Body,Title,"Study Type",Citation,AdditionalSources,"Original Publication Date","Original Publication URL","Reviewed by CLEAR","Topic Area","Subtopic Area","Study Evidence Rating","Causal Design",Conclusions,Body,"Features of Intervention","Features of the Study",Findings,"Considerations for Interpreting the Findings","Types of and Outcomes","Study Sites",ProfileFootnotes,"Conflict of Interest","Firm Characteristics","Geographic Setting","Study Type","Target Population","Topic Tags","Domain Effectiveness",Outcome,"Review Protocol"
"Reference-dependent job search: Evidence from Hungary (DellaVigna et al. 2014)","Reference-dependent job search: Evidence from Hungary (DellaVigna et al. 2014)","Study Type: Causal Impact Analysis","DellaVigna, S., Lindner, A., Reizer, B., & Schmieder, J. (2014). Reference-dependent job search: Evidence from Hungary. Unpublished working paper.",,2014,http://eml.berkeley.edu/~sdellavi/wp/HungaryUIRD_wp_2014_07_20.pdf,"February 2016","Behavioral Insights",,"Low Causal Evidence","Regression Analysis",,"The study’s objective was to empirically test a behavioral model of job search using a reform in Hungary’s Unemployment Insurance (UI) program.
The authors compared rates of exit from unemployment in groups of claimants who entered UI before and after the reform’s implementation in November 2005. The authors used social security and UI data from Hungary’s Institute of Economics and National Employment Service.
The study found support for the behavioral job search model and theorized that approximately budget-neutral transitions to two-step UI systems could speed claimants’ exit out of unemployment.
The quality of causal evidence presented in this report is low because it cannot account for factors other than the UI program that also changed over the period of study and could also have influenced the outcomes of interest. This means we are not confident that the estimated effects are attributable to Hungary’s 2005 UI reform; other factors are likely to have contributed.","Before 2005, Hungary’s UI system had two tiers. In the first tier, benefit duration depended on a worker’s previous contributions to social security, with a maximum duration of 270 days, and benefit amount depended on earnings in the previous year. After reaching their maximum duration, claimants who remained unemployed became eligible for Unemployment Assistance, which provided the same benefits to everyone regardless of past earnings.
On November 1, 2005, Hungary introduced a reformed, two-step UI system. In the first step, which lasted for half the maximum first-tier duration, or at most 90 days, benefit amounts were calculated based on the claimant’s previous earnings, with a lower replacement rate than under the previous system (60 percent instead of 65 percent), but substantially higher minimum and maximum benefit amounts than the previous system. In the second step (days 91 to 270), every claimant received the same benefit amount regardless of previous earnings. The total UI duration for each claimant and eligibility for Unemployment Assistance remained the same. The authors noted that for those with high pre-UI earnings, the overall payout under both systems would be approximately the same because lower benefits in step two (days 91-270) offset higher benefits in step one (days 1-90).","The authors proposed a “reference-dependent” model of job search where a reference point is given by consumption in the recent past and workers are loss-averse with respect to payoffs below the reference point. So, for a newly unemployed worker, the reference point would be recent earnings before becoming unemployed. Because the UI benefits are substantially below previous earnings, the worker finds this painful and tries hard to find a new job. As time goes on and the worker remains unemployed, he becomes accustomed to the lower UI benefits, and job search effort decreases. However, as the exhaustion of UI benefits approaches, the worker anticipates the even lower benefit amount he will receive, and resumes diligently job searching.
To test this theory, the authors compared the outcomes of a group of UI claimants who entered the system between February and October 2005, before the reform’s implementation in November 2005, to those of a group of claimants who entered the system between February and October 2006, after the reform went into effect. If the theory were supported, the exit rate from unemployment should be higher around the 270-day exhaustion date for workers under the pre-reform system than for workers in the post-reform system because they faced a much steeper decline in benefits upon UI exhaustion. Measuring time as days since UI entry, the authors calculated the reform’s impact on the probability of exit from UI separately in each two-week time period as the difference in probability of exit between members of the pre- and post-reform groups, controlling for characteristics such as age, gender, years of education, and previous earnings.","The study found support for the reference-dependent model of job search. Rates of exit from UI were higher in the period immediately preceding the exhaustion of regular benefits and transition to Unemployment Assistance for workers receiving UI pre-reform compared with those receiving UI post-reform. The authors claim this is consistent with the higher anticipated loss in consumption for workers receiving UI pre-reform.","The authors compared reemployment trends among UI claimants before the UI reform took place to reemployment trends among claimants after the reform took place. Even including controls for the demographic characteristics and earnings histories of the two groups, there remains a strong possibility that other changes in the economy and the UI system that could have influenced reemployment rates were simultaneously taking place. Because the study does not include a comparison group of unaffected individuals, it is not possible to attribute the observed differences in outcomes to the reforms alone.",,,,,,International,"Causal Impact Analysis","Adult, Unemployed, Dislocated or displaced worker","Unemployment Insurance, Behavioral Interventions, Job search assistance and supportive services","Public benefits receipt-Low-Favorable impacts","Public benefits receipt","Behavioral Insights Review Protocol"
"Psychological frictions and the incomplete take-up of social benefits: Evidence from an IRS field experiment (Bhargava & Manoli 2015)","Psychological frictions and the incomplete take-up of social benefits: Evidence from an IRS field experiment (Bhargava & Manoli 2015)","Study Type: Causal Impact Analysis","Bhargava, S., & Manoli, D. (2015). Psychological frictions and the incomplete take-up of social benefits: Evidence from an IRS field experiment. American Economic Review, 105(11), 3489-3529.","Bhargava, S., & Manoli, D. (2013). Why are benefits left on the table? Assessing the role of information, complexity, and stigma on take-up with an IRS field experiment. Working paper.",2015,,"February 2016","Behavioral Insights",,"High Causal Evidence","Randomized Control Trial (RCT)",,"The study’s objective was to examine the impact of complexity, program information, and stigma on the take-up rate of the Earned Income Tax Credit (EITC) among eligible taxpayers who had not previously claimed the credit.
The authors conducted a randomized controlled trial, sending different combinations of materials to likely eligible taxpayers who had not previously claimed the EITC. They used tax data from the Internal Revenue Service (IRS) for tax year 2009.
The study found that use of complex materials and a longer worksheet decreased the EITC take-up rate by 6 and 4 percentage points, respectively. Materials that displayed the potential tax credit amount increased the EITC take-up rate by 8 percentage points.
The quality of causal evidence presented in this report is high because it was based on a well-implemented randomized controlled trial. This means we are confident that the estimated effects are attributable to the experimental mailings, and not to other factors.",,"Many low-income taxpayers are eligible for the EITC, which must be claimed on tax forms submitted to the IRS each year. Eligibility for the EITC and the tax credit received, if eligible, depend primarily on the number of hours worked, total earnings, and number of dependents claimed by the taxpayer. The IRS sends taxpayers who are likely eligible for the EITC but do not claim it a notice alerting them of the EITC and a worksheet to claim the credit retroactively. Response rates to the mailing, and thus take-up of EITC, have historically been lower than desired.
The study’s objective was to examine the impact of different mailing materials on the EITC take-up rate among eligible taxpayers who did not respond to the first mailing. Through their collaboration with the IRS, the authors gained access to several years of tax data on EITC-eligible individuals. The authors manipulated three aspects of the mailing:

The envelope. A control envelope had no messaging on it, while a treatment envelope was printed with the message, “Important—Good News for You (Importante—Buenas Noticias para Usted)” to test whether encouraging recipients to open the envelope would affect take-up.
The worksheet: Each household received a worksheet assessing EITC eligibility.

