Information about http://research.yale.edu/karlan/downloads/CARES.pdf
Put Your Money Where Your Butt Is: …
Tags: behavioral economics, contract commitment, control group, esther duflo, field experiments, gin, group treatment, jel codes, karlan, nicotine, percentage points, project management assistance, rosenthal, scott nelson, seminar participants, smokers, smoking cessation, urine test, voluntary commitment, zinman,Pages: 23
Language: english
Created: Sat Aug 2 06:17:10 2008
Display cached document
Page 1
Page 2
Page 3
Page 4
Page 5
Page 6
Page 7
Page 8
Page 9
Page 10
Page 11
Page 12
Page 13
Page 14
Page 15
Page 16
Page 17
Page 18
Page 19
Page 20
Page 21
Page 22
Page 23
Put Your Money Where Your Butt Is:
A Commitment Savings Account for Smoking Cessation*
Xavier Giné
World Bank
Dean Karlan
Innovations for Poverty Action
Jonathan Zinman
Innovations for Poverty Action
August 2008
Abstract
We designed and tested a voluntary commitment product to help smokers quit
smoking. The product (CARES) offered individuals a savings account in which
they deposit funds for six months, after which they take a urine test for nicotine
and cotinine. If they pass, their money is returned; otherwise, their money is
forfeited to a charity of the bank's choosing. Smokers randomly offered CARES
were approximately 3 percentage points more likely to pass the 6-month test than
the control group. Surprise tests at 12 months, 6 months after the account was
closed, indicate an effect on lasting cessation: those offered CARES were still at
least 3 percentage points more likely to pass the surprise test than the control
group. Treatment-on-the-treated estimates suggest that CARES usage increased
the likelihood of smoking cessation by 30 percentage points or more.
Keywords: commitment contract; commitment device; public health; addictive
consumption; intertemporal choice; behavioral economics; field experiments
JEL codes: D12, I12
*
Thanks to the management and staff of Green Bank for their cooperation, and to Paulette Cha, Kareem
Haggag, Scott Nelson, and especially Tomoko Harigaya for research and project management assistance.
Thanks to Satish Chand, Stefano DellaVigna, Esther Duflo, Meredith Rosenthal, and conference and
seminar participants at the American Society of Health Economists biennial meeting, Australian National
University, Case Western, and NBER Summer Institute for comments. Thanks to the World Bank and
Innovations for Poverty Action for funding. Disclosures: Karlan is President of stickK.com, which offers
online commitment contracts. Zinman is on the Research Advisory Board of stickK.com.
I. Introduction
A rich theoretical literature shows that consumers of addictive substances who face self-control or
temptation problems will seek to voluntarily constrain their future consumption choices: they will
demand commitment devices.1 Yet there is little field evidence on the demand for or effectiveness
of such commitment devices.
We take some initial steps toward addressing the empirical viability and effectiveness of
commitment devices for smoking cessation, using evidence from a field experiment in the
Philippines. Some smokers were randomly assigned an opportunity to voluntarily sign a
commitment contract (branded Committed Action to Reduce and End Smoking, or "CARES") to
stop smoking. A smoker signing the contract pledged his own money that he would pass a
cotinine (the primary metabolite of nicotine) urine test six months later.2 If the CARES client
passed the urine test he got his money back (no interest accrued on the account). If he failed the
test the local bank offering the savings product donated the money to charity. This is essentially
the performance bond contract suggested in Gruber and Koszegi (2001). A second treatment
group received "cue cards," visually aversive wallet-sized pictures that are modeled on Canada's
mandated cigarette packaging and intended to regularly remind smokers of the health risks from
smoking.
Eleven percent of smokers offered the CARES contract signed up. This is comparable to
takeup rates for a leading "self-help" treatment: nicotine replacement medications (patch, gum,
inhaler, or nasal spray).3 The average client made a deposit every two weeks and ended up
committing 550 pesos ($11 USD) by the end of the six-month contract period. 550 pesos is about
20% of monthly income4 and roughly equal to the average out-of-pocket expense for about 6
months' worth of cigarettes incurred by CARES clients at baseline.
Our results suggest that CARES helps smokers quit. Smokers randomly offered CARES were
an estimated 3.3 to 5.8 percentage points more likely to pass the 6-month urine test than the
1
See Gruber and Koszegi (2001), Laibson (2001), O'Donoghue and Rabin (2002), Bernheim and Rangel
(2004), and Gul and Pesendorfer (2007). In contrast, standard neoclassical models of intertemporal choice
do not predict a demand for commitment. Becker and Murphy (1988) model the consumption of addictive
substances along the lines.
2
The testing protocol has limitations, detailed below, but has been used by public health campaigns and
tests of other treatments, including Volpp et al (2006) and some of the randomized trials of nicotine
replacement medications summarized in Stead et al (2008).
3
Seventeen percent of smokers U.S. smokers reported using nicotine replacement medication during the
last 12 months in a nationally representative 2001 phone survey (Bansal, Cummings, Hyland and Giovino
2004). In the only study we know of from the Philippines, only six percent of a sample of relatively heavy
smokers who had already decided to quit had ever used any form of nicotine replacement therapy in past
smoking cessation attempts (Tipones and Fernandez 2006).
