Employee financial literacy and retirement plan behavior: a case study.

Clark, Robert ; Lusardi, Annamaria ; Mitchell, Olivia S. 等

I. INTRODUCTION

Employers offer pension plans in the workplace to attract, retain, motivate, and ultimately retire their employees. Nevertheless, many workers are quite uninformed about financial matters, both in the United States and around the world (Lusardi and Mitchell 2014). Unless employees understand their plans and the incentives imbedded in them, they are unlikely to value them, save, invest, and manage their retirement portfolios appropriately. For this reason, employers have an interest in providing financial education to help workers better understand and make better decisions about their retirement savings.

Previous evidence has shown that workplace-based financial education programs can be beneficial. For instance, Allen et al. (2016) assessed employer-provided retirement seminars and showed that these seminars boosted financial literacy and influenced people's ability to plan for retirement.1 In our previous research (Clark, Lusardi, and Mitchell 2015; Lusardi, Michaud, and Mitchell 2016), we examined whether financial literacy is linked to investment returns and we found that more financially literate employees are better investors. Moreover, the higher returns earned by the more financially savvy are an important contributor to household wealth inequality.

Nevertheless, few studies to date have focused on how employer-provided learning modules can shape pension plan participation, contributions, and investment patterns. Below, we use a unique dataset on employees of the U.S. Federal Reserve (FR) System provided by the Office of Employee Benefits of the Federal Reserve System (OEB). These data were collected in connection with an on-line employer-provided educational module (hereafter, the Learning Module), which included a survey inquiring about FR employees' financial literacy. The linked data allow us to study the saving and investment patterns of the more- versus less-financially literate segments of employees who were offered the opportunity to save in a defined contribution (DC) plan. Specifically, we combine administrative records with a survey evaluating workers' financial knowledge. With these data, we can examine whether financial literacy is associated with higher participation and contribution rates in the employer plan. We also evaluate, for those who participate in the program, whether financial literacy influences saving responses after exposure to a learning module.

As one might expect, we find that FR employees perform better on the financial literacy survey compared to average Americans, and they also save at substantially higher rates. As in other studies, marital status, salary, tenure, and financial literacy are all associated with greater plan participation, and DC contribution levels are influenced by workers' age, salary, and financial literacy. We also find that more financially knowledgeable employees are much more likely to participate in their retirement saving plan, contribute a higher percentage of their salaries, and hold more equity in their DC retirement accounts. While this correlation is not definitively causal, we present some results suggesting that the relationships are real and not merely associations.

An important contribution of this research is that we show that participation in the Learning Module had large and significant effects on the retirement saving decisions of FR employees. Of those not participating in the retirement savings plan at baseline, employees who took the Learning Module had a 4.6 probability of starting to contribute post-Module, or 40% higher than their counterparts. Of those who stopped participation, those who took the Module had a 3.8 percentage point chance of stopping contributions, or half that of their counterparts. We also find that those who took the Module contributed 1.0% more of their salaries post-Module, for an improvement of 12%, and they boosted their equity share by 3.7 percentage points (compared to a baseline of 57.2%, or a 6.5% change). When we attenuate potential sample selection issues using inverse propensity weighting, the increases in contributions and equity shares are even larger. Those taking the Module contributed 11% of their salaries, more than doubling their pre-Module rates, and increased their equity share by 6.3 percentage points, for an 11% change. To summarize, the employer did have a positive impact on participation, contributions, and allocation to riskier assets as a result of implementing the Learning Module.

II. DATA AND METHODS

For this case study of FR employees, we focus on three measures of pension plan behavior: participation in the DC plan, the percent of salary contributed by those making contributions, and portfolio allocations by plan participants. All FR employees are covered by a relatively generous defined benefit (DB) plan and Social Security. In addition, the FR System offers its employees the opportunity to participate in a DC plan. The investment menu includes stock and bond index funds, lifestyle funds (conservative, moderate, aggressive), international and emerging market funds, and a real estate fund. Approximately 84% of the employees studied contributed to the DC plan. (2)

Anonymized pension plan data were drawn from administrative records for over 21,000 FR employees across the United States, as of September 2013. The items provided included each employee's contribution rates and investment allocations, as well as other items including sex, age, marital status, job tenure, salary, and DC plan balance. In October 2013, we also fielded an internet survey on financial knowledge, enabling us to link survey responses to the administrative data. Subsequently, the employer released to its workforce a Learning Module on the DC plan benefit offerings. (3) The goal for this Learning Module was to help workers assess their retirement spending needs; learn how Social Security and personal savings as well as the employer's retirement offerings interacted; and evaluate their DC plan savings. Below, we use the merged dataset to estimate empirical relationships between financial literacy, DC pension contributions, and investment allocations. We were granted access to similar administrative records in 2014, which allowed us to measure changes in retirement plan participation and contribution rates for those who participated in the Learning Module compared to those who did not. Thus, a distinguishing feature of this article is our ability to use administrative records to track actual behavioral changes before and after participation in the Learning Module, conditioned on financial literacy levels for those taking the survey. We were not able to randomize who was exposed to the Learning Module, yet our empirical work seeks to address potential selection issues to the extent possible.

III. MEASURING FINANCIAL LITERACY

With our input, the OEB designed and fielded an online survey on employee financial knowledge levels as part of a Learning Module provided to all employees via the FR's internal email system. In this section, we discuss the questions we have used to measure financial literacy while in the following section, we describe the employer's educational module. Of approximately 21.000 active workers, 16% responded to the invitation to take the survey. This response rate is similar to those in other voluntary and nonincentivized internet surveys. (4)

To compare the respondent sample with nonrespondents, we draw on information provided in the administrative data. The first column of Table 1 presents summary statistics from the administrative data for all employees. Over two-thirds of all employees contributed to the pretax plan alone, and 8% participated only in the Roth (after tax) plan. Interestingly, about 9% contributed to both types of plans.

