutch humorous tradition aptly illustrates the Netherlands’ position
in the world. All the main comic genres have foreign names: cabaret, revue,
sketch, stand-up comedy, sitcom. However, each of these imported genres
has acquired a highly Dutch flavor. The main humorous genre in the Neth-
erlands is called cabaret, which has emerged from the French tradition of
cafe entertainment (Ibo 1981). Nevertheless, Dutch cabaret generally con-

The context of Dutch humor 17

sists of long performances in a theater setting rather than in a smoky cellar
or cafe; one person is involved, doing humorous monologues, which may
be combined with songs, poetic monologues, social satire, and social cri-
tique. The best known representatives of cabaret are the late Wim Kan,
Freek de Jonge, and Youp v a n ‘ t Hek, celebrities who will be mentioned
many times in this book.

Many comedians have a rather ambivalent view of the “cabaret” label
because of its association with intellectualism and elitist social critique.
The late Toon Hermans, whose humor lies somewhere between the
clownish and the poetic, explicitly rejected cabaret. More absurdist comics
like Brigitte Kaandorp or Herman Finkers are not completely comfortable
being called “cabaretiers” either (Hanenberg and Verhallen 1996).
Younger performers like Hans Teeuwen or Theo Maassen often move back
and forth between the cabaret tradition and the faster and shorter “stand-up
comedy”, imported from America and the UK in the 1990s. This Anglo-
Saxon tradition is faster and more densely packed with jokes: the Dutch
version of stand-up comedy does not leave much room for poetic musings
or social satire.

Until the 1970s, there was a strong tradition of popular performance, va-
riety shows called revue: a combination of sketch comedy with other tradi-
tional, vaudeville disciplines like dancing and singing. Andre van Duin, the
favorite comedian of most joke lovers, comes from this revue tradition,
which was superseded in large part in the 1970s by television comedy.
Television is now one of the main sources of comedy. TV comedy shows
clear American and British influences, with most comedies following the
sitcom format. Other genres on Dutch television are transnational too and
include candid-camera shows, talk shows presented by comedians, and
sketch comedy.

Different TV stations in the Netherlands target different audiences.
RTL4, which aired Moppentoppers, is the largest commercial network,
with popular shows and comedies aimed at large audiences. However,
some of the public broadcasting corporations also aim for popular success.
Andre van Duin had a variety show aired by TROS, one of the public
broadcasting corporations, before joining RTL4, and the (formerly) social-
ist broadcasting corporation VARA has a long tradition of producing suc-
cessful sitcoms. The most explicitly intellectual or highbrow broadcaster is
called VPRO. Over the years, this corporation has produced many forms of
slightly experimental and avant-garde humor, ranging from the political
satire of van Kooten and de Bie to the absurdist sketch comedy of Jiskefet.

18 Introduction: Jokes, humor and taste

The design of this book

This book consists of three parts: following upon the introduction, one part
deals with differences in the appreciation of the genre and another with
differences in the appreciation of separate jokes. These two parts are fol-
lowed by a chapter describing the American research into social back-
ground and joke telling.

The first part is called Style and social background. This part deals with
the question of where differences in appreciating the joke as genre arise. To
begin with, Chapter 2 summarizes the coming into existence of the genre. I
go into the history of the joke and try to connect this with the development
of other humorous genres. There follow three chapters centered around the
question of why people like the joke as a genre. In Chapter 3, I connect
social differences in appreciation of jokes to differences in communication
styles. Chapter 4 deals with the relationship between the appreciation of
jokes and other forms of humor. In it, I contend that the appreciation of the
genre of the joke is connected as well to differences in humor style: ideas
about good and bad humor. Chapter 5 addresses humor styles in day-to-day
life: here humor styles and communication styles are connected.

In the second part, entitled Taste and quality, the attention shifts to the
content of separate jokes. Not everyone thinks every joke equally funny, or
every subject equally suited to the making of a joke. But perhaps every joke
is not equally funny, and differences in the appreciation of jokes are also a
question of quality. I begin this part with a chapter about the repertoire:
about the jokes being told at this moment in the Netherlands. Chapter 7
deals with the question of what a “good joke” is. In the last chapter of the
Dutch study, I look at variations between groups in the appreciation of
separate jokes. The content of the jokes is consciously placed after the
chapters dealing with differences in the diffusion of the genre as a whole; if
jokes are primarily told in certain social groups, this will also have an effect
on their content.

The final part of this book is called Comparing humor styles. This con-
sists of Chapter 9, which presents the results of the American study and
compares these with Dutch humor styles. Some theoretical implications of
this study are discussed in the concluding chapter.

While the book is not so much about jokes as about judgments about
jokes, jokes are regularly quoted. Some of these will perhaps strike the
reader as hurtful or shocking. This is intrinsic to jokes: humor always
touches on social and moral boundaries. The fact that I quote a joke does

The design of this book 19

not mean that I appreciate it or even approve of it; it means only that it
illustrates or supports my arguments appropriately.

In writing this book, I consciously chose to make my own voice heard
occasionally. In writing about tastes it is almost impossible not to reveal
something of your own judgment. No matter how hard I try, my descrip-
tions of elite humor will always sound more sympathetic than my descrip-
tions of a typical sexual joke. This has played a role not only in the descrip-
tions but in the whole research process. I am deeply persuaded of the fact
that because I did the research and all the interviews personally, this has
influenced the book. The central concept of this book makes this almost
matter-of-course: the interviews with persons who shared my taste and
conversational style went more smoothly. I simply did not always succeed
in laughing at the jokes my informants told me; I could not always tell them
jokes they thought were funny either. In research into humor a first-person
perspective is not only honest, or difficult to avoid in collecting material,
but also useful. The judgment of humor cannot be disconnected from the
person judging. Both in conversations with the interviewees and in expos-
ing myself to the humor, my own role and judgment formed an important
research resource.

The judgment of whether something is funny or not is spontaneous,
automatic, almost a reflex; people laugh almost without reflecting. This
goes for someone investigating humor too. Sense of humor thus lies very
close to self-image. My informants often found it strange and not entirely
pleasant that I was trying to connect something as personal and spontane-
ous as humor to social background: a scientific approach to humor seems to
make its authenticity dubious. Resistance was particularly focused on my
attempts to connect humor to class. While interviewees seemed quite ready
to reflect on masculine and feminine humor, on the rough humor of the
young and the respectable humor of the old, on the differing humor of be-
lievers and nonbelievers, and even enjoyed expanding on differences be-
tween people from different parts of Holland or the United States, the dis-
cussion of humor and class obviously made them ill at ease. Here a re-
searcher encounters double resistance. Neither the Dutch, nor the Ameri-
cans, like to talk about class differences. Both societies are believed to be
meritocratic and individualistic; this does not sit well with the notion that
your parents’ professions influence who you are. Thus, people do not easily
admit that their own behavior might have something to do with class, and
they are more wary still of making any statement whatsoever about the
class of others.

20 Introduction: Jokes, humor, and taste

To indulge in a sociological analysis of humor is not to opt for unquali-
fied pleasure. Not only is this opting to strip something we hold dear – our
sense of humor – of its magic, the serious tone of science is also hard to
combine with the frivolity of the subject. A scholar dealing with humor
must quickly mount an adequate defense against the lethal accusation of
not having a sense of humor. Analyzing humor is, after all, not always easy
to combine with appreciating it – let alone creating it. This is not meant as a
funny book. It is a book about jokes and not in itself any funnier than other
books in the same genre: social-scientific monographs.

TWO

A Sense of Mission

Leaders have a vision and a sense of mission that lifts up and inspires men and women to help
achieve that mission

In fact, there is in every one of us a desire to commit to something bigger
than ourselves; leaders have the ability to tap into that root of motivation, drive, and
enthusiasm that allows us to commit ourselves to achieving that vision.
As a leader, then, you have to have a goal that excites and inspires. And the only goals that

excite and inspire are goals that are qualitative. Nobody gets excited or inspired about raising
the share price or making more money or getting a raise. But we do get inspired and excited
about bringing a product or service to people who need it, and about being the best, and about
winning great success in a competitive field.

Strive to Be the Best
As a leader, the most important vision you can have for yourself is to be the best. And that
same vision must apply to your business or your organization. You will accept nothing less for
yourself or your company than to be the best at what you do. In business, that means asking:

What quality about your product or service is most relevant or important to your
customers?

Once you’ve identified that quality, focus all the energies and creativity of your employees
and managers on achieving superior performance in that area.
We need to be the best. You won’t feel great or as good as you could feel, or capable of

extraordinary performance, unless you are aligned with the best people in your field and doing
the very best job that people are capable of.

Instill Meaning and Purpose
Being dedicated to a mission gives work meaning and purpose. As human beings, we need
meaning and purpose as much as we need food and water and air. We need a sense of
significance. And leaders are those people who make us feel significant. They make us feel
important and remind us that what we are doing has value far beyond just the day-to-day work.
They make us feel that we are an integral part of the mission team.
There are four ways to make people feel important, and they each start with the letter A.

First is appreciation. Take every opportunity to thank people for the quality of their work and
their role in making the company a success. Every time you thank individuals, they are going to

Tracy, B. (2014). Leadership. Retrieved from http://ebookcentral.proquest.com
Created from asulib-ebooks on 2020-02-27 15:54:30.

C
op

yr
ig

ht
©

2
01

4.
A

M
A

C
O

M
. A

ll
rig

ht
s

re
se

rv
ed

feel more valuable and will be more motivated to justify your faith in them.
The second way to make people feel more important and valuable is by showing approval.

Praise people at every opportunity, for any accomplishment, large or small. Praise them also
for their suggestions and insight—for their thinking. People will take praise emotionally. Their
self-esteem and self-worth rises. But it’s important to praise immediately and specifically, so
that people know that it is genuine.
The third way you can build a sense of importance and value in a person is through

admiration. Continually compliment people, whether it’s on their traits, such as persistence, on
their possessions, such as clothes, or on their accomplishments.
Perhaps the most important way for people to feel important and valued is through attention.

People aren’t going to be dedicated to the goals of the organization if they are continuously
ignored. They are not going to feel like key players in the mission if they just receive
commands without having any opportunity for input or feedback. Attention means listening to
people, without interrupting. You don’t necessarily have to take their suggestions or agree to
what they are saying. But give them a chance to say it.

A Common Cause
A good goal or a good mission gives a clear sense of direction not only to the organization, but
to every person in the organization.
A good goal unifies everyone in a common cause. For example, IBM is one of the great

industrial leaders of business history. One of its goals is to give the very best customer service
of any company in the entire world. One of its missions is to be known as the company that
cares for its customers. This mission, which involves a qualitative not a quantitative goal,
excites and inspires people throughout the company because they think about it and talk about it
all the time. They believe they’re the best and that nobody takes care of customers like they do
at IBM. Everyone in the company knows that his job, one way or another, is related to taking
care of customers, and this knowledge unifies everybody in a common cause.
The mission of a company will often be encapsulated in a mission statement. A mission

statement is a clear statement of why the company exists in the first place and what its
overarching goal or purpose is. Mission statements usually involve the customer in some way
—for example, how your product or service is going to help make the customer’s life better.
YouTube founder Chad Hurley wanted people to be able to send homemade videos over the
Internet. Charles Schwab’s mission was to be the “most useful and ethical financial services
company.” Google founders Larry Page and Sergey Brin wanted to make navigating the Internet
easier.
Why does your company exist? What is its cause?

The Core Purpose of Every Business
For a business leader, there is one core purpose above all, and that is to acquire and serve a
customer. Leaders make the customer of the organization the central focus. Take the example of

Tracy, B. (2014). Leadership. Retrieved from http://ebookcentral.proquest.com
Created from asulib-ebooks on 2020-02-27 15:54:30.
C
op
yr
ig
ht
©
2
01
4.
A
M
A
C
O
M
. A
ll
rig
ht
s
re
se
rv
ed
.

Nordstrom, whose leaders think incessantly and continuously about their customers. IBM
thinks and talks about only its customers. More and more companies are becoming obsessed
with the customer. You see, once everybody agrees on who the customer is and agrees that the
purpose of the company is to satisfy that customer the very best way possible, then it’s easy to
get everybody pulling together.
As a matter of fact, I believe that you can tell how well led an organization is by applying a

very simple test. When you’re in that organization, look at and listen to how people refer to the
customer. In a good organization, the customers are always referred to with respect. They are
always referred to with pride, as though they are really important. When a customer calls, it is
an important occasion. And when a customer has a problem and is helped, it is a cause for
celebration. When a customer calls and is happy or satisfied with a product or service,
everybody takes on a tremendous feeling of pride and accomplishment.
In your organization, how do people talk about the customer?
Let’s say you run a department that services another department within the organization. That

other department is your customer. Whoever has to use what you produce in your area of
responsibility is your customer. And leaders have to be very much focused on satisfying that
customer.
If you are going to be a business leader or a leader of a department or any organization, you

have to sit down and think through what the mission or the overarching purpose or goal is
going to be for that business or department. It is the determination of a mission to be the best,
that does something to help others, that is the starting point of your ascension and rise to the top
of leadership.

Tracy, B. (2014). Leadership. Retrieved from http://ebookcentral.proquest.com
Created from asulib-ebooks on 2020-02-27 15:54:30.
C
op
yr
ig
ht
©
2
01
4.
A
M
A
C
O
M
. A
ll
rig
ht
s
re
se
rv
ed
.

LEARNING SELF-CONTROL∗

S. NAGEEB ALI

This article examines how a decision maker who is only partially aware of
his temptations learns about them over time. In facing temptations, individuals
use their experience to forecast future self-control problems and choose the
appropriate level of commitment. I demonstrate that rational learning can be
perpetually partial and need not result in full sophistication. The main result
of this article characterizes necessary and sufficient conditions for learning to
converge to full sophistication. I apply this result to a consumption-savings
environment in which a decision maker is tempted by present bias and establish
a learning-theoretic justification for assuming sophistication in this setting

JEL Codes: D03, D83.