4
Income is very roughly estimated from marketer observations of subject appearance and work activity.
control group. But this urine test is not necessarily a good indicator of a lasting spell of smoking
cessation, since the 6-month test date was scheduled up to 4 weeks in advance, and the test could
be passed by abstaining from smoking for as little as a few days before the test date. So we also
worked with the bank offering CARES to conduct surprise 12-month tests that would provide
sharper evidence on true quits (vs. short-term, strategic ones). The 12-month results show that
smokers randomly offered CARES were 3.5 to 5.7 percentage points more likely to pass the test
than the control group. The analogous treatment-on-the-treated estimates are 31 to 53 percentage
points.
The effect of CARES on smoking quits appears to be large. The sample mean pass rate for
the surprise test was only 18% in the control group. One can also compare the effect of CARES
to other treatments. Within-sample we find little evidence that the aversive cue cards affect
smoking quits, and the upper bound of the cue card 12-month treatment-on-the-treated confidence
interval implies an increased likelihood of surprise test passage that is 1/8 of our the comparable
point estimate on CARES. The results also suggest that CARES has effects that are comparable to
other treatments that have been tested using randomized trials on other samples. Volpp et al
(2006) find that modest financial bonuses offered through a U.S. Veterans Affairs hospital
increase short-term cessation but not lasting quits. Over-the-counter nicotine replacement
medications have been tested in dozens of randomized trials and generally produced treatment-
on-the-treated effects that are smaller than those found here for CARES (Stead, Perera, Bullen,
Mant and Lancaster 2008).
Despite its large treatment effects a surprisingly large proportion of smokers who voluntarily
commit with CARES, 66%, ended up failing to quit. This is consistent with various behavioral
bias in preferences and/or expectations (partial naiveté about dynamic inconsistency, projection
bias, over-confidence), and the implications of such biases for optimal contract design and
treatment effectiveness is an important topic for future research.
The results in this study are unusually direct evidence on the takeup and effectiveness of a
commitment device for managing the consumption of an addictive substance. The only
comparable studies we know are Paxton's (1979; 1980; 1982). These studies have three key
differences from ours. First, they were administered in a highly structured and clinical setting to
smokers who were already participating in a smoking cessation program. Our study includes
smokers of varying smoking intensities and ex-ante dispositions toward cessation aids. Second,
Paxton's control groups received a rich set of other smoking cessation aids, including counseling,
social pressure, and aversion therapy. Our study takes a more over-the-counter approach and
compares the effects of CARES to a control group that receives nothing other than basic
information. Third, Paxton's analysis does not exploit random assignment.5
Our study also relates to prior work on commitment devices for other decisions that may
involve self-control problems. Ariely and Wertenbroch (2002) find that 37 of 51 MBA students
elect to impose binding deadlines on themselves for completing class assignments. Deadlines
improve task performance but students do not necessarily set them optimally. Thaler and Benartzi
(2004) and Ashraf, Karlan, and Yin (2006) design new commitment products for savings and find
high takeup rates and large treatment effects.6
Our paper proceeds as follows: the next section describes the voluntary commitment savings
product that we designed for smokers who want to quit smoking. Sections III describes the cue
cards treatment. Section IV details the experimental design and implementation by Green Bank in
the Philippines. Section V reports the results of the study. Section VI concludes.
II. CARES Product Design
Committed Action to Reduce and End Smoking ("CARES") is a voluntary commitment savings
program specifically designed for smokers who want to quit smoking. The basic design of the
product allows a smoker to risk a self-selected amount of his own money that will be forfeited
unless he passes a biochemically verified test of smoking cessation, administered as a urine test
of nicotine and cotinine byproducts, at six months after signing the commitment contract. The
particular product design and study described below was implemented by the Green Bank of
Caraga, on the island of Mindanao in the Philippines.
Green Bank marketed CARES by sending bank representatives into the street to target
obvious smokers. Details on the marketing are described with the experimental design below (in
Section IV).
Green Bank required a minimum balance of 50 pesos (~= $1USD), collected by the field
marketers, to open a CARES account. Marketers encouraged smokers to deposit the money they
would normally expect to spend on cigarettes into a savings account every week for six months.
The savings account did not yield any interest-- this is an important feature for the bank to
prevent non-smokers from opening the account merely because of the convenience of deposit
collection services. The bank offered some randomly-selected individuals weekly deposit
5
Paxton randomized subjects into different arms but then estimates treatment effects by comparing those
who tookup the commitment product to the control group.
6
See DellaVigna (2007) for a more comprehensive review of field evidence on commitment devices.
collection; the remaining CARES clients had to go to a branch to make deposits beyond the
opening one.7
Clients could only make deposits, and not withdrawals, from the CARES account during the
six month commitment period. Hence all deposited funds were at risk. Clients who passed the
six-month urine test got their entire balance back. Clients who failed (or did not take) the test
forfeited their entire balance.
Trained Green Bank technicians test CARES clients' smoking status using the NicCheckTM
urine strip test for nicotine and its primary metabolite, cotinine.8 NicCheck has been used in
previous anti-smoking programs, including the Dutch Cancer Society's "Quit and Win"
campaign, and the financial bonus incentive testing in Volpp et al (2006). The test result provides
a categorical measure of recent nicotine consumption, with values ranging from zero (no
exposure) to fifteen (high exposure).9 Green Bank counts only a zero result as passing, and both
marketers emphasized that clients must stop smoking completely in order to be sure of passing
the test.
Green Bank contacts each client three to four weeks prior to his six-month deadline to set up
a urine testing appointment. If a client can not be reached initially the Bank makes repeated
attempts to set up a test date within one week of the maturity date. If a client is deemed unable to
take the test within the stipulated one-week grace period due to mitigating circumstances (e.g.,
working in another location), he is allowed an additional three weeks to take the test. If the client
was reached and refused to schedule a date, the account balance was forfeited one week after the
six-month commitment date.