Columns 2 and 3 report means for individuals who did versus did not complete the Learning Module. Module respondents contributed 2 percentage points more of their salaries to the DC plan. (5) They also had significantly larger plan balances, held 2.4 percentage points more equity, were about 4 years older, and had worked for the institution about 3 years longer than nonrespondents. Accordingly, it appears that the employees who participated in the Learning Module were likely to be more financially sophisticated than those who did not, a point to which we return below.

A. Financial Literacy Questions

We measured respondent financial knowledge using questions that have been tested in many other surveys (correct answers are provided in bold):

* Interest rate: Suppose you had $100 in a savings account and the interest rate was 2% per year. After 5 years, how much do you think you would have in the account if you left the money to grow? More than $110, Exactly $110, Less than $110, DK, RF. (6)

* Inflation: Imagine that the interest rate on your savings account was 1% per year and inflation was 2% per year. After 1 year, how much would you be able to buy with the money in this account? More than today, Exactly the same, Less than today, DK, RF.

* Risk: Is this statement True or False? Buying a single company's stock usually provides a safer return than a stock mutual fund. True, False, DK, RF.

* Tax offset: Assume you were in the 25% tax bracket (you pay $0.25 in tax for each dollar earned) and you contributed $100 pretax to an employer's 401 (k) plan. Your take-home pay (what's in your paycheck after all taxes and other payments are taken out) will then: Decline by $100, Decline by $75, Decline by $50, Remain the same, DK, RF.

* Match: Assume that an employer matched employee contributions dollar for dollar. If the employee contributed $100 to the 401 (k) plan, his account balance in the plan including his contribution would: Increase by $50, Increase by $100, Increase by $200, Remain the same, DK, RF.

The first question measures people's ability to do a simple interest rate calculation; the second tests people's understanding of inflation; and the third is a joint test of knowledge about "stocks" and "stock mutual funds" as well as knowledge of risk diversification, since the correct response requires the respondent to know both what a stock is and that a mutual fund is comprised of many stocks. The first three questions were developed by Lusardi and Mitchell (2008, 201 la) and used in the Health and Retirement Study and many other U.S. national surveys including the National Longitudinal Survey of Youth (Lusardi, Mitchell, and Curto 2010), the American Life Panel (Lusardi and Mitchell 2009), and the U.S. National Financial Capability Study (Lusardi 2011; Lusardi and Mitchell 2011b). They have also been fielded in 14 other countries. (7)

Previous research has demonstrated that very few Americans can answer all of the first three questions correctly (with similar results in other countries). Not many more know the correct answers to two of the three questions (Lusardi and Mitchell 2014). By contrast, the FR workforce is substantially more financially knowledgeable (as measured by the first three queries) than the general population, an unsurprising result given that the workforce consists of financial sector employees. Specifically, 75% of the respondents answered the interest rate question, 91% the inflation question, and 85% the risk question correctly. This can be compared to substantially lower correct response rates in other surveys. For example, Americans in the 2012 National Financial Capability Study (NFCS) averaged 75%, 61%, and 48% correct on the interest, inflation, and risk diversification questions. (8)

The last pair of financial knowledge questions, developed by Clark, Maki, and Morrill (2014), proved more difficult, even for this relatively sophisticated employee group. These queries required respondents to understand how employer and employee contributions influence pension accumulations, taking into account the tax implications of worker pretax contributions. In the tax offset question, the employee needed to understand that the tax-deferred aspect of own contributions reduces the net cost of payments into the plan. Here, only 43% of respondents answered correctly. The match question required the employee to know that a dollar-for-dollar match means that $100 of own contributions would generate an equal employer contribution. This last question was easier to answer, and 76% of FR employees scored correctly.

As is conventional, we sum the correct answers to these questions to produce a Financial Knowledge Index for each employee respondent. This Index ranges from 0 to 5, with a mean of 3.7 and a standard deviation of 1.2. (9) One-third of the FR respondents answered all five questions correctly, and just under 2% (46 respondents) could not answer any question correctly. The proportion answering each question correctly is shown in Table 2 along with the average number of correct answers.

B. Financial Literacy Heterogeneity by Employee Characteristics

Prior research has shown that the level of financial literacy is associated with a higher probability of planning for retirement, participating in retirement saving plans, and contributing to these plans. Yet little attention has been devoted to how financial literacy varies by employee characteristics and how this variation might influence econometric results. Table 3 illustrates the distribution of financial literacy as measured by our literacy index across economic and demographic variables. Interestingly, we find little difference in financial literacy across age groups, as the mean of the literacy index is 3.8 for those younger than age 60, and it only rises to 3.9 for those age 60 and over. Similarly, measures of literacy differ only modestly by years of tenure. Factors associated with larger differences include sex, marital status, and annual salary. Three-quarters of the male FR employees answered four or five of the literacy questions correctly, compared to only 56% of the females. Some 70% of married employees scored four or five correct answers, but only 60% of nonmarried individuals met this standard. By far, the largest differences are observed across levels of salary. Only a little over one-third of workers earning under $50,000 per year scored four or five right answers, and the mean number of correct answers was 2.9 of 5. By comparison, 96% of those with incomes over $250,000 correctly answered four or five questions with an average of 4.7 correct answers. This may be due to higher salary individuals being economists, but the findings related to sex, marital status, and income are overall similar to what is reported in many other papers (see Lusardi and Mitchell 2014, for a review).

