{"paper_id":"33f9fdc7-50cb-4dbc-9305-2a705a5fef54","body_text":"Behavioural messages amplify tax incentives: A nationwide megastudy of retirement savings reminders | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Behavioural messages amplify tax incentives: A nationwide megastudy of retirement savings reminders Heidi Reinson, Thomas Post, Nina Mazar, Crystal Reeck, Stylianos Syropoulos, and 3 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8008072/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted You are reading this latest preprint version Abstract Insufficient retirement savings threaten financial security worldwide as individuals increasingly bear responsibility for funding their retirement. We present the first population-level test of behaviourally informed messaging for boosting saving in retirement accounts. In collaboration with the Estonian Ministry of Finance, we conducted a preregistered randomized controlled trial among all eligible account holders (N = 127,974) testing whether email reminders about tax benefits boost voluntary pension contributions. Across nine behaviourally designed reminders versus a no-reminder control, reminders increased contribution likelihood by 10.49% and raised average contributions by 14.75% within one week before the deadline. Most framings increased participation but notably, a family-security message also increased individual contribution size (from €868 to €1,009). Exploratory causal forest analysis revealed the baseline message most strongly influenced younger, higher-income individuals with prior contributions, while the family-security message had the largest impact on high-income individuals. The intervention generated ~€1.2 million in additional retirement savings within one week. These results show that simple, low-cost behavioural reminders can meaningfully amplify tax incentives at scale, offering practical tools for policymakers. Social science/Psychology/Human behaviour Social science/Economics Figures Figure 1 Figure 2 Main text Inadequate retirement savings pose a growing challenge worldwide. For example, half of American households risk being unable to maintain their living standards in retirement (Munnell, Chen, & Siliciano, 2021). As demographic shifts and fiscal pressures limit governments' ability to provide supplementary support (Gomes et al., 2021), saving for retirement is increasingly falling on the shoulders of individuals. Understanding how to help individuals overcome psychological barriers to increasing their retirement savings has therefore become a key priority for policymakers. Saving sufficiently for retirement does not come naturally to most people. “For many people, being asked to solve their own retirement savings problems is like being asked to build their own cars,” according to Nobel laureate Richard Thaler (New York Times, 2013). Multiple psychological barriers impede effective retirement planning. Individuals heavily discount distant future outcomes (Berns, Laibson, & Loewenstein, 2007; Ruggeri et al, 2022), struggle with the self-control required to save consistently (Laibson, 1998; Kim, Lee, & Hong, 2016), and tend to view their future selves as strangers, making it difficult to prioritize their future financial needs over present desires (Hershfield, 2011). These barriers can shape behaviour in striking ways, with individuals typically spending more time choosing a television set than planning for retirement (TIAA-CREF, 2014) and preferring household chores like scrubbing floors to reviewing their savings accounts (Monzo Bank, 2024). Various policy approaches can be used to encourage retirement savings. Mandatory and auto-enrolment pension schemes can effectively steer individual behaviour, but these more paternalistic approaches can face resistance. Furthermore, the long-term benefits of auto-enrolment can be undermined by early withdrawal options and increased household debt burdens (Beshears et al, 2024; Choi et al, 2024). To avoid these downsides, policymakers may want to consider less restrictive measures such as tax benefits that encourage proactive savings while preserving individual choice. However, even well-designed tax benefits can fail if individuals either do not know about them (Chan and Stevens, 2008) or lack the attention and motivation to act on them (Karlan et al, 2016). Reminder messages offer a promising solution to these attention and motivation challenges, as they can effectively direct focus to important goals and opportunities that might otherwise be overlooked (Gravert, 2022). Reminders have proven effective across various policy domains from safety, health, and education (Austin et al, 2006; Vervloet et al, 2012; Duckworth et al, 2025) to financial decision-making (Saez, 2009; Karlan et al, 2016). Recent evidence from the United States suggests email reminders can modestly increase short-term liquid savings, with effects ranging from 0.5-1.3% (Milkman et al., 2025). However, the impact of behavioral messaging on tax-incentivized retirement savings—where the stakes, time horizon and incentives differ substantially—remains unexplored. Building on this evidence, our first aim was to test whether email reminders could enhance the impact of tax benefits on retirement savings. While reminders can be effective on their own, behavioural science suggests their impact can be amplified through careful message design (Gravert, 2022). However, the context-specific nature of behaviour change makes it difficult to predict which messaging strategies will be most effective in a particular domain. Our second aim was therefore to systematically identify the most effective messaging approaches for enhancing the impact of tax incentives on retirement savings. Using a megastudy—a single experiment comparing multiple interventions simultaneously against a common control (Milkman et al., 2021; Duckworth & Milkman, 2022; Voelkel et al., 2024), we tested eight behaviourally informed reminder messages alongside a neutral reminder message by comparing their effectiveness relative to a no-reminder control. The impact of behavioural interventions often varies across different population segments, making it crucial to understand who responds best to which approaches (Sunstein, 2022). For instance, a recent study of retirement savings reminders in Mexico found that while the interventions were effective on average, they backfired for certain subgroups (Shah et al., 2023). Our third aim was therefore to examine how message impact varied across key characteristics including age, gender, and past pension saving behaviour. We focused on variables that pension administrators can readily observe and therefore potentially use for personalized reminders in the future. These three aims—the general effectiveness of reminders, the relative impact of different messaging strategies, and the potential for heterogeneous effects—were investigated in a nationwide pre-registered megastudy targeting the entire population of eligible pension account holders in Estonia. We assessed how different email messages reminding recipients of a tax benefit deadline impacted both the likelihood and size of pension contributions. In contrast to laboratory studies (often with student populations) that comprise a large share of the behavioural intervention literature, our field experiment provides evidence from a representative decision environment, crucial for establishing the generalizability of results and their relevance for policy (List, 2007; Saccardo et al., 2024; Crum et al., 2024). Estonia is a uniquely suitable setting for this research. The country offers substantial tax benefits through its ‘3rd pillar’ voluntary pension scheme, similar to U.S. 401(k) plans, allowing individuals to claim a 20% income tax return on contributions up to 15% of their gross income (maximum 6,000€ per year). [1] Despite these incentives, the expected proportion of pension benefits to pre-retirement income (i.e., pension replacement rate) remains at just 34%, well below the OECD average of 61% (OECD, 2023). This gap creates both a clear need for interventions and an ethically sound basis for encouraging increased savings. All Estonian working-age residents with a voluntary 3rd pillar retirement account (n = 127,974), representing roughly one fifth of the working-age population were randomly assigned either to a no-reminder control group (n = 23,238) or to one of nine treatment groups (averaging 11,640 participants per group), each receiving either a neutral or a behaviourally informed reminder message. The sample yielded 80% power to detect a 0.6 percentage point increase when comparing all reminders combined against control, and a 0.9 percentage point increase for individual message comparisons (α=0.05, baseline rate=8.96%). The randomized groups were comparable across all available moderator variables (see Appendix D). The reminders were sent once one week before the end-of-year tax deadline in December 2023 by the Estonian pension registry to account holders’ private email addresses. We analysed the contributions account holders made to their 3 rd pillar accounts during that final week before tax deadline. The reminder messages, seen in Table 1, were developed through three stages: (1) systematic review of behavioural interventions in retirement savings, (2) co-creation with an international consortium of behavioural scientists and Estonian pension experts, ensuring both theoretical rigor and contextual relevance, (3) pre-testing with Estonian adults for comprehension and cultural appropriateness. Each message highlighted different behaviourally informed motivational cues, including salient tax returns, future income, social norms, loss aversion, family security, and urgency. We followed a pre-registered analysis strategy with clearly marked additions throughout the paper to address our three key research questions. Can timely reminders increase retirement savings? We examined this research question by comparing the combined treatment groups against the control group on three metrics: (a) the likelihood of making additional contributions, (b) the average contribution amount in Euros across all participants (including non-contributors), and (c) the size of contributions among those who contributed (not preregistered). How does the effectiveness of reminders vary by messaging strategy? We compared each treatment group against both the control group and other treatment groups, examining the same three outcome measures to identify the most effective messaging approaches. Do reminder effects vary across different population segments? We investigated whether message effectiveness depended on the key demographic and behavioural characteristics, including age, gender, and past saving behaviour in both the mandatory (2 nd pillar [2] , a proxy for income) and voluntary (3 rd pillar) pension schemes. Our pre-registered analysis examined these variables individually as moderators, with p-values adjusted for multiple comparisons. In a non-preregistered additional analysis, we used causal honest forest machine learning (Athey & Imbens, 2016) to examine how combinations of moderator variables might jointly influence treatment effectiveness. Table 1: Reminder conditions Condition Text Rationale and references No-reminder control n/a Status quo 1 Baseline reminder* Dear III pillar account holder, <[treatment text here for conditions 2-10]> Contributions made to the III pension pillar until December 27 will be included in the income tax refund for 2023. You can get back 20% of your III pillar payment. Check if you are taking advantage of this opportunity: Questions and answers about the III pillar tax refund The income tax refund applies to III pillar contributions up to 15% of your gross income, but not more than 6,000 euros per year. Yours sincerely, Pension Center Provides a simple reminder at the tax-relevant moment to address inattention. Evidence suggests timely prompts can increase financial actions (Karlan et al., 2016). Uses formal language shown to be effective in public sector communications (Linos et al., 2024). The main text was included in all the e-mail messages, as we were required to provide this information to all groups. Only the first sentence varied between conditions. 2 Loss aversion Only a few days left, don’t miss out on your tax refund! Frames the tax benefit as something to avoid losing rather than gaining. Loss-framed messages have shown effectiveness in some retirement savings contexts (Eberhardt et al., 2021). 3 Psychological ownership Only a few days left, your tax refund is waiting for you! Presents the tax refund as already belonging to the recipient. Ownership framing has increased engagement in government benefits and health behaviours (De La Rosa et al., 2021; Buttenheim et al., 2022). 4 Short- & long-term gain Increase your future income as well as your next tax return! Highlights both temporally close and distant gains, addressing different time preferences. Gain framing has been effectively used in pension communication (Saez, 2009; Eberhardt et al., 2021). 5 Investment boost With the tax refund, you can further increase the return on your investment! An alternative gain frame reframes pension contributions as smart investments rather than savings. Developed with Estonian financial experts to match local customer insights. 6 Pennies-a-Day Even small amounts help secure your future! Reduces perceived barriers by emphasizing small contributions. Reframing large amounts as smaller units has shown effects in charitable giving and retirement contexts (Gourville, 1999; Shah et al., 2023). 