“An individual who finds himself continuously repudiating his past plans may
learn to distrust his future behavior, and may do something about it.”

— Strotz (1955)

I. INTRODUCTION

Questions of temptation and self-control are at the forefront
of psychology and economics. When studying self-control, a mod-
eler must decide just how much individuals can be assumed to
know about their temptations because different assumptions of
self-awareness translate intodifferent behavioral predictions. The
assumption most commonly made, referred to as sophistication,
is that the decision maker perfectly anticipates future self-control
problems. Yet sophistication is often seen as an inaccurate model
of a decision maker’s awareness, especially when the decision
maker may lack experience. An alternative assumption is that of
naivete, whereby a decision maker anticipates having perfect self-
control in the future. Bridging the gap between these two extreme

∗This article is a revised version of Chapter 3 of my dissertation. I am grateful
to my advisers, Susan Athey and Doug Bernheim, for their support and encour-
agement and to Drew Fudenberg and Paul Niehaus for numerous suggestions that
greatly improved the article. I thank Ricardo Alonso, Manuel Amador, Dan Ben-
jamin, Aislinn Bohren, Juan Carrillo, Chris Chambers, Vince Crawford, Stefano
DellaVigna, Ben Ho, Shachar Kariv, Navin Kartik, Botond Köszegi, Troy Kravitz,
Jon Levin, Charles Lin, Ulrike Malmendier, Andres Santos, Josh Schwartzstein,
Shamim Sinnar, Joel Sobel, Joel Watson, Tom Wiseman, the editor (Robert Barro),
and three referees for helpful comments. I acknowledge financial support from the
UCSD Academic Senate, the UCSD Hellman Fund, the Stanford Institute for Eco-
nomic Policy Research, and the Institute for Humane Studies.

c© The Author(s) 2011. Published by Oxford University Press on behalf of the President and
Fellows of Harvard College. All rights reserved. For Permissions, please email: journals.
permissions@oup.com.
The Quarterly Journal of Economics (2011) 126, 857–893. doi:10.1093/qje/qjr014.

857

D
ow

nloaded from
https://academ

ic.oup.com
/qje/article-abstract/126/2/857/1870713 by A

rizona S
tate U

niversity user on 27 February 2020

858 QUARTERLY JOURNAL OF ECONOMICS

assumptions, O’Donoghue and Rabin (2001) propose a framework
of partial naivete in which a decision maker with(β,δ) preferences
assigns probability 1 to a particular level of present bias (β̂) less
than his true present bias (β̂ > β).

With few exceptions, most models of imperfect self-control
make one of the above assumptions. That derived behavior is sen-
sitive to the specification of beliefs has been widely recognized,
including by those that first modeled self-control (Strotz 1955;
Pollak 1968). Nevertheless, the practice has been to treat a
decision maker’s belief about his self-control as exogenous and
model choice given that exogenous belief. While this practice
makes analysis tractable, it raises two conceptual issues. First,
in models of partial sophistication, a decision maker’s beliefs may
be incompatible with what is observed over time, and so ancillary
assumptions are necessary for analysis.1 Second, given the many
different awareness assumptions that could be made, it has been
difficult to assess which assumption is appropriate for a particular
environment and why. Fudenberg (2006), in his recent discussion
of behavioral economics, voices this concern:

I think that behavioral economics would be well served by concerted

attempts to provide learning-theoretic (or any other) foundations for its equi-

librium concepts. At the least, this process might provide a better under-

standing of when the currently used concepts apply.

This article proposes a simple framework to address these
concerns. In this approach, beliefs and choices are derived en-
dogenously and jointly evolve based on the decision maker’s ex-
perience. Endogenizing beliefs in this way allows one to pose and
answer the question of whether and when sophistication closely
approximates the decision maker’s self-awareness once he has had
many opportunities to learn. Although one may be tempted to
conclude that Bayesian learning should always engender sophisti-
cation, I demonstrate that tradeoffs between commitment and flex-
ibility inherent in self-control environments may actually impede
learning. The main result of this article is to offer a necessary and

1. For example, a partially naive decision maker might see behavior that con-
tradicts his view of the world in each and every period. The literature typically
addresses this issue by assuming that his beliefs are not revised, which implicitly
assumes that he receives no feedback about payoffs or the past history.

D
ow
nloaded from
https://academ
ic.oup.com
/qje/article-abstract/126/2/857/1870713 by A
rizona S
tate U
niversity user on 27 February 20

LEARNING SELF-CONTROL 859

sufficient condition—across self-control environments—under
which learning is not impeded and inexorably leads to sophisti-
cation; this condition can be checked in applications to assess the
appropriateness of assuming sophistication.

As the main application of this approach, I analyze a standard
consumption-savings environment in which a decision maker is
tempted by immediate consumption and purchases illiquid assets
to commit toward future consumption. This setting has been the
focus of many papers in the quasi-hyperbolic literature (e.g.,
Laibson 1997; Barro 1999), almost all of which assume sophisti-
cation. I demonstrate that this canonical setting satisfies the con-
dition for adequate learning: thus, a decision maker who learns
about his tendency to overconsume from his past choices even-
tually chooses commitment as if he could perfectly forecast his
temptation tooverconsume. This result therefore offers a learning-
theoretic foundation for sophistication in consumption and
savings decisions.

The general framework that I develop builds on the Planner–
Doer approach to self-control: decisions are made by a single
long-run Planner with dynamically consistent preferences and a
myopic Doer. Prior research using such models, particularly
Fudenberg and Levine (2006), has demonstrated that such models
offer analytically simple and tractable frameworks to understand
self-control in a variety of settings,2 and this article illustrates
how this approach is useful for studying questions of self-
awareness. The Planner represents the forward-looking rational
individual who chooses how much to commit by investing in illiq-
uid assets, signing contracts, making promises that are costly to
betray, and so on. In contrast, the Doer represents the instinc-
tive response of the individual who makes the daily choices and
is presented with stimuli, situations, and commitments; the Doer
is best thought of as a short-run player or behavioral type. The
Planner is uncertain about the extent to which the Doer resists
temptation and uses the Doer’s behavior to learn over time.

To fix ideas, I describe the example studied in Section II: sup-
pose that an individual faces a consumption choice between fish

2. Thaler and Shefrin (1981) were the first to propose a dual-self approach
to imperfect self-control. The recent literature also includes Benabou and Pycia
(2002), Benhabib and Bisin (2005), Bernheim and Rangel (2004), Brocas and
Carrillo (2008), Chatterjee and Krishna (2009), Dekel and Lipman (2010), and
Loewenstein and O’Donoghue (2007).

D
ow
nloaded from
https://academ
ic.oup.com
/qje/article-abstract/126/2/857/1870713 by A
rizona S
tate U
niversity user on 27 February 2020

860 QUARTERLY JOURNAL OF ECONOMICS

and steak in each period. In the beginning of the period, the Plan-
ner makes a reservation at either a restaurant that serves only
fish or one that serves both fish and steak. The payoff from eating
fish is constant over time and across the two restaurants, but that
from eating steak depends on an i.i.d. taste shock that is realized
after the Planner selects the menu. Eating fish is ex ante optimal
given its long-term health benefits, but there are contingencies
in which steak may be preferred. As in Kreps (1979), the Plan-
ner strictly prefers the flexibility associated with the larger menu
were self-control not an issue; however, once at the restaurant,
the Doer chooses what to eat (if there is a choice to make). The
Doer’s preferences are partially aligned with that of the Planner,
and partially reflect a temptation to eat steak in contingencies in
which the Planner prefers fish. The Planner is uncertain about
the strength of this temptation and would benefit from knowing
it because that helps him plan. Were the Planner to know that in
the restaurant, temptation is resisted, he may strictly prefer the
larger menu for the sake of flexibility. In contrast, if he believes
that the Doer would too easily succumb to the temptation to eat
steak, the Planner may find it ex ante optimal to select the smaller
menu, which offers commitment value.

Learning about one’s self-control involves costly experimenta-
tion (and, in this case, a two-armed bandit): the Planner observes
the Doer’s self-control only when the Planner chooses the larger
menu and exposes himself to temptation. Accordingly, once the
Planner becomes sufficiently pessimistic about the Doer’s type,
he chooses the smaller menu because the value of flexibility and
learning no longer outweigh the expected cost of the Doer suc-
cumbing to temptation. Notably, the Planner may make this deci-
sion with positive probability even when the Doer is a “good type”
who resists temptation. Despite the infinite possibilities to learn,
the Planner may decide it not worthwhile to do so and therefore
make forever inferior commitment choices.

The direction of the skewness in these beliefs merits discus-
sion. The Planner cannot perpetually overestimate the Doer’s self-
control and undercommit relative to how he would choose in the
full information benchmark. To see why, notice that any belief
that rationalizes flexibility will allow the Planner to continue to
passively learn and therefore update his beliefs. Thus, if the Doer
is unable to resist temptation, the Planner almost-surely learns
this over time, and eventually he makes the same commitment
choice that he would in the full information benchmark.

D
ow
nloaded from
https://academ
ic.oup.com
/qje/article-abstract/126/2/857/1870713 by A
rizona S
tate U
niversity user on 27 February 2020

LEARNING SELF-CONTROL 861

On the other hand, the Planner may perpetually underesti-
mate the Doer’s self-control even if he began with an overopti-
mistic prior: once the Planner has observed the Doer choose steak
too often, flexibility appears costly, and so a Planner will choose to
actively learn as long as the future value of learning outweighs its
cost. If the Planner is not perfectly patient, he is willing to under-
take a finite number of trials, after which he might commit to the
singleton menu and never revise his belief thereafter. Thus, par-
tial awareness and learning can endogenously lead to perpetual
overregulation, whereby individuals choose rigid commitments or
lifestyles that they would not were they more self-aware. This pos-
sibility resonates with the perspective that obsessions, compul-
sions, and rigidity emerge from erroneous beliefs that one lacks
control otherwise (Baumeister, Heatherton, and Tice 1994).3

A feature of the setting described above is that the Planner
chooses between full flexibility or full commitment but has no abil-
ity to partially commit. In contrast, many settings feature par-
tial commitments that retain some flexibility for the decision
maker while offering a measure of commitment. For example, in a
savings environment, an individual can purchase illiquid assets
to ensure some minimal savings without relinquishing flexibil-
ity altogether. Contractual mechanisms, such as recent innova-
tions like stickK.com,4 can penalize certain choices and provide
incentives to counteract temptation while retaining some flexi-
bility. Social mechanisms—through promises, shame, and peer
groups—also provide partial commitment insofar as they change
the benefits and costs of particular actions but are not completely
binding. Apart from external commitments, a decision maker may
also rely on internal commitments, such as costly self-control (Gul
and Pesendorfer 2001; Fudenberg and Levine 2006), which influ-
ence the choice from a menu without relinquishing flexibility al-
together. To capture possibilities for partial commitment, I allow
the Planner to affect the Doer’s choice through menus and nudges
that restrict and influence the Doer’s choice. The framework here
demonstrates that these partial commitments play an important
role in the decision maker’s long-run behavior and beliefs.

3. A complementary channel for overregulation is explored by Benabou and
Tirole (2004), who study a self-signaling mechanism in which temptation is miti-
gated by the adverse reputation effect it induces in future incarnations.

4. StickK.com is designed to “help people achieve their goals and objec-
tives by enabling them to form Commitment Contracts.” (http://www.stickk.
com/about.php).

D
ow
nloaded from
https://academ
ic.oup.com
/qje/article-abstract/126/2/857/1870713 by A
rizona S
tate U
niversity user on 27 February 2020

862 QUARTERLY JOURNAL OF ECONOMICS

The main result of this article identifies when partial com-
mitments induce efficient learning regardless of the Planner’s pa-
tience. I identify a condition, full commitment distinguishability
(FCD), that is sufficient for learning to engender sophistication
and may also be necessary. FCD is a condition on how fully in-
formed Planners behave and relates to the richness of the set of
partial commitments. The necessity and sufficiency of FCD offers
a simple criterion to understand whether learning leads to sophis-
tication in applications: instead of solving a more intricate model
in which the Planner is uncertain and updates beliefs over time,
a modeler needs to solve only the full information model in which
the Planner knows the Doer’s type and check whether that sat-
isfies FCD. Using this characterization, I study two applications
that have been examined in prior work assuming sophistication:
consumption-savings with illiquid assets and a costly self-control
environment. Within these settings, I illustrate how to check the
validity of FCD and derive its implications for learning.

The results herein underscore the insights that emerge from
deriving the decision maker’s perception of his temptations from
his environment. In contrast to prior research that divorces a
decision maker’s self-awareness from the fundamentals of the set-
ting, whether incorrect beliefs persist is determined endogenously
within this framework. Based on these results, one may expect
that in those settings in which an individual has access to a wide
range of partial commitments, he may know more about his temp-
tations than in settings in which partial commitments are inef-
fectual or lacking.5 From a normative perspective, the role that
partial commitments play in learning has new implications for the
design of commitment and suggests when interventions enhance
learning. Endogenizing beliefs also makes it possible to connect
a decision maker’s perceptions of his temptation to other aspects
of his preference; for example, in Section III.D., I highlight how
patience fosters learning and sophistication.

Some readers may be troubled by the disappearance of
overoptimism at the limits of Bayesian learning. Although I use
the approach developed here to study asymptotic behavior, the
framework is sufficiently flexible and tractable to model a

5. Just as the nature and severity of an individual’s temptations varies across
decision problems, it is likely that his perception and awareness of his temptations
also differs across these settings. Thus someone could be sophisticated about his
tendency to procrastinate and yet still have erroneous beliefs about the extent to
which he may become addicted to particular substances.