IV. THE LEARNING MODULE

Our financial knowledge online survey was paired with a Learning Module developed by OEB entitled "Your Retirement Plan." The Module was available as a video which was introduced by a high-ranking member of the Federal Reserve Bank who chaired the Committee on Employee Benefits. It was sent to all employees as a video file via the company intranet, so employees could view it at times convenient to them. Employees were told that the video "shouldn't take more than 30 minutes to complete."

The video began with a discussion of income needs in retirement and noted that many analysts suggest that retirees should save enough to replace about 80% of their final earnings. The first section of the Learning Module discussed the importance of preparing for retirement and the likely differences in consumption needs before and after retirement. The video then went on to discuss the key components of retirement income including Social Security, the FR DB plan, the FR thrift saving plan, and personal savings. Next, the video described the DB pension plan, including how retirement benefits were calculated as a function of salary and service, along with providing information on how the DB plan protected retirees from inflation, investment, and longevity risk. Slightly different presentations were provided to employees closer to retirement versus those with a longer planning horizon. The video also discussed how to think about whether the employee needed to undertake additional retirement savings. Finally, it described distribution options and how these could be used to enhance retirement income streams. (10) Overall, the OEB's goal in delivering the Learning Module was to boost employee awareness of the need for retirement savings, so as to help its workforce build better retirement incomes.

V. BASELINE ANALYSIS

To explore how financial literacy is linked to DC plan outcomes, we first use multivariate analysis to examine plan contributions, percentage of salary saved in the DC plan, and equity allocation--all behaviors measured prior to the Learning Module. The next section examines changes postintervention.

Table 4 reports the results of three regression models where the dependent variable is equal to one if the employee contributed to the DC contribution plan (0 otherwise) in April 2013, prior to the Module. The first two columns report results using all current employees, whereas the third column uses only those individuals who completed the Learning Module. These regressions allow us to explore the determinants of participation and contribution to the DC plan, taking many factors simultaneously into account.

Results in Column 1 show that 84% of the FR employees contributed to their plan, with some differences by worker characteristics: men were 2.0 percentage points less likely to contribute to a plan compared to women, and each year of job tenure decreased the probability of contributing to a plan by 0.8 percentage points. Consistent with our prior research, we also find that those with higher earnings and married individuals were more likely to participate in the DC plan (Clark, Maki, and Morrill 2014). (11) Column 2 presents estimated coefficients from a model which adds an indicator variable taking the value of one for employees who participated in the subsequent Learning Module (0 otherwise). Inclusion of this variable in the model has no significant impact on the other estimated coefficients, but it does show that participants in the Module were 6.3 percentage points more likely to contribute to the DC plan. Evidently, those who elected to take the survey and participate in the educational module differed from their counterparts.

Column 3 reports results when we narrow the sample to only those employees who completed the Learning Module. In addition to the variables included in Column 1. two variables reflecting respondents' financial literacy levels are now included. One is an indicator for high financial knowledge (i.e., anyone who got four to five questions correct) versus medium (two to three questions correct) versus the reference category (at most one question correct). Our estimates indicate that the more financially knowledgeable employees were much more likely to contribute to the DC plan, and the estimated coefficients are strongly positive and statistically significant. The magnitudes imply that a worker scoring four to five correct answers was 13.4 percentage points more likely to be contributing, while someone scoring two to three answers correct had a 6.8 percentage point higher probability of contributing, compared to the least financially savvy employee who had at most one correct answer. (12) We are aware these estimates do not necessarily establish a causal link between financial literacy and contributions to a retirement plan. For example, unobservable factors such as patience could be driving this finding. Nevertheless, our other work has explored the casual links between financial literacy and retirement planning, and we showed that ordinary least squares (OLS) estimates (such as those reported here) tend to underestimate the true effect of financial literacy on retirement planning (Lusardi and Mitchell 2014). Moreover, in the next section, we provide some robustness analysis on this point.

Next, we examine plan contribution rates expressed as a percentage of employees' salary. Across all plan participants, the contribution rate as a percent of salary averaged 8.7% (see Table 5). Columns 1 and 2 again focus on all employees, while Column 3 limits the sample to survey participants only. The first two columns indicate that each additional 10 years of age was associated with 1.2 percentage points more pay being contributed to the DC plan. The more highly paid also contributed a larger share of their pay, and this effect was statistically significant. Including a control for who took the Learning Module in Column 2 indicates those who answered the survey contributed around 1 percentage point more of their pay, for an improvement of 12% compared to their nonrespondent counterparts (other coefficients are unaffected by including this additional variable). Finally, in the subsample of survey respondents, we see that being more financially literate is associated with higher contribution rates: that is, people who correctly answered four to five of the financial questions contributed 2.6 percentage points more of their pay, and those who answered two to three questions correctly contributed 1.4 percentage points more, compared to employees with at most one correct answer.

It is also of interest to evaluate whether financially knowledgeable DC plan participants invested differently from their counterparts. To examine how employees invest in addition to whether they contribute and how much, we have classified each fund in the DC plan menu according to its equity share, after which we compute each person's stock exposure as a function of his allocation to each of the various funds. Table 6 shows, for the full sample, that men held more of their retirement savings in equities, and an additional $10,000 in annual earnings increased the equity share by 1.4 percentage points. Not surprisingly, older workers favored less risky portfolios: an additional 10 years of age was associated with 5.7 percentage points less in risky investments. Learning Module respondents held 3.7 percentage points more of their assets in equities, and given the mean baseline of 57.2%, this is a 6.5% higher equity share. In the final column of Table 6, we report estimates for the subset of respondents to the Learning Module. Coefficient estimates here are very similar to those for the full sample. Additionally, we learn that an employee measured to be highly financially knowledgeable had 14.6 percentage points more of his retirement assets in equities compared to the least savvy. (13) Even those with intermediate knowledge held more in stocks than those who were less financially knowledgeable.