7 Family security Help secure the future for yourself and your loved ones! Expands motivation beyond self-interest to include loved ones. Legacy-oriented appeals have increased prosocial financial behaviours in multiple contexts (Zaval et al., 2015; Shrum, 2021; Syropoulos et al., 2023) and increased pension savings (Shah et al, 2023). 8 Social norms In recent years, the number of people saving in the III pillar has doubled to nearly 200,000! Provides information about peer saving behaviour. Descriptive norms can influence financial decisions, though effects vary by context (Bursztyn et al., 2014; Dur et al., 2021). 9 The power of now What is done today, you don’t have to worry about tomorrow! Creates urgency through immediate action framing. Uses familiar Estonian proverb previously effective in health behaviour reminders (Rüütsalu et al., 2023). *Full version in Estonian, Russian, and English languages in Appendix B Results We present findings according to our three research questions. Figure 1 summarizes the effects of each intervention on all outcomes. Can timely reminders increase retirement savings? Our analysis revealed that reminders significantly increased pension contributions. Logistic regressions showed that the likelihood of making a payment to one’s 3 rd pillar account during the final week before the tax deadline rose from 8.96% in the no-reminder control group to 9.90% in the reminder groups, a 10.4% significant relative increase (OR = 1.12, 95% CI [1.08, 1.16], p < .001; Fig 1 top left panel). A Tweedie GLM regression demonstrated that reminders significantly increased the average contribution by 14.75% from 77.76€ to 89.23€ (exp(b) = 1.15, 95% CI [1.05, 1.25], p = .001, Fig 1 top middle). Importantly, this average includes the majority of participants who contributed 0€. A subsequent non-preregistered Tweedie GLM regression revealed that among those who did contribute, the reminder messages did not significantly affect contribution amount (Fig 1 top right). This pattern suggests that reminders primarily worked by motivating more people to contribute rather than by increasing the size of individual contributions. How does the effectiveness of reminders vary by messaging strategy? We compared each message to the no-reminder control by using logistic regression for contribution likelihood and Tweedie regression for contribution amounts. We computed simultaneous confidence intervals to account for multiple comparisons. Most messages significantly increased the likelihood of contributing (Fig 1 bottom left). However, only the family security message (“Help secure the future for yourself and your loved ones!”) significantly increased the average contribution, showing a 28.4% increase from 77.76€ to 89.23€ (exp(b) = 1.28 95% CI [1.07, 1.54], p = .001; Fig 1 bottom middle). Additional non-preregistered analyses of average contribution size among contributors showed no significant effects, though the family security message approached significance (exp(b) = 1.16 95% CI [0.99, 1.35], p = .060; Fig 1 bottom right). In post-hoc pairwise comparisons between all reminders on all three dependent variables, we found remarkably consistent effects across different messages. The only significant difference emerged in contribution size, where the family security message (1,008.80€) outperformed the Loss Aversion message (821.80€) by 187€ (p = .046). Do reminder effects vary across different population segments? We investigated whether message effectiveness varied systematically across six key moderators: age, gender, 2 nd pension pillar membership, annual 2 nd pillar contributions (reflecting income levels), 3 rd pillar enrolment duration, and annual 3 rd pillar contributions before the treatment week (reflecting prior savings behaviour). See Supplemental Table 1 for complete variable definitions. Each moderator variable was added individually into the analysis models testing individual message effects yielding 18 analyses (6 moderators * 3 dependent variables). Given the multiple comparisons involved within and across all analyses, we corrected all interaction effect p-values from all models with the same dependent variable using the Holm method (we had preregistered combining the Holm adjustment with simultaneous confidence intervals that were used in the primary analyses but dropped the simultaneous confidence intervals to avoid correcting p-values twice.). No interaction effects reached statistical significance after p-value correction. Before correction, we found that men showed larger increases in average contributions in response to the family security message compared to women (b = 172.54, 95% CI [52.18, 292.90], p = .005). Exploratory treatment heterogeneity analyses To complement our pre-registered moderation analyses, we employed honest causal forests to explore whether combinations of moderating variables influenced message effectiveness. This not pre-registered machine learning approach can detect complex, nonlinear interaction effects among continuous variables without requiring factorization (Athey & Imbens, 2016). Among the 18 possible contrasts (comparing each of the nine messages against the control group for 2 outcomes), we found significant heterogeneity in two cases. First, the effect of the baseline message on contribution likelihood showed significant variation (differential effect estimate 1.29, p = 0.014), influenced primarily by three factors: annual 2 nd pillar contributions (33% of splits in the prediction forest used this variable), annual 3 rd pillar contributions (29.2%), and age (20.2%). According to the Friedman’s H statistic, 64% of the variance in the predicted treatment effect could be attributed to interactive rather than additive effects. The baseline message was most effective for younger individuals who had higher income levels (indicated by the 2 nd pillar contributions) and higher existing pension contributions in the 3 rd pillar. Given that this was highly sensitive to interactions, we plot the tree on Figure 2, illustrating these findings as a surrogate decision tree that reproduces 64% of the causal forest estimates. Second, the family security message’s effect on mean contributions showed significant variation (differential effect estimate 1.04, p = .003), driven predominantly by annual 2nd pillar contributions (49% of splits), with age (18.5%) and annual 3rd pillar contributions (16.6%) playing smaller roles. In this case, the variables acted mostly independently, with interactions explaining only 5.8% of predicted effect variance. The family security message was most effective for people with higher income (indicated by 2nd pillar contributions), higher 3rd pillar savings and older age. Discussion Insufficient retirement savings pose a growing global challenge as individuals increasingly bear responsibility for their financial security in retirement. While tax incentives can encourage saving, their effectiveness depends critically on individuals' awareness and motivation to act. Our nationwide megastudy demonstrates that low-cost email reminders, depending on their wording, can significantly boost voluntary pension contributions near a tax incentive deadline, with important implications for both theory and practice. The 10-15% increases we observe dwarf those found recently for non-tax-incentivized short-term liquid savings (see 0.5-1.3% in Milkman et al., 2025), suggesting tax benefits create a particularly responsive context for reminder interventions. The Estonian context provides compelling evidence for the need for such interventions. Despite tax benefits for voluntary pension contributions, the country's expected pension replacement at just 34.4% remains below the OECD average of 61.4%. This gap underscores that tax incentives alone, without behavioural interventions to promote their use, may be insufficient to overcome psychological barriers to retirement saving. The results are both statistically and economically significant. Email reminders increased contribution likelihood from 8.96% in the no-reminder condition to 9.90% among those who received a reminder, a 10.49% relative increase, generating an additional €1.2 million in voluntary pension contributions within just one week. These effects are particularly notable given several factors that likely made our estimates conservative: potential household communication spillover effects to the control group, competition from concurrent marketing campaigns and timing during peak holiday spending season (McNair et al., 2024). In addition, the contribution process required overcoming administrative burdens such as separate websites for checking income eligibility and current contribution levels. Lowering such “hassle costs” has been found to substantially increase pension savings (Daminato et al., 2024). Taken together, these considerations suggests that the potential of email reminders of tax benefits may be even larger than the effects observed here. Our findings contribute to the growing literature on behaviourally informed financial interventions. By employing a pre-registered megastudy approach (Milkman et al., 2021) on Estonia’s entire eligible population, we provide robust evidence about the relative effectiveness of different messaging strategies while contributing to insights about generalizations across populations and institutional contexts. Overall, the different message strategies we investigated were similarly effective. Only three strategies (psychological ownership, pennies-a-day, the power of now) failed to produce statistically significant increase in contribution likelihood relative to the no-message control. Nevertheless, their effects were of similar magnitude to significant effects. We only find one pairwise significant difference between messages. This suggests that, on average, the bulk of the benefits of reminders were drawn from the timely delivery of the information shared across all messages. Beyond the general effectiveness of all messages, we found that the family security message (“Help secure the future for yourself and your loved ones”) may have combined the benefits of two complementary psychological effects. Most other messages were effective in eliciting contributions from individuals who might not have otherwise contributed without significantly altering the amount that people decided to contribute. By contrast, the family security message increased not only the likelihood of making a contribution but also the average contribution size by 16.17% (from 868€ to 1,009€). The success of this message in Estonia—a highly individualistic culture—mirrors findings from culturally more collectivist Mexico (Shah et al., 2023, Beugelsdijk et al., 2017), suggesting that family-focused appeals tap into broadly shared motivations for retirement saving. Future research is needed to determined which components of this message are driving its effectiveness - the action verb “help,” the prosocial emphasis on “loved ones,” the possible “future” framing, or the legacy-orientation. The average effects of different messages can conceal important individual differences in message responsiveness. Given the large and representative sample of this study, we also thoroughly investigated potential effect heterogeneity. On the one hand, we found that most of the effects discussed above were robust as they did not vary significantly with any of the moderator variables available in our dataset. On the other hand, employing a more flexible honest causal forest approach revealed conjunctions of moderators that predicted increased sensitivity to some of the messages. Specifically, the baseline message and the family security reminders were most effective among individuals with higher incomes who had made 3 rd pillar savings also before the intervention with age slightly increasing the family security message and reducing the baseline message effects. This pattern suggests that reminders may be most effective for individuals who have the means to save and familiarity with saving while additional strategies may be needed to engage other groups. The small age effects suggest that individual characteristics can interact with message content. Taken together, our treatment heterogeneity analyses suggests that simple, timely reminders have broad effects that can be amplified when messages align with financial capacity other personal circumstances. While our study benefits from its scale and ecological validity, several limitations suggest directions for future research. Studies may want to examine the impact of timing (e.g., one-week vs. one-month before deadline) and frequency of reminders (e.g., Milkman et al., 2025) on habit formation, assess long-term effects on saving behaviour, and investigate strategies for reaching currently disengaged populations. Following Patterson and Skimmyhorn’s (2022) study among members of the U.S. Army, researchers should also explore how reminders interact with structural factors like auto-enrolment and tax incentives. Additionally, testing these interventions across different pension systems could help identify universal versus context-specific effective strategies. Overall, our findings demonstrate that low-cost email reminders increased pension contributions by 14.75%. The effectiveness of family security messaging across cultural contexts suggests fundamental psychological mechanisms that could inform future interventions. These results provide valuable guidance for policymakers seeking cost-effective ways to enhance the impact of tax incentives