D
ow
nloaded from
https://academ
ic.oup.com
/qje/article-abstract/126/2/857/1870713 by A
rizona S
tate U
niversity user on 27 February 2020

LEARNING SELF-CONTROL 863

decision maker’s initial response, and these initial responses may
reflect initial optimism. Indeed, there are numerous reasons to be-
lieve that decision makers may begin with optimistic priors and
learn slowly,6 and so findings in the field are consistent with the
short-run behavior predicted by this theory. Moreover, when a
decision maker makes choices in an environment in which he has
had little prior experience, he has to simultaneously learn about
how tempted he is by different choices in that setting and the pay-
offs of different actions. I show by example that such multidimen-
sional learning can be impeded by challenges of identification that
slow down the rate of learning. Thus, there are natural reasons to
expect decision makers to appear to undercommit in the short run
as they simultaneously learn about self-control and the benefits
and costs of actions.

Section II presents a simple example of the impediment to
learning introduced by imperfect self-control. Section III studies
a general framework with partial commitments, in which experi-
mentation takes a rich form. That section contains the main
results of this article and describes the implications of patience, of-
fers some suggestions on how a modeler might make
partial inferences about a decision maker’s awareness, and
describes the connection of the results here with the steady-state
solution concept of self-confirming equilibrium. Section IV applies
the framework to savings behavior and costly self-control.
Section V discusses the results of this article in light of the related
literature and illustrates the identification challenged induced by
multidimensional learning. The proofs for all results are collected
in an Online Appendix.

II. A SIMPLE EXAMPLE

I begin with a simple example to illustrate the mechanism
for incomplete learning in the most transparent way. Consider an
infinitely-lived individual who chooses between undertaking an
activity (at = 1) or not (at = −1) in period t = 0, 1, 2, . . . . Under-
taking the activity in period t generates a deterministic reward
b ∈

(
1
2 , 1
)

but involves a stochastic cost st uniformly drawn from

6. Optimism could emanate from a number of different sources, if a positive
belief about one’s attributes has some intrinsic value (Brunnermeier and Parker
2005; Kőszegi 2006) or induces motivation (Carrillo and Mariotti 2000; Benabou
and Tirole 2002; Compte and Postlewaite 2004).

D
ow
nloaded from
https://academ
ic.oup.com
/qje/article-abstract/126/2/857/1870713 by A
rizona S
tate U
niversity user on 27 February 2020

864 QUARTERLY JOURNAL OF ECONOMICS

[0, 1]. This activity can represent, for example, the choice to con-
sume fish from the example described in the introduction or to
exercise, in which case b captures the discounted long-run gain
from the activity.

The decision in each period is made through the conjunction
of two systems, the Planner and the Doer, who act sequentially in
each period. The Planner is a long-run agent with an exponential
discount factor δ ∈ (0, 1), and in period t, he obtains a payoff of
(b − st) if at = 1 and 0 otherwise. In the first subperiod of each
period t, the Planner chooses a menu prior to the realization of st,
after which the Doer selects an alternative from the menu. The
Planner chooses between full commitment to undertaking the ac-
tivity by selecting the singleton menu{1}; full commitment to not
undertaking the activity by selecting the singleton menu {−1};
and flexibility by selecting the menu {−1, 1}, thereby permitting
the Doer tochoose either action. When the Planner chooses tocom-
mit to an action, that action is undertaken regardless of the Doer’s
type or the realized cost.

The Doer in each period is not a strategic actor but can be
thought of as a behavioral type or short-run player. Nature se-
lects the type, θ, of the Doer from

{

θ, θ̄

}

where 0 ≤ θ < θ̄ ≤ 1, and this type persists through time. When the Planner is flexi- ble, the Doer of type θ observes st and chooses at = 1 if and only if θb exceeds st. Thus, the Doer is tempted toward inactivity and undertakes the activity in fewer contingencies than the Planner would wish to do so. This temptation might emerge from a mani- festation of a present bias when costs and rewards are temporally separated. The perspective here is that the Doer’s being myopic (representing the individual’s instincts) does not account for how his actions influence the Planner’s future commitment choices.

The Planner faces a tradeoff between commitment and
flexibility: while he would like to exploit the Doer’s informational
advantage, flexibility allows temptations to guide that choice.
Since the payoff from flexibility depends on the Doer’s suscepti-
bility to temptation, it is valuable for the Planner to try to learn
about θ to optimally choose commitment in the future. I assume
that the Planner would prefer to commit to undertaking the ac-
tivity if he were perfectly confident that θ = θ and would prefer to
remain flexible if he were confident that θ = θ̄.7

7. The relevant condition is b ∈
(

1
2 −θ

, 1
2 −θ̄

)
; when this does not hold, the

Planner’s optimal commitment choice is independent of his belief about the Doer’s
type.

D
ow
nloaded from
https://academ
ic.oup.com
/qje/article-abstract/126/2/857/1870713 by A
rizona S
tate U
niversity user on 27 February 2020

LEARNING SELF-CONTROL 865

He begins with a prior μ0 ≡ Pr
(
θ = θ̄

)
that ascribes positive

probability to both types. For simplicity, suppose that all that he
observes over time is behavior and not past realizations of costs
(I consider more general informational structures in Section III).
Because full commitments override the Doer, the Planner learns
about θ only when he chooses to be flexible. I use ht to summarize
the relevant history for the Planner when acting at time t and μt
to denote the Planner’s posterior belief that the Doer’s type is θ̄.
Given a prior belief μ, the Planner updates his beliefs to μ+ and
μ− when he is flexible and observes the Doer choose at = 1 and
at =−1 respectively. The dynamic decision problem of the Planner
is described by the value function

V (μ) = max






b − E [s] + δV(μ) ,
∑
θ∈{θ,θ̄}Pr

μ (θ)

(
θb(b − E [s|s ≤ θb] + δV (μ+))

+(1 −θb)δV (μ−)

)




,

where the first term is the value of committing to undertake the
action and the second term is the value of flexibility. The Planner
never commits to a =−1 since on expectation this is dominated by
committing to a = 1. Standard arguments ensure that the value
function exists and is unique, continuous, and nondecreasing inμ,
and thus the optimal decision takes the form of a simple threshold
rule.

PROPOSITION 1. There exists μ∗ ∈ (0, 1) such that for all μ < μ∗, the Planner’s optimal choice is to commit to the activity and for μ≥μ∗, the Planner’s optimal choice is flexibility.

When the Planner is more optimistic about the Doer’s abil-
ity to resist temptation or values the option to learn, he is more
willing to remain flexible. Because of the option-value associated
with flexibility and learning, a forward-looking Planner is will-
ing to choose flexibility even if he expects that committing yields
greater short-run payoffs. For less optimistic beliefs, whenμ<μ∗, the benefits of flexibility and learning do not offset the expected costs of temptation, and therefore the Planner chooses to commit. Because the choice of commitment shuts down the channel for learning, once he chooses to commit in one period, he finds it opti- mal to commit in every subsequent period. Because the choice to commit is endogenous, eventual beliefs are endogenous and there- fore evolve differently for each type of the Doer (assuming that the prior μ0 exceeds μ

∗).

D
ow
nloaded from
https://academ
ic.oup.com
/qje/article-abstract/126/2/857/1870713 by A
rizona S
tate U
niversity user on 27 February 2020

866 QUARTERLY JOURNAL OF ECONOMICS

Consider the case in which the Doer is of typeθ. The Planner’s
beliefs will fluctuate as long as he remains flexible and his beliefs
remain bounded away from putting probability 1 on θ̄. The Mar-
tingale Convergence Theorem ensures that the Planner’s beliefs
eventually settle but do not converge to the completely incorrect
belief ofμ = 1. Therefore, the Planner eventually commits, making
the same choice that he would in the full information benchmark;
initial partial awareness does not forge any long-lasting differ-
ences from the standard benchmark.

Now suppose that the Doer is of type θ̄. The Martingale
Convergence Theorem again implies eventual convergence of the
Planner’s beliefs, which obtains in two distinct ways: either the
Planner’s beliefs converge to the truth (μt → 1) or the Planner’s
belief μt falls below μ

∗, leading him to commit. Both events oc-
cur with positive probability. Recall that in the full information
environment, a Planner who knew that his self-control problem
corresponds to θ̄ would never choose to commit. Relative to this
benchmark, partial awareness and learning introduce a possibil-
ity for long-run inefficiency and overcommitment. The preceding
ideas are summarized below.

THEOREM 1. For any typeθ ∈
{
θ, θ̄
}

, a Planner eventually chooses
to commit with strictly positive probability.

1. If the Doer is of type θ, then almost-surely, the Planner
eventually chooses to commit.

2. If the Doer is of type θ̄, then the Planner either chooses
to commit or he learns the Doer’s type. Both events occur
with strictly positive probability if μ0 ≥μ

∗.

This result illustrates how beliefs that induce excessive flex-
ibility eventually dissipate: if the Doer is of type θ, any belief of
the Planner that rationalizes flexibility is refined over time as the
Planner infers the Doer’s type from its choices. Almost-surely, this
information leads the Planner to conclude that the Doer’s type
does not warrant flexibility, and therefore the Planner eventually
chooses to commit.8

What is the source of the friction that leads to perpetual over-
commitment? The principal challenge that the Planner faces is
that when he optimally chooses to commit (believing that he faces
a Doer of typeθ), his commitment choice leaves no opportunity for

8. While such optimism disappears eventually, it might emerge in the
short run, especially if individuals begin with priors that overestimate self-control.

D
ow
nloaded from
https://academ
ic.oup.com
/qje/article-abstract/126/2/857/1870713 by A
rizona S
tate U
niversity user on 27 February 2020

LEARNING SELF-CONTROL 867

him to retain flexibility in case the Doer proves to be type θ̄. Thus
the Planner faces an experimentation challenge that traps learn-
ing with positive probability. Augmenting this environment with
partial commitments offer an escape from this experimentation
trap.

Consider the inclusion of a simple bond commitment contract.
For simplicity, suppose that the individual has risk-neutral and
additive preferences and consider a one-period bond contract in
which the Planner pays a lump-sum amount L and a prize x is
returned to the decision maker if the Doer chooses a = 1. There
are various ways to model how such a contract affects the Doer;
for simplicity, suppose that a Doer of type θ chooses a = 1 if θb + x
exceeds the cost s. In this case, a fully informed Planner, regard-
less of the Doer’s type, implements the first-best by setting x to
equal (1 −θ)b and paying a prior lump-sum amount of (1 −θ)b2.
This bond contract leads to zero expected transfers and induces
the Doer of type θ to choose a = 1 whenever the Planner would
like to do so.

Not only does such a bond contract attain first-best, but it also
helps a partially aware Planner learn efficiently. Regardless of
the Planner’ beliefs, it is now never optimal to restrict the Doer’s
choice set to a singleton: even when the Planner becomes quite
confident that the Doer’s type isθ, he is better off, even in the short
run, by partially committing through a bond contract. Over time,
as the Planner uses bond contracts repeatedly, he can observe the
Doer’s behavior and therefore continue learning; almost-surely,
his beliefs converge to the truth.

The possibility for partial commitments to dynamically im-
prove learning has implications for commitment design. Recent
years have seen the advent and study of commitment contracts
that incentivize good behavior and mitigate self-control problems.
The following section analyzes generally when partial commit-
ments facilitate sophistication.

III. WHEN IS LEARNING ADEQUATE?

The framework builds on the above example in several ways.
The action-space can be binary or a continuum. The commitment
technology can be one of two types: first, as in the prior section,
the Planner can set incentives to nudge the Doer to take a par-
ticular action, and second, he can choose menus that restrict the

D
ow
nloaded from
https://academ
ic.oup.com
/qje/article-abstract/126/2/857/1870713 by A
rizona S
tate U
niversity user on 27 February 2020

868 QUARTERLY JOURNAL OF ECONOMICS

Doer’s choice. Both commitment technologies capture the essence
of partial commitment, albeit in different forms, and the gener-
ality of this framework is useful in applying these results to dis-
tinct but commonly studied applications. The Doer’s type is also
drawn from a continuum. Finally, the Planner observes signals
about past taste shocks.

As before, time is discrete and unbounded, with choices be-
ing made in period t = 0, 1, 2, . . . . The Doer makes choices from a
fixed, time- and history-invariant action set A: this is assumed
to be a compact subset of <. In each period, a state s is real- ized that determines the Planner’s preferences for that period: s is drawn from S ≡ [s, s̄] and is distributed i.i.d. with measure ν that has cumulative distribution function F (∙) and a continuous and strictly positive density f . The Planner’s state-dependent prefer- ences over actions, u( a, s), are bounded, strictly quasi-concave in a for each state s, and jointly continuous and satisfy the strict single-crossing condition in(a, s). Accordingly, for each state s, the Planner has a most preferred action aP(s) which is nondecreasing in s. Let â denote the action that the Planner would choose if the Planner had to prescribe a single action for every contingency; for simplicity, I assume that â is unique, and I denote by π̂ the ex- pected payoff of this action.9

The Planner cannot directly choose actions but can choose
commitment. The Planner has two forms of commitment: menus
and nudges. A menu, M, is a nonempty closed subset of A, which
restricts the Doer’s choice of action. Let F be the collection of all
nonempty closed subsets of A; this is the collection of all feasible
menus. The Planner may or may not have access to all feasible
menus: the set of menus from which the Planner can make a choice
isM, a closed subset ofF that contains full flexibility (M = A) and
the optimal full commitment (M = {â}).10 In addition to restrict-
ing choices through menus, the Planner can nudge the Doer into
taking particular actions through incentives: the Planner can sign
commitment contracts, make promises to peers that are costly to
betray, or exert self-control. A nudge, η, directly affects the Doer’s
choices between actions in a state-dependent manner that is spec-
ified below. The set of nudges is N , a closed subset of [0, 1], which

9. In other words, â is the unique solution to maxa∈A
∫

S u(a, s)dν, and π̂ is
the associated expected payoff.

10. Distance is defined by the Hausdorff metric. By Theorem 3.85 of Aliprantis
and Border (2006), F is compact, and M, being a closed subset of a compact set,
inherits compactness.

