Digitized by the Internet Archive
in
2011 with funding from
Boston Library Consortium
Member
Libraries
http://www.archive.org/details/roleofinformatioOOdufl
3B31
.M415
l^W£\
13
Massachusetts Institute of Technology
Department of Economics
Working Paper Series
THE ROLE OF INFORMATION AND SOCIAL
INTERACTIONS IN RETIREMENT PLAN DECISIONS:
EVIDENCE FROM A RANDOMIZED EXPERIMENT
Esther Duflo
Emmanel Saez
Working Paper 02-23
March 21 ,2002
Room
E52-251
50 Memorial Drive
Cambridge,
MA 021 42
This paper can be downloaded without charge from the
Social Science Research Network Paper Collection at
http://papers.ssm.com/paper.taf7abstract
id=XXXXXX
MASSACHui^'l
.
,;
v;
OFT'
JUN 1 3
2002
"libraries
<tf
Massachusetts Institute of Technology
Department of Economics
Working Paper Series
THE ROLE OF INFORMATION AND SOCIAL
INTERACTIONS IN RETIREMENT PLAN DECISIONS:
EVIDENCE FROM A RANDOMIZED EXPERIMENT
Esther Duflo
Emmanel Saez
Working Paper 02-23
March 21, 2002
Room
E52-251
50 Memorial Drive
Cambridge,
MA 021 42
This paper can be downloaded without charge from the
Social Science Research Network Paper Collection at
h1tp://papers. ssrn. com/paper. taf?abstract id=XXXXXX
The Role
of Information
and Social Interactions
in
Retirement
Plan Decisions: Evidence from a Randomized Experiment
MIT and NBER
Esther Duflo,
Emmanuel
Saez, Harvard University
March
21,
and
NBER*
2002
Abstract
This paper analyzes a randomized experiment to shed light on the role of information
and
social interactions in employees' decisions to enroll in
retirement plan within a large university.
a Tax Deferred Account (TDA)
The experiment encouraged a random sample
of employees in a subset of departments to attend a benefits information fair organized
the university, by promising a monetary reward for attendance.
The experiment more than
tripled the attendance rate of these treated individuals (relative to controls),
that of untreated individuals within departments where
enrollment 5 and
1 1
months
after the fair
was
on
TDA
enrollment
receive the
treatment
results.
is
nobody was
almost as large for individuals
encouragement as
for those
network
effects, social
who did. We
effects,
some individuals were
significantly higher in
individuals were treated than in departments where
in
by
and doubled
treated.
TDA
departments where some
treated. However, the effect
treated departments
who
did not
provide three interpretations, differential
and motivational reward
effects, to
account for these
(J EL D83, 122)
'We thank Daron Acemoglu,
Orley Ashenfelter, Josh Angrist, David Autor, Abhijit Banerjee, David Card,
Jonathan Gruber, Guido Imbens, Larry Katz, Jeffrey Kling, Botond Koszegi, Michael Kremer, Alan Krueger,
David Laibson, Sendhil Mullainathan, and seminar participants
very helpful comments and discussions.
Foundation (SES-0078535).
for their help
and support
or its benefits office.
We
in
We
at Berkeley,
MIT, Princeton, and
UCLA
for
gratefully acknowledge financial support from the National Science
are especially grateful to
all
the
members
of the Benefits Office of the University
organizing the experiment. This paper does not reflect the views of the university
1
Introduction
Low
levels of savings in the
of
what determines savings
United States have generated substantial interest
A
decisions.
in
the question
vast literature has studied the impact of
Tax Deferred
Accounts (hereafter, TDA), such as Individual Retirement Accounts (IRAs) and 401 (k)s, on
re-
tirement savings decisions, 1 and, concurrently, the impact of these plans' features on enrollment
and contribution
ployer's match), a
rates.
In addition to the tax savings and economic incentives (such as em-
number
of recent studies emphasize the role of non-economic factors, such as
social interactions, financial education, inertia,
how
individual participation in a
ticipation in one's department.
TDA
(2001a, 2001b)
plan within a large university
They obtain
influence on the decision to enroll in
show that default
TDA
When
is
affected
by average par-
suggestive evidence that peer effects have a strong
plans.
rules have
contribution, and asset allocation.
and commitment. Duflo and Saez (2000) study
Madrian and Shea (2001) and Choi
et al.
an enormous impact on employees' participation,
employees are enrolled by default in a
TDA.
very few
opt out and most employees do not change the default contribution rate or the default allocation
of assets.
Thaler and Bernatzi (2001) show that inducing employees to commit to contribute
a large fraction of their pay raises to the
TDA
dramatic positive impact on savings
Bernheim and Garett (1996) and Bayer, Bernheim,
rates.
(the "Save
More Tomorrow" program) has a
and Scholz (1996), Bernheim, Garett and Maki (1997), among others, study the
education.
They
role of financial
present evidence that financial education tends to be remedial 2 but that
increases participation in savings plans, suggesting that employees
may
it
not be able to gather
the necessary information on their own. This evidence, though suggestive, does not provide fully
convincing proof that information and financial education can have a strong impact on
participation decisions, because employers' decision to provide this information
is
TDA
endogenous.
Recently, Madrian and Shea (2002) studied the effects of benefits seminars within a large firm
and showed interesting evidence of
who
attend seminars are
self-selection in the decision to attend benefits:
much more
likely to
be recent enrollees
in
the
TDA
plan.
employees
They found
'See Poterba, Venti and Wise (1996) and Engen, Gale and Scholz (1996) for a controversial debate summarizing
the literature.
2
Employers resort to
it
when they
compensated employees are too low.
fail
discrimination testing because the contribution rates of the not highly
modest
on
positive effect of information seminars
Financial education
is
participation after a few months.
generally recognized as an potentially important avenue to improve
71%
the quality of financial decision making.
financial information sessions.
A
of the fortune 500 companies systematically hold
10% conducts them
further
(Summers, 2000) outlined a proposal to improve
financial services of lower
TDA
occasionally. 3
financial literacy
The U.S. Treasury
and increase the access to
income American households. In particular, the report stressed the
importance of information on savings instruments and the role of social interaction
decision to save.
shed
light
The
goal of this paper
on both the
is
and
social interactions
TDA plan of a large university.
enroll in the employer sponsored
random experiment
to analyze the evidence from a
role of information
effects in the
to
on the employees' decision to
Our
analysis improves
upon the
studies discussed above because the source of identification comes directly from the randomized
experiment. This allows us to overcome some of the very
presence of peer
Each
effects,
to provide information on benefits.
TDA, which
Obviously, comparing the
attend the
TDA
fair
all its
employees to a benefits
In particular, a stated goal of the fair
the university administration feels
TDA
enrollment decisions of
enrollment, because the decision to attend the
fair
TDA
This data come from a telephone survey of
In spite of these difficulties, there
which essentially focuses on
is
all
is
fair in
order
to increase the
is
too low (around 35%).
attendees to those
who
did not
of a causal effect of fair attendance
fair is
endogenous. 5
we have implemented the following experiment.
of employees not yet enrolled in the
3
and invites
would not provide convincing evidence
selection problem,
the
in
described notably in Manski (1993, 1995). 4
year, the university organizes
enrollment rate in
problems
difficult identification
To circumvent
We selected
and sent them an invitation
500 companies we conducted
in
letter
the
a
on
this
random sample
promising a $20
summer
2001.
a growing empirical literature on peer effects using observational analysis
social behavior,
and the adoption of new technologies. For example, Case and Katz
(1991) and Evans, Oates and Schwab (1992) on teenagers' behavior, Bertrand, Mullainathan and Luttner (1998)
on welfare participation, Munshi (2000a) on contraception, and Besley and Case (1994), Foster and Rosenzweig
(1995) and Munshi (2000b) on technology adoption in developing countries. Sorensen (2001) analyzes peer effects
within departments of a university
in
the choice of Employer sponsored Health Plans using a methodology related
to Duflo and Saez (2000).
For example, individuals who had already decided to
to contribute,
may
be more
likely to
attend the
decision to attend information sessions.
fair (see
enroll,
but are not sure exactly
Madrian and Shea (2002)
for
how much they wanted
evidence of selection
in
the
reward
for
attending the
fair.
This type of experiment
is
a classical encouragement design, often
who
used in medical science, where treatments are offered to a random group of patients
then
decide whether or not to take the treatment. 6 Encouragement designs are rare in economics.
An example
on
test scores
who
the study by Powers and Swinton (1984)
is
by randomly mailing
analyze the effect of hours of study
test preparation materials to students to
encourage them to
study.
The second
objective of our study
designed our experiment such that
who were
"treated" individuals
is
we
to analyze peer effects within departments.
are able to estimate social interaction effects.
sent the invitation letter were selected only from a
subset of departments (the "treated" departments)
A
.
number
is
random
effects.
Kremer and
perhaps the most closely related to our study. They analyze an experiment
design to evaluate
own and
external effects of a medical treatment against intestinal
children in schools in Kenya,
in treated schools
Namely,
of recent studies have also used
experimental or quasi-experimental situations to study social interaction
Miguel (2001)
We therefore
who
and obtain evidence of
spillover effects.
worms
They show that
for
children
did not get the medicine were positively affected. However, in their case,
variation in treatment status within a school
was not randomized but occurred because some
children were not present on treatment day.
Katz
et al.
(2001) use
random assignment
to a
housing voucher program for households living in high poverty public housing projects in the
Boston area and find improvement of treated families
Sacerdote (2001) uses random assignment of
and finds peer
social groups.
first- year
effects strongly influence levels of
These
latter
two studies on
in safety, health,
and exposure to crime.
students in Dartmouth college dorms
academic
effort as well as decisions to join
social interactions differ
from ours mainly because
they study the effect of assigning individuals to different peer groups, whereas in our study,
peer groups (departments) are fixed, and
we analyze how
individual decisions are affected by an
exogenous change on the information set of some members of the peer group.
The
6
first
stage of our study analyzes the effect of the invitation letter on fair attendance.
For example, Permutt and Hebel (1989) study the effect of maternal smoking on birth weight using randomly
assigned free smoker's counseling to encourage mothers to quit smoking.
effect of flu
shots
(recommended but not required) to a random subset
Imbens
et al.
(2000) analyze of the
of patients on flu outcomes.
'Following our previous discussion, the voucher program can be seen as an encouragement design to leave
public housing projects.
Treated individuals are three times as likely to attend the
ingly, control individuals in treated
fair as control individuals.
departments are twice as
likely to
attend the
Interest-
control
fair as
individuals in non-treated departments, despite the fact that only original letter recipient could
claim the $20 reward.
tendance rate
This shows that the invitation letters not only increased the
for individuals
who
received
The
colleagues within departments.
them but had
also a spill-over social effect
direct effect of the letter
fair at-
on
their
on attendance (purged from the
peer effects) can also be estimated by comparing the attendance rates of treated and control
individuals within treated departments only.
The second
effects
stage of the study tries to estimate the causal effect of
on the decision to
enroll in the
TDA. We show
individuals in treated departments are significantly
the
TDA
than control individuals.
successful in increasing
TDA
enrollment between those
departments who did not.
of the financial
colleagues,
and
for the former.
fair
was
is
no significant difference
propose three different interpretations, not necessarily mutually
TDA
First, this
enrollment.
might be evidence of
differential
Employees who come to the
plausible to think that the treatment effect
is
fair
fair
treatment
only because
because of their
larger for the latter
group than
Second, there might be social network effects within departments. Fair attendees
affect their subjective
information they obtain at the
literature.
fair,
same
might also be explained by motivational reward
might
have started contributing to
reward are different from those who decide to come to the
it is
after the fair,
actually received our encouragement letter and those in the
However, there
might be able to spread information obtained from the
results
months
TDA
We
attendance on
likely to
11
in
exclusive, to account for these facts.
effects of fair
and
attendance and social
This shows that our experiment, and hence the
enrollment.
who
more
that, 5
fair
fair in their
effects.
departments. Third, our
Paying individuals to attend the
motivation and therefore the perceived value or quality of the
fair.