III. Cue Cards Treatment Design
7
Clients lose the weekly deposit collection service if they miss three consecutive deposits.
8
Initially CARES clients were required to take a urine test at a nearby hospital lab. But given the costs and
delays associated with lab based testing for nicotine and cotinine (the metabolite of nicotine) levels in
blood, Green Bank employed the more feasible and cost-effective urine strips for nicotine and cotinine
analyses. NicCheck product specifications indicate that urine strips sacrifice a bit of test specificity (the
ability to detect a true negative result, which is 97% for urine strip versus 99% for lab-based cotinine
analysis), but offer equivalent test sensitivity (the ability to detect a true positive result, which is roughly
97% for both urine strips and lab-based cotinine analysis) and the ability to provide results in the field,
within 15 minutes. Green Bank found similar specificity (one false positive out of 18 self-reported non-
smokers) and much lower sensitivity in its own pilot testing, where marketers randomly approached people
on the street in our study area, asked if they were smokers, and then offered 30 pesos to take the urine strip
test.
9
Small and portable test strips are dipped into the urine sample, stimulating a chemical reaction that
changes the test strip's color. The color result ranges from white (no nicotine exposure), to light pink
(moderate nicotine exposure), to red (high nicotine exposure). The test administrator then compares the
test strip's color to a NicCheck color scale and assigns the test result a number ranging from 0 (no
exposure) to 15 (high exposure).
The cue cards are pocket-sized, graphic depictions of the negative health consequences of
smoking. Each individual received one of four pictures: a premature baby (with text "Smoking
harms unborn babies"), bad teeth (with text "Smoking causes mouth and throat cancer"), black
lung (with text "Smoking causes lung cancer"), or a child hooked up to a respirator (with text
"Don't let children breathe your smoke"). By law, such images must be featured on cigarette
packages in Australia, Canada, and New Zealand (Hoek and Gendall 2005). Smokers assigned to
the Cues treatment were offered their choice of the above cards, and encouraged by the marketers
to keep them handy and/or post them in locations where the subject tended to smoke. More than
99% of subjects offered the cue cards accepted them.
IV. Experimental Design
Our study sample consists of 2,000 smokers aged 18 or older who reside on the island of
Mindanao in southern Philippines. Green Bank marketers identified smokers by approaching
people and asking them whether they smoke regularly. If they did, the marketer then asked if
they wanted to participate in a short survey on smoking. All subjects received an informational
pamphlet on the dangers of smoking, and a tip sheet on how to quit. Since the primary objectives
of this study were to determine whether first there was demand for CARES, and second whether
CARES increased smoking cessation, the marketers only collected very quick and basic baseline
data on age and smoking status (see Section V-A for more details).
The experiment was implemented in three distinct waves of marketing. The first two waves
took place in Butuan City from August to December 2006. After completing the baseline survey
marketers revealed a sticker on the back of the survey that randomly assigned the subject to one
of four groups: (1A) CARES with deposit collection, (1B) CARES without deposit collection, (2)
Cues, or (3) Control.10 The probability of assignment to groups was initially 45%, 45%, 5%, and
5%. After establishing that there was sufficient takeup of CARES, Green Bank changed the
assignment probabilities to 15%, 15%, 30%, and 40% for the second wave. 418 smokers were
surveyed (and hence drawn into the sample frame) in the first two waves. Of the 266 assigned a
CARES offer, 34 took the product. Two individuals from the Cues group also opened an account
10
In the first wave there were 20 situations in which marketers interviewed respondents with either one or
two others present; in these cases, marketers were instructed to interview all individuals in the group before
disclosing the random assignment. All respondents in the group received the same assignment as the first
interviewee. Impact results discussed below correct standard errors for any clustering within groups of
individuals that received joint marketing.
(after hearing about the product and approaching bank staff). In our analysis we code these
individuals in the Cues group, in adherence to the random assignment.
The third marketing wave ran from February to May 2007, in the neighboring town of
Ampayon. Here Green Bank implemented new randomization procedures designed to produce
even better compliance with the randomized treatment assignment. Now marketers used a
calculator to solve an equation based on the subject's birth date (the residual of dd + mm + yy,
divided by three). The individual was then assigned to CARES group if the residual was zero, to
Cues if the residual was one, and to Control if the residual was two. Given the low takeup in the
CARES group without deposit collection in the first two waves, all respondents in the Ampayon
CARES group were offered deposit collection service. 49 of the 515 Ampayon subjects offered
CARES opened the account.
In order to validate the quality and accuracy of information provided by the marketers, field
staff from Innovations for Poverty Action conducted spot-checking visits with randomly selected
respondents who had been offered CARES. More than 90% of the clients accurately described
the main features of the product design.
Given the random assignment, we expect individuals who end up in treatment and control
groups to have statistically indistinguishable baseline characteristics on average. Table Ia presents
related evidence. The F-statistic from a regression of assignment to CARES on all baseline
covariates is 0.42 (p-value of 0.963), and for assignment to Cues is 0.54 (p-value of 0.903).
When we examine individual variables across the CARES and Control groups, 12 out of 13 are
similar statistically, and only one variable fails at the 10% level: 95.4% in the CARES group
reported experiencing specific situations that make them want to smoke, whereas only 92.8% of
control individuals reported the same. The Cues treatment individuals are similar statistically to
the control in 10 out of 13, with the significant differences found on "wanting to stop smoking
sometime in your life," "wanting to stop smoking in 1 year" and "will actually quit smoking in 6
months." These variables may also be correlated with smoking cessation, so we estimate
treatment effects with the full set of baseline covariates as control variables.