D
ow
nloaded from
https://academ
ic.oup.com
/qje/article-abstract/126/2/857/1870713 by A
rizona S
tate U
niversity user on 27 February 2020

LEARNING SELF-CONTROL 869

includes the case of no nudge (η = 0). I let C = M×N denote the
full set of commitments available and c = (M,η) denote a generic
element.

As before, the Doer is a myopic agent who does not account for
how his actions affect the Planner’s choices. The Doer’s behavior
is governed by its type θ drawn fromΘ =

[
θ, θ̄
]

. The Doer observes

the state s and nudge η and chooses the action from the menu M
that maximizes the payoff function W (a, s,θ,η); when indifferent
between two actions, the Doer chooses the action that the Plan-
ner prefers. The function W is continuous and satisfies the strict
single-crossing property in (a, s), (a,θ), and (a,η), and for each
(s,θ,η), W is strictly quasi-concave in a. The action selected by
the Doer facing a commitment c is denoted by aD(s,θ, c).

These assumptions ensure that the Doer’s preferred action is
nondecreasing in the state s for a fixed type θ and nudge η. Thus,
both the Doer and Planner prefer higher actions in higher states.
For a fixed state, the Doer’s preferred action is nondecreasing in
its typeθ and in the nudgeη chosen by the Planner. I assume that
in the absence of all commitment, the Doer prefers to choose a
lower action than the Planner.

ASSUMPTION 1. The Doer is tempted to undertake lower actions
than the Planner: u(a, s) dominates W(a, s,θ, 0) by the single-
crossing property for every type θ of the Doer.11

Assumption 1 implies that the Planner knows the direction
of temptation but is uncertain about its strength, which varies
with the type of the Doer. Such an assumption is consistent with
most applications of this model, for example, in which temptation
is driven by some form of present bias, or by identification of a
tempting object. A special case of this framework, also consistent
with most applications, accords the highest type, θ̄, with the Plan-
ner’s preferences over actions, and thus no temptation.

I denote a full information environment—in which the Plan-
ner knows the Doer’s type—byΓ = (S,ν, A, u, W,Θ; C), a constella-
tion of parameters that satisfy the above conditions. The Planner
begins with a prior μ0; throughout the article, I restrict attention
to priors μ0 with a continuous and strictly positive density on Θ.

After each period t, the Planner observes at and obtains some
information about st. The setting of Section II considered an ex-
treme case in which the Planner learns nothing about the state;

11. In other words, for every type θ, state s, and pair of actions a and a′ > a,
W(a′, s,θ, 0)− W( a, s,θ, 0)≥(>) 0 implies u( a′, s)−u(a, s)≥ (>) 0.

D
ow
nloaded from
https://academ
ic.oup.com
/qje/article-abstract/126/2/857/1870713 by A
rizona S
tate U
niversity user on 27 February 2020

870 QUARTERLY JOURNAL OF ECONOMICS

one could equally consider the other extreme in which the Planner
observes the state perfectly. Spanning these two extremes is an
imperfect attribution environment in which the Planner observes
the state with noise. Generally, suppose that the Planner obtains
a signal σt( at, st) that comes from the space S and is distributed
according to F̃(σ|s, a).

For t > 0, ht denotes the relevant history for the Planner in
period t, which comprises past commitments, actions, and signals.
The Planner’s posterior belief about the Doer’s type conditional on
history ht is denoted by μt.

I analyze two special cases of this model relevant for the ap-
plications herein.

MODEL 1—Actions are chosen from a binary set A = {−1, 1}
in which the ex ante optimal choice, â, is 1. The Planner’s payoff
satisfies u(−1, s) =0 for all s, and u(1, s) = U(s), where U is smooth,
strictly increasing in s, and satisfies U(s)< 0 < U(̄s). The Planner can choose from all possible menus—M = {{−1, 1} ,{−1} ,{1}}— and can choose nudges from N .

MODEL 2—Actions are chosen from a continuum A = [a, ā],
and the Planner’s and Doer’s state-contingent payoffs satisfy the
assumptions of the general model and are smooth. The Planner
does not have access to nudges(N = {0}) but has access to a closed
set of menus, M. For convenience, suppose that a = aD( s,θ, A, 0)
and ā = aP( s̄). Two special cases of interest are those in which
M is the set of all feasible interval menus, which I refer to as
Model 2a, and in whichM comprises all feasible menus including
nonintervals, which I refer to as Model 2b.1

Model 1 generalizes the example in Section II to a contin-
uum of types.13 In the binary choice environment, menus without
nudges offer the Planner a choice between commitment to a sin-
gleton menu or full flexibility. Nudges introduce partial commit-
ment by allowing the Planner to influence how the Doer chooses
actions in each state. Apart from its usefulness in understanding

12. Specifically, Model 2a corresponds toM = FI = {[α,β]: α,β ∈ A andβ ≥
α}, and Model 2b corresponds toM = F.

13. Section II, restricted to two types, corresponds to S = [−1, 0], u(−1, s) = 0
for all s, and u(1, s) = b + s, where b ∈

(

1
2

, 1
)
. The Doer’s payoff is W(a, s,θ,η) =

1{a = 1} (θb + η + s) and thus the ideal action a
∗
D(s,θ,η) takes on a value of 1 if

0 ≤ θb + η + s and −1 otherwise. In the absence of the bond contract, the set of
nudges is N = {0}, but with the bond contract, the set of nudges is expanded to
N ′ = [0,(1 −θ)b].

D
ow
nloaded from
https://academ
ic.oup.com
/qje/article-abstract/126/2/857/1870713 by A
rizona S
tate U
niversity user on 27 February 2020

LEARNING SELF-CONTROL 871

the impact of incentive-based commitment mechanisms,
this model provides insight into how a decision maker learns
to exert internal self-control and willpower, as studied in
Section IV.B.

Model 2 offers a continuous action space in which the Planner
commits by restricting the Doer’s choices and has no other way to
influence the Doer. This framework is particularly applicable to
study savings environments, as I demonstrate in Section IV.A.,
in which the Planner can purchase illiquid assets to commit the
Doer to some minimal savings but still retain some flexibility for
the Doer to save more than that minimum.

Since the applications of interest in the literature satisfy
either Model 1 or Model 2, I characterize learning only for these
two special cases. In both cases, the Planner faces an exercise
of optimal experimentation similar to but richer than a standard
multiarmed bandit. Naturally, my work builds on insights
derived from the experimentation literature, particularly Easley
and Kiefer (1988) and Aghion et al. (1991).

III.A. Adequate versus Partial Learning

The central issue analyzed is whether the Planner eventu-
ally attains the same payoffs that he would in the full informa-
tion benchmark in which the Doer’s type is known. Answering this
question sheds light on when an assumption of sophistication can
be justified through long-run learning.

The full information payoff is defined as follows: for each type
θ, when the Planner selects commitment c, his expected staticpay-
off is π(θ, c) =

∫
S
u( aD[s,θ, c], s) dν. Given a Doer of type θ, denote

the Planner’s full information payoff by π∗(θ) = maxc∈Cπ(θ, c),
and let C∗(θ) denote the set of commitments that attain this pay-
off. This payoff corresponds to the Planner’s optimal balance be-
tween flexibility and commitment when the Planner knows the
Doer’s type; its existence is guaranteed by continuity and com-
pactness assumptions.

When the Planner is uncertain about θ and has beliefs μ, he
optimizes with respect to his beliefs. Denote the payoff in a single
period from a commitment choice c by m(μ, c) =

∫
Θ
π(θ, c)dμ; how-

ever, when the Planner is even slightly patient, he also values
learning. Given a commitment choice c and prior μ, the Planner
may revise his beliefs when observing the Doer’s chosen action
and the realized signal; Q(μ, c) denotes the probability distribu-
tion over posteriors induced by (μ, c), and P(Θ) denotes the set of

D
ow
nloaded from
https://academ
ic.oup.com
/qje/article-abstract/126/2/857/1870713 by A
rizona S
tate U
niversity user on 27 February 2020

872 QUARTERLY JOURNAL OF ECONOMICS

Borel probability measures onΘ. The Planner solves the Bellman
Equation,

V(μ; δ) =max
c∈C

{

(1 − δ)m(μ, c) +δ
∫

P(Θ)
V(μ̃; δ) dQ(μ, c)

}
.

The Planner’s beliefs, commitment choices, and therefore,
value function evolve stochastically; the object of interest is their
long-run distribution conditional on the true θ. We can partition
the set of types based on whether learning is necessarily efficient,
yielding the full information payoffs in the longrun.

DEFINITION 1. Learning is adequate for a type θ if the Planner’s
payoffs eventually converges to the full information bench-
mark with probability 1,

Pr
(
lim

t→∞
V[μt; δ] = π

∗[θ]|θ
)

= 1,

and otherwise, learning is partial for typeθ. Learning is glob-
ally adequate for a prior μ0 and discount factor δ if the set
of types for which learning is partial has μ0-measure 0, and
otherwise, learning is inadequate for that prior and discount
factor.

Adequate learning is weaker than complete learning insofar
as the eventual belief need not identify the type, but the Planner
nevertheless obtains the same payoffs as in the full information
benchmark. Globally adequate learning requires more than the
mere existence of strategies that ensure learning; indeed, it must
be that at least one such strategy is optimal for the Planner.

III.B. Full Commitment Distinguishability

The possibility for globally adequate learning relies critically
on the commitment set, and one can assess if it is conducive to
learning by examining how a fully informed Planner behaves. Be-
fore deriving the precise condition, I offer a heuristic explana-
tion of its connection to globally adequate learning. Whenever the
Planner remains partially flexible so that the Doer can choose dif-
ferent actions in different states, the Planner eventually learns
the Doer’s type from the empirical frequency of its action choices.
It is only when the Planner chooses to constrain the Doer to a sin-
gle action in all states that learning is possibly impeded. Since it
might be optimal to fully commit some types of the Doer to choose
â, the Planner can perpetually distinguish these types from others

D
ow
nloaded from
https://academ
ic.oup.com
/qje/article-abstract/126/2/857/1870713 by A
rizona S
tate U
niversity user on 27 February 2020

LEARNING SELF-CONTROL 873

if there is a way to fully commit those types while retaining flex-
ibility for others. The presence of such expedient commitment is
formalized as the property of full commitment distinguishability.

Formally, let Ĉ(θ) ={c ∈C: aD( s,θ, c) = â for almost all s} de-
note the set of commitments that induce a Doer of type θ to choose
the action â with probability 1; for each type, this set is nonempty
because C includes the singleton menu {â}. The set of types for
which a fully informed Planner finds it optimal to fully commit to
the action â—and relinquish flexibility altogether—is denoted by
Θ̂ = {θ ∈Θ: π∗(θ) = π̂}.

DEFINITION 2. An environment Γ satisfies full commitment dis-
tinguishability (FCD) if for almost every θ̂ in Θ̂ and every θ
not in Θ̂, Ĉ(θ̂)* Ĉ(θ).

Under FCD, the Planner can carefully select commitment so
as to fully commit almost every “bad type” in Θ̂ while retaining
partial flexibility for some “good type” outside Θ̂. A sufficient but
unnecessary condition for FCD is that a fully informed Planner
never chooses full commitment, as in Section II when the bond
commitment contract was introduced. An environment fails FCD
if the only commitments that fully commit a strictly positive mea-
sure of types inΘ̂ also fully commit some types not inΘ̂. Since the
Planner’s preference for commitment is determined by how the
Doer maps taste shocks into actions, the particular representation
of the Doer’s preferences does not affect whether an environment
satisfies FCD.

To clarify what is being assumed in FCD, I detail when the
condition holds in each of these two models below. These details
are not entirely necessary to understand the connections drawn
between FCD and globally adequate learning, and so some read-
ers may wish to skip to Section III.C.

When Does FCD Hold? Consider the binary action environ-
ment of Model 1: because W satisfies the single-crossing condition
in(a,θ), any commitment that induces the Doer of typeθ toalways
select a = 1 also induces all higher types to choose that higher ac-
tion in every state. Thus, FCD can be satisfied only if whenever a
type θ is in Θ̂, all higher types are also in Θ̂.

OBSERVATION 1. In Model 1, an environment Γ satisfies FCD if
and only if for almost all θ in Θ̂, all higher types θ′ are also
in Θ̂.

D
ow
nloaded from
https://academ
ic.oup.com
/qje/article-abstract/126/2/857/1870713 by A
rizona S
tate U
niversity user on 27 February 2020

874 QUARTERLY JOURNAL OF ECONOMICS

The following examples highlight aspects of FCD in Model 1:
let the Planner’s payoff from a = 1 in period t be b + st in which
b = 34 and st is drawn uniformly from S = [−1, 0], and recall that
the Planner’s payoff from a = 0 is normalized to 0. The Doer’s type
is drawn from Θ =

[
0, 12

]
, and the Doer chooses a = 1 if and only if

θb + s + η is positive. The Planner selects nudges η from N , and
depending on N , FCD is satisfied or violated. In each of these ex-
amples, the Planner has access to all possible menus but does at
least as well by choosing the menu {−1, 1} and using the nudges
specified below.

EXAMPLE 1 (FCD with no commitment). SupposeN ⊇
[

3
8 ,

3
4

]
: then

a fully informed Planner sets η∗(θ) = (1 −θ)b to align every
type’s preference perfectly with the Planner. SinceΘ̂ =∅, FCD
is trivially satisfied.

EXAMPLE 2 (FCD with commitment). Suppose that N = {0} ∪[
3
4 , 1
]
: then a fully informed Planner uses a nudge of 34 for

each type. Such a commitment choice offers flexibility to all
types below 13 and fully commits all higher types. Thus, Θ̂ =[

1
3 ,

1
2

]
and satisfies FCD.