Such
Our experiment does not
effects
have been documented in the social psychology
allow us to separately identify these three effects but
allows us to conclude that the important decision about
how much
it
to save for retirement can
be affected by small shocks such as a very small financial reward and/or the influence of peers,
and thus does not seem to be the consequence of an elaborate decision process.
The remainder
of the paper
is
organized as follows. Section 2 describes the benefits
fair
and
the design of our experiment. Section 3 discusses the reduced form evidence. Section 4 develops
a simple model to guide the subsequent analysis of our results. Section 5 provides additional
evidence from a follow-up questionnaire and a general interpretation of our results.
Finally,
Section 6 offers a brief conclusion.
Context and Experiment Design
2
Benefits and Benefits Fair
2.1
The
university
we study has approximately 12,500 employees. About a quarter
members.
are faculty
Our study was
provides retirement benefits to
its
8
limited to non-faculty employees only.
of the employees
The
employees through a traditional pension plan and a comple-
mentary Tax Deferred Account (TDA) plan. Part of the traditional pension plan
Contribution (DC) plan whereby 3.5% of an employee's salary
tual fund account.
9
university
Employees can
is
also voluntary contribute to a
is
a Defined
put into an individual mu-
TDA
403(b) plan.
10
Every
employee can contribute to the 403(b) plan any percentage of their salary up to the IRS limit
($10,500 per year for each individual in 2001).
In both the
number
DC
and the
TDA
The
university does not
match contributions.
plans, employees can choose to invest their contributions in
any
of four different vendors.
Each
year, the university organizes a benefits fair
and learn about
university.
The
all
where
all
employees are invited to come
benefits (such as health benefits, retirement benefits, etc..) provided
fair is
held on two consecutive days in early
November
in
two
by the
different locations,
each one close to the two separate main university campuses. About one week before the
every employee receives a letter through university mail inviting her to attend the
letter also provides a brief description of
fair.
fair,
This
the event. At the same time, under separate cover, every
employee receives a packet describing in detail university benefits along with enrollment forms.
November
choices
8
is
by submitting the enrollment form.
lo
If
may change
TDA
choices are not influenced by
and vice-versa.
Non-faculty employees have an additional Defined Benefits plan in addition to the
403(b) plans are very similar to the better
firms.
her benefits
the employee does not send back the form, her
Duflo and Saez (2000) present suggestive evidence that staff employees
faculty choices
9
"open enrollment" month during which each employee
known 401 (k) plans but
their use
is
DC
plan.
restricted to not- for-profits
benefits choices are automatically carried over
free to enroll in
the
TDA
from the previous year. However, employees are
or change their contribution level or investment decision at any time
throughout the year.
In both locations, the fair
is
held in a large hotel reception room. There are a large
number
of
stands representing the university Benefits Office, and the various health and retirement benefits
The
service providers.
university Benefits Office offers information on
conversation with benefits office staff present at the
pamphlets
freely available at their stand.
The
benefits through direct
all
and through a number of information
fair,
benefits office also provides information on
how
the other stands at the fair are organized. These other stands are run by each of the specialized
service providers.
For example, each of the mutual fund vendors has a stand at which they
provide information about the
The
fair
TDA
plan and the specific services they offer within that plan.
computer program to
also offers individuals the chance to use a specially designed
come anytime during the three and
analyze their specific situation. Employees are free to
hours during which the
fair is held,
and
visit
a half
any number of stands they want.
Experiment Design
2.2
The
university organizes the annual fair in order to disseminate information about benefits and
help
its
(34%)
is
employees make better decisions. The university
too low compared to other universities, and that this
A simple comparison
to change their benefits choices
may be more
identify the causal effect of fair attendance
fair.
on
TDA
We
attend the
enrollment,
fair.
we
TDA
work (Duflo and
are very correlated
Saez, 2000),
among
we have shown
within departments, not
can thus measure peer
peers that did.
all
among
staff
to lack of information.
fair
and those who
Clearly, those
effect of the fair.
likely to
who
plan
Therefore, in order to to
set
up an "encouragement
for attending the
that the decisions to participate
individuals within departments, which suggests the
existence of social effects in enrollment decisions.
effects
may be due
by promising a random subset of employees a small amount of money
In previous
into the
the participation rate
between the benefits choices of those who attend the
do not does not provide an unbiased estimate of the
design",
feels
Therefore, in order to shed light on social
individuals within the treated departments received a letter.
effects using individuals
who
didn't receive a letter, but
who had
We used a cross-section of administrative data provided
as of
aged
We
August 2000.
less
criteria,
than 65 and
sample to
restricted the
eligible to participate in the
around 3,500 were enrolled
TDA
in the
TDA
employees
staff
TDA. 11 Of the
as of
all its
9,700 employees meeting these
From now
August 2000.
once they are enrolled, 12 we focus on the decision to start participating into the
sample of 6,200 non-enrolled individuals
In the
first step,
we randomly
is
selected
we
on,
refer
6,200 individuals were not
As very few employees stop contributing
by August 2000.
employees
non-faculty employees)
(i.e.
The remaining
to these individuals as the pre-enrolled individuals.
enrolled in the
by the university on
to the
TDA
TDA. Thus
the
our sample of primary interest.
two thirds
of the
departments of the university (220
out of a total of 330) as follows. In order to maximize the power of the experiment (in a context
in
which we know there are strong department
to their size
number
(i.e.
of employees)
effects),
we
and participation rate
separated department into deciles of participation rates
We
33 departments.
first
then ranked them by
size
matched departments according
TDA
in the
among the
staff.
before the
Each
fair.
We
decile contains
within each decile, and formed groups of three
departments by putting three consecutive departments on these
lists in
the same
triplet.
Within
each of these triplets, we randomly selected two departments to be part of the group of treated
From now
departments.
we
on,
refer to the treated
departments as the
T
departments and to
departments.
the control departments as the
In the second step, within each of the treated departments, any individual not enrolled as of
August 2000 was selected with probability one
individuals.
for
We
Treated department and
for
1
who were
where no treatments were
One week
before the
fair,
Part time employees working
we
less
it
is
denoted by
TO (T
being selected). In total, there are 4,168 individuals in
control group
selected;
composed of 2,039
being selected). The group formed by the employees in the
for not
The
is
and denote them by Tl (T
not selected contains 2,129 individuals and
Treated department and
the treated departments.
11
This treatment group
referred to this group as the Treated individuals
treated departments
for
half.
is
formed by employees
contains 2,043 individuals and
in the control
is
denoted by
departments
0.
sent a letter via university mail to the 2,039 employees in the
than 20 hours per week are not
eligible for the
TDA. Most
of these employees
are students of the university.
12
Only 80
of the 3,500 employees enrolled in the
examine. More than
five
times as
many employees
TDA
stopped contributing during the one year period we
started contributing to the
TDA
during the same period.
treatment group T\. The letter reminded them of the
receive a check for $20
letter is
reproduced
At the
their
fair,
we
from us
if
they were to come to the
up a stand
and
fair
register at our desk. This
appendix.
in facsimile in the
set
and informed them that they would
fair
for the
name. Unfortunately, the benefits
employees
office did
who
received our invitation letter to register
not authorize us to record the names of the
participants
who
stood at the
fair
coupons had
different colors according to the status of the participant (active or retired),
did not receive our letter. However,
number
We are thus confident
carefully counted.
In order to collect information on the
we organized a
had two
parts, with a
in the raffle
affiliation
raffle.
TDA
gave us half of the coupon.
fair.
and
TDA
asked
departments
versus
affiliation
and
TDA
in
An
who
status).
attendant
fair
all
the
in the
letter.
raffle
raffle
A
Out
affiliation of all
we assume
between
want to stay
raffle
was held every 30
employees
total of 1,617 active
of the remaining 1,044 employees,
is
their
department
that fair attendants
T and
who
(and hence provide their department
enough to wait
who
for the
departments as those who did
what
register. Therefore, in
1,
044/766.
In order to assess the effects of the experiment and the fair on
We
present in Section 5 evidence supporting our non-selection hypothesis.
modifying this assumption would
next
raffle.
refused
had been
Therefore,
did not register their department affiliation are distributed
the attendance recorded in each department by a factor of
1
affiliation
whether there was selection by
do not believe this was the case: most of those
at the fair long
fair
to participate
to play the raffle did so because they visited our stand just after the previous raffle
played, and did not
the
participants their department
and registered
raffle
of participants.
who wanted
TDA. The
important issue that arises
decided to play the
We
which
distributed at the entrance of the
gift certificate.
573 of them had received our
enrollment status.
The
Everybody had to
fair.
and the department
Each
twice.
about three quarters) came to play the
(i.e.,
T
status
We
entered the hall.
we accurately recorded the number
and whether they were currently enrolled
attended the
attended the
The coupons that were
number written
who
number: a student
and the few people who refused the coupon were
fair,
that
minutes, and the prize was a $50 Macy's
766
who
of active employees
pass through the narrow entrance enter the
fair
their total
entrance and distributed a coupon to each person
allowed us to count the
participants,
we recorded
fair
affect our result.
follows,
we
scale
up
13
TDA
participation, the
However, we
will discuss
how
The
university provided us three waves of data.
before the
wave from October 2001
2001.
we
wave was obtained
The second wave was from March 2001
fair.
Finally,
first
(11
months
months
(4.5
after the fair)
just
and the third
sent a short questionnaire (reproduced in the appendix) to 917 employees in April
The questionnaire was designed
to assess the intentions and evaluate the knowledge of
An
additional goal of sending out the questionnaire was
to remind those that were not yet enrolled of their
a cue to think about enrolling in the
they were enrolled in the
TDA, why
TDA.
TDA
to send back the questionnaire,
In the questionnaire,
we promised a $10 Macy's
we
we asked employees whether
they were not enrolled, whether they saved for retirement
would send back the questionnaire within 6 weeks.
questionnaire as follows. First,
and (potentially) provide them
status,
through other means, and whether they had attended the
restricted the
We
fair.
In order to induce employees
gift certificate to
any employee who
selected 917 employees to receive the
sample to those who were not enrolled
in the
by March 2001. Second, one third of employees (301) were selected among the 573
participants
selected
fair.
September 2000,
after the fair).
employees about retirement benefits.
TDA
in
who
among
The
did receive the invitation letter.
the 1,499 employees
last third (305)
invitation letter). 14
of questionnaires
We
who
The second
fair
third (311) of employees were
received the invitation letter but did not
were selected among our control group (those
who
come
to the
did not receive the
did not intentionally leave out any departments, but since the number
was not very
large, there are a
number
of
departments where we did not send
any questionnaire. 15
Results:
3
Summary
statistics
and Reduced form
In the presence of social interactions, employees
who work
themselves.
They may be more
likely to
come
14
Out
15
These departments tend to be smaller, but once we control
department belongs, the difference
16
This
is
in size
something we observed at the
is
small.
fair.
TO
if
they did not receive the
letter
to the fair themselves, because they are reminded
by others of the event, or because employees come to the
of these 305 individuals. 160 are from the
departments where some people
in
received the letter can be affected by the experiment even
differences
fair in
groups. 16
They may
control group and 145 are from the
for
the
dummy
also be
more
control group.
indicating in which group the
likely to enroll in
the
TDA
even
they do not come to the
if
fair
by the action of those who went to the
are directly influenced
share the information they gathered at the
themselves, either because they
or because these individuals
fair,
Thus, employees are potentially subjected to
fair.
two kinds of treatments: they can receive the invitation
can be in a group where some employees received the
letter
letter
themselves (group Tl), and they
(departments T). Those
who
receive
the letter are, obviously, subject to both treatments.