To test whether the length and complexity of the worksheet affected take-up, a simple version of the worksheet contained fewer eligibility screens than a more complex version.
To test whether recipients’ perceptions of potential penalties affected take-up, half of the worksheets contained a headline assuring them that mistakenly reporting incorrect information would not result in a penalty (known as indemnification).


The notice. Each household received a notice informing them of their potential eligibility for the EITC.

To test whether the simplicity of the notice affected take-up, the authors compared a simple, single-sided notice with clear font and streamlined information to a more complex, two-sided notice. The complex notice was textually dense and repeated eligibility requirements that were also discussed in the worksheet.
To test whether additional information affected take-up, a version of the notice contained information about size of benefit receipt while another version of the notice displayed information about the transaction cost, either noting that the worksheet could be completed in less than 60 or 10 minutes. To test whether general program information affected take-up, some notices included a one-page informational flyer explaining the formula for calculating benefit amount and describing program “myths and realities.”
To test whether program stigma affected take-up, one version of the notice contained a message intended to reduce personal stigma, “You may have earned a refund due to your many hours of employment,” and another version contained a message intended to reduce social stigma, “Usually, four out of every five people claim their refund.”



The 35,050 study participants were California residents who were likely eligible for EITC in tax year 2009, who did not claim an EITC benefit in tax year 2009, who had received a first reminder note about EITC, and who did not respond to the first reminder note. The authors conducted a randomized controlled trial by assigning participants to versions of the materials using three separate randomizations, one each for the notice, worksheet, and envelope. All study participants received some type of mailing. The authors estimated regression models to determine the impact of the various treatments on subsequent claiming of the EITC.","Overall, 23 percent of taxpayers receiving the control mailing—the simple notice, simple worksheet, and plain envelope—responded to the mailing.
Compared with the simple versions of the notice and the worksheet, the complex notice decreased the response rate by 6 percentage points and the longer worksheet decreased the response rate by 4 percentage points. These results were statistically significant.
Compared with the simple notice alone, the simple notice plus a potential tax credit amount display increased the response rate by 8 percentage points. However, the simple notice plus informational flyer decreased the response rate by 4 percentage points. These results were statistically significant.
Compared with the simple notice alone, the simple notice plus a note designed to reduce social stigma decreased the response rate by 4 percentage points. These results were statistically significant.
There were no effects of providing estimates of the time required to complete the worksheet, indemnifying taxpayers against errors, the message on the envelope, or messages to reduce personal stigma on the EITC take-up rate.","The study focused solely on EITC-eligible taxpayers who did not originally claim the EITC and who had not responded to a previous prompt to claim the EITC. Therefore, these results apply only to similar populations and should not be generalized to all taxpayers, or even all EITC-eligible taxpayers.",,,,,,"United States","Causal Impact Analysis","Adult, Employed","Behavioral Interventions","Public benefits receipt-Mod/high-Favorable impacts","Public benefits receipt","Behavioral Insights Review Protocol"
"Would people behave differently if they better understood Social Security? Evidence from a field experiment (Liebman & Luttmer 2011)","Would people behave differently if they better understood Social Security? Evidence from a field experiment (Liebman & Luttmer 2011)","Study Type: Causal Impact Analysis","Liebman, J., & Luttmer, E. (2011). Would people behave differently if they better understood Social Security? Evidence from a field experiment. National Bureau of Economic Research working paper no. 17287. Cambridge, MA: NBER.",,2011,http://www.nber.org/papers/w17287.pdf,"February 2016","Behavioral Finance: Retirement, Behavioral Insights",,"Moderate Causal Evidence","Randomized Control Trial (RCT)",,"The study’s objective was to examine the impact of providing information about Social Security rules and benefits on labor force participation, knowledge of Social Security, and claiming of Social Security benefits.
Workers nearing retirement age were randomly assigned into either the treatment group, which was sent an informational brochure about Social Security and invited to a 15-minute web tutorial on Social Security provisions, or to the control group, which was not offered this information but could seek other publicly available information about Social Security. The authors measured outcomes through a follow-up survey conducted 13 months after random assignment.
The study found that, on average, members of the treatment group were more likely to work for pay in the last month and were more likely to be aware that Social Security benefits were based on the number of years with the highest earnings, compared with the control group. However, there were no statistically significant differences between the two groups on any other outcomes related to earnings and employment, Social Security benefit receipt, or understanding Social Security rules.
The quality of causal evidence presented in this report is moderate because it was based on a well-conducted randomized controlled trial with high attrition and adequate control variables. This means we are somewhat confident that the estimated effects are attributable to providing information about Social Security provisions, but other factors might also have contributed.",,"The authors investigated whether a relatively low-cost intervention could impact workers’ knowledge of and choices associated with claiming Social Security benefits. Knowledge Networks, a data collection firm, identified participants for the study via random-digit dialing and collected a survey covering demographic and employment characteristics in November 2008.
To be eligible for the study, workers had to be 55 years of age or older, employed, and covered by Social Security at the time of random assignment. Approximately 90 percent of the sample was 60 to 65 years old. The 2,483 eligible individuals in the sample were randomly assigned to either the treatment or control group. Individuals randomly assigned to the treatment group were offered additional information about Social Security provisions in two ways: (1) an informational brochure that was mailed to them and (2) a 15-minute web tutorial. The brochure was generic and all members of the treatment group received the same version. The web tutorial was tailored to each individual, with specific examples that fit the characteristics of that individual, such as their age, gender, and length of time in the labor force. Individuals randomly assigned to the control group received no additional information, but could access other publicly available information about Social Security.
A follow-up survey was fielded from April to June 2010 to collect demographic and outcomes data for study participants. The authors estimated the impact of being offered additional information on Social Security provisions using regression analyses, controlling for demographic characteristics. The primary outcomes examined were labor force participation, claiming of Social Security benefits, hours worked and earned income, and multiple measures of understanding Social Security benefits rules and provisions.","The study found that the treatment group was about four percentage points more likely to have worked for pay in the past month, compared with the control group; this difference was statistically significant. The study found no statistically significant differences between the treatment and control groups on earnings, hours worked, or Social Security benefit receipt.
The study found that the treatment group was about six percentage points more likely to be aware that Social Security benefits were based on the number of years with the highest earnings, compared with the control group. This was a statistically significant difference. However, there were no statistically significant differences between the treatment and control groups on any other measures of understanding Social Security rules and provisions, or on the age at which individuals retired or started claiming Social Security benefits.","Although the study design was a randomized controlled trial, the study had high attrition because not all study participants responded to the follow-up survey. Therefore, the study is not eligible to receive a high causal evidence rating. However, the authors demonstrated that the study groups included in the analysis had equivalent demographic characteristics, educational backgrounds, and household incomes. Therefore, the study receives a moderate causal evidence rating.
Only 33 percent of individuals in the treatment group recalled receiving the informational brochure. The low recall rate suggests that the informational brochure had not made much of an impression on study participants, which is further reflected in the few statistically significant impacts of the intervention. However, the intervention was also very low-cost, so even very small impacts may be cost-effective.",,,,,,,"Causal Impact Analysis",Employed,,"Employment-Mod/high-No impacts",,"Behavioral Finance: Retirement Review Protocol"