EXAMPLE 3 (FCD not preserved under expansion). Suppose that
the nudge

{
3
16

}
is added to N above: then a fully informed

Planner uses a nudge of 316 for types in
(

5
12 ,

1
2

)
, thereby offer-

ing flexibility to those types, and a nudge of 34 for any other
type. FCD is violated because Θ̂ =

[
1
3 ,

5
12

]
.

Examples 2 and 3 highlight a distinction between FCD and
standard notions of richness: FCD in Model 1 is not necessarily
preserved when N is expanded.

In Model 2, it is difficult to offer a simple criterion for FCD
generally. One can simplify having to check every potential pair
of types θ̂ inΘ̂ andθ not inΘ̂: FCD holds if and only if it is possible
to fully commit the type with the least self-control (θ) while retain-
ing flexibility for any type not in Θ̂. FCD simplifies further in the
special cases of Models 2a−b in which M comprises all feasible
interval menus or all feasible menus respectively. In these cases,
a Planner has no reason to fully commit a Doer that would freely
pick actions higher than â in some states: by Assumption 1, the
Doer chooses actions higher than â only in those states that the
Planner prefers higher actions to â. Thus a Planner is better off by
offering the menu [â, ā] rather than committing the Doer to choos-
ing â in every state. So the set of types for which full commitment

D
ow
nloaded from
https://academ
ic.oup.com
/qje/article-abstract/126/2/857/1870713 by A
rizona S
tate U
niversity user on 27 February 2020

LEARNING SELF-CONTROL 875

is optimal is a subset of those that choose actions less than â in
every contingency. In this context, FCD is satisfied if and only if
these two sets of types are identical.

PROPOSITION 2. An environmentΓ satisfies FCD in Model 2 if and
only if Θ̂ has zero measure or for every θ not in Θ̂, it is the
case that Ĉ(θ) * Ĉ(θ). In Models 2a−b, FCD is satisfied if and
only if the set of types for which full commitment is optimal,
Θ̂, coincides with the set of all types whose preferred action
is below â in every state.

III.C. Main Results

I describe the connection between FCD and globally adequate
learning: FCD is sufficient for globally adequate learning and
might also be necessary in particular cases. When FCD is sat-
isfied, these results offer Bayesian learning foundations for so-
phistication. Simultaneously, the results indicate that individuals
might not be fully sophisticated in environments in which FCD
fails and that interventions may help even experienced individu-
als develop a better understanding of their temptations.

I begin by demonstrating sufficiency.

THEOREM 2. If an environment Γ satisfies FCD, then for all pri-
ors μ0 and discount factors δ, learning is globally adequate in
Models 1 and 2.

When FCD is satisfied, even a slightly patient Planner even-
tually chooses commitment as if he were fully sophisticated about
his imperfect self-control. Moreover, under FCD, globally adequate
learning may obtain even if the Planner is not sufficiently forward-
looking to actively learn and experiment: a passive learner who
myopically best-responds to his beliefs eventually behaves as if
he perfectly understood his temptation. As such, the introduction
of partial commitments that induce FCD can have strong impli-
cations for individuals regardless of their patience or prior.

The essence of the argument is that if FCD is satisfied, learn-
ing has no short-term costs but offers short-term and long-term
benefits; in other words, the commitment structure facilitates
cheap experimentation. Under FCD, the Planner can select com-
mitments that fully commit those types in Θ̂—for which full com-
mitment is optimal—while retaining partial flexibility to some
types not in Θ̂. By choosing commitment in this way, the Plan-
ner can learn about the Doer based on the empirical frequency of

D
ow
nloaded from
https://academ
ic.oup.com
/qje/article-abstract/126/2/857/1870713 by A
rizona S
tate U
niversity user on 27 February 2020

876 QUARTERLY JOURNAL OF ECONOMICS

his choices. Should the Doer choose different actions over time,
that frequency identifies his type. On the other hand, if the Doer
chooses the same action in every period, the Planner learns that
the Doer’s type is one that has no flexibility at that commitment
choice. Based on this updated belief, the Planner can change his
commitment choice so as to still guarantee that types in Θ̂ fully
commit while offering partial flexibility to types not in Θ̂ that re-
main in the support of his posterior. By iterating this procedure,
the Planner eventually makes choices that converge to those in
the full information benchmark.

Notice that such an argument applies independently of the
accuracy with which the Planner observes past states. The signal
σt(at, st) could be a noisy signal of the state st or uninformative al-
together, and globally adequate learning is nevertheless ensured.
Insofar as there are substantive reasons to consider noisy attri-
bution structure (prior work has focused on such cases),14 it is
noteworthy that such noise does not prevent long-run learning in
the presence of FCD, although it likely interferes with the speed
of learning.

In the absence of FCD, challenges of costly experimentation
emerge. When the cost of experimentation outweighs the bene-
fits of learning, the Planner may choose to sacrifice learning for
the sake of short-term payoffs if he is not perfectly patient. Since
sets of menus can assume many different forms in Model 2, it is
challenging to demonstrate the necessity of FCD generally. Ac-
cordingly, I demonstrate the necessity of FCD in the special cases
of Model 2a−b, in which M comprises all feasible interval menus
or all feasible menus respectively.

THEOREM 3. Suppose that an environment Γ fails FCD. Then for
every discount factor δ < 1, learning is inadequate for some open set of priors in Models 1 and 2a−b.

14. Noisy attribution has featured in, for example, Benabou and Tirole (2004),
where the assumption is that a decision maker learns about his self-control using
a “revealed preference” approach. One reason to study noise in attribution is that
if past circumstances and taste shocks are representations of subjective states and
emotions, it might be difficult for an individual with imperfect self-control toobtain
perfect information about past circumstances. Various studies highlight the diffi-
culty that individuals face in separating situational factors from aspects of one’s
innate character (Baumeister, Heatherton, and Tice 1994; Kahneman, Wakker,
and Sarin 1997; Ameriks, Caplin, and Leahy 2003). Moreover, if the Doer observe
a noisy signal of st and the Planner observes st but not the signal of the Doer,
the imperfect monitoring model that emerges is equivalent to a noisy attribution
framework.

D
ow
nloaded from
https://academ
ic.oup.com
/qje/article-abstract/126/2/857/1870713 by A
rizona S
tate U
niversity user on 27 February 2020

LEARNING SELF-CONTROL 877

The inadequacy of learning highlights that when FCD fails, a
modeler cannot preclude the possibility of partial learning. When
learning fails, it does so in a systematic way: the Planner fully
commits a Doer type that he would have offered flexibility had
he been fully informed. The likelihood with which learning fails
necessarily depends on the discount factor, the Planner’s prior,
and the informativeness of the feedback that the Planner receives;
I show in Theorem 4 below that such inefficiencies vanish as the
Planner approaches perfect patience.

III.D. Remarks

The Role of Patience Posing self-awareness as an issue of ex-
perimentation permits the study of the connection between the
decision maker’s patience and his sophistication. As the Planner
becomes more patient, he increasingly values sacrificing current
payoffs to gain information that helps improve commitment
choices in the future. This logic suggests that a more patient Plan-
ner is less likely to fully commit when the Doer’s type warrants
flexibility, and by the same token, will experiment more even if
the Doer’s type warrants full commitment. I offer a limit result in
this vein.

THEOREM 4. As the Planner approaches perfect patience, his pay-
offs and choices approach the full information benchmark
with probability 1 regardless of whether FCD is satisfied: for
every prior μ0, limδ→1V(μ0; δ) =

∫
Θ
π∗(θ)dμ0.

Theorem 4 follows from the observation that if a fully informed
Planner is forced to select from a finite number of commitments
CN in C, his payoff loss from this restriction can be made arbi-
trarily small regardless of the Doer’s type. Analogous to results
derived elsewhere in the experimentation framework,15 the Plan-
ner can then approach the full information payoffs (restricted to
a finite set of commitment choices) by conducting a large number
of experiments and then forever selecting what appears to be the
best commitment thereafter.

This result indicates the virtue of patience when it comes to
learning; a patient Planner tolerates many lapses of self-control
so as to optimally commit in the future. Generally, it is ambigu-
ous as to whether the increased experimentation takes the form

15. For example, see Aghion et al. (1991) and Fudenberg and Levine (1993b).

D
ow
nloaded from
https://academ
ic.oup.com
/qje/article-abstract/126/2/857/1870713 by A
rizona S
tate U
niversity user on 27 February 2020

878 QUARTERLY JOURNAL OF ECONOMICS

of flexibility for a greater number of periods or greater flexibility
per period. However, for the simplest commitment setting studied
in Section II, this result would suggest that the patient Planner
will choose to remain fully flexible for many periods even when he
sees the Doer choosing a = 1 with low probability. Accordingly, it
would be difficult to distinguish between naive and patient Plan-
ners merely on the basis of short-run commitment choices.

Identifying Awareness. The necessary and sufficient condition
for adequate learning, FCD, is defined in terms of how a fully
sophisticated Planner copes with the full range of possible tempta-
tions. As in the applications in Section IV, it is relatively straight-
forward to check this condition in full information models that
we typically use; in many such contexts, specifying the full infor-
mation model of temptation and commitment implicitly assumes
FCD or its absence.

When a modeler or policy maker does not have a particular
full information model, checking for FCD may be challenging be-
cause how a decision maker copes with different levels of tempta-
tion is not directly observed. Nevertheless, within some settings,
an individual’s long-run preference for commitment can help iden-
tify his eventual awareness.

To illustrate, if a decision maker strictly prefers some
nonsingleton menu c to every singleton menu, then the decision
maker is partially committing and therefore must have
adequately learned. In the opposite case, if the decision maker
weakly prefers some singleton menu to every other menu, then
he is fully committing and so he may or may not be sophisticated.
In this case, the policy maker can learn about the decision maker’s
awareness by exposing him to different commitments and observ-
ing how he responds. If temporary subsidies to commitments not
in his most preferred set have long-lasting implications on the
decision maker’s commitment choices, then learning was
inadequate.16

Relationship to Self-Confirming Equilibrium. The dynamic
process of learning that I have studied bears a resemblance to

16. Consistent with this possibility, temporary subsidies to drug rehabilitation
are often believed to have effects after the subsidy has ended (Kaper et al., 2005).
Given the experimentation challenge that individuals face in finding the appropri-
ate commitment devices and therapy for addiction, it is natural that learning is
inadequate.

D
ow
nloaded from
https://academ
ic.oup.com
/qje/article-abstract/126/2/857/1870713 by A
rizona S
tate U
niversity user on 27 February 2020

LEARNING SELF-CONTROL 879

steady-state solution concepts motivated by learning, in particu-
lar to the concept of self-confirming equilibrium (SCE), developed
by Fudenberg and Levine (1993a). One way toviewthese results is
as foundations for studying SCE (or a variant thereof) in dynamic
self-control environments. In this section, I compare the limiting
behavior and beliefs of the dynamic model with those that emerge
from the steady-state concept.

In an SCE, players do not deviate given their beliefs about
how others behave, and each player’s belief is consistent with ex-
periences on the path of play; unlike a Nash equilibrium, these
beliefs might err on how others would behave off the equilibrium
path. Adapting this concept to the framework here requires addi-
tional restrictions: the Planner’s beliefs both on and off the path
of play must be compatible with his knowledge of the Doer’s type
space, that is, how different levels of temptation lead to differ-
ent distributions of behavior. Encoding these restrictions yields an
adaptation of rationalizable self-confirming equilibria (RSCE) to
Bayesian games: an RSCE refines the set of SCE by imposing the
further restriction that any player’s belief about others’ actions,
on or off the path of play, must be rationalized by her knowledge
of their payoffs.17

Thus, the appropriate analogue to the framework here is a
Bayesian game in which: (a) Nature chooses the Doer’s type, which
is fixed once and for all, (b) the Planner and Doer have payoff func-
tions specified above, (c) the Planner knows Nature’s distribution
of states s and how the Doer chooses actions as a function of its
type, and (d) the Planner observes only the action chosen by the
Doer and the signal σ(a, s) at the end of each round. An RSCE of
this game consists of a commitment choice c rationalized by some
consistent belief μ about the Doer’s type and θ, the actual type
of the Doer. The Planner’s belief μ is consistent with the path of
play if the distribution over actions and signals generated by the
commitment choice c and the Doer’s type θ coincides with what
the Planner expects given belief μ.18 The following describes the
RSCE of this setting.

17. RSCE is developed by Dekel, Fudenberg, and Levine (1999). Dekel, Fuden-
berg, and Levine (2004) adapt SCE to Bayesian games, including those in which
a player’s type is fixed once and for all; however, they do not study RSCE in this
setting. Fudenberg and Levine (2009) offer a survey of this literature.

18. Formally, when the Planner chooses commitment c, and the Doer’s
type is θ, consistency of μ demands that if θ′ is in the support of μ, then
ν (s: aD[s,θ, c] =/ aD[s,θ

′, c]) = 0.

D
ow
nloaded from
https://academ
ic.oup.com
/qje/article-abstract/126/2/857/1870713 by A
rizona S
tate U
niversity user on 27 February 2020

880 QUARTERLY JOURNAL OF ECONOMICS

OBSERVATION 2. The set of RSCE in which the Doer’s type is θ
comprises:

1. Complete learning: a commitment c in C∗(θ), rationalized
by the belief that ascribes probability 1 toθ.

2. Full commitment: a commitment c rationalized by some be-
liefμ, such that c induces typeθ and all types in the support

ofμtofully commit, that is, c is in Ĉ(θ)∩
(
∩θ′∈Supp(μ)Ĉ(θ

′)
)

Since the Planner’s beliefs must be consistent with observa-
tions from repeated play, if the Planner retains some flexibility
for the Doer, his steady-state beliefs must ascribe probability 1 to
the true type of the Doer (because different Doer types respond
differently to flexibility). The Planner may also hold a belief that
rationalizes full commitment, which precludes the possibility for
further learning. In contrast to the dynamic setting, full commit-
ment may be chosen even if FCD is satisfied and the Doer’s type
does not warrant it. A simple illustration of this RSCE is one
in which the Planner assigns probability 1 to Θ̂ and chooses the
singleton menu {â}. Regardless of the Doer’s actual type, these
potentially incorrect beliefs are consistent with experiences on
the path of play and rationalize the commitment choice. Because
Theorem 2 demonstrates that such outcomes cannot emerge in
the dynamic framework unless the Doer’s type is indeed in Θ̂,
this inconsistency suggests that the limits of learning in the
dynamic framework correspond to a particular selection of
RSCE.