The summary
statistics are displayed in
we present the
to (3),
Table
statistics for individuals
1,
broken down into 4 groups. In columns
who belong
has the statistics for the entire group (group T), column
(1)
group of treated individuals (group Tl), and column
individuals in treated departments (group TO).
individuals
who belong
first
individuals not enrolled in the
differences in the
which includes a
level.
1
'
same
the
fair),
(1), (2), (3),
and
(2)
column
(4),
we
It is
present the statistics for
important to note that
all
and the second row of Panel B) focus only on
The
fair.
In Table
differences are estimated
and corrects standard
TO and group
for the
has the statistics for the untreated
on September 2000 before the
(4) present
has the statistics
Column
group Tl and group
presents background characteristics.
In the
first
we
present
by a regression,
errors for clustering at the
the differences between group
0,
2,
T
department
and group
0,
respectively.
wave (on September 2000 before
a very small proportion of employees started contributing to the
TDA
(the
first
wave
from September 2000, but we used data from August 2000 to construct the randomization),
but there
in
A
TDA
A
variables across groups.
group Tl and group TO, group
Panel
row of Panel
triplet fixed-effect,
Columns
In
(3)
to the untreated departments (group 0).
these statistics (except the
is
to treated departments T.
(1)
is
no apparent difference across groups
changes caused by the
were
still
fair,
not enrolled in the
we
first
in these proportions.
Since
we
are interested
focus in the remaining of the analysis on individuals
wave
(i.e.
who
by September 2000). Since the groups were chosen
randomly, the mean of observable characteristics such as sex, years of service, annual salary, and
age, are very similar across groups.
In panel B,
As
it
it is
visible
we can
As expected, none
of the differences are significant.
see that our inducement strategy
from inspecting Table
1,
had a strong
effect
on the probability
this does not affect the point estimates of the differences.
However,
reduces the standard errors, by absorbing some unexplained differences across departments of similar sizes and
pre-fair
TDA
enrollment rates.
10
of attending the fair: in treated departments, as
In control departments, fewer than
is
highly significant (Table
5%
column
2,
much
of individuals attended the
receive the letter.
is
28%, and
Thus, the direct
which are the same
for
is
enrolled.
The
account
fair
attendance.
15.1%
for
those in the treated departments
fair
group (which
is
The
(3) of
control departments
TDA
is
small and insignificant.
10%
at the
level.
1.26 percentage points
is
still,
and the
and
1.4 percentage point
The
enrollment).
difference
The
difference
people update their
difference
20%
TDA
significant as well
TDA
group
19
TDA
If
enrollment.
it is
is
parallelled
24%
be
increase
represents a
now
is
is
19%
positive, but
positive,
and
increase
still
very
significant
This impact
is
by an increased
point estimate in table 2
is
in their
large in relative
However, because
small in absolute terms (an increase of
is
tiny
compared to interventions
we made about department
we make the extreme assumption
fair participation rate for
(which would go up to 9%). In addition,
The
likely to
enrollment (such as in Marian and Shea (2001), and Choi
of course sensitive to the assumption
departments, the
group
(it
between group TO and group
status very infrequently,
did not register at our desk.
from
18
Eleven months
significant.
in the likelihood of enrollment after 11 months).
that change the default rules for
who
2).
between treated departments and
only 1.5% points of enrollment, on a base of 36%)). This effect
is
Table
Obtaining significant differences between these randomly chosen groups means
terms (an increase of
This result
and
between groups Tl and TO
that our experiment did have an impact on
18
due to
no significant difference between groups Tl and TO, however. 19
is
after the fair, enrollment is higher
in
solely
After 4.5 months, relatively few people have
participation.
between groups TO and
difference
effect,
13.8%. The
is
enrolled than employees in control departments (4.9% versus 4%). This represents a
There
a
did not
out any social
Table 2
(2) of
who
However, employees in treated departments are already significantly more
in the enrollment rate.
for
attendance rate of those
effect of receiving the letter (taking
between the TO group and the
TDA
look at
controls in the
social effects
almost as high, at 10.2%, and highly significant (see column
we
In Panel C,
shows that
groups TO and Tl) displayed on column
difference in the attendance rate
social effects)
is
1
fair.
difference, 16.5%,
on
large part of the effect of our experiment
received our letter
The
fair.
Comparing treated individuals versus
(1)).
treated departments in columns (2) and (3) of Table
who
21.4% of individuals attended the
as
TDA
TO group would
we show below
fall
that
down
affiliation of fair
non registered individuals come
all
to
11% but
that the increase in
11
t-statistic of
about
1.
be higher than
for
attendance in the
TO
still
fair
participation.
even slightly negative, with a
attendants
(2001a, 2001b)) or that offer individuals the option to allocate automatically future pay
et al.
rises to
TDA
contributions (Thaler and Bernatzi (2001)).
summary, the
In
a large
effect
on
encouragement
on
effect
TDA
who
we present
and
in Tables 1
2 suggest that the incentive
scheme had
participation of treated departments (due to a combination of the direct
fair
effect
and the multiplier
effect of social interactions), as well as
a significant
enrollment. This shows that the fair had an effect on the decision to enroll in the
TDA. However,
those
results
within treated departments, there
who
received the letter and those
is
no difference
The next
did not.
in
TDA
enrollment between
section presents simple models
to interpret these results.
Understanding the Effects of the Experiment
4
Fair attendance
4.1
Let us
analyze the decision to attend the
first
with
letter
its
As we have
fair.
promise of a $20 reward increases the probability of attending the
this increase in the probability of attending the fair
effects in the decision to attend the fair
individuals.
form
seen, receiving our invitation
A
because
TO
by
<5.
As we have
individuals are
simple way to capture these two effects
more
fair.
Denote
seen, there are peer
likely to attend
than
to posit the simple following reduced
is
specification:
f = 5U + liD + e\
1
where
U
is
/'
a
is
the
dummy
dummy
for
attending the
fair for
f.i
in (1). Actually, these
presented in Table
20
As
/*
is
2.
e'
%
is
and
a
is
number
is
random
D
l
a
individual effect,
dummy
indicator for
Table 3 presents the estimation parameters 5
in the
Taking the difference of the averages of equation
a 0-1 variable, equation (1)
to ensure that this
(1) of
e
two parameters were already estimated
reduced form results
(1) across
groups T\ and
not strictly correct. The left-hand-side should be replaced with the
probability of attending the fair for individual
of
i,
indicator for receiving the inducement letter,
being in a treated department. 20 Column
and
individual
(1)
i,
and
always between
restrictions
and
1.
imposed on the parameters and the distribution
These
technicalities can
thus are ignored to keep the presentation focused on identification questions.
12
be taken care of easily and
TO shows
that S
group Ti,
i
TO and
=
= 0,1.
fri
—
fro where fri denotes average fair attendance
shows that
=
fi
fro
—
/o-
given department decides to go to the
them information about the
her colleagues attend the
might talk about
it
sufficiently
if
Second,
fair.
for
it
can take two forms.
fair
is
First,
if
an individual in a
she might talk to her colleagues about the
them
who
We
who
received the letter divided by the
model these peer
each department influence the individual
number
fair, it
effects
must be
number
of
of employees in the department) in
attendance decision.
fair
the letter rate in the department of individual
fair
may
by assuming
Let us denote by / the average attendance rate in the department of individual
on
and thus
received the letter
ready to pay some people $20 to attend the
as well.
fair,
that the average fair attendance rate and the average "letter rate" (defined as the
employees
give
employee receiving the
she does not go herself to the
if
For example, colleagues of those
them to attend
fair,
to join her, and thus increase the probability
also conceivable that an
is
to her colleagues, even
the benefits office
important
fair,
details, or ask
also affect their attendance rate.
think that
individuals in
Similarly, taking the difference of the averages of equation (1) across groups
Peer effects in the decision to attend the
letter
among
The
i.
and by L
i,
invitation letter effect and the peer effects
participation can be captured by the simple following linear model (see e.g.
Manski
(1993))
f
where v l
is
the random individual
i
=
5Li
effect,
+AL +
and
(3\
<
1,
equation states that getting the letter increases the
and the probability of everybody
of employees in the department),
1
in the
/31
f
+v
and
own
A
i
(2)
,
are the peer effect coefficients. This
probability of attending the fair by 5,
department of attending by A/TV (N being the number
and that an exogenous
percent translates into a final increased
fair
direct increase in fair attendance of
attendance of 1/(1
—
(3\)
percent through the
multiplier peer effect.
Obviously, our experiment does not allow us to identify
because we have only two instruments: receiving the letter
being in a
T
all
L
1
three parameters
and the
dummy
5,
A, and
indicator
D
l
(3\
for
(versus 0) department. However, the following semi-reduced form of equation 2
identified,
13
is
f = 5L + 5 R L + v'\
i
Equation
(3)
can be easily derived from
by
(2)
first
(3)
averaging equation
obtain an expression for /, and then plugging this expression for / in
that 5 R
= —S +
+
(A
—
<5)/(l
/?i).
The parameters
5
our experiment 21 and can be estimated with an IV regression using
Column
number
(2) of
Table 3 presents the estimate of equation
of letters
is
0.28,
and
is
an increase
significant:
in
The
(3).
10%
U
by department to
Simple algebra shows
(2).
and 5 R of equation
(2)
(3) are identified
and
D
1
as instruments. 22
coefficient of the average
in the proportion of people
received a letter in the department lead to an increase of 2.8% in participation of those
not themselves receive the
It is
perhaps reasonable to impose the additional restriction on equation
In that case, equation (2)
fair.
is
identified
fair
and
j3\
attendance only
and
D
l
we obtain
The
as instruments.
for
large
is
/3i
results are reported on
is
1/(1
—
letter will induce,
the
TDA
4.2.1
is,
if
A=
0, i.e.,
she decides to go
column
by September 2000, using
(3) of
in
Table
3.
The estimate
attendance increases the
by 7.5%. Put another way, the multiplier peer
an additional person induced to go to the
effect,
that
(2)
fair
because of the
on average four additional individuals to attend
Participation
The Model
showed
in Section 3 that individuals in
groups Tl and
TO
This specification
externalities
22
that
through a trickle-down
individuals in group
21
4,
did
fair.
4.2
We
=
ft)
TDA
and precisely estimated: a 10% increase
probability that an individual attends the fair
effect
who
can be estimated by running the IV
regression (2) on the sample of individuals not enrolled in the
U
who
letter.
a person receiving a letter can influence her colleagues
to the
with
TO but
only equally likely to enroll in the
are in the
is
group Tl are more
TDA
same departments and thus exposed
similar to that of
Acemoglu and Angrist
likely to
(1999),
fair
than
after the fair. Individuals in
to the
who
attend the
same network
seek to estimate
effects at
human
capital
on earnings.
The average L
is
not exogenous because
it is
computed over
September 2000).
