Professional Documents
Culture Documents
Measuring Self Control Problems
Measuring Self Control Problems
American Economic Association is collaborating with JSTOR to digitize, preserve and extend access to The
American Economic Review.
http://www.jstor.org
MeasuringSelf-ControlProblems
ByJOHNAMERIKS,ANDREWCAPLIN,JOHNLEAHY,ANDTOMTYLER*
While models of self-control problems have
proliferatedin recentyears, therehave been few
corresponding advances in measurement. We
develop a survey instrumentto measure selfcontrol problems and apply it to a sample of
highly educated adults. Measured self-control
relates in the anticipatedmannerto wealth accumulation and standardpersonalitymeasures.
Yet while self-control problems are typically
seen as resulting in overconsumptionand low
wealth, we identify a significant group who
underconsume and thereby accumulate high
levels of wealth. In addition, self-control problems are found to be smaller in scale for older
than for younger respondents.Those who put
money aside in retirement accounts may be
delaying access to a point at which self-control
problems are no longer important.Continued
advances in measurementare essential to guide
development of self-control models in empirically relevant directions.
I. The Self-ControlMeasureand Its Properties
Most theories of self-controlsharea common
structure.There is an ideal action that the agent
* Ameriks: Vanguard,P.O. Box 2600, MSV37, Valley
Forge, PA 19482 (e-mail: john_ameriks@vanguard.com);
Caplin: Departmentof Economics, New York University,
19 West 4th St., New York, NY 10012 (e-mail:
andrew.caplin@nyu.edu);Leahy: Departmentof Economics, NYU, 19 West 4th St., New York, NY 10012 (e-mail:
john.leahy@nyu.edu); Tyler: Department of Psychology,
NYU, 6 WashingtonPlace, New York, NY 10003 (e-mail:
tom.tyler@nyu.edu).Caplin thanks the Center for ExperimentalSocial Science at New York Universityand the C.V.
StarrCenterat NYU for financialsupport.We would like to
thank Douglas Bernheim, Xiaohong Chen, Douglas Fore,
Faruk Gul, Guido Imbens, David Laibson, Brent Roberts,
Ariel Rubinstein,Julio Rotemberg,Yaacov Trope,andthree
anonymousreferees for their help. We gratefully acknowledge financial support for our survey provided by the
TIAA-CREF Institute and to Caplin and Leahy under
National Science Foundation grant SES-0351115. The
research described in this article was completed while
Ameriks was Senior Research Fellow at the TIAA-CREF
Institute. The opinions expressed in this article are solely
those of the authors,and do not necessarilyreflectthe views
of their currentor past employers.
966
VOL.97 NO. 3
967
THEAMERICANECONOMICREVIEW
968
Entire sample
Regression sample
EI
gap
Number
Percent
Number
Percent
5
4
3
2
1
0
-1
-2
-3
-4
-5
-6
-7
-8
-9
All
9
2
8
39
113
1,059
141
94
25
9
14
2
1
1
3
1,520
0.6
0.1
0.5
2.6
7.4
69.7
9.3
6.2
1.6
0.6
0.9
0.1
0.1
0.1
0.2
100.0
2
0
1
9
35
246
41
28
7
1
2
1
1
0
0
374
0.5
0.0
0.3
2.4
9.4
65.8
11.0
7.5
1.9
0.3
0.5
0.3
0.3
0.0
0.0
100.0
JUNE 2007
w= a +
SC+
2X +
VOL.97 NO. 3
TABLE 2-NET
Variable
Expected-idealgap
Ideal level
Log 1999 income
Zero 1999 income
Past income
Zero past income
Futureincome
Zero future income
Age
Age2
Empl. status
Working
Partiallyretired
Retired
Occupation
Faculty
Mgmt./sen. admin.
Tech./professional
Other
Education
College or below
M.A./professional
Ph.D.
R. has defined ben. plan
S. has defined ben. plan
Maritalstatus
Curr.married
Prev. married
Never married
Male respondent
Num. kids
Constant
N
Coeff.
Std. err.