This analysis of plan contributions and holdings provides key insights about the heterogeneity of employee behavior, which we summarize as follows. First, a relatively large share of FR employees participated in the DC plan, and those who did contributed an average of almost 9% of their pay. Married employees were more likely to participate and older employees were more likely to contribute, while men participated and contributed less than women. Overall, this employee group held about 57% of its retirement assets in equities, with older workers holding slightly less, and men, married, and particularly the higher-paid workers taking riskier positions.

Second, employees who responded to the Learning Module were 6 percentage points more likely than nonrespondents to participate in the plan. Also they contributed 1 percentage point more of their pay and held somewhat more equity (4 percentage points). The fact that the Learning Module respondents differed somewhat from the nonrespondent workers suggests some degree of self-selection, about which we say more below. And finally, people scoring high on the financial knowledge survey were even more likely to participate in the plan, contribute even more of their salaries, and hold even more equity in their retirement accounts.

VI. BEHAVIORAL CHANGES POSTLEARNING MODULE

Several recent studies have examined how financial literacy influences retirement savings, and our results from the FR case study are consistent. Nevertheless, we can also move beyond simply correlating how financial literacy is associated with plan behaviors by estimating the impact of the Learning Module on retirement plan behavior. Specifically, we evaluate whether there were significant changes in retirement saving patterns after the module was fielded, by drawing on additional administrative data on participation, contributions, and investment allocations gathered 1 year after the initial information examined above.

Our pre/postsample includes 20,867 individuals for whom we have retirement saving data for the presurvey wave in 2013 as well as 6 months after the completion of the Learning Module in 2014. To determine the impact of participating in the Module, we examine factors associated with the probability that a participant contributed to the DC plan at baseline and then stopped contributing 1 year later. Results appear in Table 7, where the first column in each case shows results controlling only on the Learning Module dummy, and the second includes other controls.

Prior to the fielding of the Module, 17,538 FR employees contributed to the DC plan while 3,329 did not (in 2013). A year later, 7.1% of plan participants stopped contributing and 11.6% started contributing. Taking the Learning Module deterred plan dropouts, with a 3.7 percentage point reduced plan dropout rate (Column 1) compared to those who did not take the Module. In Column 2, we see that the Learning Module dummy is not altered when we control for other sociodemographics. Married employees and the more highly paid were less likely, while those with more tenure and older were more likely to stop contributing. In sum, taking the Module likely encouraged employees to remain in the plan.

Columns 3 and 4 of Table 7 examine whether employees who were not contributing to the DC plan in 2013 began contributing 1 year later. Of the employees who were not contributing to the DC plan in 2013, 11.6% were contributing a year later, and interestingly, people who completed the Learning Module were 2.9 percentage points more likely to initiate DC contributions, relative to those who did not, in Column 3. The effect in Column 4 is larger, with a 4.6 percentage point rise in contributors. By this measure. the Module had a significant positive effect on employee retirement savings, as almost 40% more of Module participants began contributing to a retirement saving plan. We also find that older employees not enrolled in 2013 were less likely to start contributing, as were workers with more years of service. By contrast, more highly paid employees were more likely to have enrolled a year later.

In summary, employees who completed the Learning Module were more likely to start contributing and less likely to have stopped contributing to the DC plan, compared to their nonrespondents. This suggests that participating in the program had a positive effect on DC plan behavior. Thus, from the standpoint of the employer, we have learned that providing employees with access to financial education can alter their retirement saving behavior. (14)

VII. CHECKS

As noted above, we were unable to randomize treatments across different subsets of the employee population, and we have shown that employees who elected to take the Learning Module differed in observable ways from those who did not. In an effort to investigate whether our results are sensitive to sample selection, we have undertaken two additional tests. One technique uses a Heckman 2-step model to first estimate the probability that employees took the Learning Module, followed by a second-stage model indicating the estimated effects of financial knowledge on changes in contributions and changes in equity allocation pre-/post- the Module. Controlling on this selection correction confirms that the most financially knowledgeable (with an index score of 4-5) boosted their contribution amounts and equity allocations substantially, and the effects are statistically significant (see Table A1 for detail). In particular, and after correcting for selection, those scoring highest on the knowledge index boosted their contributions by 2.5% (or 25% above the baseline), and increased their equity percentage by 14.6 percentage points (or 25% above the baseline). Evidently, the employer's intervention was quite influential for those who took advantage of the Learning Module. (15)

A second way to model sample selection uses a propensity matching model with inverse probability weights, a tactic more robust to misspecification bias (Todd 2014). Yet matching models also maintain that there is no systematic unobserved difference between those who were "treated"--that is, took the Learning Module--and their counterparts. (16) This is not directly testable with our data, but when we use this framework to estimate the impact of the Learning Module on changes in contribution and equity allocations, the estimated Average Treatment Effect (ATE) of the Module proves to be strongly positive and significant for both changes in contributions and equity share (see Table A2). Specifically, those taking the Module contributed 11 percentage points more (compared to a baseline of 9 percentage points), and held 6 percentage points more equity (compared to a baseline of 57%), after doing the Module. In other words, even after controlling for sample selection, we find a strong positive association between taking the Module and boosting contributions and equity allocations. Accordingly, the employer's effort to boost awareness of the FR retirement plan appears to have worked in the anticipated direction.

VIII. DISCUSSION AND IMPLICATIONS

Our analysis of FR employee behavior is only a case study of a single employer, yet several observations can be made. First, despite the fact that this workforce is covered by both a DB plan and Social Security, a remarkably high proportion of employees participates in and contributes to the DC retirement plan. Second, the surveyed employees' level of financial literacy is quite high compared to the general population. Third, we confirm, as in other studies, that marital status, salary, tenure, and financial literacy are all predictors of participation and contribution levels to DC plans. Fourth, the most financially knowledgeable employees are more likely to participate in their pension plan, contribute a higher percent of their pay, and hold more equity in their retirement accounts. In general, higher levels of financial literacy do appear to be associated with greater retirement readiness.