on retirement savings. Method Ethics and preregistration This study was approved by the ethics board of the University of Tartu (Approval number 383/T-17). All data were anonymised by the Pension Registry (Nasdaq) before analysis and complied with GDPR (the European Union's General Data Protection Regulation). We consulted with the Estonian Data Protection Board and the regional GDPR expert of Nasdaq (who manages the pension registry) prior to implementation. The study was preregistered at the Center for Open Science (https://doi.org/10.17605/OSF.IO/29CXS). Sample and randomization We included all working-age 3 rd pillar account holders born between 1970 and 2001 (N = 127,974) in our experiment. This excluded younger account holders who are likely not earning an income yet and also older groups who are able to already decumulate the 3 rd pillar savings with a lower tax rate. Account holders were randomly assigned to one of ten groups: a no-reminder control group (n = 23,238) or one of nine treatment groups (average n = 11,640 per group), each receiving a different reminder message. Randomization was conducted using R sample function. We generated 1,000 randomizations and conducted statistical tests to estimate differences between the randomized groups on age and current 2 nd / 3 rd pillar account balance (using analysis of variance F test) and gender (using Chi squared proportions test). We selected a randomization with the largest mean p-value across these tests to ensure the groups were as similar in their profiles as possible. The reminders were sent to account holders' private email addresses one week before the end-of-year tax deadline on December 21, 2023 by the Estonian pension registry. The emails were sent in batches at 30-minute intervals between midnight and 5:00 AM to ensure overnight delivery while avoiding server congestion. Message design We designed nine distinct messages: a baseline reminder and eight behaviourally informed variations. The messages were developed iteratively with behavioural scientists (the authors of this paper as well as colleagues from the Estonian Public Sector Innovation Team) and Estonian pension experts (from the Ministry of Finance, Ministry of Social Affairs, pension fund managers, pension fund communication experts). We pretested messages with 17 Estonian adults to ensure clarity and cultural appropriateness. All messages followed a common structure but varied in their motivational component (see Table 1). Statistical analysis We analyzed three outcome variables (1) the likelihood of making additional contributions, (2) the average contribution amount across all participants, and (3) the size of contributions among those who contributed (not preregistered). For contribution likelihood, we used logistic regressions with a binary outcome variable (1 = made at least one payment during the study period; 0 = did not contribute). For contribution amounts, we used Tweedie regressions for both total contributions (including zeros) and non-zero contributions separately. We estimated suitable Tweedie parameters using Generalized Additive Models. We conducted three sets of analyses. First, we compared the combined treatment groups against the control group using a binary independent variable (1 - reminder; 0 – no reminder). Second, to evaluate how effectiveness varied by messaging strategy, we compared each treatment group against both the control group and other treatment groups on the same three outcome measures. As these analyses involved more than one effect of interest, we controlled for multiple comparisons using simultaneous confidence intervals (the glht function from the multicomp package). Third, to investigate demographic and behavioural moderators, we first conducted six multiple logistic regressions, each adding a single moderator to the models described above for evaluating individual messages: gender, age group, income level (based on 2nd pillar contributions), 2nd pillar status, 3rd pillar enrolment duration, and marginal value of year-end payments (see Supplemental Table 1 for factor levels). Categorical moderators were entered with sum coding. We applied the Holm method to correct p-values of interaction effects from all models with the same outcome variable. We conducted additional non-preregistered analyses of treatment heterogeneity, using honest causal forests to predict average treatment effects conditional on the five moderator variables. We considered significant heterogeneity to exist when the 95% confidence interval of the difference between Conditional Average Treatment Effects for median-split treatment effects excluded zero. We assessed variable importance using Friedman's H statistic and interpreted findings through partial dependence plots and surrogate trees (Athey & Imbens, 2016). All analyses were conducted using R version 4.2.1. Declarations Data availability Our experiments and analysis involve confidential financial data from Estonia that cannot be released publicly. We can arrange for individuals to work with the raw data for replication purposes on a secure computer after arranging a nondisclosure agreement. The research team will facilitate this process. Code availability R code for the analyses is publicly available at the Center for Open Science: https://doi.org/10.17605/OSF.IO/29CXS References Athey, S., & Imbens, G. (2016). Recursive partitioning for heterogeneous causal effects. Proceedings of the National Academy of Sciences, 113(27), 7353–7360. https://doi.org/10.1073/pnas.1510489113 Athey, S., Tibshirani, J., & Wager, S. (2019). Generalized random forests. Annals of Statistics, 47 (2), 1148–1178. https://doi.org/10.1214/18-AOS1709 Berns, G. S., Laibson, D., & Loewenstein, G. (2007). Intertemporal choice—Toward an integrative framework. Trends in Cognitive Sciences, 11(11), 482–488. https://doi.org/10.1016/j.tics.2007.08.011 Beshears, J., Blakstad, M., Choi, J. J., Firth, C., Gathergood, J., Laibson, D., ... & Stewart, N. (2024). Does Pension Automatic Enrollment Increase Debt? Evidence from a Large-Scale Natural Experiment (No. w32100). National Bureau of Economic Research. Beugelsdijk, S., Kostova, T., & Roth, K. (2017). An overview of Hofstede-inspired country-level culture research in international business since 2006. Journal of International business studies , 48 (1), 30-47. https://doi.org/10.1057/s41267-016-0038-8 Bursztyn, L., Ederer, F., Ferman, B., & Yuchtman, N. (2014), Understanding Mechanisms Underlying Peer Effects: Evidence From a Field Experiment on Financial Decisions. Econometrica , 82: 1273–1301. https://doi.org/10.3982/ECTA11991 Buttenheim, A., Milkman, K. L., Duckworth, A. L., Gromet, D. M., Patel, M., & Chapman, G. (2022). Effects of ownership text message wording and reminders on receipt of an influenza vaccination: a randomized clinical trial. JAMA network open, 5(2). https://doi.org/10.1001/jamanetworkopen.2021.43388 Chan, S., & Stevens, A. (2008). What You Don't Know Can't Help You: Pension Knowledge and Retirement Decision-Making, The Review of Economics and Statistics , 90 (2), 253–266. https://doi.org/10.1162/rest.90.2.253 Coyne, A. J. (2024, April 16). Monzo chief marketer: Chemistry with agencies and why brands’ obsession with pitching free ideas is wrong. Campaign . Croy, G., Gerrans, P., & Speelman, C. (2010). The role and relevance of domain knowledge, perceptions of planning importance, and risk tolerance in predicting savings intentions. Journal of Economic Psychology , 31 (6), 860–871. https://doi.org/10.1016/j.joep.2010.06.002 Crum, J., Spencer, C., Doherty, E., Richardson, E., Sherman, S., Hays, A. W., ... & Hirshfield, L. (2024). Misinformation research needs ecological validity. Nature Human Behaviour , 8 (12), 2268-2271. https://doi.org/10.1038/s41562-024-02015-2 Dalen, H. P. van, & Henkens, K. (2023). Trust in pension funds, or the importance of being financially sound. Journal of Pension Economics & Finance , 22 (4), 658–669. https://doi.org/10.1017/S147474722200021X Daminato, C., Filippini, M., & Haufler, F. (2024). Digitalization and retirement contribution behavior: evidence from administrative data. The Review of Financial Studies , 37 (8), 2510-2549. https://doi.org/10.1093/rfs/hhae015 De La Rosa, W., Sharma, E., Tully, S. M., Giannella, E., & Rino, G. (2021). Psychological ownership interventions increase interest in claiming government benefits. Proceedings of the National Academy of Sciences , 118 (35), e2106357118. https://doi.org/10.1073/pnas.2106357118 Duckworth, A. L., Ko, A., Milkman, K. L., Kay, J. S., Dimant, E., Gromet, D. M., ... & Bulte, C. V. D. (2025). A national megastudy shows that email nudges to elementary school teachers boost student math achievement, particularly when personalized. Proceedings of the National Academy of Sciences , 122 (13), e2418616122. https://doi.org/10.1073/pnas.2418616122 Dur, R., Fleming, D., van Garderen, M., & van Lent, M. (2021). A social-norm nudge to save more: A field experiment at a retail bank. Journal of Public Economics, 200 , 104443. https://doi.org/10.1016/j.jpubeco.2021.104443 Eberhardt, W., Brüggen, E., Post, T., & Hoet, C. (2021). Engagement behavior and financial well-being: The effect of message framing in online pension communication. International Journal of Research in Marketing , 38 (2), 448–471. https://doi.org/10.1016/j.ijresmar.2020.11.002 El Mekkaoui, N., & Legendre, B. (2023). Does pension information impact savings? Journal of Pension Economics and Finance , 22(3), 352-370. https://doi.org/10.1017/S1474747221000470 Fishbane, A., Ouss, A., & Shah, A. K. (2020). Behavioral nudges reduce failure to appear for court. Science , 370 (6517), eabb6591. Foster, L. (2017). Young People and Attitudes towards Pension Planning. Social Policy and Society , 16 (1), 65–80. https://doi.org/10.1017/S1474746415000627 Fuentes, O. M., Mitchell, O. S., & Villatoro, F. (2023). Early pension withdrawals in Chile during the pandemic. Journal of Pension Economics & Finance , 1–26. https://doi.org/10.1017/S1474747223000112 Gallego-Losada, R., Montero-Navarro, A., Rodríguez-Sánchez, J.-L., & González-Torres, T. (2022). Retirement planning and financial literacy, at the crossroads. A bibliometric analysis. Finance Research Letters , 44 , 102109. https://doi.org/10.1016/j.frl.2021.102109 Gneezy, U., & List, J. A. (2006). Putting behavioral economics to work: Testing for gift exchange in labor markets using field experiments. Econometrica , 74 (5), 1365-1384. https://doi.org/10.1111/j.1468-0262.2006.00707.x Gollwitzer, P. M., & Sheeran, P. (2006). Implementation intentions and goal achievement: A meta-analysis of effects and processes. Advances in Experimental Social Psychology, 38 , 69–119. https://doi.org/10.1016/S0065-2601(06)38002-1 Gourville, J. T. (1998). Pennies-a-day: The effect of temporal reframing on transaction evaluation. Journal of Consumer Research, 24 (4), 395–408. https://doi.org/10.1086/209517 Gravert, C. in Behavioral Science in the Wild (eds Mažar, N. & Soman, D.) 120–132 (Univ. Toronto Press, 2022). Karlan, D., McConnell, M., Mullainathan, S., & Zinman, J. (2016). Getting to the top of mind: How reminders increase saving. Management Science, 62 (12), 3393–3411. https://doi.org/10.1287/mnsc.2015.2296 Kim, K. T., Lee, J. M., & Hong, E. (2016). The role of self-control on retirement preparedness of US households. International Journal of Human Ecology , 17 (2), 31-42. Koehler, D. J., Langstaff, J., & Liu, W.-Q. (2015). A simulated financial savings task for studying consumption and retirement decision making. Journal of Economic Psychology , 46 , 89–97. https://doi.org/10.1016/j.joep.2014.12.004 Laibson, D. I., Repetto, A., Tobacman, J., Hall, R. E., Gale, W. G., & Akerlof, G. A. (1998). Self-control and saving for retirement. Brookings Papers on Economic Activity , 1998 (1), 91-196. Linos, E., Lasky-Fink, J., Larkin, C., Moore, L., & Kirkman, E. (2024). The formality effect. Nature Human Behaviour, 8 (2), 300–310. https://doi.org/10.1038/s41562-023-01761-z List, J. A. (2007). Field experiments: a bridge between lab and naturally occurring data. The BE Journal of Economic Analysis & Policy , 6 (2). https://doi.org/10.2202/1538-0637.1747 Mckenzie, C. R. M., & Liersch, M. J. (2011). Misunderstanding Savings Growth: Implications for Retirement Savings Behavior. Journal of Marketing Research , 48 (SPL), S1–S13. https://doi.org/10.1509/jmkr.48.SPL.S1 McNair, S., Nyhus, E. K., & Ranyard R. (2024). Propensity to spend and borrow at a time of high pressure: the role of the meaning of Christmas and other psychological factors. Frontiers in Behavioral Economics , 3 https://doi.org/10.3389/frbhe.2024.1385609 Mertens, S., Herberz, M., Hahnel, U. J. J., & Brosch, T. (2022). The effectiveness of nudging: A meta-analysis of choice architecture interventions across behavioural domains. Proceedings of the National Academy of Sciences, 119 (1), e2107346118. https://doi.org/10.1073/pnas.2107346118 Milkman, K. L., Gromet, D., Ho, H., Kay, J. S., Lee, T. W., Pandiloski, P., Park, Y., Rai, A., Bazerman, M., Beshears, J., Bonacorsi, L., Camerer, C., Chang, E., Chapman, G., Cialdini, R., Dai, H., Eskreis-Winkler, L., Fishbach, A., Gross, J. J., … Duckworth, A. L. (2021). Megastudies improve the impact of applied behavioural science. Nature , 600(7889), 478–483. https://doi.org/10.1038/s41586-021-04128-4 Milkman, K. L., Ellis, S. F., Gromet, D. M., DeMay, I. M., Graci, H. N., Jung, Y., ... & Duckworth, A. L. (2025). Can reminder emails compel Americans to save? A two-million-person megastudy. PNAS nexus , 4 (9). https://doi.org/10.1093/pnasnexus/pgaf280 