"Learning about job search: A field experiment with job seekers in Germany (Altmann et al. 2015)","Learning about job search: A field experiment with job seekers in Germany (Altmann et al. 2015)","Study Type: Causal Impact Analysis","Altmann, S., Falk, A., Jäger, S., & Zimmermann, F. (2015). Learning about job search: A field experiment with job seekers in Germany. Unpublished manuscript.",,2015,https://www.cens.uni-bonn.de/team/board/armin-falk/afjz-job-search-dp-150509.pdf,"February 2016","Behavioral Insights",,"High Causal Evidence","Randomized Control Trial (RCT)",,"The study’s objective was to examine whether providing job search information to unemployed job seekers affected their labor market outcomes.
German citizens identified as unemployed were randomly assigned to the treatment group, which was mailed an informational brochure, or to the control group, which was not mailed a brochure. The Integrated Employment Biographies, an administrative database available through the German Federal Employment Agency, was the primary data source for the study.
The study did not find any statistically significant effects of the brochure on the employment or earnings outcomes of the treatment group relative to the control group. However, the study did find positive impacts of the brochure on earnings and employment outcomes for the subgroup of individuals who had been identified as being at risk of long-term unemployment.
The quality of causal evidence presented in this report is high because it was based on a well-implemented randomized controlled trial. This means we are confident that the estimated effects are attributable to the informational brochure, and not to other factors.",,"One-hundred sixty thousand individuals who had recently registered for unemployment insurance in Germany were randomly assigned to either a treatment group, which was mailed an informational brochure, or a control group, which was not mailed a brochure. Individuals were eligible for the study if they were German citizens between the ages of 25 and 55. Individuals were excluded if they had re-registered for unemployment after having participated in a training program, were civil servants, or were self-employed. Random assignment occurred in four waves between October 2010 and January 2011; in each wave, the authors randomly assigned 10,000 eligible individuals to the treatment group and 30,000 eligible individuals to the control group.
The brochure was designed to inform and motivate unemployed individuals to find work. It consisted of four parts: (1) facts about the current labor market in Germany, (2) information on the benefits of job search efforts and the association of longer unemployment spells with lower rates of finding jobs, (3) evidence of beneficial health and other non-financial outcomes associated with employment, and (4) job search resources.
The authors analyzed data from the Integrated Employment Biographies, an administrative data source available through the German Federal Employment Agency, which contains information on individuals’ employment status, earnings, gender, age, and education level. The authors estimated regression models comparing the employment and earnings outcomes of the treatment and control group members for 52 weeks after the brochure was mailed.","The study did not find any statistically significant effects of the brochure on employment or earnings outcomes of the treatment group as a whole relative to the control group as a whole.
The brochure had a statistically significant effect on employment and earnings for the subgroup of individuals identified as being at risk of long-term unemployment; for this subgroup, individuals in the treatment group were employed for 4.78 days more and had average earnings of 446.20 EUR higher than individuals in the control group over the 52 weeks following the intervention.","The authors noted that, although the estimated impacts were small and confined to the subgroup that had been identified as being at risk of long-term employment, the intervention itself was very low-cost, and thus might have been cost-effective.",,,,,,International,"Causal Impact Analysis","Adult, Unemployed","Job search assistance and supportive services, Behavioral interventions","Earnings and wages-Mod/high-No impacts, Employment-Mod/high-No impacts","Earnings and wages, Employment","Behavioral Insights Review Protocol"
"Notes on behavioral economics and labor market policy (Babcock et al. 2012)","Notes on behavioral economics and labor market policy (Babcock et al. 2012)","Study Type: Descriptive Analysis","Babcock, L., Congdon, W., Katz, L., & Mullainathan, S. (2012). Notes on behavioral economics and labor market policy. IZA Journal of Labor Policy, 1(1), 1-14.",,2012,,"February 2016","Behavioral Insights",,"Not Rated",,,"The article’s objective was to assess how behavioral economics can inform labor policy reforms to increase policy efficiency.
The authors used behavioral economic theory to explain barriers to successful policies related to unemployment insurance, job search assistance, and job training. They also proposed policy solutions to overcome these barriers.
For unemployment insurance, the authors recommended implementing wage loss insurance that subsidizes reemployment wages to counteract individuals’ tendencies to set high reservation wages when seeking employment, which slows their return to work. They also recommended providing small, immediate, and high frequency reminders and incentives to search for work.
To increase participation in job search assistance, the authors recommended policies that automatically put unemployed individuals into services and/or to make job search assistance more user-friendly and personalized. Similarly, to increase participation in job training programs, the authors recommended streamlining or simplifying training offerings and providing guidance to reduce the burden on the prospective users to select and start a program. Simplifying the decision-making process by creating a competitive training market could also be beneficial to prospective users.",,,,,,,,,,"United States","Descriptive Analysis","Adult, Unemployed","Job search assistance and supportive services, Unemployment Insurance, Other training and education",,Employment,"Behavioral Insights Review Protocol"
"Signing at the beginning makes ethics salient and decreases dishonest self-reports in comparison to signing at the end (Shu et al. 2012)","Signing at the beginning makes ethics salient and decreases dishonest self-reports in comparison to signing at the end (Shu et al. 2012)","Study Type: Causal Impact Analysis","Shu, L., Mazar, N., Gino, F., Ariely, D., & Bazerman, M. (2012). Signing at the beginning makes ethics salient and decreases dishonest self-reports in comparison to signing at the end. Proceedings of the National Academy of Sciences, 109(38), 15197-15200.",,2012,,"February 2016","Behavioral Insights",,"High Causal Evidence","Randomized Control Trial (RCT)",,"The study’s objective was to examine whether placing a signature line at the beginning of a form, rather than at the end, which is more common, discouraged dishonesty.
The authors conducted three separate randomized experiments. In experiments 1 and 2, individuals completed math puzzles and were asked to report the number of puzzles they completed, as well as their expenses for traveling to the site where the experiment was conducted. In experiment 3, individuals acquiring a new car insurance policy were asked to report their current cars’ odometer readings on an insurance form. In each experiment, individuals were randomly assigned either to receive a form requiring a signature at the top, or to receive a form requiring a signature at the bottom.
The study found that, across the three experiments, participants who signed at the top of the form were less likely to respond dishonestly than participants who signed the form at the bottom or did not sign the form at all.
The quality of causal evidence presented in this report is high because it was based on three well-implemented randomized controlled trials. This means we are confident that the estimated effects are attributable to the placement of the signature line on the forms, and not to other factors.",,"The authors conducted three experiments, two in a laboratory setting and one in the field. In experiments 1 and 2, university students and employees completed math puzzles and then reported on a form designed to look like a typical tax form how many they completed. Participants received a cash incentive for each puzzle that they reported completing. They also reported their time and commuting costs for traveling to the lab and were reimbursed for some of these expenses.
In experiment 1, 101 individuals were randomly assigned to one of three conditions: (1) tax form with a signature line at the top; (2) tax form with a signature line at the bottom; or (3) tax form with no signature line. In experiment 2, 60 individuals were randomly assigned to one of two conditions: (1) tax form with a signature line at the top, or (2) tax form with a signature line at the bottom.
In experiments 1 and 2, the authors compared the actual number of completed puzzles and the number of puzzles that individuals reported completing on the tax form. The authors considered any difference between the actual and reported number of puzzles as a result of dishonesty. They also compared the average reported expenses across experimental groups; they assumed that if one group reported higher expenses than the other, resulting in higher reimbursement rates, that group was dishonestly inflating their actual expenses.
In the third experiment, 13,488 existing car insurance customers who were acquiring a new car insurance policy had to report current odometer readings of all vehicles under the existing policy. The authors randomly assigned individuals to either sign next to the statement: “I promise that the information I am providing is true” on the top of the form or on bottom of the form. Previously reported odometer readings for the cars (from insurance records) were subtracted from the newly-reported odometer readings to calculate average mileage driven per car, which was then compared with the group that signed the statement at the top of the form and the group that signed the statement at the bottom of the form. The authors assumed that if one group reported lower average mileage than the other, this indicated the group was dishonestly misreporting its mileage.","The study found that, across the three experiments, participants who signed at the top of the form were less likely to respond dishonestly than participants who signed the form at the bottom or did not sign the form at all.