We also examine the effects of a Learning Module offered to the employees. When we compare behavioral changes among employees who took the Learning Module versus their counterparts who did not, we show that a significantly higher proportion of non-contributors who completed the Module subsequently enrolled in the retirement saving plan, compared to those who did not. Very few of those who viewed the Module stopped contributing to their DC plan, compared to those who did not view the Module. Moreover, the most financially knowledgeable were significantly more likely to boost contributions and equity allocation after having taken the employer's Learning Module. In other words, employer-provided information about the need to save for retirement increased workers' willingness to participate in and contribute to the DC vehicle. Accordingly, employers seeking reasons to develop such educational programs can be more confident that these programs will boost employee awareness and enhance retirement readiness. ABBREVIATIONS ATE: Average Treatment Effect DB: Defined Benefit DC: Defined Contribution DK: Do Not Know FR: Federal Reserve IPW: Inverse Propensity Weighting NFCS: National Financial Capability Study OEB: Office of Employee Benefits OLS: Ordinary Least Squares RF: Refuse to Answer

doi: 10.1111/ecin.12389

Online Early publication August 17, 2016

APPENDIX TABLE A1 Two-Stage Heckit Models Correcting for Sample Selection with Respect to Taking the Learning Module Took % Salary % Equity Learning Contribution Allocation Module 1 2 Probit Age 0.017 *** -0.001 -0.035 *** (0.001) (0.002) (0.009) Male 0.010 -0.005 * 0.024 (0.024) (0.003) (0.012) Married 0.032 -0.002 -0.033 (0.026) (0.004) (0.021) Salary ($ 10k) 0.017 *** -0.003 ** -0.020 ** (0.003) (0.002) (0.009) Tenure 0.004 *** 0.000 -0.009 *** (0.001) (0.000) (0.002) Med Fin. Knowl. 0.014 ** 0.054" Index (2-3) (0.006) (0.026) Hiah Fin. Knowl. 0.025 *** 0.146 *** Index (4-5) (0.005) (0.026) Lambda -0.172 -2.132 *** (0.120) (0.661) N 17,791 3,008 3.008 [R.sup.2] 0.070 0.065 0.149 Mean of dep var 0.169 0.101 0.587 SD of dep var 0.375 0.062 0.302 Note: Probit model in column 1, OLS in columns 2 and 3. *** Significant at the 0.01 level; ** significant at the 0.05 level; * significant at the 0.10 level. TABLE A2 Inverse Propensity Weighting Models: ATE Refers to Average Treatment Effect in the Population (Treatment Is Took Learning Module, ATE Reported) % Salary % Equity Contribution 1 Allocation 2 ATE 0.009 *** 0.036 *** (0.001) (0.006) POmean 0.085 *** 0.566 *** Learning Module (0) (0.000) (0.002) ATEa 0.106 *** 0.065 *** (0.015) (0.010) (a) ATE reported as a percentage. Also see Table A1. *** Significant at 0.01 level. TABLE A3 Weighted Comparisons Using Inverse Propensity Scores Variable Diff. Signif. (Nonpart-Part) % Participants in pretax only -3.02 *** % Participants in Roth only -1.03 * % Participants in both -2.67 *** % Salary contribution -2.02 *** Total balance ($ 100k), pretax only -0.89 *** % Balance in equity, pretax only -2.40 *** % Contribution in equity -1.77 ** Age -4.22 *** Male 0.00 Married -0.08 *** Salary ($ 10k) -0.87 *** Tenure (years) -3.05 Variable Diff. Signif. (Nonpart-Part) % Participants in pretax only 1.56 % Participants in Roth only -3.01 *** % Participants in both -5.64 *** % Salary contribution -1.43 *** Total balance ($ 100k), pretax only -0.26 *** % Balance in equity, pretax only -4.41 *** % Contribution in equity -3.59 *** Age -2.90 *** Male 0.02 Married -0.02 Salary ($ 10k) -0.79 *** Tenure (years) 0.07 Variable Notes % Participants in pretax only % Participants in Roth only % Participants in both % Salary contribution Total balance ($ 100k), pretax only % Balance in equity, pretax only % Contribution in equity Age Male Male excluded from IPW model Married Married excluded from IPW model Salary ($ 10k) Salary excluded from IPW model Tenure (years) Tenure excluded from IPW model Notes: The first column restates results from Table 1, testing for differences in characteristics of plan participants who took the Learning Module versus those who did not. The second column repeats the exercise but now weighting observations based on each respondent's Inverse Propensity Weighting (IPW) score. Also see Table A1. *** Significant at the 0.01 level: ** significant at the 0.05 level; * significant at the 0.10 level.

REFERENCES

Allen. S.. R. Clark, J. Maki, and M. Morrill. "Golden Years or Financial Fears: How Plans Change after Retirement Seminars." Journal of Retirement, 3(3). 2016, 96-115.

Austin, P. C. "An Introduction to Propensity Score Methods for Reducing the Effects of Confounding in Observational Studies." Multivariate Behavioral Research, 46, 2011, 399-424.

Bayer, P., D. Bernheim, and J. K. Scholz. "The Effects of Financial Education in the Workplace: Evidence from a Survey of Employers." Economic Inquiry, 47(4), 2009, 605-24.

Bernheim, D., and D. Garrett. "The Effects of Financial Education in the Workplace: Evidence from a Survey of Households." Journal of Public Economics, 87, 2003, 1487-519.

Cattaneo, M. D. "Efficient Semiparametric Estimation of Multi-Valued Treatment Effects under Ignorability." Journal of Econometrics, 155, 2010. 138-54.

