Estimating Biased Beliefs in Smoking Cessation
Estimating Biased Beliefs in Smoking Cessation
Tuesday, June 12, 2018: 8:00 AM
2001 - Second Floor (Rollins School of Public Health)
Discussant: Jing Li
Cost-benefit analyses of tobacco control policies offset the benefits of quitting with lost
consumer surplus to smokers. The behavioral biases of smokers may reduce this offset,
although the degree to which smokers hold biases has not been empirically derived. We
undertook a randomized field experiment to estimate smokers’ beliefs regarding their
willingness to abstain from smoking in the future and the effect of intervening abstinence
on their future willingness to abstain (n = 397). Over 12 weeks, subjects received varying
incentives to abstain from smoking. In a random week of both Month 1 and Month 3,
subjects received an incentive payment for abstaining (amount selected at random).
Treated subjects received weekly payments for abstaining during Month 2. Subjects were
paid for reporting abstinence and testing negative for cotinine using a saliva test. A
“truth-telling” lottery incentivized subjects to report their smoking status accurately. We
elicited beliefs about future smoking at multiple points using an incentive-compatible
mechanism. Comparing predicted versus actual smoking behavior in response to the
incentives, we measure the extent to which participants mispredict their behavior and
exhibit present bias and projection bias.
We find that smokers overestimated their likelihood of future abstinence by nearly 100%
across all incentive levels, consistent with present-biased preferences. The abstinence
intervention during Month 2 increases the likelihood of future abstinence, but ex-ante
subjects do not anticipate any effect, and ex-post subjects believe the effect to be
(marginally) negative. Our results point to a pernicious pattern of biased beliefs: 55% of
subjects are na¨ive regarding future self-control problems; subjects fail to reject complete
projection bias over long-run addiction; and after the intervention subjects appear to
project craving and withdrawal. We discuss the implications of our findings for
1) designing policies and interventions to address the biased beliefs of smokers and
2) estimating lost consumer surplus in cost-benefit analyses of tobacco control policies. We
also explore heterogeneity of the treatment effects by pre-specified characteristics, including
smoking intensity, nicotine dependence, dual use of e-cigarettes, and degree of present bias
based on a monetary discounting task. Finally, we obtain structural estimates of behavioral
parameters using a pre-specified model.
consumer surplus to smokers. The behavioral biases of smokers may reduce this offset,
although the degree to which smokers hold biases has not been empirically derived. We
undertook a randomized field experiment to estimate smokers’ beliefs regarding their
willingness to abstain from smoking in the future and the effect of intervening abstinence
on their future willingness to abstain (n = 397). Over 12 weeks, subjects received varying
incentives to abstain from smoking. In a random week of both Month 1 and Month 3,
subjects received an incentive payment for abstaining (amount selected at random).
Treated subjects received weekly payments for abstaining during Month 2. Subjects were
paid for reporting abstinence and testing negative for cotinine using a saliva test. A
“truth-telling” lottery incentivized subjects to report their smoking status accurately. We
elicited beliefs about future smoking at multiple points using an incentive-compatible
mechanism. Comparing predicted versus actual smoking behavior in response to the
incentives, we measure the extent to which participants mispredict their behavior and
exhibit present bias and projection bias.
We find that smokers overestimated their likelihood of future abstinence by nearly 100%
across all incentive levels, consistent with present-biased preferences. The abstinence
intervention during Month 2 increases the likelihood of future abstinence, but ex-ante
subjects do not anticipate any effect, and ex-post subjects believe the effect to be
(marginally) negative. Our results point to a pernicious pattern of biased beliefs: 55% of
subjects are na¨ive regarding future self-control problems; subjects fail to reject complete
projection bias over long-run addiction; and after the intervention subjects appear to
project craving and withdrawal. We discuss the implications of our findings for
1) designing policies and interventions to address the biased beliefs of smokers and
2) estimating lost consumer surplus in cost-benefit analyses of tobacco control policies. We
also explore heterogeneity of the treatment effects by pre-specified characteristics, including
smoking intensity, nicotine dependence, dual use of e-cigarettes, and degree of present bias
based on a monetary discounting task. Finally, we obtain structural estimates of behavioral
parameters using a pre-specified model.