Six months and 12 months after the initial marketing, the bank attempted to administer the
urine test to all study subjects (testing procedures are detailed in Section II). CARES clients had
to take the six-month test or automatically forfeit their deposit balance. Non-clients (including
those assigned to the cues and control groups) were paid 30 pesos (60 cents US) for taking the
six-month test, and everyone in the sample frame was paid 30 pesos for taking the 12-month test.
Table Ib Panel A shows that the bank reached 63% of those in the baseline for the six-month
urine test, with no difference in contact rate across the three treatment and control groups). Of
those contacted 95% agreed to take the test. Since we find lower agreement in the CARES group
(93% vs. 97% in the control) we report six-month treatment effects under alternative assumptions
about the smoking status of those who refused to take the test.
Table Ib Panel B shows that the bank reached 60% of those in the baseline for the 12-month
urine test, with no difference in contact rate across the three treatment and control groups). Of
those contacted 95% agreed to take the test, again with no differences across groups.
V. Results
A. CARES Takeup
In total, 83 out of 781 (11%) individuals offered CARES signed a contract. Table Ib Columns 7-9
shows univariate analysis of the takeup decision from data on the limited set of characteristics
marketers collected in the quick baseline survey administered prior to treatment assignment and
marketing.11 The following baseline characteristics were positively correlated with taking up
CARES: wanting to quit (at some pont in life, or now), optimism about quitting (as indicated by
responding yes to "will you quit smoking in the next year?"), and pre-existing strategic behavior
in managing one's cravings (as indicated by responding yes to "do you try to avoid areas or
situations that make you want to smoke?"). Negative correlates with CARES takeup were:
wanting to quit smoking more than a year in the future (perhaps an indicator of procrastination)
and smelling like cigarettes (likely an indicator of heavy smoking). Table II shows multivariate
estimation of takeup correlates.12 The main results here are that the full set of baseline
characteristics are jointly significant but explain only about 10% of the variation in the takeup
decision.
B CARES Usage
Table 3 shows some summary statistics on CARES deposits.
Opening balances were 57 pesos on average: this is four times the monetary value of the
number of cigarettes the client reported smoking per week. Ninety percent of clients opened with
the minimum amount of 50 pesos. Eighty percent of clients then made additional contributions.
On average CARES clients made a deposit every two weeks, and by six months the average
balance grew to 553 pesos. Given self-reported smoking intensity and a per-cigarette cost of one
11
Only a handful of the 2,000 subjects were existing Green Bank clients. Marketers did not elicit income
directly, but their observation of subject appearance and work activity indicate the majority were self-
employed.
12
All takeup and impact regressions include indicator variables for the three marketing waves.
peso, the average CARES client committed roughly six months worth of cigarette spending to the
account.
Not surprisingly CARES clients who used the account more intensively were more likely to
pass the urine tests. We show results for the 6-month test in Table 3 and the 12-month test in
Appendix Table 1. Successful clients made more deposits, were more likely to retain deposit
collection services by making regular deposits, and had larger balances at contract maturity.
These differences were more pronounced for 6-month test passage than for 12-month passage. Of
course, since contract terms and deposit requirements were not randomized, we can not conclude
a causal relationship between deposit amount, deposit regularity and success.
C. Treatment Effects on Smoking Cessation
We estimate intent-to-treat (ITT) effects of CARES and cue cards on test passage using the OLS
specification:
(1): passit = + caresi + cuesi + Xi + Wi + i
Where i indexes individuals, t refers to the 6-month or 12-month test, pass, cares and cues are all
binary variables, X is the vector of baseline covariates, and W is a vector of dummies for the three
marketing waves. We report these results in Table 4, Panel A. We also estimate (1) using probit
instead of OLS (Appendix Table 2), and after dropping the baseline covariates (Appendix Table
3), and find very similar results.
Each table reports results on 6-month test passage in odd columns, and on 12-month test
passage in even columns. We estimate effects under three different assumptions on clients for
whom we do not have a test result: i) these clients would have failed the test (Columns 1 and 2),
ii) these clients have the average pass rate; i.e., we drop these clients (Columns 3 and 4), iii) these
clients have the average pass rate, unless they were found by the technician and refused to take
the test, in which case we assume they would have failed (Columns 5 and 6).13
Table 4 Panel A shows CARES ITT effects on 6-month test passage of 3 to 6 percentage
points under these assumptions. These effects are large relative the control group sample mean
passage rates of 0.08 to 0.12. The effects on 12-month test passage, which as discussed above are
probably a better measure of effects on a lasting quit spell, range from 4 to 6 percentage points.
Again these effects are large relative to the control group sample mean passage rates of 0.10 to
0.18. We do not find any significant effects of the cue cards.
13
Six test strips turned blue (off the NicCheck results spectrum) in each of the six- and twelve-month
follow-up pools. This is the likely due to the TB medicine Isoniazid. We coded these blue strips as failures,
but Green Bank returned the commitment balance to the one CARES client with a blue result.
Table 4 Panel B shows treatment-on-the-treated (ToT) results, using random assignment to
CARES as an instrument for takeup. The ToT estimates imply 30 to 65 percentage point increases
in test passage. This suggests that CARES usage increases by several fold the probability of test
passage and a lasting quit spell.14
Appendix Table 4 reports the same specifications for the sub-sample of smokers that reported
wanting to quit smoking at some point in their life in the baseline survey (Appendix Table 5
reports summary statistics for this sub-sample). The CARES point estimates suggest somewhat
larger treatment effects for this sample. We also find some significant increases in 6-month test
passage from the cue cards, but no significant effects at 12 months.