14
all
employees (enrolled or not in the
TDA
by
the department
The only
level.
between the 7T and TO groups
difference
received the inducement letter and hence are
that the direct fair effect
is
zero for those
Reduced form evidence from Section
than individuals
phenomena can explain these
First, as individuals in
is
who
likely to
attend the
have attended the
fair
and to enroll
group TO are more
likely to
attend the
(compared to group
for
individuals)
is
measured
for the
for those individuals
treatment
effect of
TO
same population. The
who come because
the
attend the
fair for
fair just for
thus do not get
much out
This suggests
group TO are more
in the
TDA
afterward. Three
had a
than group
individuals,
positive effect
on their
latter effect is the
treatment
effect of
TDA
the
of the inducement letter while the former effect
those individuals
it.
it
group TO individuals
effect for
who
are induced to
It is
come
is
fair
the
to the fair because
plausible to think that individuals
the $20 might not be very interested in the content of the
of
likely
individuals) because these treatment effects
they have been influenced to attend by their colleagues.
who
fair.
not necessarily contradictory with the zero treatment effect
group Tl individuals (compared to group
are not
in
fair
important to note that this positive treatment
is
T\ individual
results.
plausible to think that for this group, attending the fair has
participation. It
that
because of the $20 reward.
fair just
showed that individuals
both to attend the
group
in
3 also
more
is
In contrast, individuals induced to
come by
fair
and
their colleagues
(with no financial reward) are likely to be more interested by the event and thus end up being
more
influenced
treatment
effect
by what they learn
We
at the fair.
will
develop and formalize this differential
below using the theory of Local Average Treatment Effects (LATE) developed
by Imbens and Angrist (1994), and Angrist, Imbens, and Rubin (1996).
The second reason why group TO
individuals
is
individuals are
that, because of our experiment,
and individuals may be influenced by
could influence
TDA
social
more
likely to enroll in the
T departments
network
effects.
TDA
We
assume, however, that only
start enrolling in the
TDA.
fair
attendees
who
Peer effects within departments
who
are not enrolled in the
Individuals already enrolled in the
TDA
who
attend the
decides to enroll in the
TDA, can
fair
fair.
TDA
induce their colleagues to
presumably influence
through the second channel and not through the information collected at the
15
departments
with their colleagues and thus increase the
23
Second, an individual
enrollment rate in their department.
23
are different from
enrollment through two channels. First, individuals
might share the information obtained on the
TDA than group
their colleagues directly
might also discuss her decision with colleagues, and induce some of them to enroll as
Third and more subtle,
it is
conceivable that, even for an individual
well.
who would have come
to the fair with no external inducement, receiving the letter offering the $20 reward modifies her
psychological motivation for attending the
the
fair,
she might convince herself that she
really interested in the content of the fair.
but there
is
of rewards.
and
This literature
and Cooper
incentives for acting as
if
direction than providing
Lepper
et al. (1973)
.
This type of
just for the $20
effect is
now paid
to attend
and thus that she
not
is
not standard in economic models
These
is
summarized
in
Ross and Nisbett (1991) (pp. 65-67). Festinger
(1978) showed that providing people with small financial
et al.
they hold a given belief promotes greater change in the "rewarded"
them with
large incentives. Perhaps
most
closely related to our setting,
showed that school children who are rewarded to play with magic markers
are less likely to enjoy
"work"
coming
is
is
substantial evidence in the psychology literature on the motivational consequences
(1959)
al.
Because the individual
fair.
it
than children
who
are not, as
results generated a substantial
amount
if
"play" had subjectively turned into
of interest because they
go against the
conventional reinforcement theory that would appear more intuitive.
These three
effects,
namely the
differential treatment effect, the social
network
and
effect,
the motivational reward effect can be captured the following simple linear model as follows.
Let us assume that
i
fair
attendance increases the probability of
by 71 Let us denote by y % the
.
dummy
TDA
participation of individual
for individual participation in the
TDA. We
posit the
following specification
i
y
The
fact that 7*
treatment
effect.
The
the treatment effect
social
network
/*
effect
is
Finally, the motivational
-y
1
is
+ l7+ti*.
(4)
captured by the average
reward
effect
fair
participation rate /
can be captured by assuming that
potentially (negatively) correlated with the letter treatment
order to simplify the presentation,
It
i
can vary from individual to individual captures the potentially differential
in the department.
24
=7
let
.
In
us assume that -f takes to following simple form
would be possible to develop a more general non-linear model but
strategy and interpretation. Therefore,
U
we consider only the simple
16
linear
this
would not change our estimation
framework.
.
= 7& -
7*
where 75
is
independent of L
the motivational reward
assuming that v
=
!
(this is the
and thus that
attendance on
outcomes
fair
7'
TDA
literature
we observe only one
on
differential
independent of
is
enrollment,
effect
amounts
to simply
V
In order to define treatment
0.
useful to introduce the notion of potential
it is
i,
had he been
treatment
1 Monotonicity
in
group Tl, TO, or
0.
Obviously, for each
As the
of the three potential outcomes for fair attendance.
effects
has recognized (see Imbens and Angrist (1994)), in
order to be able to identify parameters of interest,
Assumption
component), and v represents
effect
of the groups Tl, TO, or
attendance decision of individual
i,
standard treatment
For each individual, we denote by /'(Tl), /'(TO), and /'(0) the
for fair attendance.
individual
(5)
Assuming no motivational reward
effect.
Each individual belongs to one
effects of fair
vL\
we need
make
to
Assumption: for each individual
the following assumption:
/'(Tl)
i,
>
/'(TO)
>
/'(0).
This assumption states that receiving the letter can only encourage individuals to attend the
fair
(and in no case deter them), and that having one's colleagues receive the letter can also only
encourage an individual to attend the
fair (relative
to the situation where no colleagues receive
the letter). This assumption sounds very plausible in the situation
we
analyze.
The Monotonicity
assumption implies that the population can be partitioned into four different types.
First, the never takers are individuals
=
such that /'(Tl)
/'(TO)
=
/'(0)
=
0.
These
individuals do not attend the fair and would not attend regardless of the group to which they
belong. Second,
1
>
/'(TO)
financial
/'(0)
define the financial reward compliers type as individuals such that /'(Tl)
=
0.
These individuals attend the
=
/'(TO)
=
1
>
/'(0)
department receives the
=
only
if
they receive the
These individuals would not attend the
0.
letter,
but attend the
(whether or not they themselves receive the
individuals such that /'(Tl)
of the
fair
letter
=
with the
reward promise. Third, we define the social interaction compliers as individuals such
that /'(Tl)
in their
=
we
=
/'(TO)
=
/'(0)
letter).
=
1.
group to which they belong.
We make
the following additional assumption.
17
fair if
fair if
nobody
they are in a treated department
Finally,
we
define the always takers as
These individuals attend the
fair
regardless
.
Assumption
2 Exclusion restriction assumption: u l
The assumption
means that the
letter inviting the
decisions of those
fair
that the error term
attendance).
who do
25
The
we
is
fair
To
assess the extent to
appendix).
There
is
no
assignment status
has no direct
effect
on
fair
its effect
sent did not mention
TDA
TDA
but only benefits in general, and
status (see the facsimile in appendix).
TDA
significant difference in
we send the
affect decisions,
TDA
participation after 6
status (see
months between
departments to which we sent the questionnaire and departments to which we did not (the
ference
is
dif-
actually negative at -0.093 percentage points with a standard error of 1.3 percentage
Within departments to which the questionnaire was
points).
l
TDA participation
which asked detailed questions about
2,
L
on individual and departmental
which written communication could
questionnaires described in section
U
letter
(beyond
did not contain any mention of the employee's
independent of
independent of the
employee to the
not attend the
letter
ul
is
sent, the difference
is
only 0.90
percentage points (with a standard error of 0.94 percentage points) and not statistically significant either. Therefore, the targeted questionnaire
participation to the
TDA.
It is
thus
is
who
sent
that employees
no more
did not
seem to
TDA
affect individuals'
Assumption
1,
a
fair
enrollment. This echoes the results in Choi et al
two versions of a questionnaire to randomly selected employees, and found
who
likely to
TDA
plausible to think that, as stated in
invitation letter does not directly affect
(2001a),
on
received a questionnaire with
subsequently enroll in the
more questions about retirement savings were
TDA
than those who received a version without
those questions. 26
Taking the average of equation
(4)
over groups
Tl and TO, and taking the
difference,
we
obtain
yT1
-
y T0
=
E\y'\T\]
- £[y
2
|T0]
= E\^s f - vf + Tf + u
Using the exclusion assumption stating that u r
is
l
\Tl]
- E[Ys f + Tf +
u*\T0}.
independent of L*, we have
yTi-yro = E\^ {f{T\) - f(T0))} - vE[f\T\].
5
However note that assumption
those
who attended the
fair,
1
does not rule out the possibility that the letter can affect the
by reducing the
fair's effectiveness
(through the motivational reward
TDA
effect
status of
described
above)
"
This
is
also evidence that information
conveyed through mailing
18
may
not have a great impact on final decisions.
Using the monotonicity assumption, we then obtain
VTi
~
= E{fs \f(Tl) - f (TO) =
Vto
As P{f(Tl)
=
1)
= hi
1
and P{p{Tl)
~
f
f°
JT\ — JTO
Thus comparing
individuals
-/
1]
l
T(f (Tl) -
•
(T0)
= E[fs \f(Tl) -
=
/'(TO)
= fn -
1)
/'(TO)
=
1]
=
/to,
we
individuals
TO
-
-
(6)
.
- JTO
sum
provides an estimate of the
of the direct
average treatment effect for financial reward compilers and the motivational reward
effect.
that the social network effects (term Tf) cancel out in the comparison of groups
because they are
common
Individuals in group
in
TO and
effects,
TO do
TO
individuals in
effect involved in this
over groups
TO and
y T0
0,
Tl and TO
do not receive the inducement
individuals in group
TO. As none of the individuals in groups
reward
Note
to both groups.
but some of the peers of individuals in
because of network
1)
finally obtain
/T1
-„
JT1
T\ and
- uP(f(Tl) =
1)
receive the letter.
are
more
TO and
As we have seen
attend the
likely to
fair
receive the letter, there
is
letter
in Section 3,
than individuals
no motivational
comparison. More precisely, taking the average of equation
and taking the
difference,
-go = Etf\T0] - E\y
l
\0]
(4)
we have
= E\Y(f(T0) - f(0))}+T\fT -
/„].
Hence, we finally obtain
vto
JTO
-yo =
— 70
Thus comparing group TO
treatment
Our
to group
effect for social interaction
analysis has
treatment
E tf\fi {T0) _ r (0) =
+r
.
fr ~ fo
JTO
-
provides an estimate of the
(7)
JO
sum
compilers and the social network
shown that there are
effect for financial
i]
of the direct average
effect.
four parameters of interest in the model: the average
reward compilers, the average treatment
effect for social interaction
compilers, the social network effect parameter T, and the motivational reward effect v.
experiment provides us with only two instruments
identify
all
four parameters together.
these four parameters can
Only
if
L and
l
D
l
,
thus
it is
clear that
we
Our
cannot,
we make additional assumptions about two
we estimate the remaining two parameters. In the next
19
subsection,
of
we
discuss three alternative assumptions under which the remaining parameters of the
be estimated.
is
proceed not with the intention of claiming any particular set of assumptions
correct, but rather to explore the implications of each assumption.
Interpretation under Alternative Identification Assumptions
4.2.2
•
We
model could
No
motivational reward and no social network effects
In that situation, both parameters
T and
v are equal to zero, and
we can
identify both
average treatment effects for financial reward compilers and social interaction compilers. Under
these assumption, equation (6) reduces to:
V
~
f
P
JT\ ~ JTO
1
Thus, the average treatment
IV regression
of
TDA
departments using
U
effect for financial
enrollment on
JTO
of
TDA
enrollment on
-
fair
=E[ 1
i
i
\f
(T0)-f
i
(0)
we have
=
l}.
months and 11 months
and
(9)
JO
compliers can be obtained by an
attendance on the sample of individuals
(2)
(8)
attendance on the sample of individuals in treated
effect for social interaction
an instrument. Column
4.5
fair
1].
reward compilers can be obtained by a simple
as an instrument. Similarly, using (7),
yT°- yo
The average treatment
= EtflfiTl) - f (TO) =
(3) in
in
TO
or
Table 4 present these IV estimates,
after the fair.