In experiments 1 and 2, participants who signed at the top of the form were less likely to misreport the number of completed math puzzles than participants who signed the form at the bottom or did not sign the form at all. They also reported lower expenses (and, hence, lower reimbursement rates) than participants who signed the form at the bottom or did not sign the form at all.
In experiment 3, participants who signed at the top of the form reported driving more average miles per car than participants who signed the form at the bottom.","The authors did not explicitly provide information to calculate attrition. However, the authors conducted experiments 1 and 2 in a laboratory setting, implying no attrition. For experiment 3, the authors used an administrative database to compare mileage reported on the experimental form to previous mileage; the analysis reported by the authors suggests low attrition.",,,,,,"United States","Causal Impact Analysis",Adult,"Behavioral Interventions","Attitudes-Mod/high-Favorable impacts",Attitudes,"Behavioral Insights Review Protocol"
"Can compulsory dialogues nudge sick-listed workers back to work? (Markussen et al. 2015)","Can compulsory dialogues nudge sick-listed workers back to work? (Markussen et al. 2015)","Study Type: Causal Impact Analysis","Markussen, S., Røed, K., & Schreiner, R. C. (2015). Can compulsory dialogues nudge sick-listed workers back to work? (IZA Discussion Paper No. 9090). Bonn, Germany: Institute for the Study of Labor (IZA).",,2015,http://ftp.iza.org/dp9090.pdf,"March 2016","Behavioral Insights",,"Low Causal Evidence","Regression Analysis",,"The study’s objective was to examine the impact of a compulsory dialogue meeting (DM) on the return to partial or full-time employment for long-term sick-listed workers in Norway.
The authors compared differences in outcomes between long-term sick-listed workers whose absence started in July 2004 through June 2006 and those whose absence started in January 2009 through December 2010, during which time the DM program became compulsory. The authors used salary and social insurance data from the Norwegian Social Security Administration (SSA) to measure program impacts.
The study found that increasing the probability of a sick-listed worker attending a DM resulted in a significant increase in base salary, for those who returned to full-time employment, of $1,436. It also resulted in a significantly lower probability of receiving temporary disability benefits two years later for those who returned to full-time employment (1.4 percent) and those who returned to either part- or full-time employment (2.3 percent).
The quality of causal evidence presented in this report is low because the authors did not establish the comparability of the treatment and comparison groups. This means we are not confident that the estimated effects are attributable to compulsory DMs; other factors are likely to have contributed.","Workers in Norway who go on long-term sick leave are paid their base salary for the first year of their absence. The employer covers the salary for the first 16 days, and the Norwegian SSA provides coverage for the remainder of the year. Approximately 6 months (26 weeks) into the absence, the SSA organizes a compulsory DM between the worker, the employer, and if necessary, the physician who authorized the sick leave. During this meeting, the parties discuss whether arrangements can be made at the workplace to facilitate the worker’s full or partial return to work.","Because of differences in the operations of local SSA offices, the timing of the DMs and issuing of exemptions from them varied across counties, such that some sick-listed workers were not invited to a DM during the first 52 weeks of their absence. The authors incorporated this local variation into their design to estimate the probability that a sick-listed worker attended a DM. Specifically, the authors modeled the effect of a change in the probability of attending a DM on earnings and insurance receipt, moving from the lowest to highest average probabilities observed across counties.
Long-term earnings and social insurance receipt were measured two years after the beginning of an individual’s spell of absence. Long-term sick-listed workers who began a spell of full-time absence in July 2004 through June 2006 were compared to those beginning spells in January 2009 through December 2010, during which time DMs became compulsory. In both periods, workers with absence spells lasting long enough to lead to an invitation to a DM were compared to those with shorter spells who were not likely to experience a DM. The impacts that the authors estimated can be interpreted as a comparison of the difference in outcomes before and after the introduction of compulsory DMs for those with long spells (and likely to be affected by the policy) relative to those with short spells (and not likely to be affected by the policy) across counties with varying probabilities of inviting sick-listed workers to a DM.
The study included 129,656 and 151,911 individuals from the non-compulsory and compulsory DM periods, respectively, across all 19 counties in Norway. Approximately 60 percent of long-term sick-listed workers were women, with an average age of 44 and a base salary of $29,000. On average, workers were absent for a period of 40 weeks.","The study found that an increase in the probability of a sick-listed worker attending a DM, from the lowest county-level probability to the highest county-level probability, resulted in a significant increase ($1,436) in base salary for those who returned to full-time employment.
An increase in the probability of attending a DM from the lowest county-level probability to the highest county-level probability resulted in a significant decrease in the probability of receiving temporary disability benefits two years later by 1.4 percent for those who returned to full-time employment, and 2.3 percent for those who returned to either part-time or full-time employment.","The authors controlled for differences between the treatment and comparison groups in many individual demographic and employment characteristics. However, they did not show equivalent trends in baseline earnings or benefit receipt outcomes between groups before the compulsory DMs came into effect. This is important because, among other things, the introduction of compulsory DMs could have led workers to change the timing and length of their absences from work, introducing systematic differences in earnings or benefit receipt between the treatment and comparison groups. Workers might have extended their absences to receive the treatment, making them systematically different from those in the comparison group with short spells. This means that the estimated impacts may reflect differences between the groups in addition to the effect of compulsory DMs.",,,,,,International,"Causal Impact Analysis","Unemployed, Adult","Behavioral Interventions","Earnings and wages-Low-Favorable impacts, Public benefits receipt-Low-Favorable impacts","Earnings and wages, Public benefits receipt","Behavioral Insights Review Protocol"
"Details matter: The impact of presentation and information on the take-up of financial incentives for retirement saving (Saez 2009)","Details matter: The impact of presentation and information on the take-up of financial incentives for retirement saving (Saez 2009)","Study Type: Causal Impact Analysis","Saez, E. (2009). Details matter: The impact of presentation and information on the take-up of financial incentives for retirement saving. American Economic Journal: Economic Policy, 1(1), 204-228.",,2009,,"March 2016","Behavioral Finance: Retirement, Behavioral Insights",,"High Causal Evidence","Randomized Control Trial (RCT)",,"The study’s objective was to examine the impact of contribution matches, credit rebates, and advance notification on tax filers’ decisions about opening an individual retirement account (IRA) during the tax preparation process and the amount they contributed to the IRA.
The authors randomly assigned tax filers at 60 H&R Block locations in St. Louis, Missouri, to treatment conditions, defined by whether the filers were offered a 50 percent one-time match on IRA contributions, a 33 percent credit rebate on IRA contributions, or a 50 percent match on one-time and monthly IRA contributions. H&R Block provided tax filing information from the 2005 and 2006 tax years as well as background information on the filers.
The study found that offering a 50 percent match on one-time IRA contributions and offering a 33 percent credit rebate increased the likelihood of opening an IRA and the amount contributed, but the effect on the likelihood of opening an IRA was larger for the 50 percent match treatment group.
The quality of causal evidence is high for some outcomes because they were based on a well-implemented randomized controlled trial. This means we are confident that the estimated effects are attributable to the treatment under study, and not to other factors. However, the quality of causal evidence for other outcomes is low because the analyses were based on a nonrandom subset of the randomized sample, and the author did not use sufficient controls when estimating impacts.",,"The author designed a randomized controlled trial in collaboration with 60 H&R Block locations in St. Louis, Missouri. Individuals who had filed their taxes at one of these locations in 2005 and returned in 2006 were eligible to participate. Members of the study sample were randomly assigned to one of several conditions, depending on the location at which they filed their taxes. The author constructed four study scenarios, in which all sample members, regardless of treatment condition, received a waiver for the IRA set-up fee:



