What will my account really be worth? An experiment on exponential growth bias and retirement saving (Goda et al. 2012)
Goda, G.S., Manchester, C.F., & Sojourner, A. (2012). What will my account really be worth? An experiment on exponential growth bias and retirement saving. National Bureau of Economic Research working paper 17927. Cambridge, MA: NBER.
By clicking the links below, you are leaving CLEAR and are subject to the privacy and security policies of the owners/sponsors of the external site. CLEAR does not control or guarantee the accuracy, relevance, timeliness, or completeness of information contained in a linked site. We also do not endorse the organizations or individuals maintaining sites that we link to, any views they express, or any products/services they offer.
The original publication was found at:
If this link no longer works, you can try a Google search for the citation.
Goda, G.S., Manchester, C.F., & Sojourner, A. (2014). What will my account really be worth? Experimental evidence on how retirement income projections affect saving. Journal of Public Economics, 119, 80-92.
- The study’s objective was to determine how general or personalized information on projected retirement savings account balances and annual retirement income affects employees’ retirement savings decisions.
- A group of university employees were randomly assigned to one of four conditions:
- Participants in the planning treatment group received general retirement savings information, including enrollment instructions for the university’s voluntary retirement plan.
- Participants in the balance treatment group received all planning treatment materials plus personalized projections of retirement fund balances and access to an online customization tool.
- Participants in the income treatment group received all planning and balance treatment materials plus a personalized projection of income during retirement.
- Participants in the control group received no intervention and were in departments with only control group members.
- The study analyzed retirement-fund contribution data from the university’s Office of Human Resources using fixed-effects regressions. The study also collected and analyzed additional survey data regarding financial literacy, tolerance for delayed gratification, tendency to procrastinate, and beliefs about saving.
- Participants in the income treatment group contributed $85.42 more on average, annually, to their retirement savings accounts than did members of the control group. Both this group and those in the balance treatment group exhibited statistically significant increases in the probability of changing their contribution amounts of 1.2 and 1.4 percentage points, respectively.
- The quality of causal evidence provided in this study is high. This means that we are confident that the estimated effects are attributable to the different levels of the intervention and not to other factors.
The study considered three different interventions: (1) an informational intervention, in which people received general retirement savings information and information on the options provided by, how to sign up for, and how to change contributions to the university’s voluntary retirement plan; (2) an intervention that provided personalized projections estimating the impact of additional contributions on assets at retirement and access to an online planning tool; and (3) an intervention that provided personalized projections translating additional contributions into changes in annual retirement income.
The study took place at several University of Minnesota campuses from October 2010 to May 2011 and involved university employees. University departments were grouped into strata and randomly assigned to one of four groups: a control group that received no intervention, a planning treatment group that received only the first intervention, a balance treatment group that received the first and second interventions, and an income treatment group that received all three interventions.
The study examined intervention impacts using administrative data on retirement fund contributions from the university’s Office of Human Resources. A supplemental follow-up survey was also fielded to explore behavioral factors related to savings decisions, but response rates were very low (22 percent). Survey questions gauged respondents’ tolerance for delayed gratification; tendency to procrastinate; financial literacy; and beliefs regarding expected retirement income, rates of return, and retirement ages.
The authors assessed the effect of each treatment level on retirement savings contributions using a linear regression including stratum-level fixed effects and individual-level controls for age, tenure, salary, change in salary over the course of the intervention, gender, employment type (faculty versus staff), and campus site.
- Participants in the income treatment group contributed significantly more annually than the control group, with an average additional contribution of $85.42.
- The balance and income treatment groups were 1.4 and 1.2 percentage points, respectively, more likely to change their contributions—that is, either enroll in the plan or modify their contribution levels, with both differences statistically significant at the 1 percent level.
- Despite apparent gains in retirement savings contributions among the treatment groups, the authors did not reject the joint hypothesis that all three treatments had no effect on contribution amounts. However, all three treatments, taken together, had a statistically significant effect on participants’ propensity to change their contribution levels.
- Based on responses to the follow-up survey, participants in the income treatment group were 12 percent more likely to have determined how much they need to save for retirement. Treatment groups did not differ in their reported understanding of how savings today affect retirement income.
Considerations for Interpreting the Findings
Contribution rate analyses relied on administrative data from a randomized controlled trial (RCT) with low attrition; however, response rates to the follow-up survey were so low that these outcomes were derived from an RCT with high attrition. The authors established that individual characteristics, such as income and age, do not predict survey response differently in the treatment and comparison groups. Additionally, there is little reason to believe that the composition of survey respondents differed systematically across treatment groups. Thus, attrition likely did not lead to large biases in the survey data analyses.
The authors note that the participants in their study might be more educated and have a higher level of financial literacy compared with the average individual. This suggests that the results of this study might not apply to broader populations of workers.
The study authors estimated multiple related impacts for both changes in contributions and indicators suggesting any change in contributions. Performing multiple statistical tests on related outcomes makes it more likely that some impacts will be found statistically significant purely by chance. Although the authors correct for multiple comparisons within a single model by testing the joint hypothesis that all three treatment effects—the effects of the planning, balance, and income conditions—are equal to zero, this adjustment does not correct for the large number of models presented in the study. Thus, some of both the joint and individual hypothesis tests are likely to exceed the statistical significance threshold by chance alone.
Causal Evidence Rating
Based on the analysis of administrative data, the quality of causal evidence provided in this study is high. This means that we are confident that the estimated effects on contribution rates are attributable to the three retirement savings interventions and not to other factors. Low response rates to the follow-up survey result in a moderate causal evidence rating for this component of the study. This means we are somewhat confident that the estimated effects are attributable to the informational interventions, but other factors might also have contributed.