IV
regression
groups using
for
TDA
D
%
as
enrollment
Consistent with the reduced form evidence, the IV
estimates suggest a positive treatment effect on social interaction compliers, and no effect on
financial
reward compliers. The results
in
column
(3)
can be thought of as an upper bound on
the effects of the fair itself for the population of social interaction compliers. These individuals
are not affected
in
(9)
by the motivational reward, but
would be an upper bound of the direct
if
peers effects are present, the
effect of the fair.
IV estimate
This upper bound
is
13.5
percentage points after 4.5 months, and 14.8 percentage points after 11 months. These effects
are of comparable size (slightly higher) than those estimated
by Madrian and Shea (2002)
non-experimental set-up.
•
Constant Treatment Effects with no motivational reward
20
effect
in a
If
there
is
no motivational reward
same across individuals and equal to
and equation
are identified
=
7,
and the standard treatment
the
both parameters, 7 and T of the structural equation
(4)
0)
reduces to
(6)
7
§^f°=
JT\ ~ JTO
This ratio
is
(10)
.
the IV estimate presented in column (2) in table
Therefore, under these restrictive assumptions,
is
we can conclude
4,
which we discuss above.
that the direct effect of the
=
Therefore the overall effect of
(7
fair
-
+ r)/ + u,
yT
attendance on
TDA
(7
fair
and T, we obtain
zero for everyone. Taking the average of equation (4) over departments
yo
effect 7'
is
effect (u
+ r)/r + «•
(n)
participation, taking into account
all
the
social effects, is the ratio
=7+
f^f
It -
r.
(12)
Jo
This overall effect of one additional person attending the
the direct causal effect 7 of the
fair,
and the
months and
(for participation after 4.5
made
here, the difference of
The implied
•
estimates of
T
columns
(1)
are 10.14%
social effects
after 11
In both cases, the overall effect of the
4.
(2) gives
and 6.7%,
on
TDA
T from equation
and
after 4.5
and
11
T =
0.
who
the
sum
of
These estimates
(4).
an estimate of the social
In this case, the standard treatment effect 75 (for those
(7), since
is
(1) in table
Under the assumptions
significant.
Constant Treatment Effects and no social network
be obtained directly from
participation
months) are presented in column
fair is positive
and
fair
months
effect
parameter.
respectively.
27
effects
did not receive the letter) can
In turn, equation (6), with the
75, can be used to recover v. Using the estimates of fri, and fro from table
first
2,
term
set to
we obtain an
estimate of v of 0.0927 after 4.5 months, and 0.0620 after 11 months. Under these assumptions,
receiving the letter reduces the treatment effect of the fair by
4.5
2
months, and 42%
for
TDA
69%
for
TDA
participation after
participation after 11 months.
'Estimates of the direct and social effects parameters (and their standard errors) can also be directly obtained
by an IV estimation of equation
(4)
(where 7
1
=
7), using -D,
21
and L, as instruments.
The
distinction between differential treatment effects, social network effects,
tional reward effects
apart.
Thus,
it
is
is
clear conceptually but our experiment does not allow us to tell
useful to describe
needed to separate these
is
a
a
first
what type
effects. Differential
of alternative experimental designs
them
would be
treatment effects arise in our setting because there
stage in our experiment where individuals decide whether or not to attend the
result, only a self-selected fraction of individuals attends the fair. Motivational
arise
and motiva-
because individuals receive a monetary payment
attending the
for
fair.
reward
As
effects
fair.
Social network effects could be identified with the following experiment. Within a subsample
of the "treated" departments, a
subsample of employees would
all
attend automatically an
information session. This could be done by making attendance a job requirement of employees.
One
could then test whether the
TDA
among
participation rises
relative to that of individuals in untreated departments.
the colleagues of the treated
Motivational reward effects could be
estimated by paying people for attending an information session in a situation where everybody
is
supposed to attend.
For example, in
information sessions about benefits.
In
many
new
firms,
some departments,
presented as a normal process through which
all
hires are often invited to attend
this information session could
new employees
go.
be
In other departments,
attending this information session could be presented as voluntary but a financial reward could
be offered
for
attendance (large enough to induce virtually everybody to attend).
If
everybody
attends in both cases, the average treatment effect would be expected to be the same in both
groups in the absence of a motivational reward
effect.
28
Evidence of differential treatment
effects
could potentially be obtained by using non-monetary incentives of various intensity to attend the
fair.
fair.
For example, some employees could be sent a letter simply reminding them of the benefits
Others could be sent a more pointed
be obtained
at the fair.
person to attend the
fair.
groups of compilers and
28
in
Note that
this setting
One
them that important information can
could also use emails, personal phone
These
may
letter telling
different
calls or
even remind them in
encouragement designs are associated with
thus allow estimation of different
would be close to the experiments carried out
Ross and Nisbett (1991).
22
in
fair
treatment
different
effects.
the social psychology literature reviewed
Direct and Overall effects of the
4.3
The model developed
clarifies
comparisons with naive estimates
fair:
when
the errors that can be done
ignoring social effects in experi-
mental data. The data also allow us to compare experimental results with observational
Table 6 presents alternative estimates of the
Columns
TDA
and
(1)
(2) are, respectively, the
effect of the fair
The OLS estimate
is
OLS
estimate
is
IV
fair.
(3)
TDA
who
dummy
and
for
participation on fair at-
Given
as large as the
IV
estimate. This
in benefits to
is
not surprising,
be more likely to attend
versus
IV estimates
and 0.082
is
we do not know the
estimate
attendees
TDA
who
is
fair
by an
is
dummy
in
earlier,
information on average than non
column
(4),
TDA
the
who came
effect
and the
effect
based on the
fair
among
attend the
TDA
all
on
TDA
enrollment.
who
who
received the invitation
received the letter.
participants
is
much
higher
among the sample
of fair
employees (above 50% versus around 30%). This shows
fair,
participants (see
fair
paper would lead to a
in this
to the fair except for those
the sample of those
the fraction of
likely to
we
as an instrument, in the complete
we have developed
effect of
identity of those
more
significant. In
actually treated) and overestimate the direct effect on those
did not receive the letter than
participants are
The
naive estimate would underestimate the overall effect of the
group
obtained
we noted
letter
between the estimate of the overall
lies
misguided causal interpretation of the
OLS
estimate the overall effect of the
and 11 months and are
received the letter. Ignoring the analysis
the
TDA
on
treated as an instrument for average participation.
after 4.5
TO comparison. The
For example, as
we
(3),
IV estimate that uses the
(since part of the "control"
Since
of the overall effect of the fair
participation on average fair participation in each department, using the
sample. This estimate
that
direct
sample (treated and control departments) of employees not enrolled
full
whether the department
present the "naive",
letter,
TDA
for receiving the letter as instrument.
more interested
(4) present alternative
TDA
coefficients are 0.057
30
are
by September 2000. In column
regression of
29
dummy
more than three times
participation using the
in the
As we explained above, the
T
30
Columns
who
of the direct effect on
estimate, the two estimates are statistically indistinguishable, but
given that one would expect those
T\
IV and the OLS estimates
can be estimated by running an IV regression of
the lack of precision of the
IV
fair.
0.052 and significant.
tendance in treated departments, using the
the
the
participation of attending the fair after 11 months; they are limited to employees in
departments. 29
the
effect of
results.
probably because they are more interested
Madrian and Shea (2002)
23
for
in benefits
evidence of self-selection).
Interpretation and Additional evidence
5
The
did the experiment influence
fact that, despite the large
participation?
remaining difference
in fair attendance, there is
no difference
TDA participation between the treated and untreated individuals within treated departments,
while there
TDA
a significant difference in
is
partments. As we discussed above, the
first
participation between treated and untreated de-
stage results are a clear indication of social effects
in the decision to attend the fair, while the interpretation of the
more
delicate:
reward
a
TDA
striking results of the experiment are the large spillover effects at the fair attendance stage,
and the
in
Why
Interpretation:
5.1
they could be due to social
effects, or
common
effects, differential
TDA
participation results
treatment
effects,
is
motivational
a combination of the three. These three different explanations have, however,
feature.
They
suggests that an individual's decision to participate in the
TDA
is
and not only by the information content of the
affected by small changes in the environment,
fair.
If
the results can be entirely explained by social effects, they suggest very strong peer effects,
compared to the
fair
direct effect of the fair. This could be true in
two models.
conveys useful information, but any information obtained by a
diffused to the entire department he belongs to. This
not participate in the
more than group
effect of
at least
TDA
much
fair,
that at least one
in large
is
member
=
0),
who
is
entirely
due to the increase
fair.
departments (93%) than
thus that
of the
if,
completely
in
turn participate
in the probability that
Indeed, according to the registration
the probability that at least one department
larger in treated
is
This model has an additional testable implication: the
one member of the department attends the
implication of the model
than
0).
being in a treated department
data we collected at the
fair is
>
participant
would explain why group Tl individuals do
any more than group TO individuals (7
individuals (F
fair
In the first model, the
as one
in untreated
TDA
attends the
departments (55%).
would expect, the difference
department attends the
departments, the difference in
member
fair is larger in
An
in the probability
small departments
participation after 4.5 or 11
months between
treated and control departments should also be larger within the smaller departments. Indeed,
the difference between treated and untreated departments in the probability that at least one
person attends the
fair is
59%
in the
department of 81 employees or
24
less
(department
size for
16%
the median employee), and
we show
A
in panel
identical in the
in the
departments with more than 81 employees. However, as
form differences
of table 5, the reduced
two
after 4.5
and
11
months are
virtually
This rejects the hypothesis of complete diffusion of
sets of departments.
information.
Under the second model when individuals
letter inviting others to
tive of
attend the
more people attending the
they are directly induce to enroll
fair),
what those who went to the
see
fair learnt at
in
fair (or
the
receiving a
TDA
(irrespec-
the fair or decided to do). Those peer effects
thus do not seem to stem from a rational herd behavior in an environment where information
is
scarce or difficult to obtain (as in the models of Banerjee (1992) or Bikhchandani, Hirshleifer
and Welch
(1992)).
At the same time, there
for
Another explanation
groups of compilers:
it
for the results
is
is
fair
is
social pressure to
conform to
the case, for example, in the decision of
that the treatment effects are different for different
positive for social interaction compilers, but zero for the financial
5,
we explore
variations in the effect of the treatment.
the
(as
no strong
Bengladeshi women, as in Munshi (2000b)).
reward compilers. In Table
subgroup,
clearly
TDA
the decisions of the majority regarding the
adopt contraception
is
among those who
various observable characteristics which
Column
(1) reports
know
received the letter (we
average
may
lead to
fair participation in
the identity of those
who
each
attended
only for this group). Fair participation was larger in small departments than in large
departments, and
for
women than
for
men. In column
(2)
and
(3),
we show
the difference in
TDA
enrollment between treated and control departments after 4.5 and 11 months, respectively. After
4.5
months, the treatment
rate before the experiment
after 11
that
it
effect
months, this difference shrunk
takes
more time
for those in
men and women.
TDA
Overall, there
it
social interaction compilers
is
salaries are high (panel D).
departments with low
is
Panel
C
initial participation
shows that the
no evidence that treatment
Any
However,
panel B) or disappeared (in panel D). This suggests
participation.
across groups defined by observables.
and
(in
departments where the participation
in
was high (panel B) and average
lower salaries to adjust their
for
seems somewhat larger
differential
treatment
effects are the
effects are
effect
and those with
same
widely different
between financial reward
thus not attributable to observable characteristics. Of course,
could be due to an unobservable attribute uncorrected with these observable characteristic
(like interest in
the benefits).