In 19 offices (15,852 individuals), 10 percent of filers were offered a 50 percent match on one-time IRA contributions and 10 percent were offered a 33 percent credit rebate on an IRA contribution. Those who took up the credit rebate received 33 percent of their contribution amount as cash back. In these offices, 80 percent of filers were assigned to the control group.
In 20 offices (17,598 individuals), 10.4 percent of filers were offered a 50 percent match on one-time IRA contributions and received advanced notification of the match and fee waiver offer; 24.1 percent were offered a 50 percent match on one-time IRA contributions but did not receive advanced notification; 10.2 percent were assigned to the no-match group but did receive advanced notice of the IRA fee waiver; and 55.3 percent of filers were assigned to a control group that received no match and no advanced notification.
In 21 offices (14,878 individuals), 11 percent of filers were offered a 50 percent match on both one-time and monthly IRA contributions. The remaining 89 percent of filers were assigned to the control group.



Tax filers were randomly assigned to study groups using the last two digits of their Social Security numbers; both the tax filer and the tax preparer were blind to treatment assignment. The author estimated impacts on the IRA take-up rate and contribution amounts using H&R Block tax filing records.","The author estimated impacts for several contrasts:



Fifty percent match on one-time IRA contribution versus no match or credit control group: The average take up of an IRA was 5.72 percentage points higher and the average amount contributed was $50 greater in the 50 percent match group than the control group. Both impacts were statistically significant.
Thirty-three percent credit rebate on one-time IRA contribution versus no match or credit control group: The average take-up of an IRA was 2.4 percentage points higher and the average amount contributed was $19.85 greater in the 33 percent credit rebate group than the control group. Both impacts were statistically significant.
Fifty percent match versus 33 percent credit rebate on one-time IRA contributions: The offer of a 50 percent match increased the likelihood of making IRA contributions by a statistically significant 3.68 percentage points relative to the offer of a 33 percent credit rebate. However, there was no difference in the contribution amount.
Fifty percent match versus no match or credit control group on one-time and monthly IRA contributions: The match offer increased the likelihood of opening a one-time and systematic IRA by a statistically significant 1.44 percentage points relative to no match or credit offered.","The author randomly assigned all sample members to each treatment condition. However, in the treatment group that received advance notification, the author restricted the analytic sample to those who actually appeared at H&R Block to file their taxes in 2006. Because this is a nonrandom subset of those assigned to the advance notification group, the analyses involving this group must meet CLEAR evidence guidelines for nonexperimental studies, which require controls for age, gender, and income level. The author did not include controls for age or gender, so this analysis receives a low causal evidence rating.
The author noted that, because of tax preparer discretion, people in the study might not have received the treatment to which they were assigned. The author analyzed sample members based on the assigned treatment rather than the received treatment, so the effects presented in the study might underestimate the interventions’ impacts on those who were actually treated.",,,,,,,"Causal Impact Analysis",Adult,,"Employment-Mod/high-Favorable impacts",,"Behavioral Finance: Retirement Review Protocol"
"The effects of goal orientation on job search and reemployment: A field experiment among unemployed job seekers (van Hooft & Noordzij 2009)","The effects of goal orientation on job search and reemployment: A field experiment among unemployed job seekers (van Hooft & Noordzij 2009)","Study Type: Causal Impact Analysis","van Hooft, E., & Noordzij, G. (2009). The effects of goal orientation on job search and reemployment: A field experiment among unemployed job seekers. Journal of Applied Psychology, 94(6), 1581-1590.",,2009,,"April 2016","Behavioral Insights",,"High Causal Evidence","Randomized Control Trial (RCT)",,"The study’s objective was to examine the effects of goal orientation—the way individuals frame their objectives in a given situation—on job search and reemployment among unemployed job seekers.
The study was a randomized controlled trial (RCT) in which participants were assigned to one of three workshops: a learning goal-oriented (LGO) workshop, a performance goal-oriented (PGO) workshop, and a control workshop. The authors compared the job-search and reemployment outcomes of participants in the LGO or PGO workshops to the outcomes of those in the control workshop.
The study found that participants in the LGO workshop were 22 percentage points more likely to be reemployed eight weeks after the workshop than were participants in the control workshop, a statistically significant difference.
The quality of causal evidence presented in this report is high for the comparison of PGO versus control participants’ reemployment statuses because these outcomes derived from a well-conducted RCT with low attrition. This means we are confident that the estimated effects are attributable to workshop type and not to other factors. However, the quality of causal evidence presented in this report is low for the comparison of LGO and control participants’ job-search behaviors and reemployment statuses and the comparison of PGO and control participants’ job-search behaviors because these outcomes derived from an RCT with high attrition that did not control adequately for potential differences in the analytic samples. This means we are not confident that the estimated effects are attributable to workshop type; other factors are likely to have contributed.","Three types of workshops were included in the study. All lasted two to three hours and were held in groups of about five people.