Mrkva, K., Posner, N. A., Reeck, C., & Johnson, E. J. (2021). Do Nudges Reduce Disparities? Choice Architecture Compensates for Low Consumer Knowledge. Journal of Marketing , 85(4), 67-84. https://doi.org/10.1177/0022242921993186 Munnell, A. H., Chen, A., & Siliciano, R. L. (2021). The National Retirement Risk Index: An Update from the 2019 SCF, Issues in Brief 21-02, Center for Retirement Research. OECD. (2023). Pensions at a Glance 2023: OECD and G20 indicators . OECD Publishing. https://doi.org/10.1787/4757bf20-en Patterson, R. W., & Skimmyhorn, W. L. (2022). How do Behavioral Approaches to Increase Savings Compare? Evidence from Multiple Interventions in the U.S. Army. NBER working paper #30697. https://doi.org/10.3386/w30697 Raghubir, P., & Srivastava, J. (2009). The denomination effect. Journal of Consumer Research, 36 (4), 701–713. https://doi.org/10.1086/599222 Riitsalu, L., Uusberg, A. (2021). To double, quadruple, or keep? Semi-automated service increases micro-investments. Journal of Behavioral and Experimental Economics , 95 , 101774. https://doi.org/10.1016/j.socec.2021.101774 Ruggeri, K., Panin, A., Vdovic, M., Većkalov, B., Abdul-Salaam, N., Achterberg, J., ... & Toscano, F. (2022). The globalizability of temporal discounting. Nature Human Behaviour, 6(10), 1386-1397. Rüütsalu, K., Reinson, H., Tillmann, T., Uusberg, A. Short- and Long-Term Effects of SMS Reminders on Men`s Participation in Colorectal Cancer Screening. [Manuscript in preparation] Saccardo, S., Dai, H., Han, M. A., Vangala, S., Hoo, J., & Fujimoto, J. (2024). Field testing the transferability of behavioural science knowledge on promoting vaccinations. Nature Human Behaviour, 8 (5), 878–890. https://doi.org/10.1038/s41562-023-01813-4 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): 204–28. https://doi.org/10.1257/pol.1.1.204 Shah, A. M., Osborne, M., Lefkowitz Kalter, J., Fertig, A., Fishbane, A., & Soman, D. (2023). Identifying heterogeneity using recursive partitioning: Evidence from SMS nudges encouraging voluntary retirement savings in Mexico. PNAS Nexus , 2 (5), pgad058. https://doi.org/10.1093/pnasnexus/pgad058 Sunstein, C. R. (2022). The distributional effects of nudges. Nature Human Behaviour, 6 (1), 9–10. https://doi.org/10.1038/s41562-021-01236-z Thaler, R. H. (2013, April 6). Shifting Our Retirement Savings Into Automatic. The New York Times. https://www.nytimes.com/2013/04/07/business/an-automatic-solution-for-the-retirement-savings-problem.html TIAA-CREF (2014). TIAA-CREF Survey finds Americans spend less time planning their IRA investment than choosing a restaurant. Retrieved from https://www.tiaa.org/public/about-tiaa/news-press/press-releases/pressrelease495.html Tomar, S., Kent Baker, H., Kumar, S., & Hoffmann, A. O. I. (2021). Psychological determinants of retirement financial planning behavior. Journal of Business Research , 133 , 432–449. https://doi.org/10.1016/j.jbusres.2021.05.007 Van Den Akker, M., & Sunstein, C. R. (2023). Do People Like Financial Nudges? SSRN Electronic Journal . https://doi.org/10.2139/ssrn.4636905 van Rooij, M. C. J., Lusardi, A., & Alessie, R. J. M. (2011). Financial literacy and retirement planning in the Netherlands. Journal of Economic Psychology , 32 (4), 593–608. https://doi.org/10.1016/j.joep.2011.02.004 Vervloet, M., Linn, A. J., van Weert, J. C. M., de Bakker, D. H., Bouvy, M., & van Dijk, L. (2012). The effectiveness of interventions using electronic reminders to improve adherence to chronic medication: A systematic review. Journal of the American Medical Informatics Association, 19 (5), 696–704. https://doi.org/10.1136/amiajnl-2011-000748 Wager, S., & Athey, S. (2018). Estimation and inference of heterogeneous treatment effects using random forests. Journal of the American Statistical Association, 113 (523), 1228–1242. https://doi.org/10.1080/01621459.2017.1319839 Zaval, L., Markowitz, E. M., & Weber, E. U. (2015). How will I be remembered? Conserving the environment for the sake of one’s legacy. Psychological Science, 26 (2), 231–236. https://doi.org/10.1177/0956797614561266 Footnotes [1] The 3rd pillar offers investment flexibility, allowing contributions to be allocated across a broad range of mutual funds with varying asset classes and management styles, from low-cost index funds to actively managed portfolios, enabling individuals to align their investments with their risk preferences and financial goals. [1] The 2nd pillar is an auto-enrolment retirement account that complements the 3rd pillar. While both offer tax advantages, 2nd pillar contributions are automatically set at 6% of gross income, making them a reliable proxy for income levels. Additional Declarations There is NO Competing Interest. Supplementary Files Supplementarymaterial.docx Cite Share Download PDF Status: Under Review Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {\"props\":{\"pageProps\":{\"initialData\":{\"identity\":\"rs-8008072\",\"acceptedTermsAndConditions\":true,\"allowDirectSubmit\":false,\"archivedVersions\":[],\"articleType\":\"Article\",\"associatedPublications\":[],\"authors\":[{\"id\":550066679,\"identity\":\"d4027902-b104-4d27-a5c3-94e2769635bc\",\"order_by\":0,\"name\":\"Heidi Reinson\",\"email\":\"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA1UlEQVRIiWNgGAWjYBACAwbGBokEIIMNzK0gXcsZBgkitDAgqWJsI0KLOXtz440HDPfk+KTbHz4unHe4joG9+QBeLZY9B5stEhiKjdlkzhgbz9x2WIKB51gCfofdSGwD+iUBSOawSfOCtEjkGODXcv8hWEt9m0T6M2neOUAt8u8/ELCFEawlgU0iwUyatwFkCw9eHQwGZxKBfjFIMAQ6zNiY51i6ZBtPGgGHHT/+8OaPigR5+RnpDx/z1Fjz87MffoDfGohGJDYbEepHwSgYBaNgFBAAAEBzPeYIlNa5AAAAAElFTkSuQmCC\",\"orcid\":\"https://orcid.org/0009-0004-6700-3844\",\"institution\":\"University of Tartu\",\"correspondingAuthor\":true,\"prefix\":\"\",\"firstName\":\"Heidi\",\"middleName\":\"\",\"lastName\":\"Reinson\",\"suffix\":\"\"},{\"id\":550066680,\"identity\":\"d5e21be7-9a47-4b48-9fe7-a788241d577f\",\"order_by\":1,\"name\":\"Thomas Post\",\"email\":\"\",\"orcid\":\"\",\"institution\":\"Maastricht University\",\"correspondingAuthor\":false,\"prefix\":\"\",\"firstName\":\"Thomas\",\"middleName\":\"\",\"lastName\":\"Post\",\"suffix\":\"\"},{\"id\":550066681,\"identity\":\"d9972bc0-bdea-4bdd-a0e2-4d46dd55e093\",\"order_by\":2,\"name\":\"Nina Mazar\",\"email\":\"\",\"orcid\":\"https://orcid.org/0000-0001-8248-654X\",\"institution\":\"Boston University\",\"correspondingAuthor\":false,\"prefix\":\"\",\"firstName\":\"Nina\",\"middleName\":\"\",\"lastName\":\"Mazar\",\"suffix\":\"\"},{\"id\":550066682,\"identity\":\"c1f74dc0-daf1-45f7-be0c-bc12bca3db4e\",\"order_by\":3,\"name\":\"Crystal Reeck\",\"email\":\"\",\"orcid\":\"https://orcid.org/0000-0002-1540-5321\",\"institution\":\"Temple\",\"correspondingAuthor\":false,\"prefix\":\"\",\"firstName\":\"Crystal\",\"middleName\":\"\",\"lastName\":\"Reeck\",\"suffix\":\"\"},{\"id\":550066683,\"identity\":\"5fb6e7c6-eccd-4aad-a9c0-3f2fe6c795fe\",\"order_by\":4,\"name\":\"Stylianos Syropoulos\",\"email\":\"\",\"orcid\":\"https://orcid.org/0000-0001-5622-1417\",\"institution\":\"Arizona State University\",\"correspondingAuthor\":false,\"prefix\":\"\",\"firstName\":\"Stylianos\",\"middleName\":\"\",\"lastName\":\"Syropoulos\",\"suffix\":\"\"},{\"id\":550066684,\"identity\":\"2a728f9d-1eb3-4d45-bc8d-39b340959269\",\"order_by\":5,\"name\":\"Andris Saulitis\",\"email\":\"\",\"orcid\":\"\",\"institution\":\"Collegio Carlo Alberto\",\"correspondingAuthor\":false,\"prefix\":\"\",\"firstName\":\"Andris\",\"middleName\":\"\",\"lastName\":\"Saulitis\",\"suffix\":\"\"},{\"id\":550066685,\"identity\":\"2cf84ea5-b089-473c-a03e-f46f2ffd0655\",\"order_by\":6,\"name\":\"Avni Shah\",\"email\":\"\",\"orcid\":\"\",\"institution\":\"University of Toronto\",\"correspondingAuthor\":false,\"prefix\":\"\",\"firstName\":\"Avni\",\"middleName\":\"\",\"lastName\":\"Shah\",\"suffix\":\"\"},{\"id\":550066686,\"identity\":\"16a5beb7-e057-4602-a0a7-74eea4cd2cea\",\"order_by\":7,\"name\":\"Andero Uusberg\",\"email\":\"\",\"orcid\":\"https://orcid.org/0000-0002-7327-9503\",\"institution\":\"University of Tartu\",\"correspondingAuthor\":false,\"prefix\":\"\",\"firstName\":\"Andero\",\"middleName\":\"\",\"lastName\":\"Uusberg\",\"suffix\":\"\"}],\"badges\":[],\"createdAt\":\"2025-11-01 22:46:12\",\"currentVersionCode\":1,\"declarations\":\"\",\"doi\":\"10.21203/rs.3.rs-8008072/v1\",\"doiUrl\":\"https://doi.org/10.21203/rs.3.rs-8008072/v1\",\"draftVersion\":[],\"editorialEvents\":[],\"editorialNote\":\"\",\"failedWorkflow\":false,\"files\":[{\"id\":99625185,\"identity\":\"d1122da3-5348-4275-8631-c4586c2990b0\",\"added_by\":\"auto\",\"created_at\":\"2026-01-06 14:54:37\",\"extension\":\"png\",\"order_by\":1,\"title\":\"Figure 1\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":167528,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003e\\u003cem\\u003eImpact of reminder messages on pension contributions.\\u003c/em\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cem\\u003eNotes\\u003c/em\\u003e. Results from regression analyses showing both aggregate reminder effects (top row) and individual reminder effects across three outcomes: likelihood of making any contribution (left), average contribution across all participants (middle), and contribution sum among contributors (right). Points show regression coefficients with 95% simultaneous confidence intervals. Vertical dashed lines represent no reminder control group means. Color coding indicates statistical significance of differences from control group means.\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"1.png\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-8008072/v1/f1b7ee739180d330df5bc92b.png\"},{\"id\":99625184,\"identity\":\"d891a8a9-77f2-47f3-935d-af3980935817\",\"added_by\":\"auto\",\"created_at\":\"2026-01-06 14:54:37\",\"extension\":\"png\",\"order_by\":2,\"title\":\"Figure 2\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":115681,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003e\\u003cem\\u003eSurrogate decision tree representing a causal forest predicting the Baseline message effect on contribution likelihood for participant subgroups.\\u003c/em\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003eNotes. Box and whisker plots show the distribution of predicted treatment effects within each subgroup. Numbers along the connecting lines are monetary values in euros used to form the given subgroups. n - group size; 3.pillar sum - prior 3\\u003csup\\u003erd\\u003c/sup\\u003e pillar contributions in 2023; 2.pillar sum - prior 2\\u003csup\\u003end\\u003c/sup\\u003e pillar contributions in 2023.\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"2.png\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-8008072/v1/0a7cd745552ee6b5673964ef.png\"},{\"id\":99625194,\"identity\":\"5ebde9e8-8438-462a-829a-19e3b3f53b5c\",\"added_by\":\"auto\",\"created_at\":\"2026-01-06 14:54:44\",\"extension\":\"pdf\",\"order_by\":0,\"title\":\"\",\"display\":\"\",\"copyAsset\":false,\"role\":\"manuscript-pdf\",\"size\":1149557,\"visible\":true,\"origin\":\"\",\"legend\":\"\",\"description\":\"\",\"filename\":\"manuscript.pdf\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-8008072/v1/e586a4b1-6ea2-4711-ad4d-79b68aee4792.pdf\"},{\"id\":99625186,\"identity\":\"15819e05-01d0-465a-91a5-64b5698b1742\",\"added_by\":\"auto\",\"created_at\":\"2026-01-06 14:54:37\",\"extension\":\"docx\",\"order_by\":1,\"title\":\"\",\"display\":\"\",\"copyAsset\":false,\"role\":\"supplement\",\"size\":1868836,\"visible\":true,\"origin\":\"\",\"legend\":\"\",\"description\":\"\",\"filename\":\"Supplementarymaterial.docx\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-8008072/v1/90005d870077899cd10e5021.docx\"}],\"financialInterests\":\"There is \\u003cb\\u003eNO\\u003c/b\\u003e Competing Interest.\",\"formattedTitle\":\"Behavioural messages amplify tax incentives: \\nA nationwide megastudy of retirement savings reminders\",\"fulltext\":[{\"header\":\"Main text\",\"content\":\"\\u003cp\\u003eInadequate retirement savings pose a growing challenge worldwide. For example, half of American households risk being unable to maintain their living standards in retirement (Munnell, Chen, \\u0026amp; Siliciano, 2021). As demographic shifts and fiscal pressures limit governments\\u0026apos; ability to provide supplementary support (Gomes et al., 2021), saving for retirement is increasingly falling on the shoulders of individuals. Understanding how to help individuals overcome psychological barriers to increasing their retirement savings has therefore become a key priority for policymakers.\\u003c/p\\u003e\\n\\u003cp\\u003eSaving sufficiently for retirement does not come naturally to most people. \\u0026ldquo;For many people, being asked to solve their own retirement savings problems is like being asked to build their own cars,\\u0026rdquo; according to Nobel laureate Richard Thaler (New York Times, 2013). Multiple psychological barriers impede effective retirement planning. Individuals heavily discount distant future outcomes (Berns, Laibson, \\u0026amp; Loewenstein, 2007; Ruggeri et al, 2022), struggle with the self-control required to save consistently (Laibson, 1998; Kim, Lee, \\u0026amp; Hong, 2016), and tend to view their future selves as strangers, making it difficult to prioritize their future financial needs over present desires (Hershfield, 2011). These barriers can shape behaviour in striking ways, with individuals typically spending more time choosing a television set than planning for retirement (TIAA-CREF, 2014) and preferring household chores like scrubbing floors to reviewing their savings accounts (Monzo Bank, 2024).