VI. Conclusion
We designed a commitment product to help people quit smoking and tested it in cooperation with
Green Bank using a randomized controlled trial in the Philippines.
The results suggest that Committed Action to Reduce and End Smoking ("CARES") helps
smokers quit. At the end of the commitment contract period (6-months), subjects offered CARES
contract were 3 to 6 percentage points more likely to pass a urine test for short-term smoking
cessation than the control group. This intent-to-treat effect persisted at a surprise urine yet six
months later (12 months after the contract offer): smokers offered CARES were 4 to 6 percentage
points more likely to pass the 12-month test. Treatment-on-the-treated estimates suggest that
those who signed a CARES commitment increased their probability of test passage and a lasting
quit spell by several fold.
These results suggest that the CARES product may be an unusually effective treatment for
smoking cessation. We do not know of any comparable trials on other treatments in the
Philippines, but the CARES treatment effects compare favorably to those found for nicotine
replacement therapy in randomized controlled trials in other settings (Stead, Perera, Bullen, Mant
and Lancaster 2008). The CARES takeup rate (11%) also compares well to nicotine replacement
therapy (Bansal, Cummings, Hyland and Giovino 2004; Tipones and Fernandez 2006),
suggesting that commitment contracts could help public health efforts to address the "under-use"
of smoking cessation treatments (Cokkinides, Ward, Jemal and Thun 2005; Orleans 2007).
Nevertheless the majority of CARES clients in our study failed to quit, suggesting that there is
still much to be done in improving the effectiveness of smoking cessation treatments.
14
The cue card treatment-on-the-treated estimates are insignificant and nearly identical to the intent-to-treat
because of nearly 100% takeup of the cue cards.
We suggest four main areas for further research. One is estimating longer-term treatment
effects. Two is testing whether commitment contracts complement or substitute for other smoking
cessation treatments. Three is studying the optimal design of an anti-smoking commitment
contract.15 To highlight just one aspect of product design, note that in our study CARES was
largely bundled with deposit collection services. Hence we cannot yet unpack how much of the
treatment effect was due to the financial punishment, and how much was due to frequent contact
with the deposit collector (a sub-question here is the necessary frequency of such contact in order
to change behavior). A fourth and closely related question is what drives the takeup decision. If
behavioral biases such as loss aversion, partial naiveté, projection bias, and/or over-optimism
play a key role then there may be implications for product design (e.g., strong defaults) and
marketing (e.g., framing, information on failure rates). Strong interplay between theory and
empirics will be needed to continue developing and disseminating commitment products.
15
For theories of optimal contracting with consumption commitments see, e.g., DellaVigna and
Malmendier (2004), and Eliaz and Spiegler (2006).
REFERENCES
Ariely, D. and K. Wertenbroch (2002). "Procrastination, Deadlines, and Performance:
Self-Control by Pre-Commitment." Psychological Science 13(3): 219-224.
Ashraf, N., D. Karlan and W. Yin (2006). "Tying Odysseus to the Mast: Evidence from a
Commitment Savings Product in the Philippines." Quarterly Journal of
Economics 121(2): 635-672.
Bansal, M., M. Cummings, A. Hyland and G. Giovino (2004). "Stop-smoking
medications: Who uses them, who misuses them, and who is misinformed about
them?" Nicotine & Tobacco Research 6(Supplement 3): S303-S310.
Becker, G. and K. M. Murphy (1988). "A Theory of Rational Addiction." Journal of
Political Economy 96(4): 675-700.
Bernheim, B. D. and A. Rangel (2004). "Addiction and Cue-Triggered Decision
Processes." American Economic Review 94(5): 1558-1590.
Cokkinides, V., E. Ward, A. Jemal and M. Thun (2005). "Under-Use of Smoking
Cessation Treatments: Results from the National Health Interview Survey, 2000."
American Journal of Preventive Medicine 28(1): 119-122.
DellaVigna, S. (2007). Psychology and Economics: Evidence from the Field. Working
Paper.
DellaVigna, S. and U. Malmendier (2004). "Contract Design and Self-Control: Theory
and Evidence." Quarterly Journal of Economics 119(2): 353-402.
Eliaz, K. and R. Spiegler (2006). "Contracting with Diversely Naive Agents." Review of
Economic Studies 73(3): 689-714.
Gruber, J. and B. Koszegi (2001). "Is Addiction "Rational"? Theory and Evidence." The
Quarterly Journal of Economics 116(4): 1261-1303.
Gul, F. and W. Pesendorfer (2007). "Harmful Addiction." Review of Economic Studies
74(1): 147-72.
Hoek, J. and P. Gendall (2005). Pictorial Health Warnings: A Review of Research
Evidence. Research Report Prepared for Ministry of Health.
Laibson, D. (2001). "A Cue-Theory of Consumption." Quarterly Journal of Economics
116(1): 81-119.
O'Donoghue, T. and M. Rabin (2002). Addiction and Present-Biased Preferences.
Working Paper.
Orleans, C. T. (2007). "Increasing the Demand for and use of Effective Smoking
Cessation Treatments." American Journal of Preventive Medicine 33(6S): S340-
S348.
Paxton, R. (1979). "The Effects of a Deposit Constract as a Component in a Behavioural
Programme for Stopping Smoking." Behaviour Research and Therapy 18.