The learning goal-oriented (LGO) workshop aimed for participants to use the workshop as an aid to improve their job seeking skills and master something new. Participants were asked to set learning goals and were given information and tools (an exercise book) to help them. Participants evaluated their competence according to whether they had mastered the task or developed their skills.
The performance goal-oriented (PGO) workshop aimed for participants to obtain the best possible results in their job search. Participants were asked to set performance goals and were given instruction and tools to help them, but were also encouraged to compete with one another and reward themselves when performing well. Participants evaluated their competence according to how they performed compared to others.
The control workshop had a similar structure to the LGO and PGO workshops but focused on exploring how participants' personalities related to their job search. At the workshop, participants completed a personality questionnaire. They then discussed the results of the questionnaire and how their personalities may be related to their job search.","The study took place at a reemployment counseling agency in the Netherlands. The study sample included 109 unemployed job seekers registered with the agency. The sample was 52.3 percent female with an average age of 46. Sample members had relatively little education: 56.0 percent had completed primary school or lower vocational training, 36.7 percent had graduated from high school, and 7.3 percent had completed college. Participants were randomly assigned to an LGO workshop, a PGO workshop, or a control workshop.
Participants completed a baseline questionnaire just before the workshop began and a similar questionnaire just after it ended. Participants also completed telephone interviews two and eight weeks after the workshop. The authors used regression analyses to compare the job-search and reemployment outcomes of participants in the LGO and PGO workshops with those in the control workshop, controlling for measures of job-search behavior and goal orientation collected in the pre-workshop survey.","The study found that participants in the LGO workshop were 22 percentage points more likely to be reemployed eight weeks after the workshop than were participants in the control workshop, a statistically significant difference.
Immediately following the workshops, participants in the LGO workshop demonstrated higher job-search intentions than did participants in the control workshop, a statistically significant difference. Participants in the LGO workshop also demonstrated significantly higher job-search intentions than those in the PGO workshop.
Two weeks following the workshops, participants in the LGO workshop exhibited greater job-search behaviors than did participants in the control workshop, a statistically significant difference.","The authors implemented an RCT, so outcomes that were measured through follow-up questionnaires administered immediately after the workshop, which all sample members completed, received a high causal evidence rating. However, not all sample members responded to the two- and eight-week follow-up interviews. In particular, attrition was high for the comparisons of LGO versus control and PGO versus control on job-search behavior outcomes and LGO versus control on reemployment outcomes. RCTs with high attrition can receive a moderate causal evidence rating if the authors either demonstrated that the analytic samples were equivalent before the intervention began or included adequate controls in the empirical model. In this case, the authors did not demonstrate equivalence on the analytic sample and did not control for age or a pre-intervention measure of employment, as CLEAR requires. As a result, the comparisons received a low causal evidence rating.",,,,,,International,"Causal Impact Analysis","Unemployed, Adult","Behavioral Interventions","Employment-Mod/high-Favorable impacts",Employment,"Behavioral Insights Review Protocol"
"Lessons from the US unemployment insurance experiments (Meyer 1995)","Lessons from the US unemployment insurance experiments (Meyer 1995)","Study Type: Descriptive Analysis","Meyer, B. (1995). Lessons from the US unemployment insurance experiments. Journal of Economic Literature, 33(1), 91-131.",,1995,,"April 2016","Behavioral Insights",,"Not Rated",Descriptive,,"The article provided an overview of experimental evaluations of unemployment insurance (UI) reforms conducted from 1977 to 1992 in the United States. These reforms typically tried to improve the employment prospects of beneficiaries and reduce UI costs.
The author reviewed 10 randomized controlled trials of UI reforms—4 cash bonus experiments and 6 job-search experiments—and provided a comparative analysis on how these reforms affected average weeks of UI benefits paid, UI program costs, and recipients’ earnings, when available. The cash bonus experiments offered a cash payment to participants who quickly found and kept a job for an indicated length of time, although the details varied greatly across experiments. Job-search assistance reforms also varied but typically involved changes in the way job-search programs were implemented in terms of services offered and reporting and participation requirements.
The review found that some cash bonus reforms reduced the average number of weeks participants spent on UI by a statistically significant margin, compared with the control group. The review also found that some job-search assistance reforms reduced participants’ average number of weeks on UI and increased average quarterly earnings compared with the control group; however, not all findings for all reforms were statistically significant.",,,,,,,,,,"United States","Descriptive Analysis","Adult, Dislocated or displaced worker, Unemployed","Disability insurance, Job search assistance and supportive services",,"Earnings and wages, Public benefits receipt","Behavioral Insights Review Protocol"
"Expressive writing and coping with job loss (Spera et al. 1994)","Expressive writing and coping with job loss (Spera et al. 1994)","Study Type: Causal Impact Analysis","Spera, S., Buhrfeind, E., & Pennebaker, J. (1994). Expressive writing and coping with job loss. Academy of Management Journal, 37(3), 722-733.",,1994,,"April 2016","Behavioral Insights",,"High Causal Evidence","Randomized Control Trial (RCT)",,"The study’s objective was to examine the impact of expressive writing activities on the reemployment of people who had recently lost their jobs.
The authors randomly assigned 41 people who had recently been laid off to either a treatment group, which was instructed to write daily about their feelings on their job loss, or to a control group, which was instructed to write about their plans for the day, but not about their feelings. Another 22 participants were assigned (non-randomly) to a comparison group that did not participate in a writing activity. The authors compared outcomes for the three study groups collected from administrative records and four surveys.
The study found that participants who wrote about their feelings and attitudes about being laid off were more likely to find employment eight months after the program than those who wrote about other topics and those who did no writing activities.
The quality of causal evidence presented in this report is high for comparisons between the treatment group and the writing control group because it was based on a well-implemented randomized controlled trial. This means we are confident that the estimated effects are attributable to expressive writing, and not to other factors. However, the quality of causal evidence is moderate for comparisons between the treatment group and the non-writing comparison group because there was non-random assignment to this group. This means we are somewhat confident that the estimated effects are attributable to expressive writing, but other factors might also have contributed.",,"The authors used a randomized controlled trial to test the impact of expressive writing, compared with writing about other topics, on reemployment outcomes for people who had recently lost their jobs. They also used a nonexperimental design to compare outcomes for the treatment group with those for a non-random comparison group that did not participate in any writing activities. Participants were recruited to the study five months after they had been laid off from the same large computer and electronics company.