-0.146***
-0.019
0.198
1.555**
0.469***
1.304*
-0.047
-0.190
0.216***
-0.001***
0.048
0.033
0.179
0.776
0.161
0.707
0.109
0.467
0.046
0.000
Omitted
0.068
0.267
0.224
0.264
Omitted
-0.185
0.003
-0.134
0.155
0.147
0.174
-0.236
0.172
Omitted
0.051
-0.222*
-0.087
0.128
0.127
0.157
Omitted
-0.601***
-0.345**
-0.061
0.013
-3.356***
374
0.169
0.158
0.113
0.063
1.127
969
970
THEAMERICANECONOMICREVIEW
JUNE 2007
TABLE3-REGRESSIONS
FORWEALTH
CATEGORIES
Non-ret. fin. assets
Variable
Actual-ideal gap
Ideal level
Log 1999 income
Zero 1999 income
Past income
Zero past income
Futureincome
Zero future income
Age
Age2
Empl. status
Working
Partiallyretired
Retired
Occupation
Faculty
Mgmt./sen. admin.
Tech./professional
Other
Education
College or below
M.A./professional
Ph.D.
R. has DB plan
S. has DB plan
Maritalstatus
Curr.married
Prev. married
Never married
Male respondent
Num. kids
Constant
N
R2
Retirementassets
Coeff.
S.E.
Coeff.
S.E.
-0.285***
-0.006
0.059
1.336
0.856***
3.297*
-0.033
0.366
-0.112
0.001
0.079
0.057
0.306
1.601
0.300
1.743
0.181
0.797
0.100
0.001
-0.081
0.018
0.091
1.492
0.540**
1.272
-0.054
-0.119
0.281***
-0.002***
0.055
0.040
0.216
1.130
0.211
1.230
0.128
0.562
0.071
0.001
0.383
0.510
0.430
-0.038
0.259
0.250
0.300
-0.077
0.076
-0.302
0.289
-0.264
Omitted
-0.219
-0.299
Omitted
Omitted
Omitted
0.152
-0.003
-0.021
-0.794***
0.219
0.222
0.269
0.091
-0.270*
-0.024
0.291
0.275
0.190
0.106
2.255
-0.544***
-0.347*
0.200
0.000
-5.595***
362
0.179
0.203
0.154
0.156
0.190
Omitted
Omitted
-0.207
-0.500*
-0.144
-0.079
2.296
362
0.078
0.182
0.176
0.211
Omitted
Omitted
-0.353
-0.022
0.134
0.270
0.359
0.205
0.194
0.134
0.074
1.591
Notes: Dependentvariables are naturallogarithmsof the quantitieslisted at head of each set of columns. Asterisks indicate
the level of statisticalconfidence for rejectionof the hypothesis that the relevantcoefficient is (independently)equal to zero:
*** indicates rejection at better than a 1 percent level of confidence, ** indicates rejection at better than a 5 percent level,
and * indicates rejection at better than a 10 percent level.
Source: Authors' calculationsbased on 2000, 2001, and 2003 survey data.
conscientiousness. The data reveal a strong relationship between the conscientiousness questions and the absolute value of the El gap, and
no relationshipwith the level of the El gap. For
those who are conscientious, there is a lower
divergence between expected and ideal consumption,regardlessof sign.
A particularlyinterestingfindingin Table 4 is
the profound reduction in the scale of selfcontrol problems as individuals age, which
shows up only when one uses the absolutevalue
of the self-control measure. Older individuals
experience fewer self-control problems, either
of overconsumptionor underconsumption,than
do their younger counterparts.This finding is
certainlyconsistent with the common view that
temptation falls with age, and has important
connections with actual consumption behavior
over the life cycle. Models that allow for such a
time-changingself-controlparameterretirement
may be necessary to explain the absence of a
spike in consumption spending at the point
when retirementassets become fully liquid.
These results hold if we condition separately
on a nonpositiveor a nonnegativeEl gap. Each is
separatelyrelatedto conscientiousnessand age.