Paxton, R. (1980). "Deposit Contracts with Smokers: Varying Frequency and Amount of
Repayments." Behaviour Research and Therapy 19: 117-123.
Paxton, R. (1982). "Prolonging the Effects of Deposit Contracts with Smokers."
Behaviour Research and Therapy 21(4): 425-433.
Stead, L., R. Perera, C. Bullen, D. Mant and T. Lancaster (2008). "Nicotine Replacement
Therapy for Smoking Cessation." Cochrane Database of Systematic Reviews(1).
Thaler, R. and S. Benartzi (2004). "Save More Tomorrow: Using Behavioral Economics
to Increase Employee Saving." Journal of Political Economy 112(1, Part 2
Supplement): S164-87.
Tipones, R. and L. Fernandez (2006). "Predictors of Smoking Cessation 1 Year After
Enrollment in a Smoking Cessation Program in a Tertiary Hospital." Philippine
Journal of Internal Medicine 44: 7-12.
Volpp, K. G., A. G. Levy, D. A. Asch, J. A. Berlin, J. J. Murphy, A. Gomez, H. Sox, J.
Zhu and C. Lerman (2006). "A Randomized Controlled Trial of Financial
Incentives for Smoking Cessation." Cancer Epidemiology, Biomarkers &
Prevention 15(1): 12-18.
Table 1a. Summary Statistics, Baseline Variables
Baseline Measures
CARES Group
t-test of (2) t-test of (3) Did Not t-test of
All CARES Cues Control vs (4) vs (4) Took up Takeup (7) vs (8)
(1) (2) (3) (4) (5) (6) (7) (8) (9)
Female 0.058 0.061 0.599 0.053 0.525 0.606 0.072 0.069 0.905
(0.005) (0.009) (0.010) (0.009) (0.029) (0.010)
Age 36.571 36.951 35.667 36.972 0.978 0.101 38.341 37.181 0.465
(0.310) (0.493) (0.547) (0.576) (1.367) (0.520)
Number of cigarettes per day in the past 7 days 14.531 14.184 15.051 14.461 0.611 0.344 14.122 14.067 0.962
(0.234) (0.350) (0.463) (0.416) (1.105) (0.369)
Estimated amount spent on cigarettes per week (pesos) 101.715 99.287 105.351 101.227 0.611 0.344 98.854 98.472 0.962
(1.637) (2.453) (3.239) (2.915) (7.732) (2.586)
Tried to stop smoking in the past 12 months 0.457 0.446 0.452 0.476 0.277 0.417 0.422 0.427 0.927
(0.011) (0.018) (0.020) (0.020) (0.055) (0.019)
Wants to stop smoking sometime in life 0.723 0.725 0.690 0.754 0.219 0.013 0.855 0.723 0.010
(0.010) (0.016) (0.019) (0.017) (0.039) (0.017)
Wants to stop smoking now 0.168 0.178 0.144 0.179 0.957 0.099 0.289 0.159 0.003
(0.008) (0.014) (0.014) (0.015) (0.050) (0.014)
Wants to stop smoking in 1 year 0.426 0.431 0.393 0.452 0.420 0.037 0.494 0.426 0.234
(0.011) (0.018) (0.020) (0.020) (0.055) (0.019)
Wants to stop smoking after 1 year 0.106 0.095 0.126 0.100 0.721 0.159 0.036 0.113 0.030
(0.007) (0.010) (0.014) (0.012) (0.021) (0.012)
Will actually quit smoking in 6 months 0.523 0.537 0.473 0.555 0.493 0.004 0.741 0.483 0.000
(0.011) (0.018) (0.020) (0.020) (0.049) (0.019)
Respondent smells like cigarettes 0.403 0.423 0.379 0.400 0.377 0.469 0.277 0.461 0.001
(0.011) (0.018) (0.020) (0.020) (0.049) (0.019)
There are situations that make him/her want to smoke 0.933 0.954 0.911 0.927 0.042 0.290 0.927 0.888 0.285
(0.006) (0.008) (0.012) (0.010) (0.029) (0.012)
Tries to avoid areas that make him/her want to smoke 0.571 0.565 0.578 0.573 0.783 0.857 0.658 0.505 0.010
(0.011) (0.018) (0.020) (0.020) (0.054) (0.019)
So addicted that s/he needs help to stop smoking 0.524 0.530 0.510 0.532 0.943 0.443 0.582 0.504 0.700
(0.011) (0.018) (0.020) (0.020) (0.055) (0.019)
F-statistic [p-value] from regression of assigned group 0.410 0.540
on all of the above baseline variables. [0.9686] [0.8999]
Number of observations 2000 781 603 616 83 698
Standard errors in parentheses. Summary statistics in columns (1)-(4) are weighted to account for the change in probability of assignment to treatment across the three waves of marketing.