Clark, R., and A. Lusardi. "How Employers Can Help New Hires Save for Retirement: Best Practices that Build Long-Term Financial Security." FINRA Investor Education Foundation, 2012. Accessed May 2015. http://www.finrafoundation.org/web/groups/foundation/@foundation/documents/foundation/p 188046.pdf.

Clark, R., M. d'Ambrosio, A. McDermed, and K. Sawant. "Retirement Plans and Saving Decisions: The Role of Information and Education." Journal of Pension Economics and Finance, 5(1), 2006, 45-67.

Clark, R., M. Morrill, and S. Allen. "Effectiveness of Employer-Provided Financial Information: Hiring to Retiring." American Economic Review, 102(3), 2012a, 314-18.

--. "The Role of Financial Literacy in Determining Retirement Plans." Economic Inquiry, 50(4), 2012b, 851-66.

Clark, R., A. Lusardi. and O. Mitchell. "Financial Literacy and Retirement Plan Saving and Investment among Federal Reserve System Employees." Final Report of the OEB, Federal Reserve Bank. November, 2014.

Clark, R., J. A. Maki. and M. S. Morrill. "Can Simple Informational Nudges Increase Employee Participation in a 401 (k) Plan?" Southern Economics Journal, 80(3), 2014, 677-701.

Clark. R., A. Lusardi, and O. Mitchell. "Financial Knowledge and 401 (k) Investment Performance." Journal of Pension Economics and Finance, Nov 25, 2015. Epub ahead of print. DOI: 10.1017/S1474747215000384.

Lusardi, A. "Americans' Financial Capability." NBER Working Paper 17103, 2011.

Lusardi, A., and O. S. Mitchell. "Financial Literacy and Retirement Preparedness. Evidence and Implications for Financial Education." Business Economics, 42(1), 2007. 35-44.

--. "Planning and Financial Literacy: How Do Women Fare?" American Economic Review, 98(2), 2008, 413-17.

--. "How Ordinary Consumers Make Complex Economic Decisions: Financial Literacy and Retirement Readiness." NBER Working Paper 15350, 2009.

--. "Financial Literacy and Planning: Implications for Retirement Wellbeing," in Financial Literacy: Implications for Retirement Security and the Financial Marketplace, edited by O. S. Mitchell and A. Lusardi. Oxford, UK: Oxford University Press, 201 la, 17-39.

--. "Financial Literacy and Retirement Planning in the United States." Journal of Pension Economics and Finance, 10(4), 2011b, 509-25.

--. "Financial Literacy around the World: An Overview." Journal of Pension Economics and Finance, 10(4), 201 lc, 497-508.

--. "The Economic Importance of Financial Literacy: Theory and Evidence." Journal of Economic Literature, 52(1). 2014, 5-44.

--. "Financial Literacy and Economic Outcomes. Evidence and Policy Implications." Journal of Retirement Summer, 3(1), 2015, 107-14.

Lusardi, A., O. S. Mitchell, and V. Curto. "Financial Literacy among the Young." Journal of Consumer Affairs, 44(2), 2010.358-80.

Lusardi, A.. P.-C. Michaud, and O. S. Mitchell. Forthcoming "Optimal Financial Knowledge and Wealth Inequality." Journal of Political Economy, 2016.

Simon, C.. J. Warner, and S. Pleeter. "Discounting, Cognition, and Financial Awareness: New Evidence from a Change in the Military Retirement System." Economic Inquiry, 53(1), 2015, 318-34.

Todd. P. E.. "Advanced Microeconometric Techniques and Applications." Unpublished Manuscript, University of Pennsylvania. 2014.

Clark: Professor, Department of Economics and Department of Management, Innovation, and Entrepreneurship, North Carolina State University, Raleigh, NC 27695. Phone 919-515-4568, Fax 919-515-6943, E-mail [email protected]

Lusardi: Professor, Department of Accountancy, The George Washington University School of Business, Washington. DC 20052. Phone 202-994-8410, Fax 202-994-8289. Email [email protected]

Mitchell: Professor, Department of Business Economics/ Policy, Wharton School University of Pennsylvania, Philadelphia, PA 19104. Phone 215-898-0424, Fax 215-898-0310, E-mail [email protected]

(1.) Other studies on workplace financial education programs include Bayer, Bernheim, and Scholz (2009), Bernheim and Garrett (2003), Clark et al. (2006), and Clark, Morrill, and Allen (2012a, 2012b, 2014). Clark and Lusardi (2012) examine the business case for such programs and present best practices used in employer-provided programs. For recent reviews, see Lusardi and Mitchell (2007, 2014, 2015).

(2.) See also Clark, Lusardi and Mitchell (2014, 2015).

(3.) For additional detail on implementation, see Clark, Lusardi, and Mitchell (2014).

(4.) This response rate is in line with what is found in many other online, nonmandatory, and nonincentivized surveys (cf. Clark, Maki, and Morrill 2014; Constant Contact http:// support2.constantcontact.com/articles/FAQ/2344; Benchmark http://www.benchmarkemail.com/help-FAQ/answer/ what-is-a-typical-survey-response-rate: Surveygizmo, http:// www.surveygizmo.com/survey-blog/survey-response-rates/).

(5.) Allen et al. (2016) also find that participation in retirement seminars significantly increased the financial literacy and knowledge of retirement plans of older workers and this greater knowledge affected retirement planning.

(6.) DK refers to "Do not know" and RF to "Refuse to answer."

(7.) For the list of countries, see http://gflec.org/ initiatives/flat-world/. See also the discussion in Lusardi and Mitchell (201 Ic).