The source of this tension is that an RSCE permits the
Planner to ascribe zero probability to the truth in contrast to the
dynamic framework in which the Planner begins with a full sup-
port prior and almost-surely includes the truth within the support
of his limit beliefs. Selecting those RSCE that contain a “grain
of truth” is sufficient to reconcile the two approaches. Define an
RSCE to be absolutely continuous (AC-RSCE) if the commitment
choice c is rationalized by at least one belief that is both consistent
and includes in its support the Doer’s actual type θ. Although the
full implications of this selection are better left for future work,
it captures the implications of FCD in a steady-state
setting.

THEOREM 5. If an environment Γ satisfies FCD in Models 1 and
2, then in every AC-RSCE in which the Doer’s type is θ, the

D
ow
nloaded from
https://academ
ic.oup.com
/qje/article-abstract/126/2/857/1870713 by A
rizona S
tate U
niversity user on 27 February 2020

LEARNING SELF-CONTROL 881

Planner’s commitment choice is from C∗(θ) yielding the full
information payoffs, regardless ofθ. If an environmentΓ fails
FCD in Models 1 and 2a−b, then there exist AC-RSCEs for
a strictly positive measure of θ in which the Planner fails to
attain full information payoffs.

IV. APPLICATIONS

IV.A. Consumption-Savings

The canonical application of self-control models is toconsump-
tion and savings decisions. Evidence suggests that individuals are
tempted by immediate consumption, and a recent literature has
studied how access to an illiquid asset offers a powerful instru-
ment for commitment to such individuals.19 Holding wealth in
illiquid form effectively sets a minimal level of savings and thereby
protects some wealth from consumption splurges. At the same
time, excessive illiquidity impedes a decision maker from respond-
ing touninsurable risk, and soa delicate balance between commit-
ment and flexibility must be maintained. The burgeoning
literature on these issues has largely focused on decision mak-
ers who are fully sophisticated about their imperfect self-control.
While this is the natural starting point, it raises the question of
whether experience would lead a decision maker who is initially
uncertain about his self-control toeventually act as if he were fully
sophisticated. This section answers the question in the affirma-
tive in this canonical setting: when a decision maker is uncertain
about his tendency to overconsume and has access to an illiquid
asset, he eventually makes commitment choices as if he were fully
informed. I demonstrate this positive learning result first in the
setting studied in the prior literature—in which commitment cor-
responds to the purchase of illiquid assets—and then show that it
also extends to a setting in which singleton consumption sets are
also available.

The simple savings environment that I study is a decision
maker who faces a tradeoff between commitment and flexibility

19. The role of illiquid assets as commitment was discussed in some of the
earliest work on self-control—for example, Strotz (1955) and Thaler and Shefrin
(1981)—and studied in a number of subsequent articles, including Laibson (1997),
Barro (1999), Gul and Pesendorfer (2005), Fudenberg and Levine (2006), and
Amador, Werning, and Angeletos (2006). Harris and Laibson (2003) offer a sur-
vey and discussion.

D
ow
nloaded from
https://academ
ic.oup.com
/qje/article-abstract/126/2/857/1870713 by A
rizona S
tate U
niversity user on 27 February 2020

882 QUARTERLY JOURNAL OF ECONOMICS

in each period. His expected payoff from a consumption stream
{ct}

∞
t = 0 is E

[∑∞
t = 0 δ

tutU( ct)
]
, where U is a Constant Relative Risk

Aversion (CRRA) utility function with a coefficient of relative risk
aversion σ > 0, and ut is a taste shock. The taste shock in each
period affects the marginal utility of current consumption and fol-
lows an i.i.d. process represented by a continuous density on [u, ū]
with u > 0, and normalized so that E [u] = 1. The Planner begins
with an initial wealth y0, and subsequent wealth is generated by
the returns from savings in the prior period at gross interest rate
R. To ensure that the transversality condition is satisfied, I as-
sume thatδR1−σ < 1. The restriction to CRRA utility ensures that in the absence of a self-control problem, a decision maker would have a consumption rate that depends only on the taste shock: in period t, if he has wealth yt and faces a taste shock ut, he consumes cP( ut) yt.

In this dual self framework, consumption choices are made
as follows: each period is divided into two subperiods. Wealth at
the beginning of the period is denoted by yt, which can be in-
vested by the Planner in illiquid assets or kept in a liquid form in
the first subperiod. The Planner effectively selects a minimal sav-
ings rate, st, by investing an amount styt in illiquid assets. In the
second sub-period, the taste shock is realized, then the Doer se-
lects consumption ct from the feasible set—which is [0, ( 1 − st)yt]—
and invests the remainder in the asset. For expository con-
venience, assume that returns from saving occur at the same rate
regardless of whether the choice was made by the Planner or Doer.

As before, the Planner has the standard long-run preferences,
whereas the Doer is a short-run behavioral type that is tempted
by immediate consumption. Formally, a Doer of type θ will aim
to consume c(u,θ)y when total wealth is y, and he selects the fea-
sible consumption closest to this ideal. Doer types θ are in [θ, 1]
drawn according to μ0. I assume that c(u,θ) is strictly positive,
smooth in (u,θ), strictly decreasing in θ, and strictly increasing
in u; morever, c(u, 1) coincides with the Planner’s ideal, cP(u). Ac-
cordingly, for every taste shock, every Doer type has a constant
saving rate that is (weakly) less than the Planner’s optimal sav-
ing rate and strictly increasing in the Doer’s type. One special
case is a linear specification c(u,θ)=

(
1
θ

)
cP(u); another special case

inheres to standard formulations of present bias in which c(u,θ)

20. A guess and verify approach to solving the stochastic Euler equation yields
that cP(u)=

u1/σC
u1/σC + δ1/σR(1−σ)/σ

, where C solves

E
[(

Cu1/σ + δ1/σR(1−σ)/σ
)σ]

= 1.

D
ow
nloaded from
https://academ
ic.oup.com
/qje/article-abstract/126/2/857/1870713 by A
rizona S
tate U
niversity user on 27 February 2020

LEARNING SELF-CONTROL 883

corresponds to how the Planner would have chosen had his dis-
count factor been θδ.21

Since my goal is to establish the inevitability of globally ad-
equate learning, I make the most stark of informational assump-
tions: the Planner observes only past consumption and not past
taste shocks. Since learning emerges here, it continues do so in
richer observational environments.

The setting here is similar to Model 2a of Section III in which
partial commitment takes the form of interval menus. By invest-
ing in illiquid assets, the Planner specifies a consumption cap (or
a savings floor), and for every taste shock, the Planner would wish
to consume less than the Doer. The contrast to Section III emerges
in the introduction of a state variable other than beliefs, namely
the Planner’s wealth, although the restriction to CRRA utility af-
fords tractability. Although the set of feasible consumption choices
changes over time, the preferred savings and consumption rates
for the Planner and Doer are time and history invariant. To apply
the methods developed earlier, one can formulate actions as sav-
ings rates and states as (u)−1; with such transformations, both
the Planner and Doer prefer higher savings rates in higher states,
the Doer’s preferred savings rate in each state is increasing in its
type, and the Planner prefers a higher savings rate than the Doer
in every state.

Directly solving the model in which the Planner is uncertain
about the Doer’s temptation is challenging because the specifica-
tion of any menu induces different separating and bunching re-
gions for different types of the Doer. Accordingly, the selection
of a menu affects the Planner’s learning in intricate ways. The
tools developed in Section III sidestep these complications alto-
gether: I demonstrate that the simpler full information model (in
which the Planner knows the Doer’s type) satisfies FCD and can
then apply Theorem 2. Verifying FCD takes the following
steps.

First, suppose that the Planner could not offer any flexibil-
ity and were forced to commit to a particular consumption path
irrespective of taste shocks; this selection corresponds to finding
the analogue of the optimal full commitment â from Section III.
The optimal binding consumption path in this case is one in

21. Specifically, c( u,θ) = u
1/σC(θ)

u1/σC(θ) + (θδ)1/σR(1−σ)/σ
, where C(θ) solves

E
[(

Cu1/σ + (θδ)1/σ R(1−σ)/σ
)σ]

= 1.
D
ow
nloaded from
https://academ
ic.oup.com
/qje/article-abstract/126/2/857/1870713 by A
rizona S
tate U
niversity user on 27 February 2020

884 QUARTERLY JOURNAL OF ECONOMICS

which the Planner ensures that the fraction ŝ = δ1/σR(1−σ)/σ is
saved in each period and consumes the remainder.

Second, I solve for how the Planner resolves the tradeoff be-
tween commitment and flexibility when he knows that the Doer
is of type θ. In the Online Appendix, I demonstrate that he pur-
chases the same fraction of illiquid assets, s∗ (θ), in each period to
induce some commitment. Accordingly, the Doer can never con-
sume more than 1 − s∗(θ) of the current wealth but can choose to
consume less when consumption is less valuable. Without temp-
tation toward present consumption (θ = 1), the Planner chooses to
keep all wealth liquid and affords the Doer maximal flexibility;
otherwise, the Planner may use a strictly positive level of illiquid
assets as commitment. If it so happens that for all taste shocks,
the Doer wish to consume more than the amount of wealth that is
kept in a liquid form, the Doer is fully committed to a particular
consumption path.

Using the above characterizations, one can then partition Θ
into Θ̂, the set of types that a fully informed Planner chooses to
leave with no flexibility, and its complement, to wit, types that
the Planner offers (partial) flexibility. In the Online Appendix,
I demonstrate that Θ̂ comprises all types that consume at least
(1 − ŝ)yt for every realization of ut; to effectively fully commit such
types, the Planner only keeps that level of wealth liquid, and holds
illiquid assets of ŝyt in each period. In contrast, if the Doer con-
sumes less than (1 − ŝ)yt for some taste shocks, the Planner ben-
efits from maintaining some flexibility. When facing this milder
temptation tooverconsume, the Planner purchases a smaller hold-
ing of illiquid assets and thereby retains greater flexibility for the
Doer. Thus, the types that are fully committed, Θ̂, corresponds to
the set {θ ∈Θ: c(u,θ)≥ 1 − ŝ}.

Figure I illustrates these consumption caps. If a Planner faces
a Doer of type θ, he sets the consumption cap so as to permit the
Doer to select its most preferred consumption for all taste shocks
u < û, and constrains the Doer to consume(1 − s(θ)) otherwise. In contrast, when facing a Doer of type θ̂, who has less self-control, the Planner’s optimally chosen consumption cap binds the Doer for every taste shock, inducing it to select 1 − ŝ for every taste shock; this latter type is in Θ̂.

It is straightforward to see how FCD is satisfied in this set-
ting: when the Planner purchases ŝyt of illiquid assets in period t,
he simultaneously fully commits every type in Θ̂ while retaining
flexibility for types not in Θ̂. As shown in Figure I, when choosing

D
ow
nloaded from
https://academ
ic.oup.com
/qje/article-abstract/126/2/857/1870713 by A
rizona S
tate U
niversity user on 27 February 2020

LEARNING SELF-CONTROL 885

FIGURE I
Consumption Caps

Dotted curves indicate the Doer’s ideal consumption for each taste shock, and
solid lines indicate the Doer’s actual consumption when the Planner selects com-
mitment optimally.

from this menu, a Doer of type θ′ consumes the remaining wealth
with probability 1, whereas the typeθ chooses below the consump-
tion cap with strictly positive probability. Thus, even when se-
lecting this commitment, the Planner can distinguish types in Θ̂
(for which such full commitment is optimal) from types not in Θ̂
ensuring globally adequate learning.

THEOREM 6. Learning is globally adequate in this consumption-
savings framework for all priors and discount factors.

This positive learning result demonstrates that if a decision
maker begins with uncertainty about his self-control, eventually,
he purchases the same fraction of illiquid assets that he would
if he could forecast his temptations perfectly. Thus, assuming so-
phistication in this important setting can be justified by Bayesian
learning. Of course, while learning, the Planner may make
(ex post) inefficient choices that adversely affect his future income
distribution relative to the full information benchmark.

The commitment mechanism studied so far in this applica-
tion is empirically relevant to savings decisions and has been the
focus of prior work. However, an inconsistency with the model of
Section III is that this mechanism excludes singleton menus,
which fully commit every type of the Doer. Establishing that the
above results extend to a setting that includes both illiquid assets

D
ow
nloaded from
https://academ
ic.oup.com
/qje/article-abstract/126/2/857/1870713 by A
rizona S
tate U
niversity user on 27 February 2020

886 QUARTERLY JOURNAL OF ECONOMICS

and singleton menus requires additional analysis; in the Online
Appendix, I demonstrate that learning is globally adequate in this
richer setting if the coefficient of relative risk aversion, σ, is at
least 1.

The full information model to which behavior eventually con-
verges shares many predictions of sophistication in the
quasi-hyperbolic savings model but is analytically distinct. It has
a unique solution and a constant commitment rate with CRRA
utility (similar to Fudenberg and Levine 2006). In contrast, the
quasi-hyperbolic savings model in discrete time has a continuum
of sophisticated solutions, even after assuming Markovian behav-
ior (Krusell and Smith, 2003). Thus, equilibrium selection remains
a challenge in the quasi-hyperbolic model, thereby making it dif-
ficult to study learning in that framework but does not pose an
issue here.