Importantly, even
25
if
the results are entirely due to differential
treatment
effects,
and
they are responsible
social interactions take
Thus, in this case as
well, social
take steps which ultimately led to change their
If
in explaining the
second stage results,
attendance among the untreated individual in
for the variation in fair
treated departments.
no part
TDA
network
caused some people to
effects
participation decision.
the results are in part explained by the motivational reward
would
effect, this
also pro-
vide evidence that individuals' decisions are influenced by small non-economic factors:
attending the
fair
on their own, they are influenced by
induced to go by the $20 reward.
thus influences their final decision.
making
In
fair
Again, this suggests that individuals' process of decision
influenced by small changes in the environments.
is
these potential explanations
all
influenced by things other than
combined with the
effects at
the
new information about
is
TDA
effect of
the
fair
participation stage
on the
is
that the participation
costs and benefits of the
fact that the effect of the information fair itself
terms (the upper bound of the
no peer
but are not when they have been
small perturbation in their motivation to attend the
summary, a common thread to
decision
This,
is
A
it,
when
was modest
TDA.
in absolute
social interaction compilers,
assuming
an increase of 14.8 percentage points
in the
participation rate after 11 months) suggests that an individual's decision to participate in the
TDA
is
not taken as the outcome of a sophisticated decision process of information gathering
and careful considerations of the
alternatives. This
is
consistent with a growing
on retirement savings behavior, showing that individuals believe that
low (Choi et
al.
et al. (2001a),
(2001a)), but that their project to increase
Madrian and Shea
(2002)),
it
body
of evidence
their savings rate
are rarely followed
is
too
by action (Choi
and that retirement decision are characterized by very
strong inertia and adherence to default rules (Madrian and Shea (2001), Choi et
al.
(2001b)).
Thaler and Bertazi (2001) show that savings rates increase dramatically when individuals are
offered to enroll in a
program
We now
directly
in earnings.
in our experiment,
employees.
We
in
which they commit now to save a portion of their future increase
examine the relationship between information and decision-making
by examining responses to a follow-up questionnaire we sent to a sample of
find results consistent with the above-mentioned literature.
26
Follow up questionnaires
5.2
A follow
up questionnaire sent
measure the employees' knowledge of the retirement benefits sj'stem
questions to
two questions designed to
to 917 employees after the fair included
in the university, as well as
alternative retirement savings options available to employees
elicit
and to measure
the extent of procrastination.
Analysis of survey data presents an additional challenge, as the response rate to our questionnaire
people
TDA
error
was
less
than 50%. 31 Clearly, people who respond form a selected group:
who respond
after 6
is
to the questionnaires are 8 percentage points
months than those who received
0.017).
As we have shown
participation: thus this difference
it
in section 4.2, the questionnaire itself
is
entirely
likely to enroll in the
due to
selection.
had no causal
Moreover, those
TDA
after 6
who
effect
on
received the
months than those
did not get the questionnaire. 32 In addition, the selection seems different in treated versus
control departments.
it is
example,
but did not return the survey (the standard
questionnaire and did not respond are less likely to enroll in the
who
more
for
only
35%
The response
in control
responses across samples.
rate in treated departments
departments.
On
It
may
45%
(Table
1,
panel D), while
thus not be very informative to compare the
the other hand, network effects within departments seem to have
played an important role here too: the response rates
within treated departments are essentially identical.
had received the
is
fair invitation letter
were able to
among
A
tell
treated and untreated individuals
plausible explanation
is
their colleagues that
that those
we had
who
delivered
on our promise of sending the reward. Since the response rates are the same, the assumption
that the selection process
is
the
same
is
reasonable. Thus,
we can compare
treated and untreated individuals within treated department.
the response
among
These responses are not repre-
sentative of the population in general, but representative of the segment of the population that
tends to respond to this type of questionnaires.
The
results are presented in Table 7. People
to have attended the fair than people
to the questionnaire attended (while
the control group,
31
32
This
Since
is
a
common
29%
who
who answered
the questionnaire are more likely
did not: in the treated group,
28%
43%
of the respondents
of the entire treated population attended),
of the respondents attended (compared to 15.1%).
problem: the survey on savings intention by Choi et
we have shown above
al.
is
in
difference in
(2001a) had a response rate of
that the questionnaire had no causal effect on enrollment, this
27
The
and
33%
a sign of selection.
,
attendance (14%)
whole (13.1%), which we had recorded
rates with the
at the fair.
Yet, the satisfaction
fair.
is
33
attendance in this sample:
our reward was less likely to find the
treatment
is
fair useful
for
almost as large as the difference in
fair
participant induced by
(thus supporting the hypothesis of differential
having received the letter reduces
effects), or that
group than
significantly higher for the control
suggests either that the marginal
it
as a
Respondents report very high satisfaction
the treatment group (95% against 85%). This difference
fair
two groups
similar to the difference in fair attendance between the
is
fair satisfaction
(supporting the
motivational reward effect hypothesis).
we
In panel B,
who
for those
report the response to the question "why are you not enrolled in the
could check as
many
likely to report that
10%
the
them
report that they are not enrolled (none of
answers as were applicable.
are
more
likely to
They
Individuals in the treatment group are less
is
significant at
say that they want to enroll soon, but have not found
the time yet (45% versus 36%), although the t-statistic
is
just 1.3. 34 All the other reasons for
not contributing are mentioned equally often in both groups.
is
are actually enrolled).
they lack information (20% versus 30%). The difference
They
level.
TDA?"
The reason
"plan to enroll soon"
the single most often cited reason for not contributing in both groups. In panel C, we match
this
answer with their future behavior. Actual behavior
nobody who did not
Among
are also
more
Panel
D
TDA?" Not
33
10% do
so.
35
Thus,
surprisingly, those in the treatment
difference
is
This
for
only
75%
in
in the
likely to
so.
Among
treated
have good intentions, but
1,
likely to
say that they obtain
way we recorded departments
at the fair-even
of the participants.
in fair participation:
want to enrol on whether an individual went to the
who were not
is
more
group are more
9%, almost as large as the difference
thus give a coefficient very close to
attendee
planned to enroll do
letter individuals are
This similarity suggests that there was no systematic bias
The
well short of intention.
shows the answer to the question "where do you obtain information about the
to report that one
35
who
falls
likely to procrastinate.
though we recorded them
34
correlated with intention (virtually
declare that they intended to enroll did so) but
untreated individuals, 16.7% of those
individuals,
is
which
is
also
fair,
a simple IV on the probability
using the letter as instrument, would
what Madrian and Shea (2002) obtain:
all
seminar
14%
of those
virtually
yet enrolled in the plan were intending to enroll soon after the seminar.
the ballpark of other studies. Following the survey conducted by Choi et
who
intended to enroll in the
14%
of the attendees
(who
all
TDA
did.
al.
(2001a),
Following the financial education session in Madrian and Shea (2002),
intended to enroll) did.
28
it
from the
fair
(and the difference, 11%,
However, they are
less likely
is
close to the
14%
difference in fair attendance).
to obtain information from the benefits fair information packet
(77% versus 93%). Those two sources of information thus appear
to
be substitutes. The other
sources of information seem to be used equally by both groups.
E
Panel
employee
is
reports answers to the knowledge questions.
or
is
not enrolled in the
we asked them whether they know
(DC)
benefits are invested.
TDA
the
(when we sent the
number
If
Treatment and control groups are about as
they are contributing:
total
60%
74% and 71%,
likely to
give the correct answer
part, since this letter
in the
knowing
their
(89% versus 94%).
37
DC plan and can choose
have more than one
randomly allocates them to one vendor.
know the number
status
This could
of vendors with
who
in
received the letter are
(94% versus 99%), and
reflect
whom
answer the question, and,
However, those
TDA
Defined Contribution
Many employees
respectively, ventured to
36
of each group gave the right answer.
significantly less likely to report
office
whether the
is
none of them were). Second,
whom their
Employees are automatically enrolled
they do not make a choice, the benefits
question
first
letter,
of vendors with
to invest their contributions with four different vendors.
vendor.
The
less likely to
some over-confidence on
their
was sent only to those who were not contributing. This lends some support
was high, the
to the motivational reward hypothesis: in this group where the fair attendance
treated group has less knowledge than the group that was not directly treated.
In summary, participation in the fair did not seem to have a large impact on the information
set of those
who
received the letter: they
In fact, they are
research.
more
seem to have substituted
likely to
fair
attendance
be unsure about their actual
TDA
for individual
status,
and to
wrongly report themselves as contributing even though they are not. However, they are
likely to think
soon.
Of
of those
36
less
they they suffer from a lack of information, and more likely to plan to enroll
course,
it
who went
does not imply that the
fair
did not have an impact on the information set
to the fair without the letter (used here as the control group).
Those who did not answer are counted
as having given the
wrong answer.
^Incidentally, this level of misclassification underscores the importance of working with administrative data
when studying
TDA
savings behavior.
29
Conclusion
6
This paper has attempted to identify the causal effects of information and social interaction on
employee decisions to enroll in an employer sponsored Tax Deferred Account retirement plan.
Our encouragement strategy
successfully induced treated employees to attend a benefits
The experimental design allowed us
to demonstrate that peer effects are an important factor in
determining whether employees attend the
In the second stage of the study,
fair.
we presented
evidence that individuals affected by the experiment are indeed more likely to enroll in the
we
after the fair. Interestingly,
among those whom we
to be very small
We
find that the direct causal effect of fair attendance
directly induced to attend the fair
compared to the
effect of
proposed three different interpretations,
and motivational reward
effects, to
us to distinguish unambiguously
by means of a
differentia]
treatment
on an attendee
financial
being in a department with high
TDA
reward seems
fair participation.
effects, social
network
effects,
account for these findings. Our experiment does not allow
among
of a simple experiment in a social
fair.
these interpretations, thus illustrating
how
the analysis
and economic context may be substantially more complicated
than expected.
We
are, nevertheless, able to provide
enrollment: attending the fair increases
15%
11
(in
a sample of people
months
is
who were
an upper bound to the
TDA
participation 11
effect of
months
not enrolled). Average
initially
only 7.5% in the control group (of which
5%
the benefits
later
TDA
attended the
by a
on
fair
maximum
of
participation after
fair).
Mandatory
fair
participation might thus produce a non-negligible increase in the enrollment flow, comparable to
the effect of introducing a
it
25% employer matching
contribution (Choi et
al.
2001a). However,
remains small compared to changing default enrollment rules (Madrian and Shea (2001)) or
offering delayed enrollment, as in the enrollment the "Save
More Tomorrow" program (Thaler
and Bernatzi, 2001).
This paper also provides experimental evidence that social interactions are a powerful mech-
anism
in
the process of information acquisition
(i.e.,
the decision to seek additional information).
Individuals do not instantly learn about economic opportunities, and their informational envi-
ronment has a strong
effect
on their economic
decisions.
Low household
savings levels in the
United States have concerned academics and policy makers. Recognizing that savings decisions
30
are influenced by peers' savings decisions could improve our understanding of
enroll in
TDAs, and may
why
individuals
provide a rationale for organizing 401 (k)s through the workplace.
large effect of a small reward on fair attendance, amplified
by
The
social effects, also suggests that
individuals do not optimally seek out and process information on their own. While the motivational
reward
effect
must be addressed, encouraging employees to attend
benefits fairs
may be
a useful complement to automatic enrollment.
Finally, this
study has shown that
it
is
relatively simple
and inexpensive to carry out an
experiment within a large firm to study important economic research questions.
Moreover,
organizational divisions within a firm provide an excellent structure in which to study the effects
of social interaction in the workplace.
process and induce
more economists
In particular, our analysis raised
from
this
experiment as a
We
hope that our study
encourage this research
to tackle questions in labor economics using experiments.
more questions than we were
first step,
will
able to answer.
Using results
one could think of several alternative experimental designs
that could precisely identify the effects
we have
described.
31
References
"How Large
Acemoglu, Daron and Joshua Angrist, 1999.
are
Human
capital Externalities:
Evidence from Compulsory Schooling Laws'". Mimeo, MIT.