Table 1b. Summary Statistics, Outcome Variables
Outcome Measures
CARES Group
t-test of (2) t-test of (3) Did Not t-test of
All CARES Cues Control vs (4) vs (4) Took up Takeup (7) vs (8)
(1) (2) (3) (4) (5) (6) (7) (8) (9)
Panel A: Outcome Measures, Full Sample, Six Months
Found by surveyor for follow-up measurement 0.634 0.642 0.629 0.629 0.596 0.982 0.723 0.547 0.002
(0.011) (0.017) (0.020) (0.019) (0.049) (0.019)
Agreed to take urine test, conditional on being found 0.952 0.932 0.963 0.968 0.015 0.737 0.700 0.958 0.000
(0.006) (0.012) (0.009) (0.009) (0.060) (0.010)
Found and agreed to test urine test 0.604 0.598 0.604 0.608 0.709 0.942 0.506 0.524 0.752
(0.011) (0.018) (0.011) (0.020) (0.055) (0.019)
Passed urine test (omitted missing respondents) 0.153 0.181 0.153 0.124 0.023 0.316 0.690 0.128 0.000
(0.010) (0.019) (0.010) (0.016) (0.072) (0.018)
Passed urine test 0.093 0.108 0.093 0.075 0.033 0.355 0.349 0.067 0.000
(assumes all respondents who did not take the test are smokers) (0.007) (0.011) (0.006) (0.011) (0.053) (0.009)
Passed urine test 0.146 0.168 0.146 0.120 0.041 0.330 0.483 0.123 0.000
(assumes all respondents who were found but refused the test are smokers) (0.010) (0.018) (0.010) (0.016) (0.065) (0.017)
# of CARES accounts 85 83 2 0
Number of observations 2000 781 603 616 83 698
Panel B: Outcome Measures, Full Sample, One Year
Found by surveyor for follow-up measurement 0.596 0.615 0.578 0.590 0.339 0.670 0.723 0.547 0.001
(0.011) (0.017) (0.201) (0.020) (0.049) (0.019)
Agreed to take urine test, conditional on being found 0.949 0.948 0.941 0.958 0.489 0.280 0.984 0.939 0.157
(0.006) (0.010) (0.012) (0.010) (0.016) (0.012)
Found and agreed to test urine test 0.565 0.582 0.544 0.565 0.515 0.451 0.723 0.532 0.001
(0.011) (0.018) (0.020) (0.020) (0.049) (0.019)
Passed urine test (omitted missing respondents) 0.181 0.203 0.155 0.178 0.372 0.389 0.350 0.175 0.002
(0.011) (0.019) (0.019) (0.020) (0.062) (0.020)
Passed urine test 0.103 0.118 0.084 0.101 0.296 0.313 0.253 0.093 0.000
(assumes all respondents who did not take the test are smokers) (0.007) (0.012) (0.011) (0.012) (0.048) (0.011)
Passed urine test 0.172 0.192 0.145 0.171 0.414 0.337 0.344 0.165 0.001
(assumes all respondents who were found but refused the test are smokers) (0.011) (0.018) (0.018) (0.019) (0.061) (0.019)
# of CARES accounts 85 83 2 0
Number of observations 2000 781 603 616 83 698
Standard errors in parentheses. Summary statistics in columns (1)-(4) are weighted to account for the change in probability of assignment to treatment across the three waves of marketing.
Table 2. Multivariate Analysis of CARES Take-up
OLS, Probit
Estimator: OLS Probit
(1) (2)
Female -0.034 -0.024
(0.041) (0.028)
Age (/100) 0.894** 0.858**
(0.405) (0.398)
Age squared (/100) -.010** -0.010**
(0.005) (0.005)
Number of cigarettes per day in the past 7 days (/100) 0.153 0.103
(0.321) (0.252)
Number of cigarettes per day squared (/100) -0.002 -0.001
(0.007) (0.005)
Having tried to stop smoking in the past 12 months -0.034 -0.025
(0.025) (0.019)
Wanting to stop smoking sometime in life 0.085 0.062
(0.085) (0.039)
Wanting to stop smoking now 0.034 (0.019)
(0.038) (0.028)
Wanting to stop smoking in 1 year 0.076 0.080
(0.080) (0.127)
Wanting to stop smoking after 1 year -0.002 -0.003
(0.037) (0.050)
Will actually quit smoking in 6 months 0.116*** 0.114***
(0.036) (0.041)
Respondent smells like cigarettes -0.073** -0.056***
(0.024) (0.019)
There are situations that make him/her want to smoke 0.031 0.037
(0.039) (0.033)
Try to avoid areas that make him/her want to smoke 0.043 0.039*
(0.027) (0.022)
So addicted that s/he needs help to stop smoking 0.034 0.026
(0.027) (0.022)
probability (all variables above = 0) 0.002 0.001
Observations 781 775
(pseudo-)R-squared 0.101 0.142
Number of CARES accounts opened 83 83
Mean of dependent variable 0.106 0.107
Robust standard errors in parentheses; * significant at 10%; ** significant at 5%; *** significant at 1%. Standard errors are clustered
by the marketing group if the respondents were surveyed in group. All regressions control for 3 phases of randomization and use
marketer fixed effects. Probit specification reports marginal effects.
Table 3. Usage of CARES Bank Account by 6-Month Urine Test Result
Summary Statistics, Philippine Pesos (P50 = US$1)
# of Accounts Min Average Max Std. Dev
(1) (2) (3) (4) (5)
Opening balance 85 50 57.18 410 40.49
Success (i.e., those who passed 6-month urine test) 29 50 71.03 410 67.95
Failures (i.e., those who failed 6-month urine test) 56 50 50.00 50 0.00
# of deposits made into CARES account 85 1 11.75 29 9.35
Success (i.e., those who passed 6-month urine test) 29 7 20.90 26 5.47
Failures (i.e., those who failed 6-month urine test) 56 1 7.02 29 7.17
Proportion of clients who missed 3 deposits & lost deposit collection service 85 0 0.64 1 0.48
Success (i.e., those who passed 6-month urine test) 29 0 0.14 1 0.35
Failures (i.e., those who failed 6-month urine test) 56 0 0.89 1 0.31
Balance at 6 months 85 50 551.12 3410 651.01
Success (i.e., those who passed 6-month urine test) 29 282.75 1079.58 3410 703.37
Failures (i.e., those who failed 6-month urine test) 56 50 277.45 2657.75 414.62
Notes: Minimum account opening deposit was 50 pesos. Of the 83 CARES clients, 75 were from CARES with deposit collection group; 6 were from CARES without deposit collection
group; and 2 were from CUES group. Although respondents in CUES group were not offered CARES product, marketers opened the accounts for 2 respondents who approached them
after finding out about CARES. All takeup and impact analysis codes these 2 individuals into the CUES group in accordance with the random assignment.