(8.) The NFCS is also an online survey; see http:// www.usfinancialcapability.org/downloads/NFCS_2012_ Report_Natl_Findings.pdf

(9.) Inasmuch as respondents to the Learning Module are more likely to be financially knowledgeable than nonrespondents. it is safe to infer that the average level of financial literacy for the entire FR labor force is lower than the respondent average.

(10.) As with all videos and webinars, one cannot be certain whether the individuals watched some or all of the program or whether all individuals completing the financial literacy quiz also viewed the entire Module.

(11.) The dataset does not include measures of employees' defined benefit (DB) pension wealth, but since the DB benefit formula depends on salary and years of service, our analysis controls on employee age, salary, and tenure, to proxy for DB pension wealth.

(12.) Simon, Warner, and Pleeter (2015) extend the analysis of the choice of retirement plan distributions by incorporating cognition, financial literacy, and personal discount rates in those decisions. They conclude their analysis by stating that "More work is needed to examine the link between cognition, financial literacy, and investment choices. Hopefully, too, subsequent surveys will contain questions to elicit information about financial literacy." Our survey of FR employees provides new evidence on financial literacy and how it influences retirement saving decisions.

(13.) A more detailed analysis of how financial literacy influences the investment behavior and portfolio allocation of FR employees is available in Clark, Lusardi, and Mitchell (2015). This analysis shows that individuals with higher levels for financial literacy tend to have less idiosyncratic risk and earn higher expected risk-adjusted returns.

(14.) Table A3 shows results testing for differences in characteristics of plan participants who took the Learning Module versus those who did not, adopting the weighting based on each respondent's Inverse Propensity Weighting (IPW) score from Table A2. Overall, several factors become insignificant including tenure, male, and the probability of being in a pretax account only. Still significant but quantitatively smaller are the effects of contribution rates, total balance, age, and salary. Correcting for the sample selection in this way raises the estimated effect on the equity share.

(15.) We acknowledge that this test for sample selection approach relies on functional form for identification.