Angrist, Joshua,
Guido Imbens, and Donald Rubin, 1996,
"Identification of Causal Effects using
Instrumental Variables.", Journal of the American Statistical Association, 91, 444-455.
Banerjee, A.V., 1992.
A
simple model of herd behavior. Quarterly Journal of Economics 107,
797-817.
"The
Bayer, Patrick B., Bernheim, B.D., Scholz, K., 1996.
the Workplace: Evidence from a Survey of Employers."
Effects of Financial Education in
NBER
Working Paper No. 5655.
"The Determinants and Consequences of Financial
Bernheim, B.D., Garrett, D.M., 1996.
Education in the Workplace: Evidence from a Survey of Households."
NBER
Working
Paper No. 5667.
Bernheim, B. Douglas, Daniel M. Garrett and Dean M. Maki (1997). "Education and Saving:
The Long-Term
Effects of
High School Financial Curriculum Mandates."
NBER
Working
Paper No. 6085. Forthcoming Journal of Public Economics.
Bertrand, M., Mullainathan,
S.,
Luttner, E., 2000. "Network Effects and Welfare Cultures."
Quarterly Journal of Economics, 115, 1019-1057.
Besley, T., Case, A., 1994.
Discussion Paper 174,
Bikhchandani,
"Diffusion as a Learning Process: Evidence from
RPDS,
S., Hirshleifer,
Ann and Lawrence
D., Welch,
I.,
1992.
Katz, 1991.
"A theory of
fads, fashion, custom,
J.,
Economy,
"The Company you Keep: The
Neighborhood on Disadvantaged Youths."
Choi,
Cotton."
Princeton University.
cultural change as informational cascades" Journal of Political
Case,
HYV
NBER
100, 992-1026.
Effect of
Working Paper No.
and
Family and
3705.
Laibson, D., Madrian, B. and Metrick A., 2001a "For Better or for Worse: Default
Effect
and 401 (k) Savings Behavior",
NBER
32
Working Paper No. 8651.
Choi.
Laibson. D., Madrian, B. and Metrick A., 2001b "Defined Contributions Pensions:
J.,
Plan Rules, Participants Decisions, and the Path of Least Resistance"
,
NBER
Working
Paper No. 8655.
Cooper,
J.,
M.
P.
Zanna, and
P. A. Taves, 1978,
Change Following Induced Compliance."
"Arousal as a Necessary Condition for Attitude
Journal of Personality and Social Psychology,
36, 1101-1106.
Duflo, Esther
and Emmanuel Saez, 2000. "Participation and Investment Decisions
tirement Plan:
The
Influence of Colleagues' Choices."
NBER
in a
Re-
Working Paper No. 7735,
forthcoming Journal of Public Economics.
Engen, E.M., Gale, W.G., Scholz, J.K., 1996.
"The
Illusory Effect of
Saving Incentives on
Saving." Journal of Economic Perspective 10, 113-138.
Evans, W., Oates, W., Schwab, R., 1992. "Measuring Peer Group Effects:
A
Model
of
Teenage
Behavior." Journal of Political Economy, 100, 966-991.
Festinger, L. and
J.
M. Carlsmith,
1959.
"Cognitive Consequences of Forced Compliance."
Journal of Abnormal and Social Psychology, 58, 203-210.
Foster, A.D., Rosenzweig,
MR.,
1995. "Learning by
Capital and Technical Change in Agriculture."
Doing and Learning from Others:
Journal of Political
Economy
Human
103, 1176-
1209.
Imbens, Guido, and Joshua Angrist, 1994.
Treatment
"Identification
and Estimation of Local Average
Effects." Econometrica, 62, 467-476.
Imbens, G., Irano, K., D. Rubin, A. Zhou, 2000.
"Estimating the Effect of Flu Shots in a
Randomized Encouragement Design", Bio statistics,
1(1), 69-88.
Katz, Lawrence F., Jeffrey R. Kling, Jeffrey B. Liebman, 2001.
in
"Moving to Opportunity
Boston: Early Results of a Randomized Mobility Experiment.", Quarterly Journal of
Economics, 116(2), 607-654.
33
Kremer, Michael and Edward Miguel, 2001 "Worms: Education and Health Externalities
Kenya."
NBER Working
Paper No. 8481.
Lepper, M. R., D. Greene, and R. E. Nisbett, 1973.
terest with Extrinsic
in
Reward:
A
"Undermining Children's
Intrinsic In-
Test for the Overjustification Hypothesis."
Journal of
Psychology and and Social Psychology, 37, 688-714.
Madrian, Brigitte and Dennis F. Shea, 2001.
Participation and Savings Behavior."
"The Power of Suggestion: Inertia
in 401(k)
Quarterly Journal of Economics, 116(4), 1149-1187.
Madrian, Brigitte and Dennis F. Shea, 2002. "Preaching to the Converted and Converting those
Taught: Financial Education
in
Mimeo Graduate
the Workplace."
School of Business,
University of Chicago.
Manski,
C,
1993.
"Identification of
Exogenous Social
Effects:
The
Reflection Problem."
Re-
view of Economic Studies 60, 531-542.
Manski, C. 1995 Identification Problems in the Social Sciences.
Harvard University Press,
Cambridge.
Munshi, K., 2000a.
"Social Learning in a Heterogenous Population: Technology Diffusion in
the Indian Green Revolution." University of Pennsylvania Mimeo.
Munshi, K., 2000b. "Social Norms and Individual Decisions During a Period of Change:
An
Application to the Demographic Transition." University of Pennsylvania Mimeo.
Permutt, Thomas, and
J.
Richard Hebel.
Clinical Trial of the Effect of
1989.
Smoking on Birth Weight." Biometrics
Poterba, J.M., Venti, S.F., Wise, D.A., 1996.
Saving." Journal of
Economic Perspectives
Powers, Donald E., and Spencer
"Simultaneous Equation Estimation in a
S.
45, 619-22.
"How Retirement Saving Programs
Increase
10, 91-112.
Swinton, 1984.
"Effects of Self-Study for Coachable Test
Item Types," Journal of Educational Psychology 76, 266-278.
Ross, Lee and Richard E. Nisbett, 1991 The Person and the Situation: Perspectives of Social
Psychology, Temple University Press: Philadelphia.
34
Sacerdote, Bruce, 2001, "Peer Effects with
Random
Assignment: Results
for
Dartmouth Room-
mates." Quarterly Journal of Economics, 116(2), 681-704.
Sorensen, Alan, 2001, "Social Learning in the
ance."
UCSD
Demand
for
Employer-Sponsored Health Insur-
Mimeo, 2001
Summers, Lawrence, 2000, "Helping America
Lawrence H. Summers"
.
to Save More.
Remarks by Treasury Secretary
Treasury News Press Release LS-524, April 4th.
Thaler, Richard H. and Shlomo Bernatzi, 2001,
"Save
More Tomorrow: Using Behavioral
Economics to Increase Employee Saving." Mimeo Graduate School of Business, University
of Chicago.
35
Table
1:
Descriptive Statistics, by groups
Treat;ed departments
Treated
All
(group T)
(group Tl) (group TO)
TDA
0.010
0.009
0.011
0.012
(.0021)
(.0022)
(.0024)
4168
2039
2129
2043
of observations
0.398
0.400
0.396
0.418
(.0076)
(.0109)
(.0107)
(.011)
Years of Service
Salary
5.898
5.864
5.930
6.008
(.114)
(.161)
(.16)
(-157)
38547
38807
38297
38213
(304)
(438)
(422)
(416)
38.3
38.4
38.2
38.7
(.17)
(.24)
(.24)
(.24)
4126
2020
2106
2018
Age
Number
of observations
PANEL
B:
FAIR
ATTENDANCE (REGISTRATION DATA)
0.214
0.280
0.151
0.049
(.0064)
(.01)
(.0078)
(.0048)
4126
2020
2106
2018
Fair attendance
among non-TDA enrollees
Number of observations
0.192
0.063
(.0132)
(.0103)
6687
3311
Fair attendance
all
(4)
(.0015)
participation
Sex (% male)
for
(group 0)
BACKGROUND CHARACTERISTICS
A:
before the fair (Sept. 2000)
Number
Departments
(3)
(2)
(1)
PANEL
Untreated
Untreated
employes
staff
Number of observations
PANEL C: TDA PARTICIPATION (ADMINISTRATIVE DATA)
0.045
0.053
0.049
TDA participation after
4.5
months
Number
1 1
(.0049)
(.0051)
(.0045)
3726
1832
1894
1861
of observations
TDA participation
after
months
0.088
0.089
0.088
0.075
(.005)
(.0071)
(.007)
(.0065)
3246
1608
1638
1633
Number of observations
PANEL
Number
RESPONSE RATE TO THE ADDITIONAL QUESTIONNAIRE
D:
Response
0.040
(.0035)
rate
0.452
0.440
0.464
0.352
(.018)
(.0201)
(.0405)
(.0402)
765
6F2
1_53
142
of observations
Notes:
1
-Standard errors in parentheses.
2-The
first
part of Panel
B
includes
all
individuals not enrolled in the
September 2000. The second part includes
3-The average
collected
at
the
fair participation in the
fair.
Since only
registered, the participation
75%
all
in the
TDA)
non-treated department was obtained from the registration information
of the participants
was adjusted by
4-Demographic information and
TDA by
employes (enrolled or not
TDA
a proportionality factor.
participation are
all
obtained from administrative data
Table
PANEL
2:
A-
Differences in background characteristics, fair attendance and
Treated vs Untreated
in treated depts.
vs untreated depts.
vs untreated depts.
X T -X
X Tr X T
X T0-X
Xti-X
(1)
(2)
il)
(4)
-0.023
0.003
-0.027
-0.017
(.024)
(.015)
(.022)
(.025)
-0.169
-0.061
-0.089
-0.205
(.386)
(.252)
(.384)
(.362)
524
369
208
760
(964)
(561)
(847)
(1002)
-0.56
0.16
-0.52
-0.48
(.53)
(.33)
(.52)
(.49)
6144
4126
4124
4038
Age
Number of observations
0.158
0.138
0.090
0.231
(.021)
(.019)
(.02)
(.022)
6144
4126
4124
4038
Number of observations
C:
TDA PARTICIPATION
participation after
months
4.5
0.0097
-0.0068
0.0126
0.0075
(.0043)
(.0063)
(.0053)
(.0048)
5587
3726
3755
3693
Number of observations
TDA participation
1 1
after
months
0.0141
0.0023
0.0133
0.0153
(.0063)
(.0103)
(.0082)
(.007)
4879
3246
3271
3241
0.0070
0.1655
0.1465
Number of observations
PANEL
D:
Treated
FAIR ATTENDANCE
Fair attendance
PANEL
in treated dcpts.
BACKGROUND CHARACTERISTICS
Salary
TDA
Untreated
Treated depts.
Years of Service
B:
participation, by treatment status
vs untreated depts.
Sex (% male)
PANEL
TDA
RESPONSE TO THE QUESTIONNAIRE
0.1516
response rate to the questionnaire
(.0366)
(.0519)
(.0521)
(.0376)
907
765
295
754
Number of observations
Notes:
1
-Regression adjusted differences in means: departments were matched according to size and participation,
and
triplets
of departments of similar contribution
The regressions
control for the triplet to
2-Standard errors (reported
in
fair.
Since only
registered, the participation
75%
is
in the non-treated
at the
of the participants
TDA
participation are
restricted to individuals
who were
all
department level
department was obtained from the registration information
was adjusted by a proportionality
4-Demographic information and
5-The sample
and size were formed.
parentheses below the coefficient) are corrected for clustering
3-The average attendance or participation
collected at the
rate
which the department belongs.
factor.
obtained from administrative data
not enrolled in the
TDA before
the
fair.