\\u0026nbsp;\\u003c/p\\u003e\\n\\u003cp\\u003eVarious policy approaches can be used to encourage retirement savings. Mandatory and auto-enrolment pension schemes can effectively steer individual behaviour, but these more paternalistic approaches can face resistance. Furthermore, the long-term benefits of auto-enrolment can be undermined by early withdrawal options and increased household debt burdens (Beshears et al, 2024; Choi et al, 2024). To avoid these downsides, policymakers may want to consider less restrictive measures such as tax benefits that encourage proactive savings while preserving individual choice. However, even well-designed tax benefits can fail if individuals either do not know about them (Chan and Stevens, 2008) or lack the attention and motivation to act on them (Karlan et al, 2016).\\u003c/p\\u003e\\n\\u003cp\\u003eReminder messages offer a promising solution to these attention and motivation challenges, as they can effectively direct focus to important goals and opportunities that might otherwise be overlooked (Gravert, 2022). Reminders have proven effective across various policy domains from safety, health, and education (Austin et al, 2006; Vervloet et al, 2012; Duckworth et al, 2025) to financial decision-making (Saez, 2009; Karlan et al, 2016). Recent evidence from the United States suggests email reminders can modestly increase short-term liquid savings, with effects ranging from 0.5-1.3% (Milkman et al., 2025). However, the impact of behavioral messaging on tax-incentivized retirement savings\\u0026mdash;where the stakes, time horizon and incentives differ substantially\\u0026mdash;remains unexplored. Building on this evidence, our first aim was to test whether email reminders could enhance the impact of tax benefits on retirement savings.\\u0026nbsp;\\u003c/p\\u003e\\n\\u003cp\\u003eWhile reminders can be effective on their own, behavioural science suggests their impact can be amplified through careful message design (Gravert, 2022). However, the context-specific nature of behaviour change makes it difficult to predict which messaging strategies will be most effective in a particular domain. Our second aim was therefore to systematically identify the most effective messaging approaches for enhancing the impact of tax incentives on retirement savings. Using a megastudy\\u0026mdash;a single experiment comparing multiple interventions simultaneously against a common control (Milkman et al., 2021; Duckworth \\u0026amp; Milkman, 2022; Voelkel et al., 2024), we tested eight behaviourally informed reminder messages alongside a neutral reminder message by comparing their effectiveness relative to a no-reminder control.\\u0026nbsp;\\u003c/p\\u003e\\n\\u003cp\\u003eThe impact of behavioural interventions often varies across different population segments, making it crucial to understand who responds best to which approaches (Sunstein, 2022). For instance, a recent study of retirement savings reminders in Mexico found that while the interventions were effective on average, they backfired for certain subgroups (Shah et al., 2023). Our third aim was therefore to examine how message impact varied across key characteristics including age, gender, and past pension saving behaviour. We focused on variables that pension administrators can readily observe and therefore potentially use for personalized reminders in the future.\\u003c/p\\u003e\\n\\u003cp\\u003eThese three aims\\u0026mdash;the general effectiveness of reminders, the relative impact of different messaging strategies, and the potential for heterogeneous effects\\u0026mdash;were investigated in a nationwide pre-registered megastudy targeting the entire population of eligible pension account holders in Estonia. We assessed how different email messages reminding recipients of a tax benefit deadline impacted both the likelihood and size of pension contributions. In contrast to laboratory studies (often with student populations) that comprise a large share of the behavioural intervention literature, our field experiment provides evidence from a representative decision environment, crucial for establishing the generalizability of results and their relevance for policy (List, 2007; Saccardo et al., 2024; Crum et al., 2024).\\u003c/p\\u003e\\n\\u003cp\\u003eEstonia is a uniquely suitable setting for this research. The country offers substantial tax benefits through its \\u0026lsquo;3rd pillar\\u0026rsquo; voluntary pension scheme, similar to U.S. 401(k) plans, allowing individuals to claim a 20% income tax return on contributions up to 15% of their gross income (maximum 6,000\\u0026euro; per year).\\u003csup\\u003e\\u003csup\\u003e[1]\\u003c/sup\\u003e\\u003c/sup\\u003e Despite these incentives, the expected proportion of pension benefits to pre-retirement income (i.e., pension replacement rate) remains at just 34%, well below the OECD average of 61% (OECD, 2023). This gap creates both a clear need for interventions and an ethically sound basis for encouraging increased savings.\\u0026nbsp;\\u003c/p\\u003e\\n\\u003cp\\u003eAll Estonian working-age residents with a voluntary 3rd pillar retirement account (n = 127,974), representing roughly one fifth of the working-age population were randomly assigned either to a no-reminder control group (n = 23,238) or to one of nine treatment groups (averaging 11,640 participants per group), each receiving either a neutral or a behaviourally informed reminder message. The sample yielded 80% power to detect a 0.6 percentage point increase when comparing all reminders combined against control, and a 0.9 percentage point increase for individual message comparisons (\\u0026alpha;=0.05, baseline rate=8.96%). The randomized groups were comparable across all available moderator variables (see Appendix D). \\u0026nbsp;The reminders were sent once one week before the end-of-year tax deadline in December 2023 by the Estonian pension registry to account holders\\u0026rsquo; private email addresses. We analysed the contributions account holders made to their 3\\u003csup\\u003erd\\u003c/sup\\u003e pillar accounts during that final week before tax deadline. The reminder messages, seen in Table 1, were developed through three stages: (1) systematic review of behavioural interventions in retirement savings, (2) co-creation with an international consortium of behavioural scientists and Estonian pension experts, ensuring both theoretical rigor and contextual relevance, (3) pre-testing with Estonian adults for comprehension and cultural appropriateness. Each message highlighted different behaviourally informed motivational cues, including salient tax returns, future income, social norms, loss aversion, family security, and urgency.\\u003c/p\\u003e\\n\\u003cp\\u003eWe followed a pre-registered analysis strategy with clearly marked additions throughout the paper to address our three key research questions.\\u003c/p\\u003e\\n\\u003col start=\\\"1\\\" type=\\\"1\\\"\\u003e\\n \\u003cli\\u003e\\u003cstrong\\u003eCan timely reminders increase retirement savings?\\u0026nbsp;\\u003c/strong\\u003eWe examined this research question by comparing the combined treatment groups against the control group on three metrics: (a) the likelihood of making additional contributions, (b) the average contribution amount in Euros across all participants (including non-contributors), and (c) the size of contributions among those who contributed (not preregistered).\\u0026nbsp;\\u003c/li\\u003e\\n \\u003cli\\u003e\\u003cstrong\\u003eHow does the effectiveness of reminders vary by messaging strategy?\\u0026nbsp;\\u003c/strong\\u003eWe compared each treatment group against both the control group and other treatment groups, examining the same three outcome measures to identify the most effective messaging approaches.\\u003c/li\\u003e\\n \\u003cli\\u003e\\u003cstrong\\u003eDo reminder effects vary across different population segments?\\u003c/strong\\u003e We investigated whether message effectiveness depended on the key demographic and behavioural characteristics, including age, gender, and past saving behaviour in both the mandatory (2\\u003csup\\u003end\\u003c/sup\\u003e pillar\\u003csup\\u003e\\u003csup\\u003e[2]\\u003c/sup\\u003e\\u003c/sup\\u003e, a proxy for income) and voluntary (3\\u003csup\\u003erd\\u003c/sup\\u003e pillar) pension schemes. Our pre-registered analysis examined these variables individually as moderators, with p-values adjusted for multiple comparisons. In a non-preregistered additional analysis, we used causal honest forest machine learning (Athey \\u0026amp; Imbens, 2016) to examine how combinations of moderator variables might jointly influence treatment effectiveness.\\u003c/li\\u003e\\n\\u003c/ol\\u003e\\n\\u003cp\\u003e\\u003cstrong\\u003eTable 1: Reminder conditions\\u003c/strong\\u003e\\u003c/p\\u003e\\n\\u003ctable border=\\\"1\\\" cellspacing=\\\"0\\\" cellpadding=\\\"0\\\" width=\\\"663\\\"\\u003e\\n \\u003ctbody\\u003e\\n \\u003ctr\\u003e\\n \\u003ctd valign=\\\"top\\\" style=\\\"width: 4.37406%;\\\"\\u003e\\n \\u003cp\\u003e\\u003cstrong\\u003e\\u0026nbsp;\\u003c/strong\\u003e\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd valign=\\\"top\\\" style=\\\"width: 12.9713%;\\\"\\u003e\\n \\u003cp\\u003e\\u003cstrong\\u003eCondition\\u003c/strong\\u003e\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd valign=\\\"top\\\" style=\\\"width: 41.3273%;\\\"\\u003e\\n \\u003cp\\u003e\\u003cstrong\\u003eText\\u003c/strong\\u003e\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd valign=\\\"top\\\" style=\\\"width: 41.3273%;\\\"\\u003e\\n \\u003cp\\u003e\\u003cstrong\\u003eRationale and references\\u003c/strong\\u003e\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003c/tr\\u003e\\n \\u003ctr\\u003e\\n \\u003ctd valign=\\\"top\\\" style=\\\"width: 4.37406%;\\\"\\u003e\\n \\u003cp\\u003e\\u0026nbsp;\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd valign=\\\"top\\\" style=\\\"width: 12.9713%;\\\"\\u003e\\n \\u003cp\\u003eNo-reminder control\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd valign=\\\"top\\\" style=\\\"width: 41.3273%;\\\"\\u003e\\n \\u003cp\\u003en/a\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd valign=\\\"top\\\" style=\\\"width: 41.3273%;\\\"\\u003e\\n \\u003cp\\u003eStatus quo\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003c/tr\\u003e\\n \\u003ctr\\u003e\\n \\u003ctd valign=\\\"top\\\" style=\\\"width: 4.37406%;\\\"\\u003e\\n \\u003cp\\u003e1\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd valign=\\\"top\\\" style=\\\"width: 12.9713%;\\\"\\u003e\\n \\u003cp\\u003eBaseline reminder*\\u003c/p\\u003e\\n \\u003cp\\u003e\\u0026nbsp;\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd valign=\\\"top\\\" style=\\\"width: 41.3273%;\\\"\\u003e\\n \\u003cp\\u003eDear III pillar account holder,\\u003c/p\\u003e\\n \\u003cp\\u003e\\u003cstrong\\u003e\\u0026lt;[treatment text here for conditions 2-10]\\u0026gt;\\u003c/strong\\u003e\\u003c/p\\u003e\\n \\u003cp\\u003eContributions made to the III pension pillar until December 27 will be included in the income tax refund for 2023. You can get back 20% of your III pillar payment.\\u003c/p\\u003e\\n \\u003cp\\u003eCheck if you are taking advantage of this opportunity: Questions and answers about the III pillar tax refund\\u003c/p\\u003e\\n \\u003cp\\u003eThe income tax refund applies to III pillar contributions up to 15% of your gross income, but not more than 6,000 euros per year.\\u0026nbsp;\\u003c/p\\u003e\\n \\u003cp\\u003eYours sincerely,\\u003cbr\\u003e\\u0026nbsp;Pension Center\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd valign=\\\"top\\\" style=\\\"width: 41.3273%;\\\"\\u003e\\n \\u003cp\\u003eProvides a simple reminder at the tax-relevant moment to address inattention. Evidence suggests timely prompts can increase financial actions (Karlan et al., 2016). Uses formal language shown to be effective in public sector communications (Linos et al., 2024).\\u003c/p\\u003e\\n \\u003cp\\u003e\\u0026nbsp;\\u003c/p\\u003e\\n \\u003cp\\u003eThe main text was included in all the e-mail messages, as we were required to provide this information to all groups. Only the first sentence varied between conditions.\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003c/tr\\u003e\\n \\u003ctr\\u003e\\n \\u003ctd valign=\\\"top\\\" style=\\\"width: 4.37406%;\\\"\\u003e\\n \\u003cp\\u003e2\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd valign=\\\"top\\\" style=\\\"width: 12.9713%;\\\"\\u003e\\n \\u003cp\\u003eLoss aversion\\u0026nbsp;\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd valign=\\\"top\\\" style=\\\"width: 41.3273%;\\\"\\u003e\\n \\u003cp\\u003e\\u003cstrong\\u003eOnly a few days left, don\\u0026rsquo;t miss out on your tax refund!\\u003c/strong\\u003e\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd valign=\\\"top\\\" style=\\\"width: 41.3273%;\\\"\\u003e\\n \\u003cp\\u003eFrames the tax benefit as something to avoid losing rather than gaining. Loss-framed messages have shown effectiveness in some retirement savings contexts (Eberhardt et al., 2021).\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003c/tr\\u003e\\n \\u003ctr\\u003e\\n \\u003ctd valign=\\\"top\\\" style=\\\"width: 4.37406%;\\\"\\u003e\\n \\u003cp\\u003e3\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd valign=\\\"top\\\" style=\\\"width: 12.9713%;\\\"\\u003e\\n \\u003cp\\u003ePsychological ownership\\u0026nbsp;\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd valign=\\\"top\\\" style=\\\"width: 41.3273%;\\\"\\u003e\\n \\u003cp\\u003e\\u003cstrong\\u003eOnly a few days left, your tax refund is waiting for you!