IV.B. Costly Self-Control

Fudenberg and Levine (2006) offer a simple but powerful
Planner–Doer framework in which the Doer’s preferences are com-
pletely characterized by the present and the Planner exerts costly
self-control to induce the Doer to take actions more consistent
with the Planner’s preferences. Here I analyze how a
Planner who is uncertain about the cost or efficacy of self-control
learns over time how much self-control to use, building on the bi-
nary action environment studied in Model 1 of Section III.

Consider a decision maker who faces an infinite sequence of
tasks. Completing a task in any period (at = 1) leads to a future
benefit with a present value of b, which is not valued by the Doer.
Not completing the task (at = −1) has the benefit of immediate
leisure st, where st is uniform drawn from [0, 1]. I assume that b is
in
(

1
2 , 1
)
, generating a tradeoff between commitment and flexibil-

ity. Prior to the realization of st, the Planner chooses a menu from
{−1, 1}. After the realization of st, the Planner can exert costly
self-control to nudge the Doer to complete the task. If the Plan-
ner exerts self-control of magnitude Γ ≥ 0, self-control is success-
ful for a Doer of type θ if Γ is at least

st
θ

; otherwise, it fails. The

Planner’s payoff from completing the task at date t is b −Γ , and
the payoff from not completing the task is st −Γ . I assume that
θ is drawn from an interval

[
θ, θ̄
]

in which θ > 0; lower types re-
quire the Planner exert greater self-control to complete the task,
whereas higher types require less self-control.

D
ow
nloaded from
https://academ
ic.oup.com
/qje/article-abstract/126/2/857/1870713 by A
rizona S
tate U
niversity user on 27 February 2020

LEARNING SELF-CONTROL 887

The full information model, in which the Planner knows θ, is
similar to that considered by Fudenberg and Levine (2006). If θ is
sufficiently high, then the Planner chooses the menu {−1, 1} and
has an associated cutoff s∗ such that he uses costly self-control
to complete the task in period t if and only if st is less than s

∗.
Such a choice has the benefit of flexibility because the Planner
can choose whether to exert costly self-control depending on the
value of leisure. In contrast, when the cost of self-control is suffi-
ciently high, the Planner may prefer tocommit through the single-
ton menu. While such a choice relinquishes flexibility, it obviates
costly self-control because the Doer no longer needs to be nudged
to complete the task. The Planner prefers flexibility to full
commitment if θ∗ is at least (1− b)

2b − 1 .
When the Planner is uncertain about the cost of self-control,

he may use too much or too little self-control to influence the Doer.
Employing too little self-control leads the agent to not complete
the task, although the costs of self-control are still borne. In con-
trast, too much self-control guarantees that the task is completed
but at unnecessary expense. Over time, the Planner learns about
the efficacy of self-control based on experience. As before, it is pos-
sible to assess eventual beliefs and choices by studying the full
information model rather than the more intricate model that em-
beds uncertainty about the Doer’s type.

Before applying the tools developed for Model 1 in Section III,
let me highlight two differences between costly self-control and
nudges. In Section III, a nudge is costly if it induces the Doer to
choose a = 1 toooften but otherwise is intrinsically costless. In con-
trast, the self-control studied in this section is intrinsically costly.
Although this distinction is of conceptual importance, it does not
create any difficulty in applying the earlier results. The challenge
arises from the distinction in timing. In the earlier analyses, the
Planner’s choice of nudges and menus is made before the reso-
lution of uncertainty; as such, if the Planner partially commits,
his commitment choice induces a nontrivial distribution of be-
havior that varies by type. In contrast, in the setting here, the
Planner’s choice to exert costly self-control is made after uncer-
tainty about the state is resolved. Thus, if the Planner chooses
to exert costly self-control of magnitude Γ whenever the value of
leisure is less than some level s and this is successful for types
θ and θ′ of the Doer, the Planner cannot distinguish these types
by behavior even though he is partially committing for both. To
generate distinguishability with partial commitment, I assume

D
ow
nloaded from
https://academ
ic.oup.com
/qje/article-abstract/126/2/857/1870713 by A
rizona S
tate U
niversity user on 27 February 2020

888 QUARTERLY JOURNAL OF ECONOMICS

that if the Planner remains flexible and exerts costly self-control,
he observes some signal σ with support in < and whose distri- bution is increasing in Γ − s

θ
(in terms of first order stochastic

dominance). The interpretation of this signal is that the Planner
obtains partial information about the impact of self-control be-
yond the action choice. With this modification, results analogous
to Theorems 2 and 3 hold.

THEOREM 7. If θ > θ∗, then a fully informed Planner always
chooses a flexible menu and exercises self-control. Thus, no
type is fully committed, and FCD is satisfied. Learning is
globally adequate for all priors and discount factors. In con-
trast, if θ < θ∗, then the Planner chooses the menu {1} for sufficiently low types. FCD is violated, and for each δ > 0,
learning may be inadequate for an open set of priors.

V. DISCUSSION

This article offers a simple and tractable nonequilibrium
approach tostudying the evolution of beliefs, self-control, and com-
mitment choices as a decision maker learns from experience. I de-
rive a condition that reveals whether such learning inexorably
leads to sophistication. When FCD is satisfied, even a passive
learner eventually makes choices as if he were in the full informa-
tion benchmark. Learning-based foundations for sophistication in
these settings, such as a savings environment, can help address
the criticism that it involves an unrealistic level of rationality.22

In contrast, when FCD fails, then an impatient decision maker
may fail to learn adequately and make commitment choices that
are inefficient relative to the full information benchmark. In these
cases, policy interventions that subsidize experimentation may fa-
cilitate greater awareness.

In this article, I have adopted a dual-self approach to un-
derstand learning. By treating a Doer as nonstrategic, dual-self

22. Rubinstein (2005) offers such a criticism:

Sophistication is unrealistic since it suffers from the problems
of subgame perfection. An agent is super-rational in the sense that
he perfectly anticipates his future selves and arrives at equilibrium
between them. Present bias is a realistic phenomenon, but the
combination of the β,δ preferences with naivete or sophistication
assumptions makes the model even more unrealistic than time
consistency models.

D
ow
nloaded from
https://academ
ic.oup.com
/qje/article-abstract/126/2/857/1870713 by A
rizona S
tate U
niversity user on 27 February 2020

LEARNING SELF-CONTROL 889

models are analytically simple and generally have unique solu-
tions when a decision maker is sophisticated. From the perspec-
tive of learning, this simplicity permits the investigation of
whether the unique solution under partial awareness converges to
the unique solution under sophistication. In contrast, the quasi-
hyperbolic approach has multiple solutions even after imposing
sophistication and a Markovian restriction on behavior. This chal-
lenge of multiplicity is exacerbated when beliefs are endogenized
because in the resulting intrapersonal game, uncertainty about
self-control induces a motive for self-signaling. Multiplicity thus
impedes the study of learning in the quasi-hyperbolic approach,
especially in settings such as consumption-savings in which the
entire set of sophisticated solutions has not been characterized.
Since the dual-self framework captures issues of temptation and
commitment with greater parsimony, this approach facilitates a
more direct study of learning.

When learning fails at the limit, I find that it occurs in a
particular direction, namely that individuals underestimate their
self-control. It is important to understand the extent to which
these results can be reconciled with findings in the field. Some
of the evidence on self-awareness comes from individuals’
contractual choices and their subsequent decisionmaking.
DellaVigna and Malmendier (2006) demonstrate that many indi-
viduals choose monthly gym memberships but, given their usage
of the gym, would have paid less per visit had they selected the
pay-per-visit option. Similarly, Shui and Ausubel (2005) find that
individuals accept introductory credit card offers with lower in-
terest rates for a shorter duration rather than a higher interest
rate with a longer duration though they might be better off with
the latter choice. In both cases, behavior is consistent with indi-
viduals being sophisticated about their self-control—valuing the
commitment offered by a monthly membership or a shorter du-
ration period—as well as them overestimating their self-control.
In a model with overlapping generations, it is quite likely that
both effects are present insofar as less experienced customers who
have not had opportunities to learn may overestimate self-control.
DellaVigna and Malmendier (2006) argue that delays in cancel-
ing contracts are more consistent with partial naivete. However,
even when a decision maker does not overestimate self-control,
he has an incentive to delay canceling because there is an “op-
tion value” from learning, particularly in this sample that com-
prises first-time customers. This option value increases with the

D
ow
nloaded from
https://academ
ic.oup.com
/qje/article-abstract/126/2/857/1870713 by A
rizona S
tate U
niversity user on 27 February 2020

890 QUARTERLY JOURNAL OF ECONOMICS

decision maker’s patience making it difficult to distinguish
patience from naivete. Moreover, it would be natural to assume
that many such individuals may be learning both about their self-
control and the costs and benefits of going to a health club (as also
argued by Fudenberg and Levine 2006).

Indeed, this point raises the tension of assuming that a de-
cision maker is partially sophisticated about his self-control but
somehow fully sophisticated about the payoffs from his choices.
Although this article, like the prior literature, abstracts from this
issue, it is surely realistic that individuals often have to learn both
the payoffs of choices and whether their decisions are vulnerable
to temptation. In settings in which feedback about payoffs are de-
layed, we should expect learning to be slow. Although I defer a
fuller treatment of this issue, I illustrate its potential implications
through the following example.

Consider the two-type model from Section II, and suppose
for simplicity that δ = 0. Instead of the distribution of costs be-
ing known by the Planner, suppose that it depends on the real-
ization of a random variable ξ; in particular, let st be distributed
uniformly with support [ξ, 1 + ξ], and suppose thatξ itself is 0 with
probability 1 −λ0 and

(
θ̄−θ

)
b with probability λ0. This particu-

lar support is assumed for simplifying reasons and is unnecessary
for the analysis.23 To understand the implications of delayed feed-
back about payoffs transparently, consider a limit model in which
information about past costs does not emerge.24 In this setting, a
complete breakdown of learning occurs because a Planner who ob-
serves the Doer choose a = 1 with frequencyθb cannot distinguish
his lack of self-control from a high average cost of taking the ac-
tion. Prior beliefs then determine his choice to remain flexible or
commit (in the long-run limit), and under certain conditions, such
a Planner may perpetually undercommit relative to a full infor-
mation benchmark.

PROPOSITION 3. If μ0 andλ0 are sufficiently high and (θ, 0) is the
realized state, then a partially aware Planner always chooses
flexibility. If, however, a sophisticated Planner knew that
ξ = 0 or that θ = θ, then almost-surely, the Planner would
choose to commit.

23. It suffices if ξ is drawn from a continuum with values in its support that
are

(
θ̄−θ

)
b apart.

24. This stark setting is similar to cases studied by Dekel, Fudenberg, and
Levine (2004) in which payoffs are not observed and tothe identification challenges
that arise in Acemoglu, Chernozhukov, and Yildiz (2009).

D
ow
nloaded from
https://academ
ic.oup.com
/qje/article-abstract/126/2/857/1870713 by A
rizona S
tate U
niversity user on 27 February 2020

LEARNING SELF-CONTROL 891

The assumption that a Planner never observes any ex post in-
formation about payoffs is critical and instructive: when the Plan-
ner obtains informative signals about payoffs, this sequence of
signals identifies ξ in the long-run limit, eliminating the possi-
bility for undercommitment. As such, if feedback is delayed but
eventually forthcoming, one should expect undercommitment to
persist for some length of time but to not be perpetual. More gen-
erally, if the benefits and costs of an action are uncertain and take
time to be realized and a decision maker has limited experience
with the setting, learning should not be expected to be fast.

Finally, it is worth highlighting that the nature of long-run
beliefs does not imply that a decision maker should never
undercommit. Commitment often has costs apart from monetary
ones or its loss of flexibility; in many contexts, such as addiction,
the choice to commit is observable, and so a decision maker who
recognizes his imperfect self-control may nevertheless choose not
to commit to hide it from others. Moreover, to the extent that
commitment requires self-control (Noor 2007), even a fully so-
phisticated decision maker may not commit. Allowing for these
considerations can explain why individuals might not commit to
the extent that our models (which omit these features) predict,
even when individuals learn about their temptations over time.

In spite of all the potential reasons for people not to com-
mit, many recent studies highlight the demand for commitment
that individuals exhibit.25 Such demand for commitment is con-
sistent with both sophistication and the failure of learning that
can emerge in tradeoffs between commitment and flexibility.

UNIVERSITY OF CALIFORNIA, SAN DIEGO

REFERENCES

Acemoglu, D., V. Chernozhukov, and M. Yildiz, “Fragility of Asymptotic Agree-
ment under Bayesian Learning” Mimeo, MIT, 2009.

Aghion, P., P. Bolton, C. Harris, and B. Jullien, “Optimal Learning by Experimen-
tation,” Review of Economic Studies, 58 (1991), 621–654.

Aliprantis, C., and K. Border, Infinite Dimensional Analysis: A Hitchhiker’s Guide,
(New York, NY: Springer Verlag, 2006).

Amador, M., I. Werning, and G.-M. Angeletos, “Commitment Vs. Flexibility,”
Econometrica, 74 (2006), 365–396.

25. For instance, see Angeletos, Laibson, Repetto, Tobacman, and Weinberg
(2001); Ariely and Wertenbroch (2002); Ashraf, Karlan, and Yin (2006); Giné,
Karlan, and Zinman (2009); Thaler and Benartzi (2004); Wertenbroch (1998);
DellaVigna (2009) offers a survey.

D
ow
nloaded from
https://academ
ic.oup.com
/qje/article-abstract/126/2/857/1870713 by A
rizona S
tate U
niversity user on 27 February 2020

892 QUARTERLY JOURNAL OF ECONOMICS

Ameriks, J., A. Caplin, and J. Leahy, “Wealth Accumulation and the Propensity to
Plan,” Quarterly Journal of Economics, 118 (2003), 1007–1047.

Angeletos, G. M., D. Laibson, A. Repetto, J. Tobacman, and S. Weinberg, “The Hy-
perbolic Consumption Model: Calibration, Simulation, and Empirical Evalu-
ation,” Journal of Economic Perspectives, 15 (2001), 47–68.