Sixty-three participants volunteered for the study, which they were told involved participating in writing activities to help them with their job search. The authors randomly assigned 41 participants to either the writing treatment group (20) or the writing control group (21). Participants in the treatment group wrote about their thoughts and feelings about being laid off for 20 minutes every day for five consecutive days. These participants were instructed to engage deeply with their emotions during the writing. Those in the writing control group also wrote for 20 minutes every day for five consecutive days. However, they were instructed to write about their plans for their day and avoid writing about their opinions or feelings about being laid off. An additional group of 22 participants who volunteered for the study but were unable to attend the writing sessions were assigned to a non-writing comparison group. This group did not participate in any daily writing activities known to the researchers. Subjects in all three study groups had access to outplacement job services and an outplacement consultant throughout the project.
Participants ranged from 40 to 68 years old (with a mean age of 54), were engineers or other professionals, and had an average tenure at their former employer of 20 years. Statistical tests revealed no differences among the three groups at baseline in terms of their age, gender, race, and job search behavior.
The authors analyzed data collected from administrative records and four surveys with study participants to compare outcomes for the three study groups. They conducted statistical analyses to test for differences in the reemployment outcomes of the three groups eight months after the intervention.","More than half (53 percent) of the expressive writing treatment group had found full-time employment in the eight months after the study, compared with 24 percent of the writing control group and 14 percent of the non-writing comparison group.
More than two-thirds (68 percent) of participants in the expressive writing treatment group found some type of employment—including full-time, part-time, or contract employment—eight months after the study, compared with 27 percent of the non-writing control group. There was no statistically significant difference in the rate of any employment between the expressive writing treatment group and writing control group.","All of the study participants had access to reemployment services, so the study’s findings reflect the added benefit of expressive writing in addition to those services. In addition, the study included a relatively small number of participants from one company who volunteered to participate. All were professionals. Therefore, generalizability to other recently laid-off employees might be limited.",,,,,,"United States","Causal Impact Analysis","Unemployed, Adult","Behavioral Interventions","Employment-Mod/high-Favorable impacts",Employment,"Behavioral Insights Review Protocol"
"Framing and claiming: How information-framing affects expected social security claiming behavior (Brown et al. 2016)","Framing and claiming: How information-framing affects expected social security claiming behavior (Brown et al. 2016)","Study Type: Causal Impact Analysis","Brown, J. R., Kapteyn, A., & Mitchell, O. S. (2016). Framing and claiming: How information-framing affects expected social security claiming behavior. Journal of Risk and Insurance, 83(1), 139-162.","Brown, J., Kapteyn, A., & Mitchell, O. (2011). Framing effects and expected Social Security claiming behavior (Working paper no. 17018). Cambridge, MA: National Bureau of Economic Research.",2016,,"September 2016","Behavioral Finance: Retirement, Behavioral Insights",,"Low Causal Evidence","Randomized Control Trial (RCT)",,"The study’s objective was to examine the impact of different ways of framing retirement information on the age at which individuals intended to claim Social Security benefits.
The authors randomly assigned people who had been employed for at least 10 years and had yet to claim Social Security benefits to receive different messages about claiming the benefits. Each message contained the same financial information, but the information was presented—or framed—in different ways. There were 10 different messages—9 treatment conditions and one control condition that presented neutral information. Each study participant received 6 of the 10 messages.
The study found that, on average, those presented with the break-even treatment were more likely to indicate a younger intended age to receive benefits compared with those presented with the control condition, whereas those who received 4 of the other treatment conditions expressed an older intended age compared with the control condition.
The quality of causal evidence presented in this report is low because it was based on a randomized controlled trial with unknown attrition and the authors did not ensure that the groups being compared were similar before the intervention. This means we are not confident that the estimated effects are attributable to the framing of information about the age at which Social Security is collected; other factors are likely to have contributed.",,"To be eligible for the study, participants must not have already collected Social Security, must have worked for at least 10 years, and must not have retired yet (the study excluded those older than 55 who self-identified as retired). Of the baseline sample of 1,437 participants, 58.4 percent were women and 83.9 percent had some college, an associate’s degree, or higher education. Furthermore, 27.0 percent of the sample were ages 18 to 40, 28.2 percent were 41 to 50, 19.1 percent were 51 to 55, and 25.7 percent were older than 55. About one-fifth (21.1 percent) of the baseline sample had less than $35,000 in household income; 41.2 percent had household income from $35,000 to $74,999; and 37.7 percent had household income greater than $75,000.
The authors investigated whether the way in which information about Social Security benefits was presented could affect the age at which people intended to collect the benefits. RAND American Life Panel conducted a panel survey over three waves to test 10 different messages containing this information, including one neutral message that served as the control condition. Survey respondents received the same financial information in the following framing scenarios: the age of Social Security claiming as a break-even (anchored age 62) analysis, in which people who lived long enough would recoup foregone benefits before the age of claiming; eight different combinations of Social Security collection as the loss or gain of consumption or investment power anchored at different ages (62, 66, or 70); or a control condition with the same information presented in a symmetric and neutral fashion. Each survey respondent was randomly assigned to receive 6 different messages, 2 for each of the three waves of the survey (conducted about every two weeks). The authors examined the expected age at which respondents would collect Social Security benefits, measured as the number of months after the respondent turned 62 years old.","Framing information about the age at which Social Security is collected as a break-even analysis caused individuals to report they intended to claim benefits 22 months earlier, on average, than those exposed to the control condition.
Framing information about the age at which Social Security is collected as either a consumption or investment gain anchored at 66 years of age caused individuals to report they intended to claim benefits three months later, on average, than those exposed to the control condition.
Framing information about the age at which Social Security is collected as a consumption or investment loss anchored at 70 years of age caused individuals to report they intended to claim benefits about two to three months later, on average, than those exposed to the control condition.","Because respondents were presented with multiple frames within and across survey waves, there was the potential of spillover effects if respondents remembered their answers from the previous wave and adjusted their answers to be consistent.",,,,,,"United States","Causal Impact Analysis",Employed,"Federal retirement benefits, Behavioral interventions","Public benefits receipt-Low-Favorable impacts","Public benefits receipt","Behavioral Finance: Retirement Review Protocol"
"What do workplace wellness programs do? Evidence from the Illinois workplace wellness study (Jones, Molitor, & Reif, 2019)","What do workplace wellness programs do? Evidence from the Illinois workplace wellness study (Jones, Molitor, & Reif, 2019)","Study Type: Causal Impact Analysis","Jones, D., Molitor, D., & Reif, J. (2019). What do workplace wellness programs do? Evidence from the Illinois workplace wellness study. The Quarterly Journal of Economics, 134(4), 1747–1791.",,2019,https://academic.oup.com/qje/article/134/4/1747/5550759,"May 2021","Behavioral Insights",,"High Causal Evidence","Randomized Control Trial (RCT)",,"The study’s objective was to examine the impact of iThrive, a workplace wellness program developed at the University of Illinois at Urbana-Champaign, on employment, productivity, employer benefits receipt, and health and safety. 
	