\\u003c/strong\\u003e\\u003c/p\\u003e\\n \\u003cp\\u003e\\u003cstrong\\u003e\\u0026nbsp;\\u003c/strong\\u003e\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd valign=\\\"top\\\" style=\\\"width: 41.3273%;\\\"\\u003e\\n \\u003cp\\u003ePresents the tax refund as already belonging to the recipient. Ownership framing has increased engagement in government benefits and health behaviours (De La Rosa et al., 2021; Buttenheim et al., 2022).\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003c/tr\\u003e\\n \\u003ctr\\u003e\\n \\u003ctd valign=\\\"top\\\" style=\\\"width: 4.37406%;\\\"\\u003e\\n \\u003cp\\u003e4\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd valign=\\\"top\\\" style=\\\"width: 12.9713%;\\\"\\u003e\\n \\u003cp\\u003eShort- \\u0026amp; long-term gain\\u0026nbsp;\\u003c/p\\u003e\\n \\u003cp\\u003e\\u0026nbsp;\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd valign=\\\"top\\\" style=\\\"width: 41.3273%;\\\"\\u003e\\n \\u003cp\\u003e\\u003cstrong\\u003eIncrease your future income as well as your next tax return!\\u003c/strong\\u003e\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd valign=\\\"top\\\" style=\\\"width: 41.3273%;\\\"\\u003e\\n \\u003cp\\u003eHighlights both temporally close and distant gains, addressing different time preferences. Gain framing has been effectively used in pension communication (Saez, 2009; Eberhardt et al., 2021).\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003c/tr\\u003e\\n \\u003ctr\\u003e\\n \\u003ctd valign=\\\"top\\\" style=\\\"width: 4.37406%;\\\"\\u003e\\n \\u003cp\\u003e5\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd valign=\\\"top\\\" style=\\\"width: 12.9713%;\\\"\\u003e\\n \\u003cp\\u003eInvestment boost\\u0026nbsp;\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd valign=\\\"top\\\" style=\\\"width: 41.3273%;\\\"\\u003e\\n \\u003cp\\u003e\\u003cstrong\\u003eWith the tax refund, you can further increase the return on your investment!\\u003c/strong\\u003e\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd valign=\\\"top\\\" style=\\\"width: 41.3273%;\\\"\\u003e\\n \\u003cp\\u003eAn alternative gain frame reframes pension contributions as smart investments rather than savings. Developed with Estonian financial experts to match local customer insights.\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003c/tr\\u003e\\n \\u003ctr\\u003e\\n \\u003ctd valign=\\\"top\\\" style=\\\"width: 4.37406%;\\\"\\u003e\\n \\u003cp\\u003e6\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd valign=\\\"top\\\" style=\\\"width: 12.9713%;\\\"\\u003e\\n \\u003cp\\u003ePennies-a-Day\\u0026nbsp;\\u003c/p\\u003e\\n \\u003cp\\u003e\\u0026nbsp;\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd valign=\\\"top\\\" style=\\\"width: 41.3273%;\\\"\\u003e\\n \\u003cp\\u003e\\u003cstrong\\u003eEven small amounts help secure your future!\\u003c/strong\\u003e\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd valign=\\\"top\\\" style=\\\"width: 41.3273%;\\\"\\u003e\\n \\u003cp\\u003eReduces perceived barriers by emphasizing small contributions. Reframing large amounts as smaller units has shown effects in charitable giving and retirement contexts (Gourville, 1999; Shah et al., 2023).\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003c/tr\\u003e\\n \\u003ctr\\u003e\\n \\u003ctd valign=\\\"top\\\" style=\\\"width: 4.37406%;\\\"\\u003e\\n \\u003cp\\u003e7\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd valign=\\\"top\\\" style=\\\"width: 12.9713%;\\\"\\u003e\\n \\u003cp\\u003eFamily security\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd valign=\\\"top\\\" style=\\\"width: 41.3273%;\\\"\\u003e\\n \\u003cp\\u003e\\u003cstrong\\u003eHelp secure the future for yourself and your loved ones!\\u003c/strong\\u003e\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd valign=\\\"top\\\" style=\\\"width: 41.3273%;\\\"\\u003e\\n \\u003cp\\u003eExpands motivation beyond self-interest to include loved ones. Legacy-oriented appeals have increased prosocial financial behaviours in multiple contexts (Zaval et al., 2015; Shrum, 2021; Syropoulos et al., 2023) and increased pension savings (Shah et al, 2023).\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003c/tr\\u003e\\n \\u003ctr\\u003e\\n \\u003ctd valign=\\\"top\\\" style=\\\"width: 4.37406%;\\\"\\u003e\\n \\u003cp\\u003e8\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd valign=\\\"top\\\" style=\\\"width: 12.9713%;\\\"\\u003e\\n \\u003cp\\u003eSocial norms\\u0026nbsp;\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd valign=\\\"top\\\" style=\\\"width: 41.3273%;\\\"\\u003e\\n \\u003cp\\u003e\\u003cstrong\\u003eIn recent years, the number of people saving in the III pillar has doubled to nearly 200,000!\\u003c/strong\\u003e\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd valign=\\\"top\\\" style=\\\"width: 41.3273%;\\\"\\u003e\\n \\u003cp\\u003eProvides information about peer saving behaviour. Descriptive norms can influence financial decisions, though effects vary by context (Bursztyn et al., 2014; Dur et al., 2021).\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003c/tr\\u003e\\n \\u003ctr\\u003e\\n \\u003ctd valign=\\\"top\\\" style=\\\"width: 4.37406%;\\\"\\u003e\\n \\u003cp\\u003e9\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd valign=\\\"top\\\" style=\\\"width: 12.9713%;\\\"\\u003e\\n \\u003cp\\u003eThe power of now\\u0026nbsp;\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd valign=\\\"top\\\" style=\\\"width: 41.3273%;\\\"\\u003e\\n \\u003cp\\u003e\\u003cstrong\\u003eWhat is done today, you don\\u0026rsquo;t have to worry about tomorrow!\\u003c/strong\\u003e\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd valign=\\\"top\\\" style=\\\"width: 41.3273%;\\\"\\u003e\\n \\u003cp\\u003eCreates urgency through immediate action framing. Uses familiar Estonian proverb previously effective in health behaviour reminders (R\\u0026uuml;\\u0026uuml;tsalu et al., 2023).\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003c/tr\\u003e\\n \\u003c/tbody\\u003e\\n\\u003c/table\\u003e\\n\\u003cp\\u003e*Full version in Estonian, Russian, and English languages in Appendix B\\u003c/p\\u003e\\n\\u003ch3\\u003eResults\\u003c/h3\\u003e\\n\\u003cp\\u003eWe present findings according to our three research questions. Figure 1 summarizes the effects of each intervention on all outcomes.\\u0026nbsp;\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cstrong\\u003eCan timely reminders increase retirement savings?\\u003c/strong\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003eOur analysis revealed that reminders significantly increased pension contributions. Logistic regressions showed that the likelihood of making a payment to one\\u0026rsquo;s 3\\u003csup\\u003erd\\u003c/sup\\u003e pillar account during the final week before the tax deadline rose from 8.96% in the no-reminder control group to 9.90% in the reminder groups, a 10.4% significant relative increase (OR = 1.12, 95% CI [1.08, 1.16], p \\u0026lt; .001; Fig 1 top left panel).\\u003c/p\\u003e\\n\\u003cp\\u003eA Tweedie GLM regression demonstrated that reminders significantly increased the average contribution by 14.75% from 77.76\\u0026euro; to 89.23\\u0026euro; (exp(b) = 1.15, 95% CI [1.05, 1.25], p = .001, Fig 1 top middle). Importantly, this average includes the majority of participants who contributed 0\\u0026euro;. A subsequent non-preregistered Tweedie GLM regression revealed that among those who did contribute, the reminder messages did not significantly affect contribution amount (Fig 1 top right). This pattern suggests that reminders primarily worked by motivating more people to contribute rather than by increasing the size of individual contributions.\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cstrong\\u003eHow does the effectiveness of reminders vary by messaging strategy?\\u0026nbsp;\\u003c/strong\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003eWe compared each message to the no-reminder control by using logistic regression for contribution likelihood and Tweedie regression for contribution amounts. We computed simultaneous confidence intervals to account for multiple comparisons. Most messages significantly increased the likelihood of contributing (Fig 1 bottom left). However, only the family security message (\\u0026ldquo;Help secure the future for yourself and your loved ones!\\u0026rdquo;)\\u003cstrong\\u003e\\u0026nbsp;\\u003c/strong\\u003esignificantly\\u003cstrong\\u003e\\u0026nbsp;\\u003c/strong\\u003eincreased the average contribution, showing a 28.4% increase from 77.76\\u0026euro; to 89.23\\u0026euro; (exp(b) = 1.28 95% CI [1.07, 1.54], p = .001; Fig 1 bottom middle). Additional non-preregistered analyses of average contribution size among contributors showed no significant effects, though the family security message approached significance (exp(b) = 1.16 95% CI [0.99, 1.35], p = .060; Fig 1 bottom right).\\u003c/p\\u003e\\n\\u003cp\\u003eIn post-hoc pairwise comparisons between all reminders on all three dependent variables, we found remarkably consistent effects across different messages. The only significant difference emerged in contribution size, where the family security message (1,008.80\\u0026euro;) outperformed the Loss Aversion message (821.80\\u0026euro;) by 187\\u0026euro; (p = .046).\\u0026nbsp;\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cstrong\\u003eDo reminder effects vary across different population segments?\\u003c/strong\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003eWe investigated whether message effectiveness varied systematically across six key moderators: age, gender, 2\\u003csup\\u003end\\u003c/sup\\u003e pension pillar membership, annual 2\\u003csup\\u003end\\u003c/sup\\u003e pillar contributions (reflecting income levels), 3\\u003csup\\u003erd\\u003c/sup\\u003e pillar enrolment duration, and annual 3\\u003csup\\u003erd\\u003c/sup\\u003e pillar contributions before the treatment week (reflecting prior savings behaviour). See Supplemental Table 1 for complete variable definitions. Each moderator variable was added individually into the analysis models testing individual message effects yielding 18 analyses (6 moderators * 3 dependent variables). Given the multiple comparisons involved within and across all analyses, we corrected all interaction effect p-values from all models with the same dependent variable using the Holm method (we had preregistered combining the Holm adjustment with simultaneous confidence intervals that were used in the primary analyses but dropped the simultaneous confidence intervals to avoid correcting p-values twice.). No interaction effects reached statistical significance after p-value correction. Before correction, we found that men showed larger increases in average contributions in response to the family security message compared to women (b = 172.54, 95% CI [52.18, 292.90], p = .005).\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cstrong\\u003eExploratory treatment heterogeneity analyses\\u003c/strong\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003eTo complement our pre-registered moderation analyses, we employed honest causal forests to explore whether combinations of moderating variables influenced message effectiveness. \\u0026nbsp;This not pre-registered machine learning approach can detect complex, nonlinear interaction effects among continuous variables without requiring factorization (Athey \\u0026amp; Imbens, 2016).\\u003c/p\\u003e\\n\\u003cp\\u003eAmong the 18 possible contrasts (comparing each of the nine messages against the control group for 2 outcomes), we found significant heterogeneity in two cases. First, the effect of the baseline message on contribution likelihood showed significant variation (differential effect estimate 1.29, p\\u003cem\\u003e\\u0026nbsp;\\u003c/em\\u003e= 0.014), influenced primarily by three factors: annual 2\\u003csup\\u003end\\u003c/sup\\u003e pillar contributions (33% of splits in the prediction forest used this variable), annual 3\\u003csup\\u003erd\\u003c/sup\\u003e pillar contributions (29.2%), and age (20.2%). According to the Friedman\\u0026rsquo;s H statistic, 64% of the variance in the predicted treatment effect could be attributed to interactive rather than additive effects. The baseline message was most effective for younger individuals who had higher income levels (indicated by the 2\\u003csup\\u003end\\u003c/sup\\u003e pillar contributions) and higher existing pension contributions in the 3\\u003csup\\u003erd\\u003c/sup\\u003e pillar. Given that this was highly sensitive to interactions, we plot the tree on Figure 2, illustrating these findings as a surrogate decision tree that reproduces 64% of the causal forest estimates.\\u003c/p\\u003e\\n\\u003cp\\u003eSecond, the family security message\\u0026rsquo;s effect on mean contributions showed significant variation (differential effect estimate 1.04, p = .003), driven predominantly by annual 2nd pillar contributions (49% of splits), with age (18.5%) and annual 3rd pillar contributions (16.6%) playing smaller roles. In this case, the variables acted mostly independently, with interactions explaining only 5.8% of predicted effect variance. The family security message was most effective for people with higher income (indicated by 2nd pillar contributions), higher 3rd pillar savings and older age.\\u003c/p\\u003e\"},{\"header\":\"Discussion \",\"content\":\"\\u003cp\\u003eInsufficient retirement savings pose a growing global challenge as individuals increasingly bear responsibility for their financial security in retirement. While tax incentives can encourage saving, their effectiveness depends critically on individuals' awareness and motivation to act. Our nationwide megastudy demonstrates that low-cost email reminders, depending on their wording, can significantly boost voluntary pension contributions near a tax incentive deadline, with important implications for both theory and practice.