(16.) See Austin (2011) and Cattaneo (2010). TABLE 1 Comparison of Observables for Learning Module Respondents and Nonrespondents Variable Total Nonrespondents % Participants in pretax only 66.86 66.39 % Participants in Roth only 8.13 7.97 % Participants in both 8.96 8.53 % Salary contribution 7.32 7.00 Total balance ($100k), pretax only 1.51 1.37 % Balance in equity, pretax only 52.54 52.16 % Contribution in equity 57.19 56.89 Age 44.34 43.67 Male 0.57 0.57 Married 0.62 0.60 Salary ($10k) 9.82 9.68 Tenure (years) 12.26 11.78 N 21.192 17,835 Variable Respondents Diff. (Nonpart.-Part.) % Participants in pretax only 69.41 -3.02 % Participants in Roth only 9.00 -1.03 % Participants in both 11.20 -2.67 % Salary contribution 9.02 -2.02 Total balance ($100k), pretax only 2.26 -0.89 % Balance in equity, pretax only 54.56 -2.40 % Contribution in equity 58.66 -1.77 Age 47.90 -4.22 Male 0.57 0.00 Married 0.69 -0.08 Salary ($10k) 10.55 -0.87 Tenure (years) 14.83 -3.05 N 3.357 Variable T-Test for Significance % Participants in pretax only *** % Participants in Roth only * % Participants in both *** % Salary contribution *** Total balance ($100k), pretax only *** % Balance in equity, pretax only *** % Contribution in equity ** Age *** Male Married *** Salary ($10k) *** Tenure (years) *** N Notes: This table summarizes key variables for all 2013 Federal Reserve System 401 (k) plan participants who responded to the financial knowledge questions in the Learning Module as well as for those who did not. All data are taken from the institution's administrative records and refer to active employees (i.e., not retired, vested terminated, or deceased). *** Significant at the 0.01 level; **significant at the 0.05 level; * significant at the 0.10 level. TABLE 2 Results of Financial Literacy Assessment Financial Literacy Questions % Correct SD Suppose you had $100 in a savings account and the 0.75 0.44 interest rate was 2% per year. After 5 years, how much do you think you would have in the account if you left the money to grow? More than $110, Exactly $110, Less than $110, DK, RF Imagine that the interest rate on your savings 0.91 0.29 account was 1 % per year and inflation was 2% per year. After 1 year, how much would you be able to buy with the money in this account? More than today, Exactly the same , Less than today, DK, RF True or False? Buying a single company's stock 0.85 0.35 usually provides a safer return than a stock mutual fund. True, False, DK, RF Assume you were in the 25% tax bracket (you pay 0.43 0.50 $0.25 in tax for each dollar earned) and you contributed $100 pretax to an employer's 401(k) plan. Your take-home pay (what's in your paycheck after all taxes and other payments are taken out) will then: Decline by $100. Decline by $75, Decline by $50, Remain the same, DK. RF Assume that an employer matched employee 0.76 0.43 contributions dollar for dollar. If the employee contributed $100 to the 401(k) plan, his account balance in the plan including his contribution would: Increase by $50, Increase by $100. Increase by $200, Remain the same, DK, RF Financial Knowledge Index Score (number of 3.71 1.23 questions answered correctly) Note: DK refers to "Do not know" and RF to "Refuse to answer." TABLE 3 Heterogeneity in Financial Literacy Scores % Survey Respondents Index by Financial Knowledge Index (Row Percentages Reported) Financial Literacy Variable 0-1 2-3 4-5 Mean SD N Age <30 4.5 28.1 67.3 3.8 1.1 199 30-39 5.0 27.2 67.8 3.8 1.2 519 40-49 5.4 27.7 66.9 3.8 1.2 795 50-59 4.6 30.4 65.0 3.8 1.2 1.115 [greater than or equal to] 60 2.9 27.9 69.2 3.9 1.1 380 Sex Men 2.9 22.5 74.7 4.0 1.1 1.741 Women 7.1 37.2 55.7 3.5 1.2 1,267 Marital status Married 4.0 26.5 69.5 3.9 1.1 2.079 Nonmarried 6.1 33.5 60.4 3.6 1.2 929 Tenure < 10 years 4.8 27.7 67.5 3.8 1.2 1.348 10-14 years 3.1 28.0 68.9 3.9 1.1 354 15-19 years 5.3 31.3 63.4 3.7 1.2 374 [greater than or equal to] 20 years 4.7 29.3 66.0 3.8 1.2 932 Salary <$50,000 15.6 48.3 36.1 2.9 1.3 263 $50,000-199,999 6.2 36.7 57.2 3.5 1.2 1,235 $100,000-$149.999 1.7 21.0 77.4 4.1 1.0 1,007 $150,000-$199,999 1.7 15.4 82.8 4.2 1.0 344 $200,000-$249,999 0.0 14.7 85.3 4.3 0.8 109 >$250,000 0.0 4.0 96.0 4.7 0.6 50 TABLE 4 Determinants of Pension Participation: All Employees, April 2013 Full Sample Full Sample Surveyed in 2013 Age 0.0004 0.0002 0.0004 (0.0003) (0.0003) (0.0006) Male -0.0199 *** -0.0191 *** -0.0044 (0.0047) (0.0047) (0.0093) Married 0.0094 * 0.0090 * 0.0037 (0.0051) (0.0051) (0.0101) Salary ($ 10k) 0.0183 *** 0.0177 *** 0.0078 *** (0.0007) (0.0007) (0.0015) Tenure -0.0080 *** -0.0080 *** -0.0045 *** (0.0003) (0.0003) (0.0005) Learning Module 0.0627 *** (0.0054) Med. Fin. Knowl. 0.0675 *** Index (2-3) (0.0122) High Fin. Knowl. 0.1343 *** Index (4-5) (0.0225) N 21,192 21,192 3,357 [R.sup.2] 0.142 0.147 0.133 Mean of dep var 0.840 0.840 0.896 SD of dep var 0.367 0.367 0.305 Note: Probit models, marginal effects reported. * p <.10; ** p <.05; ** p <.01. TABLE 5 Determinants of Employee Contributions in 2013 (as a % of Salary) Plan Contribution Rate Plan Plan Surveyed and Participants Participants in Plan in 2013 Age 0.0012 *** 0.0012 *** 0.0012 *** (0.0001) (0.0001) (0.0001) Male -0.0020 ** -0.0021 ** -0.0033 (0.0009) (0.0009) (0.0024) Married 0.0000 0.0000 0.0022 (0.0009) (0.0009) (0.0026) Salary ($ 10k) 0.0004 *** 0.0004 *** -0.0008 *** (0.0001) (0.0001) (0.0002) Tenure 0.0001 0.0000 0.0002 * (0.0001) (0.0001) (0.0001) Learning Module 0.0104 *** (0.0012) Med Fin. Knowl. 0.0143 ** Index (2-3) (0.0057) High Fin. Knowl. 0.0255 *** index (4-5) (0.0055) N 17,791 17,791 3.008 [R.sup.2] 0.073 0.078 0.064 Mean of dep var 0.087 0.087 0.101 SD of dep var 0.057 0.057 0.062 Note: OLS models. * p <.10, ** p < .05, ** p <.01. TABLE 6 Determinants of Portfolio Share in Equities: Employees in 2013 Plan Plan Surveyed and Participants Participants in Plan in 2013 Age -0.0057 *** -0.0058 *** -0.0065 *** (0.0002) (0.0002) (0.0006) Male 0.0498 *** 0.0497 **** 0.0393 *** (0.0043) (0.0043) (0.0108) Married 0.0160 *** 0.0159 *** 0.0230 ** (0.0047) (0.0047) (0.0117) Salary ($ 10k) 0.0139 *** 0.0137 *** 0.0070 *** (0.0005) (0.0005) (0.0014) Tenure -0.0035 *** -0.0035 *** -0.0033 *** (0.0003) (0.0003) (0.0006) Learning Module 0.0369 *** (0.0059) Med Fin. Knowl. 0.0559 ** Index (2-3) (0.0264) High Fin. Knowl. 0.1463 *** index (4-5) (0.0260) N 17,791 17.791 3.008 [R.sup.2] 0.107 0.109 0.146 Mean of dep var 0.572 0.572 0.587 SD of dep var 0.298 0.298 0.302 Note: OLS models. * p <.10, ** p < .05, *** p < .01. TABLE 7 Probability of Stopping or Starting Contributions in 2014, Relative to 2013, as a Function of Taking the Learning Module and Other Variables Stop Participating Plan Participants in 2013 Learning Module -0.0370 *** -0.0377 *** (0.0042) (0.0041) Age 0.0005" (0.0002) Male 0.0129 *** (0.0038) Married -0.0189 (0.0042) Salary ($ 10k) -0.0021 *** (0.0005) Tenure 0.0008 *** (0.0002) N 17,538 17,538 [R.sup.2] 0.009 0.018 Mean of dep var 0.071 0.071 SD of dep var 0.257 0.257 Start Participating Nonparticipants in 2013 Learning Module 0.0289* 0.0463 *** (0.0175) (0.0177) Age -0.0021 *** (0.0005) Male -0.0167 * (0.0085) Married -0.0002 (0.0087) Salary ($ 10k) 0.0061 *** (0.0010) Tenure -0.0050 *** (0.0005) N 3,329 3,329 [R.sup.2] 0.040 0.168 Mean of dep var 0.116 0.116 SD of dep var 0.320 0.320 Note: Probit models, marginal effects reported. *p < . 10; **p < .05, ***p <.01.