Ariely, D., and K. Wertenbroch, “Procrastination, Deadlines and Performance:
Self-Control by Precommitment,” Psychological Science, 13 (2002), 219–224.

Ashraf, N., D. Karlan, and W. Yin, “Tying Odysseus to the Mast: Evidence from a
Commitment Savings Product in the Philippines,” Quarterly Journal of Eco-
nomics, 121 (2006), 635–672.

Barro, R., “Ramsey Meets Laibson in The Neoclassical Growth Model*,” Quarterly
Journal of Economics, 114 (1999), 1125–1152.

Baumeister, R. F., T. F. Heatherton, and D. M. Tice, Losing Control: How and Why
People Fail at Self-regulation (San Diego, CA: Academic Press, 1994).

Benabou, R., and M. Pycia, “Dynamic Inconsistency and Self-Control:
A Planner-Doer Interpretation,” Economics Letters, 77 (2002), 419–424.

Benabou, R., and J. Tirole, “Self-Confidence and Personal Motivation,” Quarterly
Journal of Economics, 117 (2002), 871–915.

———, “Willpower and Personal Rules,” Journal of Political Economy, 112 (2004),
848–886.

Benhabib, J., and A. Bisin, “Modeling internal commitment mechanisms and self-
control: A neuroeconomics approach toconsumption–saving decisions,” Games
and Economic Behavior, 52 (2005), 460–492.

Bernheim, B. D., and A. Rangel, “Addiction and Cue-Triggered Decision
Processes,” American Economic Review, 94 (2004), 1558–1590.

Brocas, I., and J. D. Carrillo, “The Brain as a Hierarchical Organization,” American
Economic Review, 98 (2008), 1312–1346.

Brunnermeier, M. K., and J. A. Parker, “Optimal Expectations,” The American
Economic Review, 95 (2005), 1092–1118.

Carrillo, J. D., and T. Mariotti, “Strategic Ignorance as a Self-Disciplining Device,”
The Review of Economic Studies, 67 (2000), 529–544.

Chatterjee, K., and R. Krishna, “A “Dual Self Representation for Stochastic Temp-
tation,” American Economic Journal: Microeconomics, 1 (2009),
148–167.

Compte, O., and A. Postlewaite, “Confidence-Enhanced Performance,” American
Economic Review, 94 (2004), 1536–1557.

Dekel, E., D. Fudenberg, and D. K. Levine, “PayoffInformation and Self-Confirming
Equilibrium” Journal of Economic Theory, 89 (1999), 165–185.

———, “Learning to Play Bayesian Games,” Games and Economic Behavior, 46
(2004), 282–303.

Dekel, E., and B. L. Lipman, “Costly Self Control and Random Self
Indulgence” Mimeo, Boston University, 2010.

DellaVigna, S., “Psychology and Economics: Evidence from the Field,” Journal of
Economic Literature, 47 (2009), 315–372.

DellaVigna, S., and U. Malmendier, “Paying Not to Go to the Gym,” The American
Economic Review, 96 (2006), 694–719.

Easley, D., and N. M. Kiefer, “Controlling a Stochastic Process with Unknown
Parameters,” Econometrica, 56 (1988), 1045–1064.

Fudenberg, D., “Advancing Beyond Advances in Behavioral Economics,” Journal
of Economic Literature, 44 (2006), 694–711.

Fudenberg, D., and D. K. Levine, “Self-Confirming Equilibrium,” Econometrica, 61
(1993a), 523–545.

———, “Steady State Learning and Nash Equilibrium,” Econometrica, 61 (1993b),
547–573.

———, “A Dual Self Model of Impulse Control,” American Economic Review, 96
(2006), 1449–1476.

———, “Learning and Equilibrium,” Annual Review of Economics, 1 (2009),
385–420.

Giné, X., D. Karlan, and J. Zinman, “Put Your Money Where Your Butt Is: A
Commitment Contract for Smoking Cessation” Mimeo, Yale University, 2009.

Gul, F., and W. Pesendorfer, “Temptation and Self-Control,” Econometrica, 69
(2001), 1403–1435.

D
ow
nloaded from
https://academ
ic.oup.com
/qje/article-abstract/126/2/857/1870713 by A
rizona S
tate U
niversity user on 27 February 2020

LEARNING SELF-CONTROL 893

———, “The Revealed Preferences Theory of Changing Tastes,” Review of Eco-
nomic Studies, 72 (2005), 429–448.

Harris, C., and D. Laibson, “Hyberbolic Discounting and Consumption,” in Ad-
vances in Economics and Econometrics: Theory and

Applications

: Eighth
World Congress, eds. M. Dewatripont, L. P. Hansen, and S. J. Turnovsky,
(Econometric Society, 2003, 258–297).

Kahneman, D., P. P. Wakker, and R. Sarin, “Back to Bentham? Explorations of
Experienced Utility,” The Quarterly Journal of Economics, 112 (1997), 375–
405.

Kaper, J., E. Wagena, M. Willemsen, and C. Van Schayck, “Reimbursement for
smoking cessation treatment may double the abstinence rate: results of a ran-
domized trial.” Addiction, 100 (2005), 1012.

Kőszegi, B., “Ego utility, overconfidence, and task choice,” Journal of the
European Economic Association, 4 (2006), 673–707.

Kreps, D., “A Representation Theorem for “Preference for Flexibility”,”
Econometrica, 47 (1979), 565–578.

Krusell, P., and A. A. Smith, Jr., “Consumption-Savings Decisions with
Quasi-Geometric Discounting,” Econometrica, 71 (2003), 365–375.

Laibson, D., “Golden Eggs and Hyperbolic Discounting,” Quarterly Journal of Eco-
nomics, 112 (1997), 443–477.

Loewenstein, G., and T. O’Donoghue, “The Heat of the Moment: Modeling Inter-
actions Between Affect and Deliberation” Mimeo, Cornell University, 2007.

Noor, J., “Commitment and Self-Control,” Journal of Economic Theory, 135 (2007),
1–34.

O’Donoghue, T., and M. Rabin, “Choice and Procrastination,” Quarterly Journal of
Economics, 116 (2001), 121–160.

Pollak, R. A., “Consistent planning,” The Review of Economic Studies, 35 (1968),
201–208.

Rubinstein, A., “

Discussion

of ‘Behavioral Economics’,” in Advances in
Economics and Econometrics, Theory and Applications: Ninth World Congress
of the Econometric Society, eds. R. Blundell, W. Newey, and T. Persson, (Econo-
metric Society, 2005, 246–254).

Shui, H., and L. M. Ausubel, “Time Inconsistency in the Credit Market” Mimeo,
University of Maryland–College Park, 2005.

Strotz, R. H., “Myopia and Inconsistency in Dynamic Utility Maximization,” Re-
view of Economic studies, 23 (1955), 165–180

Thaler, R. H., and S. Benartzi, “Save More Tomorrow: Using Behavioral Eco-
nomics toIncrease Employee Saving,” Journal of Political Economy, 112 (2004),
164–182.

Thaler, R. H., and H. M. Shefrin, “An Economic Theory of Self-Control,” Journal
of Political Economy, 89 (1981), 392–406.

Wertenbroch, K., “Consumption Self-Control by Rationing Purchase Quantities of
Virtue and Vice,” Marketing Science, 17 (1998), 317–337.

D
ow
nloaded from
https://academ
ic.oup.com
/qje/article-abstract/126/2/857/1870713 by A
rizona S
tate U
niversity user on 27 February 2020

Introduction
A Simple Example
When Is Learning Adequate?

Adequate versus Partial Learning
Full Commitment Distinguishability
When Does FCD Hold?
Main Results
Remarks
The Role of Patience
Identifying Awareness.
Relationship to Self-Confirming Equilibrium.

Applications
Consumption-Savings
Costly Self-Control
Discussion

What Will You Get?

We provide professional writing services to help you score straight A’s by submitting custom written assignments that mirror your guidelines.

Premium Quality

Get result-oriented writing and never worry about grades anymore. We follow the highest quality standards to make sure that you get perfect assignments.

Experienced Writers

Our writers have experience in dealing with papers of every educational level. You can surely rely on the expertise of our qualified professionals.

On-Time Delivery

Your deadline is our threshold for success and we take it very seriously. We make sure you receive your papers before your predefined time.

24/7 Customer Support

Someone from our customer support team is always here to respond to your questions. So, hit us up if you have got any ambiguity or concern.

Complete Confidentiality

Sit back and relax while we help you out with writing your papers. We have an ultimate policy for keeping your personal and order-related details a secret.

Authentic Sources

We assure you that your document will be thoroughly checked for plagiarism and grammatical errors as we use highly authentic and licit sources.

Moneyback Guarantee

Still reluctant about placing an order? Our 100% Moneyback Guarantee backs you up on rare occasions where you aren’t satisfied with the writing.

Order Tracking

You don’t have to wait for an update for hours; you can track the progress of your order any time you want. We share the status after each step.

Order Now Talk to Us

Areas of Expertise

Although you can leverage our expertise for any writing task, we have a knack for creating flawless papers for the following document types.

Essay

Thesis

Presentation

Dissertation

Term Paper

Research Paper

Book Review

Assignment

Report

Case Study

Letter

Article

Coursework

Speech

Q & A

Critical Thinking

Areas of Expertise

Although you can leverage our expertise for any writing task, we have a knack for creating flawless papers for the following document types.

Essay

Thesis

Presentation

Dissertation

Term Paper

Research Paper

Book Review

Assignment

Report

Case Study

Letter

Article

Coursework

Speech

Q & A

Critical Thinking

Trusted Partner of 9650+ Students for Writing

From brainstorming your paper's outline to perfecting its grammar, we perform every step carefully to make your paper worthy of A grade.

Preferred Writer

Hire your preferred writer anytime. Simply specify if you want your preferred expert to write your paper and we’ll make that happen.

Grammar Check Report

Get an elaborate and authentic grammar check report with your work to have the grammar goodness sealed in your document.

One Page Summary

You can purchase this feature if you want our writers to sum up your paper in the form of a concise and well-articulated summary.

Plagiarism Report

You don’t have to worry about plagiarism anymore. Get a plagiarism report to certify the uniqueness of your work.

Free Features $66FREE

Most Qualified Writer $10FREE
Plagiarism Scan Report $10FREE
Unlimited Revisions $08FREE
Paper Formatting $05FREE
Cover Page $05FREE
Referencing & Bibliography $10FREE
Dedicated User Area $08FREE
24/7 Order Tracking $05FREE
Periodic Email Alerts $05FREE

Our Services

Join us for the best experience while seeking writing assistance in your college life. A good grade is all you need to boost up your academic excellence and we are all about it.

On-time Delivery
24/7 Order Tracking
Access to Authentic Sources

Academic Writing

We create perfect papers according to the guidelines.

Professional Editing

We seamlessly edit out errors from your papers.

Thorough Proofreading

We thoroughly read your final draft to identify errors.

Thorough Proofreading

We thoroughly read your final draft to identify errors.

Delegate Your Challenging Writing Tasks to Experienced Professionals

Work with ultimate peace of mind because we ensure that your academic work is our responsibility and your grades are a top concern for us!

Check Out Our Sample Work

Dedication. Quality. Commitment. Punctuality

It May Not Be Much, but It’s Honest Work!

Here is what we have achieved so far. These numbers are evidence that we go the extra mile to make your college journey successful.

0+

Happy Clients

0+

Words Written This Week

0+

Ongoing Orders

0%

Customer Satisfaction Rate

Process as Fine as Brewed Coffee

We have the most intuitive and minimalistic process so that you can easily place an order. Just follow a few steps to unlock success.

Call Us +1 (877) 657-8180 Discuss Order Details Now

See How We Helped 9000+ Students Achieve Success

We Analyze Your Problem and Offer Customized Writing

We understand your guidelines first before delivering any writing service. You can discuss your writing needs and we will have them evaluated by our dedicated team.

Clear elicitation of your requirements.
Customized writing as per your needs.

We Mirror Your Guidelines to Deliver Quality Services

We write your papers in a standardized way. We complete your work in such a way that it turns out to be a perfect description of your guidelines.

Proactive analysis of your writing.
Active communication to understand requirements.

We Handle Your Writing Tasks to Ensure Excellent Grades

We promise you excellent grades and academic excellence that you always longed for. Our writers stay in touch with you via email.

Thorough research and analysis for every order.
Deliverance of reliable writing service to improve your grades.

Place an Order Start Chat Now

thoughts

What Will You Get?

Premium Quality

Experienced Writers

On-Time Delivery

24/7 Customer Support

Complete Confidentiality

Authentic Sources

Moneyback Guarantee

Order Tracking

Areas of Expertise

Essay

Thesis

Presentation

Dissertation

Term Paper

Research Paper

Book Review

Assignment

Report

Case Study

Letter

Article

Coursework

Speech

Q & A

Critical Thinking

Areas of Expertise

Essay

Thesis

Presentation

Dissertation

Term Paper

Research Paper

Book Review

Assignment

Report

Case Study

Letter

Article

Coursework

Speech

Q & A

Critical Thinking

Trusted Partner of 9650+ Students for Writing

Preferred Writer

Grammar Check Report

One Page Summary

Plagiarism Report

Free Features $66FREE

Our Services

Academic Writing

Professional Editing

Thorough Proofreading

Thorough Proofreading

Delegate Your Challenging Writing Tasks to Experienced Professionals

Check Out Our Sample Work

It May Not Be Much, but It’s Honest Work!

0+

Happy Clients

0+

Words Written This Week

0+

Ongoing Orders

0%

Customer Satisfaction Rate

Process as Fine as Brewed Coffee

Share Your Requirements

Place Order & Deposit Funds

Release Payment to Your Writer

See How We Helped 9000+ Students Achieve Success

We Analyze Your Problem and Offer Customized Writing

We Mirror Your Guidelines to Deliver Quality Services

We Handle Your Writing Tasks to Ensure Excellent Grades