	
	This study is a randomized controlled trial. The authors used a statistical model to compare the outcomes of treatment and control group members, using data from multiple sources including university administrative records, health insurance claims, online survey data from baseline and follow-up surveys, and health screening data.  
	
	
	The study found that the iThrive program significantly increased health screening rates at both 12 and 30 months after random assignment. The authors also found a statistically significant increase in perception of management’s priority on health and safety at 12—but not 30—months after random assignment. They did not find statistically significant changes in individuals’ medical spending, participation in runs, or number of campus gym visits. 
	
	
	The quality of causal evidence presented in this report is high for most outcomes because it was based on a well-implemented randomized controlled trial. This means we are confident that the estimated effects are attributable to iThrive and not to other factors. The quality of causal evidence is moderate for the long-term perception of management’s priority on health and safety outcome because attrition was high and could have led to imbalances across the experimental groups. This means we would be somewhat confident that any estimated effects would be attributable to iThrive, but other factors might also have contributed. However, the study did not find statistically significant effects on this outcome.  ","iThrive is a workplace wellness program developed and implemented at the University of Illinois at Urbana-Champaign. The program was designed to incorporate expert-recommended components of successful wellness programs, including a biometric screening, a health risk assessment (HRA), wellness activities, monetary incentives, and paid time off. 

The iThrive program was administered in two steps over one academic year. Authors ran the program twice over the span of two academic years with the same study participants. Step 1 included a five-week time frame for participants to schedule a biometric health screening at multiple locations on campus. After the screening, participants received an email invitation to complete an online HRA. They then received a scorecard that recommended areas of improvement based on the two assessments. Only participants who completed both assessments were eligible for the second step. In Step 2, iThrive invited individuals to participate in one wellness activity for the fall semester and one for the spring semester. Participants were encouraged to take paid time off to attend the wellness programs. Wellness activities included in-person classes on chronic disease management, weight management, tai chi, physical fitness, financial wellness, and healthy workplace habits as well as a tobacco quitline and an online wellness challenge. Depending on the treatment group participants were assigned to, they received a monetary reward of either $0, $100, or $200 for completing the biometric screening and the HRA and either $25 or $75 for each wellness activity completed. The total reward ranged from $50 to $350.   ","The study is a randomized controlled trial. The authors invited 12,459 benefit-eligible employees from the University of Illinois at Urbana-Champaign to participate in the iThrive wellness program. A total of 4,834 individuals enrolled by completing a 15-minute online survey on baseline health and wellness measures. Individuals received a $30 gift card upon completion. The authors randomly assigned 3,300 participants to six treatment groups and 1,534 to a control group. Treatment groups differed only in the amount of the monetary reward participants received, with rewards ranging from $50 to $350 per year of participation, and analysis was conducted by pooling across treatment groups. The authors used a statistical model to compare the outcomes of treatment and control group members at 12 and 30 months after random assignment. The analysis used data from multiple sources, including university administrative records; data from the Illinois marathon, 10k, and 5k runs; health insurance claims; baseline and follow-up surveys; health screening data; health questionnaire data; and online HRA and wellness activities data.  ","Health and safety. The iThrive program showed a statistically significant increase in employees’ health screening rates at both 12 months and 30 months after random assignment. The program also showed a statistically significant increase in perception of management’s priority on health and safety at 12 months, but results at 30 months were insignificant. There were no statistically significant impacts on any or amount of medical spending, participation in runs, or number of campus gym visits at 12 or 30 months. 
	


 


	
	Employment. The iThrive program showed statistically insignificant impacts on job promotion and job termination at both 12 months and 30 months after random assignment.  
	


 


	
	Employer benefits receipt. The iThrive program showed statistically insignificant impacts on days of sick leave taken at both 12 months and 30 months after random assignment. ","Random assignment was conducted within strata defined by type of employee (faculty, staff, or civil service), age, sex, race, and annual salary. The probability of assignment to each experimental group was equal across participants.  

In a separate analysis, the authors examined whether those who participated in iThrive differed from those who were invited to participate but did not. They found that participants had lower medical spending and healthier behaviors than non-participants before the intervention. ",,,,"Absence of conflict of interest.",,,"Causal Impact Analysis",,"Other health and safety, Other wages and benefits, Paid leave, Behavioral Interventions","Health and safety-Mod/high-Favorable impacts","Employer benefits receipt, Employment, Health and safety, Productivity","Behavioral Insights Review Protocol"