\\u0026nbsp;The 10-15% increases we observe dwarf those found recently for non-tax-incentivized\\u0026nbsp;short-term\\u0026nbsp;liquid savings (see 0.5-1.3% in Milkman et al., 2025), suggesting tax benefits create a particularly responsive context for reminder interventions.\\u003c/p\\u003e\\n\\u003cp\\u003eThe Estonian context provides compelling evidence for the need for such interventions. Despite tax benefits for voluntary pension contributions, the country's expected pension replacement at just 34.4% remains below the OECD average of 61.4%. This gap underscores that tax incentives alone, without behavioural interventions to promote their use, may be insufficient to overcome psychological barriers to retirement saving.\\u0026nbsp;\\u003c/p\\u003e\\n\\u003cp\\u003eThe results are both statistically and economically significant. Email reminders increased contribution likelihood from 8.96% in the no-reminder condition to 9.90% among those who received a reminder, a 10.49% relative increase, generating an additional €1.2 million in voluntary pension contributions within just one week. These effects are particularly notable given several factors that likely made our estimates conservative: potential household communication spillover effects to the control group, competition from concurrent marketing campaigns and timing during peak holiday spending season (McNair et al., 2024). In addition, the contribution process required overcoming administrative burdens such as separate websites for checking income eligibility and current contribution levels. Lowering such “hassle costs” has been found to substantially increase pension savings (Daminato et al., 2024). Taken together, these considerations suggests that the potential of email reminders of tax benefits may be even larger than the effects observed here.\\u003c/p\\u003e\\n\\u003cp\\u003eOur findings contribute to the growing literature on behaviourally informed financial interventions. By employing a pre-registered megastudy approach (Milkman et al., 2021) on Estonia’s entire eligible population, we provide robust evidence about the relative effectiveness of different messaging strategies while contributing to insights about generalizations across populations and institutional contexts. Overall, the different message strategies we investigated were similarly effective. Only three strategies (psychological ownership, pennies-a-day, the power of now) failed to produce statistically significant increase in contribution likelihood relative to the no-message control. Nevertheless, their effects were of similar magnitude to significant effects. We only find one pairwise significant difference between messages. This suggests that, on average, the bulk of the benefits of reminders were drawn from the timely delivery of the information shared across all messages.\\u0026nbsp;\\u003c/p\\u003e\\n\\u003cp\\u003eBeyond the general effectiveness of all messages, we found that the family security message (“Help secure the future for yourself and your loved ones”) may have combined the benefits of two complementary psychological effects. Most other messages were effective in eliciting contributions from individuals who might not have otherwise contributed without significantly altering the amount that people decided to contribute. By contrast, the family security message increased not only the likelihood of making a contribution but also the average contribution size by 16.17% (from 868€ to 1,009€). The success of this message in Estonia—a highly individualistic culture—mirrors findings from culturally more collectivist Mexico (Shah et al., 2023, Beugelsdijk et al., 2017), suggesting that family-focused appeals tap into broadly shared motivations for retirement saving. Future research is needed to determined which components of this message are driving its effectiveness - the action verb “help,” the prosocial emphasis on “loved ones,” the possible “future” framing, or the legacy-orientation.\\u003c/p\\u003e\\n\\u003cp\\u003eThe average effects of different messages can conceal important individual differences in message responsiveness. Given the large and representative sample of this study, we also thoroughly investigated potential effect heterogeneity. On the one hand, we found that most of the effects discussed above were robust as they did not vary significantly with any of the moderator variables available in our dataset. On the other hand, employing a more flexible honest causal forest approach revealed conjunctions of moderators that predicted increased sensitivity to some of the messages. Specifically, the baseline message and the family security reminders were most effective among individuals with higher incomes who had made 3\\u003csup\\u003erd\\u003c/sup\\u003e pillar savings also before the intervention with age slightly increasing the family security message and reducing the baseline message effects. This pattern suggests that reminders may be most effective for individuals who have the means to save and familiarity with saving while additional strategies may be needed to engage other groups. The small age effects suggest that \\u0026nbsp; individual characteristics can interact with message content. Taken together, our treatment heterogeneity analyses suggests that simple, timely reminders have broad effects that can be amplified when messages align with financial capacity other personal circumstances.\\u0026nbsp;\\u003c/p\\u003e\\n\\u003cp\\u003eWhile our study benefits from its scale and ecological validity, several limitations suggest directions for future research. Studies may want to examine the impact of timing (e.g., one-week vs. one-month before deadline) and frequency of reminders (e.g., Milkman et al., 2025) on habit formation, assess long-term effects on saving behaviour, and investigate strategies for reaching currently disengaged populations. Following Patterson and Skimmyhorn’s (2022) study among members of the U.S. Army, researchers should also explore how reminders interact with structural factors like auto-enrolment and tax incentives. Additionally, testing these interventions across different pension systems could help identify universal versus context-specific effective strategies.\\u003c/p\\u003e\\n\\u003cp\\u003eOverall, our findings demonstrate that low-cost email reminders increased pension contributions by 14.75%. The effectiveness of family security messaging across cultural contexts suggests fundamental psychological mechanisms that could inform future interventions. These results provide valuable guidance for policymakers seeking cost-effective ways to enhance the impact of tax incentives on retirement savings.\\u003c/p\\u003e\"},{\"header\":\"Method\",\"content\":\"\\u003cp\\u003e\\u003cstrong\\u003eEthics and preregistration\\u003c/strong\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003eThis study was approved by the ethics board of the University of Tartu (Approval number 383/T-17). All data were anonymised by the Pension Registry (Nasdaq) before analysis and complied with GDPR (the European Union's General Data Protection Regulation). We consulted with the Estonian Data Protection Board and the regional GDPR expert of Nasdaq (who manages the pension registry) prior to implementation. The study was preregistered at the Center for Open Science (https://doi.org/10.17605/OSF.IO/29CXS).\\u0026nbsp;\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cstrong\\u003eSample and randomization\\u003c/strong\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003eWe included all working-age 3\\u003csup\\u003erd\\u003c/sup\\u003e pillar account holders born between 1970 and 2001 (N = 127,974) in our experiment. This excluded younger account holders who are likely not earning an income yet and also older groups who are able to already decumulate the 3\\u003csup\\u003erd\\u003c/sup\\u003e pillar savings with a lower tax rate. Account holders were randomly assigned to one of ten groups: a no-reminder control group (n = 23,238) or one of nine treatment groups (average n = 11,640 per group), each receiving a different reminder message. Randomization was conducted using R sample function. We generated 1,000 randomizations and conducted statistical tests to estimate differences between the randomized groups on age and current 2\\u003csup\\u003end\\u003c/sup\\u003e / 3\\u003csup\\u003erd\\u003c/sup\\u003e pillar account balance (using analysis of variance F test) and gender (using Chi squared proportions test). We selected a randomization with the largest mean p-value across these tests to ensure the groups were as similar in their profiles as possible.\\u003c/p\\u003e\\n\\u003cp\\u003eThe reminders were sent to account holders' private email addresses one week before the end-of-year tax deadline on December 21, 2023 by the Estonian pension registry. The emails were sent in batches at 30-minute intervals between midnight and 5:00 AM to ensure overnight delivery while avoiding server congestion.\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cstrong\\u003eMessage design\\u003c/strong\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003eWe designed nine distinct messages: a baseline reminder and eight behaviourally informed variations. The messages were developed iteratively with behavioural scientists (the authors of this paper as well as colleagues from the Estonian Public Sector Innovation Team) and Estonian pension experts (from the Ministry of Finance, Ministry of Social Affairs, pension fund managers, pension fund communication experts). We pretested messages with 17 Estonian adults to ensure clarity and cultural appropriateness. All messages followed a common structure but varied in their motivational component (see Table 1).\\u0026nbsp;\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cstrong\\u003eStatistical analysis\\u003c/strong\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003eWe analyzed three outcome variables (1) the likelihood of making additional contributions, (2) the average contribution amount across all participants, and (3) the size of contributions among those who contributed (not preregistered). For contribution likelihood, we used logistic regressions with a binary outcome variable (1 = made at least one payment during the study period; 0 = did not contribute). For contribution amounts, we used Tweedie regressions for both total contributions (including zeros) and non-zero contributions separately. We estimated suitable Tweedie parameters using Generalized Additive Models.\\u0026nbsp;\\u003c/p\\u003e\\n\\u003cp\\u003eWe conducted three sets of analyses. First, we compared the combined treatment groups against the control group using a binary independent variable (1 - reminder; 0 – no reminder). \\u0026nbsp;Second, to evaluate how effectiveness varied by messaging strategy, we compared each treatment group against both the control group and other treatment groups on the same three outcome measures. As these analyses involved more than one effect of interest, we controlled for multiple comparisons using simultaneous confidence intervals (the \\u003cem\\u003eglht\\u003c/em\\u003e function from the multicomp package).\\u003c/p\\u003e\\n\\u003cp\\u003eThird, to investigate demographic and behavioural moderators, we first conducted six multiple logistic regressions, each adding a single moderator to the models described above for evaluating individual messages: gender, age group, income level (based on 2nd pillar contributions), 2nd pillar status, 3rd pillar enrolment duration, and marginal value of year-end payments (see Supplemental Table 1 for factor levels). Categorical moderators were entered with sum coding. We applied the Holm method to correct p-values of interaction effects from all models with the same outcome variable.\\u003c/p\\u003e\\n\\u003cp\\u003eWe conducted additional non-preregistered analyses of treatment heterogeneity, using honest causal forests to predict average treatment effects conditional on the five moderator variables. We considered significant heterogeneity to exist when the 95% confidence interval of the difference between Conditional Average Treatment Effects for median-split treatment effects excluded zero. We assessed variable importance using Friedman's H statistic and interpreted findings through partial dependence plots and surrogate trees (Athey \\u0026amp; Imbens, 2016).\\u003c/p\\u003e\\n\\u003cp\\u003eAll analyses were conducted using R version 4.2.1.\\u0026nbsp;\\u003c/p\\u003e\"},{\"header\":\"Declarations\",\"content\":\"\\u003cp\\u003e\\u003cstrong\\u003eData availability\\u003c/strong\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003eOur experiments and analysis involve confidential financial data from Estonia that cannot be released publicly. We can arrange for individuals to work with the raw data for replication purposes on a secure computer after arranging a nondisclosure agreement. The research team will facilitate this process.