Table
3: Individual
and
social effects
on
fair
attendance
Attended the
ols
Dummy
for received letter
Dummy
for treated
rv
iv
(2)
(3)
0.138
0.132
0.133
(.019)
(.019)
(.019)
0.090
department
(.022)
Average number of
letters in the
0.285
department
Average participation
fair in the
fair
(.072)
0.753
to the
department
(.094)
Observations
6144
6144
6144
Notes:
1- All
regressions control for the triplet to which the department belongs
2-Standard errors (reported
3-In the
is
in
parentheses) are adjusted for clustering
IV regressions, the instrument
treated, a
dummy
set consists
of a
dummy
for
at
the department level
whether the department
for whether the individual received the letters, and the triplet to
the department belongs.
which
Table
4:
IV
regressions: Effect of fair attendance on
T
vs
TDA participation
4.5
after
months
Number of observations
TDA participation
1 1
after
months
Number of observations
depts.
Letter vs non-letter Non-letter in
depts.
(y-ryo)/(frfo)
(1)
in
vs
depts.
T
depts.
Letter indiv.
vs
depts.
depts.
(yTi-yToVlfrrfro)
(y-n-yo)/(frrfo)
(2)
(3)
(4)
0.0568
-0.0446
0.1348
0.0300
(.0257)
(.0402)
(.0625)
(.0195)
5587
3726
3755
3693
0.0817
0.0142
0.1488
0.0599
(.0399)
(.0641)
(.102)
(.029)
4879
3246
3271
3241
Treated department
Instruments
T
(y-ro-yoVCfro-fo)
TLT0
T1,T0,0
Sample
TDA participation
Received
T0,0
letter
T1,0
Treated department
Received
Notes:
1-
Dependent variables are individual enrollment
2- Independent variable
is
in the
TDA 4.5 months and
1
lmonths
individual fair attendance
3- All regressions control for the triplet of the department
4- Standard errors (in parentheses) are corrected for clustering
at
the department level
after the fair
letter
Table
5:
Fair attendance and treatment effect in different groups
Difference Group T-Group
TDA
Fair attendance in Tl
participation
after 4.5
(letter recipients)
A:
DEPARMENT
PANEL
B:
after
1 1
months
(3)
SIZE
Below median (81)
Above median
TDA participation
(2)
(1)
PANEL
months
(81)
0.328
0.009
0.013
(.015)
(.0071)
(.0106)
985
2797
2403
0.235
0.009
0.015
(.0132)
(.0047)
(.0079)
1035
2790
2476
DEPARTMENT AVERAGE PARTICIPATION
Below median (34%)
Above median (34%)
IN
THE TDA BEFORE THE EXPERIMENT
0.259
0.006
0.013
(.0134)
(.0059)
(.009)
1062
2929
2523
0.304
0.013
0.016
(.0149)
(.0064)
(.0094)
958
2658
2356
PANEL C: GENDER
Women
Men
PANEL
D:
0.320
0.012
0.014
(.0134)
(.0071)
(.01 12)
1213
3298
2843
0.221
0.007
0.011
(.0146)
(.0071)
(.0086)
807
2289
2036
SALARY
Below Median ($34021)
Above Median ($34021)
0.269
0.001
0.015
(.0141)
(.006)
(.0088)
983
2745
2291
0.291
0.018
0.015
(.0141)
(.0065)
(.0104)
2842
2588
1037
1-The sample
2-Columns
2
in
column
and
3:
1
is
composed of individuals
in
group TI
Regression adjusted differences in means: department were matched according to size
and participation, and
triplets
The regressions control
of departments of similar contribution rate and size were formed.
for the triplet to
3-Standard errors (reported
in
which the department belongs.
paretheses below the coefficient) corrected for clustering
at the
department level
Table
6:
Comparison with naive estimates:
effect of the fair
on
TDA enrollment
Overall effect
Individual effect
IV
OLS
IV
"Naive" IV
(1)
(2)
(3)
(4)
-0.045
0.016
0.057
0.001
(.04)
(.011)
(.026)
(.026)
3726
1832
5587
5587
0.014
0.052
0.082
0.042
(.064)
(.018)
(.04)
(.039)
A. Participation after 4.5 months
Fair attendance
B. Participation after 11
months
Fair attendance
3246
Sample
1608
Treated
individuals
departments
Received
Instrument
who
4879
4879
Complete
Complete
received the letter
sample
NONE
Treated department
letter
sample
Received
letter
Notes:
1-
Dependent variables are individual enrollment
2- Independent variable
is
in the
TDA 4.5
months and
1 1
months
after the fair
individual fair attendance
3- All regressions control for the triplet of the department
4- Standard errors (in parentheses) are corrected for clustering at the department level
5-
The sample
in
column
(1)
is
limited to individuals (not enrolled in the
departments. The sample in column (2)
TDA by
by
Sept. 2000).
Sept. 2000.
The sample
in
is
limited to individuals
column
(3)
and
TDA by Sept. 2000) in Treated
who received the
letter (not enrolled in the
(4) is limited to individuals not enrolled in the
TDA
Table
of the fair on attitudes and knowledge
7: Effect
Treated departments
Control
(Received invitation)
Treament
(1)
Difference
(2)
(3)
A. Fair participation and impressions
Fair participation
Number of observations
Fair satisfaction (for those
who
attended the fair
B. Response to the question
Not enough information
Cannot afford
"Why are you
it
0.286
0.140
(.054)
(.064)
301
70
371
0.849
0.950
-0.101
(.027)
(.05)
(.047)
0.306
-0.107
currently not enrolled in the
0.200
to save for retirement
Plan to enroll soon but no time to do
0425
(.029)
yet
Other ways to save for retirement
TDA?"
(.025)
(.059)
(063)
0.328
0.371
-0.043
(.029)
(.062)
(.075)
0.446
0.355
0.091
(.031)
(.061)
(.07)
0.220
0.242
-0.022
(.026)
(.055)
(.063)
255
62
317
C. Enrollment 6 months after the questionnaires
Individuals
who
report that
they plan to enroll soon
Individuals who did not report that
they plan to enroll
0.099
0.167
-0.067
(.029)
(.09)
(.096)
0.020
0.000
0.020
(.013)
(.01)
D. Response to the question "where do you obtain information about benefits?"
Benefits fair
Benefits information packet
Personal
visit to the
BO
Other information seminar
Colleagues
Family or friends
Administrative officer
E.
Knowledge about
TDA status
Reported that she knew the
number of vendors with which she
Gave the correct answer about TDA status
the correct
0.254
0.117
(.052)
(.054)
0.771
0.930
-0.158
(.024)
(.031)
(.039)
0.123
0.085
0.038
(.019)
(.033)
(.05)
0.204
0.211
-0.007
(.023)
(.049)
(.049)
0.252
0.310
-0.058
(.025)
(.055)
(.053)
0.265
0.239
0.026
(.026)
(.051)
(.051)
0.049
0.014
0.035
(.012)
(.014)
(.025)
300
71
371
benefits
Reported that she knew her
Gave
0.370
(.028)
answer about pension plan
0.938
0.986
-0.048
(.014)
(.014)
(.022)
0.738
0.714
0.024
(.029)
(.061)
(.058)
-0.056
0.887
0.944
(.018)
(.028)
(.033)
0.603
0.607
-0.004
(.032)
(.066)
(.069)
235
56
291
Notes
1-A11 statistics are weighted by population weight
2-Standard errors of the difference corrected for clustering
3-Sample
is
restricted to treated
deparments
at
the department level
October 31, 2000
Name
Line
1
Line 2
City state zip
Dear Name:
You have just
received your
Open Enrollment packet from
the Benefits Services Group, inviting
you
to the
Benefits Fair 2001.
The
Fair will be held in
two
locations:
November
llam-2:30pm
7,
ADDRESS ERASED
November
8,
1
lam - 2:30pm
ADDRESS ERASED
This year, as part of a study (conducted jointly by the Benefits Services Group and economics researchers)
impact of the Fair on benefits choices, we are offering a reward of $20 to 2,000
employees, just for attending the Fair. Funding for these rewards was contributed from a research grant.
We selected those employees by a simple lottery, and your name was among those drawn.
to better understand the
In order to receive this
name
at
weeks following the
you
that the
you have
a note
find the Fair helpful in
reward
is
to
do
is
to
Sincerely yours,
of the Benefits Office
Associate Director
come
main
and give your
check within the two
to the Fair with this letter,
hall.
You
will receive a
making your
benefits choices.
However, we want to
completely independent of your benefits decisions.
of these dates (November 7 or November 8)
there.
Name
all
Fair.
We hope that you will
emphasize
Make
$20 reward,
the registration table that will be located in the
in
your calendar, and
we
look forward to seeing
st
April
1
,
2001
Name
Line
1
Line 2
City state zip
Dear Name:
We are
currently studying whether benefits fairs, along with other
information, convey the necessary information to
we would
members of the
way of obtaining
university community.
you a few questions about your
experience in obtaining information on the university retirement plans. If you could take
a few minutes to complete the questionnaire attached to this letter, your response would
be greatly appreciated. Your responses will be strictly confidential and will not be used
In the context of our study,
for
any purpose other than the study.
like to ask
You may
mail your responses in the envelope
provided.
we
$10 Macy's
As
a token of our appreciation,
we
receive the completed questionnaire. Please return the questionnaire on or before
15.
Sincerely,
will send
you
a
gift certificate
when
May
First
name, Last name
Please answer the following 6 simple questions.
if you are
be used
not sure of an answer.
no purpose other than
for
(1) In addition to
Your answers
You
will
can check the "don't know" answer
remain strictly confidential and will
this study.
your Basic Retirement Account, the university makes a monthly
contribution of 3.5% of your monthly salary to an Individual Investment Account(s).
decide
how
this contribution
should be invested from a
Through how many investment companies
are
list
You
of four investment companies.
you currently investing
this contribution?
-One....
-Two....
-Three
-Four
-Don't
(2)
know
The university
offers a supplemental retirement plan called the Tax-Deferred
(TDA) program. Through
Account
TDA
program, you can add to your retirement savings by
contributing a portion of your salary on a pre-tax basis. You pay no taxes on these
savings or the investment income until you withdraw your funds. You decide how much
to contribute
choose
how
the
and the university deducts your contributions from your paycheck. You
to invest your savings from a wide range of funds offered by four different
vendors
You
are not automatically enrolled in the
Are you currently enrolled
-Yes
in the
TDA program.
Tax-Deferred Account (TDA)?
(go to question 4)
-No
-Don't
(3) [To
be
Why
know
only if you are not currently enrolled in the TDA]
you currently not enrolled in the TDA (check all answers
filled out
are
-You do not have enough information on the TDA:
-Right now, you cannot afford to save for your retirement:
-You plan to enroll soon, but did not have the occasion to do
-You save for your retirement through other means:
(NEXT PAGE, PLEASE)
it
yet:
that apply)?
First
name,
last
name.
(3b) If you check the last answer,
which other means are you using
to save for
retirement:
-TDA through
spouse's employer:
-Individual Retirement
Account (IRA):
-Employer provided pension plan (own):
-Employer provided pension plan (spouse):
-Other mutual funds:
-Other
(4)
[To be
filled out
From which of the
(check
all
by everybody]
following sources do you get information about the retirement plans
that apply)?
-The benefits information
fair:
-Benefits information packet:
-You came in person to the Benefits office:
-You attended an information seminar:
-Colleagues:
-Family or
friends:
-The Administrative Officer of your department:
-None
(5)
Did you attend the
benefits information fair in the fall?
-Yes:
-No:
(6) If
you
-Yes:
-No:
did, did
you
find
it
useful?
MIT LIBRARIES
3 9080 02246 2334