Table 4. Impact of CARES on Passing Cotinine Urine Test
OLS, IV
Everyone That Was Found But
Everyone That Did Not Take The Test Refused To Take The Test Still
Assumption: Continues Smoking Drop If Did Not Take The Test Smokes
Outcome Measurement Timing: Six Months One Year Six Months One Year Six Months One Year
(1) (2) (3) (4) (5) (6)
Panel A: Intent-to-Treat Estimates, OLS
CARES Treatment 0.033* 0.035** 0.058** 0.057** 0.041* 0.054**
(0.017) (0.018) (0.026) (0.028) (0.024) (0.027)
Cue cards 0.015 0.009 0.022 0.019 0.021 0.019
(0.016) (0.016) (0.024) (0.026) (0.023) (0.025)
# of observations 2000 2000 1226 1161 1287 1218
F-test p-value: CARES = Cues 0.302 0.142 0.162 0.184 0.408 0.194
R-squared 0.048 0.057 0.068 0.083 0.056 0.081
Mean of dependent variable 0.083 0.089 0.123 0.147 0.119 0.140
Sampling weights no no yes yes yes yes
Panel B: Treatment on the Treated Estimates, IV
CARES Treatment 0.296** 0.312** 0.646** 0.533** 0.522* 0.509**
(0.151) (0.159) (0.270) (0.266) (0.293) (0.253)
Cue cards 0.014 0.008 0.022 0.017 0.021 0.017
(0.016) (0.016) (0.024) (0.026) (0.023) (0.025)
# of observations 2000 2000 1226 1161 1287 1218
F-test p-value: CARES = Cues 0.051 0.053 0.016 0.045 0.077 0.044
Mean of dependent variable 0.083 0.089 0.123 0.147 0.119 0.140
Sampling weights no no yes yes yes yes
Robust standard errors in parentheses; * significant at 10%; ** significant at 5%; *** significant at 1%. All regressions control for the 3 waves of marketing and include
covariates (all independent variables from take-up regressions in Table 3). Panel B shows the results of IV regressions with assignment to treatment group as an instrument for
CARES take-up. Cue cards take-up is not instrumented by CUES group assignment, because only two respondents rejected the cue cards. Models estimated in columns (3)-(6) are
weighted to reflect the different likelihood of a subject taking a urine test between CARES clients and non-clients and across CARES, Cues, and control groups.
Appendix Table 1. Usage of CARES Bank Account by 12-Month Urine Test Result
Summary Statistics, Philippine Pesos (P50 = US$1)
# of Accounts Min Average Max Std. Dev
(1) (2) (3) (4) (5)
Opening balance 61 50 60.00 410 11.04
Success (i.e., those who passed 12-month urine test) 21 50 74.29 410 78.97
Failures (i.e., those who failed 12-month urine test) 40 50 52.50 100 47.61
# of deposits made into CARES account 61 1 12.70 26 9.22
Success (i.e., those who passed 12-month urine test) 21 1 15.86 26 9.81
Failures (i.e., those who failed 12-month urine test) 40 1 11.05 25 8.56
Proportion of clients who missed 3 deposits & lost deposit collection service 61 0 0.57 1 0.50
Success (i.e., those who passed 12-month urine test) 21 0 0.43 1 0.51
Failures (i.e., those who failed 12-month urine test) 40 0 0.65 1 0.48
Balance at 6 months 61 50 585.58 3410 673.33
Success (i.e., those who passed 12-month urine test) 21 50 786.76 1886.6 617.58
Failures (i.e., those who failed 12-month urine test) 40 50 479.96 3410 684.60
Minimum account opening deposit was 50 pesos.
For this table we we drop the 24 clients who were not found for the surprise 12-month test, and code the 1 client who was found and refused to take the test as a failure.
Appendix Table 2: Impact of CARES
Same as Table 4, except using a probit model
Probit, IV-Probit
Everyone That Was Found But
Everyone That Did Not Take The Refused To Take The Test Still
Assumption: Test Continues Smoking Drop If Did Not Take The Test Smokes
Outcome Measurement Timing Six Months One Year Six Months One Year Six Months One Year
(1) (2) (3) (4) (5) (6)
Panel A: Intent-to-Treat Estimates, Probit
CARES Treatment 0.033** 0.033* 0.061** 0.059** 0.044* 0.055**
(0.016) (0.017) (0.027) (0.029) (0.025) (0.028)
Cue cards 0.015 0.009 0.023 0.020 0.022 0.020
(0.016) (0.017) (0.027) (0.029) (0.025) (0.028)
# of observations 1993 1989 1225 1155 1286 1212
F-test p-value: CARES = Cues 0.232 0.140 0.140 0.178 0.355 0.192
Mean of dependent variable 0.083 0.089 0.123 0.147 0.119 0.140
Sampling weights no no yes yes yes yes
Panel B: Treatment on the Treated Estimates, IV Probit
CARES Treatment 0.385 0.509 0.736*** 0.702*** 0.690* 0.689***
(0