VOL.97 NO. 3
TABLE 4--CONSCIENTIOUSNESS
AND SELF-CONTROL
Variable
El gap
Absolute El gap
Age
0.003
(0.002)
0.048
(0.063)
0.016
(0.029)
0.057
(0.029)
-0.306
(0.169)
1489
0.005
-0.008***
(0.002)
-0.129**
(0.056)
0.070***
(0.026)
1.101***
(0.026)
0.682***
(0.150)
1489
0.039
Male
Not dependable
Not organized
Constant
N
R2
B. Temptationand Self-Control
We define the temptation-ideal(TI) gap as
the differencebetween the answersto questions
(c) and (a), the most tempting choice and the
ideal choice. The correlationbetween the TI gap
and the El gap is about 0.4. The TI gap is also
correlatedwith wealth, but loses all predictive
power if the El gap is includedin the regression.
The TI gap appearsto work throughthe El gap.
Most self-controltheories suggest that the El
gap shouldlie somewherebetweenthe TI gap and
zero. This is true for 1,173 of the 1,448 respondents for whom we can constructboth measures.
Among the others,235 reporta TI gap of zero,yet
a nonzeroEl gap. Interestingly,the vast majority
of the violations(211) occur when the El gap is
do not
negative.It is possiblethatunderconsumers
fit into the ideal-temptation
framework.It may be
thattemptationlacksmeaningfor thisgroup(what
does it mean to be temptedto consume less0) or
they may have troubleconsumingat all, possibly
becausethey arebusy at workor at home.It is also
possible that the El gap is capturingsomething
other than self-controlin these cases. When we
restrictthe sample in the wealth regressionsto
those that fit the TI framework,that is, those for
whom the El gaplies betweenthe TI gap andzero,
the effect of the El gap tends to be stronger.The
coefficienton the El gap rises in absolutevalue to
-0.19 with a t-statisticof 2.34 on 295 individuals,
56 of whom have nonzeroEl gaps. For nonretirementfinancialassets,the coefficientis -0.46 and
971
the t-statisticis 3.51, whereasfor retirementfinancial assets it is -0.12 with a t-statisticof 1.23.
Both these regressionshave 329 observations.
C. Commitmentand Self-Control
An implicationof most theoriesof self-control
is that agents would like to precommitto their
desiredaction.Followingour main questions,we
asked responderswhetherthey would use an option to restrictsome of the certificatesfor use only
in the firstyear and/orthe second year, and if so
how many certificatesthey would like to restrict.
We dropped29 observationsdue to missing data,
19 observationsthat restrictedmore than the allotted 10 meals, and 103 whose restrictionsmade
it impossiblefor themto consumetheirideallevel.
This left 1,369 responses.For this group,we define the signed distance between the expected
choice and the constraintset to be the revealed
preference(RP) gap, a possible alternativemeasure of self-control.
In manyways, the RP gapreinforcesourearlier
findings.The correlationwith the El gap is 0.5.
Like the El gap, the absolutevalue of the RP gap
is positivelyrelatedto our measuresof conscientiousnessand negativelyrelatedto age, although
the correlationwith age is significantonly at the 6
percentlevel. Like the El gap, the RP gap has a
largeeffect on wealth,althoughagainthe effect is
less statisticallysignificant.The p-value is 7 percent. As with the TI gap, people with overconsumptionproblemsaccordingto the El gap are
morelikely to have a nonzeroRP gap thanpeople
with underconsumption problems, indicating
again that there might be something different
about underconsumption.
In other ways, however, the RP gap presents
a different and more complex picture. On the
one hand, self-control problems appearless severe from the perspectiveof the RP gap. As was
mentionedabove, the correlationwith wealth is
less significant. Surprisingly,there is no correlation between the RP gap and liquid assets.
Partlythis is because few are willing to impose
binding constraintson themselves. Fewer than
10 percent of agents impose strictly binding
constraints,while 30 percent have self-control
problems accordingto the EI gap. On the other
hand, while binding constraints are rare, nonbinding constraintsare common. Thirtypercent
of the respondentswith a zero RP gap choose to
restrict some certificates to one period or the
THEAMERICANECONOMICREVIEW
972
JUNE 2007