\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cstrong\\u003eCode availability\\u003c/strong\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003eR code for the analyses is publicly available at the Center for Open Science: \\u0026nbsp;https://doi.org/10.17605/OSF.IO/29CXS\\u0026nbsp;\\u003c/p\\u003e\"},{\"header\":\"References\",\"content\":\"\\u003col\\u003e\\n\\u003cli\\u003eAthey, S., \\u0026amp; Imbens, G. (2016). Recursive partitioning for heterogeneous causal effects. Proceedings of the National Academy of Sciences, 113(27), 7353\\u0026ndash;7360. https://doi.org/10.1073/pnas.1510489113 \\u003c/li\\u003e\\n\\u003cli\\u003eAthey, S., Tibshirani, J., \\u0026amp; Wager, S. (2019). Generalized random forests. \\u003cem\\u003eAnnals of Statistics, 47\\u003c/em\\u003e(2), 1148\\u0026ndash;1178. https://doi.org/10.1214/18-AOS1709 \\u003c/li\\u003e\\n\\u003cli\\u003eBerns, G. S., Laibson, D., \\u0026amp; Loewenstein, G. (2007). Intertemporal choice\\u0026mdash;Toward an integrative framework. Trends in Cognitive Sciences, 11(11), 482\\u0026ndash;488. https://doi.org/10.1016/j.tics.2007.08.011 \\u003c/li\\u003e\\n\\u003cli\\u003eBeshears, J., Blakstad, M., Choi, J. J., Firth, C., Gathergood, J., Laibson, D., ... \\u0026amp; Stewart, N. (2024). Does Pension Automatic Enrollment Increase Debt? Evidence from a Large-Scale Natural Experiment (No. w32100). National Bureau of Economic Research.\\u003c/li\\u003e\\n\\u003cli\\u003eBeugelsdijk, S., Kostova, T., \\u0026amp; Roth, K. (2017). An overview of Hofstede-inspired country-level culture research in international business since 2006. \\u003cem\\u003eJournal of International business studies\\u003c/em\\u003e, \\u003cem\\u003e48\\u003c/em\\u003e(1), 30-47. https://doi.org/10.1057/s41267-016-0038-8 \\u003c/li\\u003e\\n\\u003cli\\u003eBursztyn, L., Ederer, F., Ferman, B., \\u0026amp; Yuchtman, N. (2014), Understanding Mechanisms Underlying Peer Effects: Evidence From a Field Experiment on Financial Decisions. \\u003cem\\u003eEconometrica\\u003c/em\\u003e, 82: 1273\\u0026ndash;1301. https://doi.org/10.3982/ECTA11991 \\u003c/li\\u003e\\n\\u003cli\\u003eButtenheim, A., Milkman, K. L., Duckworth, A. L., Gromet, D. M., Patel, M., \\u0026amp; Chapman, G. (2022). Effects of ownership text message wording and reminders on receipt of an influenza vaccination: a randomized clinical trial. JAMA network open, 5(2). https://doi.org/10.1001/jamanetworkopen.2021.43388\\u003c/li\\u003e\\n\\u003cli\\u003eChan, S., \\u0026amp; Stevens, A. (2008). What You Don\\u0026apos;t Know Can\\u0026apos;t Help You: Pension Knowledge and Retirement Decision-Making, \\u003cem\\u003eThe Review of Economics and Statistics\\u003c/em\\u003e, 90 (2), 253\\u0026ndash;266. https://doi.org/10.1162/rest.90.2.253 \\u003c/li\\u003e\\n\\u003cli\\u003eCoyne, A. J. (2024, April 16). Monzo chief marketer: Chemistry with agencies and why brands\\u0026rsquo; obsession with pitching free ideas is wrong. \\u003cem\\u003eCampaign\\u003c/em\\u003e.\\u003c/li\\u003e\\n\\u003cli\\u003eCroy, G., Gerrans, P., \\u0026amp; Speelman, C. (2010). The role and relevance of domain knowledge, perceptions of planning importance, and risk tolerance in predicting savings intentions. \\u003cem\\u003eJournal of Economic Psychology\\u003c/em\\u003e, \\u003cem\\u003e31\\u003c/em\\u003e(6), 860\\u0026ndash;871. https://doi.org/10.1016/j.joep.2010.06.002\\u003c/li\\u003e\\n\\u003cli\\u003eCrum, J., Spencer, C., Doherty, E., Richardson, E., Sherman, S., Hays, A. W., ... \\u0026amp; Hirshfield, L. (2024). Misinformation research needs ecological validity. \\u003cem\\u003eNature Human Behaviour\\u003c/em\\u003e, \\u003cem\\u003e8\\u003c/em\\u003e(12), 2268-2271. https://doi.org/10.1038/s41562-024-02015-2 \\u003c/li\\u003e\\n\\u003cli\\u003eDalen, H. P. van, \\u0026amp; Henkens, K. (2023). Trust in pension funds, or the importance of being financially sound. \\u003cem\\u003eJournal of Pension Economics \\u0026amp; Finance\\u003c/em\\u003e, \\u003cem\\u003e22\\u003c/em\\u003e(4), 658\\u0026ndash;669. https://doi.org/10.1017/S147474722200021X\\u003c/li\\u003e\\n\\u003cli\\u003eDaminato, C., Filippini, M., \\u0026amp; Haufler, F. (2024). Digitalization and retirement contribution behavior: evidence from administrative data. \\u003cem\\u003eThe Review of Financial Studies\\u003c/em\\u003e, \\u003cem\\u003e37\\u003c/em\\u003e(8), 2510-2549. https://doi.org/10.1093/rfs/hhae015 \\u003c/li\\u003e\\n\\u003cli\\u003eDe La Rosa, W., Sharma, E., Tully, S. M., Giannella, E., \\u0026amp; Rino, G. (2021). Psychological ownership interventions increase interest in claiming government benefits. \\u003cem\\u003eProceedings of the National Academy of Sciences\\u003c/em\\u003e, \\u003cem\\u003e118\\u003c/em\\u003e(35), e2106357118. https://doi.org/10.1073/pnas.2106357118\\u003c/li\\u003e\\n\\u003cli\\u003eDuckworth, A. L., Ko, A., Milkman, K. L., Kay, J. S., Dimant, E., Gromet, D. M., ... \\u0026amp; Bulte, C. V. D. (2025). A national megastudy shows that email nudges to elementary school teachers boost student math achievement, particularly when personalized. \\u003cem\\u003eProceedings of the National Academy of Sciences\\u003c/em\\u003e, \\u003cem\\u003e122\\u003c/em\\u003e(13), e2418616122. https://doi.org/10.1073/pnas.2418616122 \\u003c/li\\u003e\\n\\u003cli\\u003eDur, R., Fleming, D., van Garderen, M., \\u0026amp; van Lent, M. (2021). A social-norm nudge to save more: A field experiment at a retail bank. \\u003cem\\u003eJournal of Public Economics, 200\\u003c/em\\u003e, 104443. https://doi.org/10.1016/j.jpubeco.2021.104443\\u003c/li\\u003e\\n\\u003cli\\u003eEberhardt, W., Br\\u0026uuml;ggen, E., Post, T., \\u0026amp; Hoet, C. (2021). Engagement behavior and financial well-being: The effect of message framing in online pension communication. \\u003cem\\u003eInternational Journal of Research in Marketing\\u003c/em\\u003e, \\u003cem\\u003e38\\u003c/em\\u003e(2), 448\\u0026ndash;471. https://doi.org/10.1016/j.ijresmar.2020.11.002\\u003c/li\\u003e\\n\\u003cli\\u003eEl Mekkaoui, N., \\u0026amp; Legendre, B. (2023). Does pension information impact savings? \\u003cem\\u003eJournal of Pension Economics and Finance\\u003c/em\\u003e, 22(3), 352-370. https://doi.org/10.1017/S1474747221000470 \\u003c/li\\u003e\\n\\u003cli\\u003eFishbane, A., Ouss, A., \\u0026amp; Shah, A. K. (2020). Behavioral nudges reduce failure to appear for court. \\u003cem\\u003eScience\\u003c/em\\u003e, \\u003cem\\u003e370\\u003c/em\\u003e(6517), eabb6591.\\u003c/li\\u003e\\n\\u003cli\\u003eFoster, L. (2017). Young People and Attitudes towards Pension Planning. \\u003cem\\u003eSocial Policy and Society\\u003c/em\\u003e, \\u003cem\\u003e16\\u003c/em\\u003e(1), 65\\u0026ndash;80. https://doi.org/10.1017/S1474746415000627\\u003c/li\\u003e\\n\\u003cli\\u003eFuentes, O. M., Mitchell, O. S., \\u0026amp; Villatoro, F. (2023). Early pension withdrawals in Chile during the pandemic. \\u003cem\\u003eJournal of Pension Economics \\u0026amp; Finance\\u003c/em\\u003e, 1\\u0026ndash;26. https://doi.org/10.1017/S1474747223000112\\u003c/li\\u003e\\n\\u003cli\\u003eGallego-Losada, R., Montero-Navarro, A., Rodr\\u0026iacute;guez-S\\u0026aacute;nchez, J.-L., \\u0026amp; Gonz\\u0026aacute;lez-Torres, T. (2022). Retirement planning and financial literacy, at the crossroads. A bibliometric analysis. \\u003cem\\u003eFinance Research Letters\\u003c/em\\u003e, \\u003cem\\u003e44\\u003c/em\\u003e, 102109. https://doi.org/10.1016/j.frl.2021.102109\\u003c/li\\u003e\\n\\u003cli\\u003eGneezy, U., \\u0026amp; List, J. A. (2006). Putting behavioral economics to work: Testing for gift exchange in labor markets using field experiments. \\u003cem\\u003eEconometrica\\u003c/em\\u003e, \\u003cem\\u003e74\\u003c/em\\u003e(5), 1365-1384. https://doi.org/10.1111/j.1468-0262.2006.00707.x \\u003c/li\\u003e\\n\\u003cli\\u003eGollwitzer, P. M., \\u0026amp; Sheeran, P. (2006). Implementation intentions and goal achievement: A meta-analysis of effects and processes. \\u003cem\\u003eAdvances in Experimental Social Psychology, 38\\u003c/em\\u003e, 69\\u0026ndash;119. https://doi.org/10.1016/S0065-2601(06)38002-1 \\u003c/li\\u003e\\n\\u003cli\\u003eGourville, J. T. (1998). Pennies-a-day: The effect of temporal reframing on transaction evaluation. \\u003cem\\u003eJournal of Consumer Research, 24\\u003c/em\\u003e(4), 395\\u0026ndash;408. https://doi.org/10.1086/209517 \\u003c/li\\u003e\\n\\u003cli\\u003eGravert, C. in Behavioral Science in the Wild (eds Mažar, N. \\u0026amp; Soman, D.) 120\\u0026ndash;132 (Univ. Toronto Press, 2022).\\u003c/li\\u003e\\n\\u003cli\\u003eKarlan, D., McConnell, M., Mullainathan, S., \\u0026amp; Zinman, J. (2016). Getting to the top of mind: How reminders increase saving. \\u003cem\\u003eManagement Science, 62\\u003c/em\\u003e(12), 3393\\u0026ndash;3411. https://doi.org/10.1287/mnsc.2015.2296 \\u003c/li\\u003e\\n\\u003cli\\u003eKim, K. T., Lee, J. M., \\u0026amp; Hong, E. (2016). The role of self-control on retirement preparedness of US households. \\u003cem\\u003eInternational Journal of Human Ecology\\u003c/em\\u003e, \\u003cem\\u003e17\\u003c/em\\u003e(2), 31-42.\\u003c/li\\u003e\\n\\u003cli\\u003eKoehler, D. J., Langstaff, J., \\u0026amp; Liu, W.-Q. (2015). A simulated financial savings task for studying consumption and retirement decision making. \\u003cem\\u003eJournal of Economic Psychology\\u003c/em\\u003e, \\u003cem\\u003e46\\u003c/em\\u003e, 89\\u0026ndash;97. https://doi.org/10.1016/j.joep.2014.12.004\\u003c/li\\u003e\\n\\u003cli\\u003eLaibson, D. I., Repetto, A., Tobacman, J., Hall, R. E., Gale, W. G., \\u0026amp; Akerlof, G. A. (1998). Self-control and saving for retirement. \\u003cem\\u003eBrookings Papers on Economic Activity\\u003c/em\\u003e, \\u003cem\\u003e1998\\u003c/em\\u003e(1), 91-196.\\u003c/li\\u003e\\n\\u003cli\\u003eLinos, E., Lasky-Fink, J., Larkin, C., Moore, L., \\u0026amp; Kirkman, E. (2024). The formality effect. \\u003cem\\u003eNature Human Behaviour, 8\\u003c/em\\u003e(2), 300\\u0026ndash;310. https://doi.org/10.1038/s41562-023-01761-z \\u003c/li\\u003e\\n\\u003cli\\u003eList, J. A. (2007). Field experiments: a bridge between lab and naturally occurring data. \\u003cem\\u003eThe BE Journal of Economic Analysis \\u0026amp; Policy\\u003c/em\\u003e, \\u003cem\\u003e6\\u003c/em\\u003e(2). https://doi.org/10.2202/1538-0637.1747 \\u003c/li\\u003e\\n\\u003cli\\u003eMckenzie, C. R. M., \\u0026amp; Liersch, M. J. (2011). Misunderstanding Savings Growth: Implications for Retirement Savings Behavior. \\u003cem\\u003eJournal of Marketing Research\\u003c/em\\u003e, \\u003cem\\u003e48\\u003c/em\\u003e(SPL), S1\\u0026ndash;S13. https://doi.org/10.1509/jmkr.48.SPL.S1\\u003c/li\\u003e\\n\\u003cli\\u003eMcNair, S., Nyhus, E. K., \\u0026amp; Ranyard R. (2024). Propensity to spend and borrow at a time of high pressure: the role of the meaning of Christmas and other psychological factors. \\u003cem\\u003eFrontiers in Behavioral Economics\\u003c/em\\u003e, 3 https://doi.org/10.3389/frbhe.2024.1385609 \\u003c/li\\u003e\\n\\u003cli\\u003eMertens, S., Herberz, M., Hahnel, U. J. J., \\u0026amp; Brosch, T. (2022). The effectiveness of nudging: A meta-analysis of choice architecture interventions across behavioural domains. \\u003cem\\u003eProceedings of the National Academy of Sciences, 119\\u003c/em\\u003e(1), e2107346118. https://doi.org/10.1073/pnas.2107346118 \\u003c/li\\u003e\\n\\u003cli\\u003eMilkman, K. L., Gromet, D., Ho, H., Kay, J. S., Lee, T. W., Pandiloski, P., Park, Y., Rai, A., Bazerman, M., Beshears, J., Bonacorsi, L., Camerer, C., Chang, E., Chapman, G., Cialdini, R., Dai, H., Eskreis-Winkler, L., Fishbach, A., Gross, J. J., \\u0026hellip; Duckworth, A. L. (2021). Megastudies improve the impact of applied behavioural science. \\u003cem\\u003eNature\\u003c/em\\u003e, 600(7889), 478\\u0026ndash;483. https://doi.org/10.1038/s41586-021-04128-4\\u003c/li\\u003e\\n\\u003cli\\u003eMilkman, K. L., Ellis, S. F., Gromet, D. M., DeMay, I. M., Graci, H. N., Jung, Y., ... \\u0026amp; Duckworth, A. L. (2025). Can reminder emails compel Americans to save? A two-million-person megastudy. \\u003cem\\u003ePNAS nexus\\u003c/em\\u003e, \\u003cem\\u003e4\\u003c/em\\u003e(9). https://doi.org/10.1093/pnasnexus/pgaf280 \\u003c/li\\u003e\\n\\u003cli\\u003eMrkva, K., Posner, N. A., Reeck, C., \\u0026amp; Johnson, E. J. (2021). Do Nudges Reduce Disparities? Choice Architecture Compensates for Low Consumer Knowledge. \\u003cem\\u003eJournal of Marketing\\u003c/em\\u003e, 85(4), 67-84. https://doi.org/10.1177/0022242921993186\\u003c/li\\u003e\\n\\u003cli\\u003eMunnell, A. H., Chen, A., \\u0026amp; Siliciano, R. L. (2021). The National Retirement Risk Index: An Update from the 2019 SCF, Issues in Brief 21-02, Center for Retirement Research.\\u003c/li\\u003e\\n\\u003cli\\u003eOECD. (2023). \\u003cem\\u003ePensions at a Glance 2023: OECD and G20 indicators\\u003c/em\\u003e. OECD Publishing. https://doi.org/10.1787/4757bf20-en \\u003c/li\\u003e\\n\\u003cli\\u003ePatterson, R. W., \\u0026amp; Skimmyhorn, W. L. (2022). How do Behavioral Approaches to Increase Savings Compare? Evidence from Multiple Interventions in the U.S. Army. NBER working paper #30697. https://doi.org/10.3386/w30697 \\u003c/li\\u003e\\n\\u003cli\\u003eRaghubir, P., \\u0026amp; Srivastava, J. (2009). The denomination effect. \\u003cem\\u003eJournal of Consumer Research, 36\\u003c/em\\u003e(4), 701\\u0026ndash;713. https://doi.org/10.1086/599222 \\u003c/li\\u003e\\n\\u003cli\\u003eRiitsalu, L., Uusberg, A. (2021). To double, quadruple, or keep? Semi-automated service increases micro-investments. \\u003cem\\u003eJournal of Behavioral and Experimental Economics\\u003c/em\\u003e, \\u003cem\\u003e95\\u003c/em\\u003e, 101774. https://doi.org/10.1016/j.socec.2021.101774\\u003c/li\\u003e\\n\\u003cli\\u003eRuggeri, K., Panin, A., Vdovic, M., Većkalov, B., Abdul-Salaam, N., Achterberg, J., ... \\u0026amp; Toscano, F. (2022). The globalizability of temporal discounting. \\u003cem\\u003eNature Human Behaviour, \\u003c/em\\u003e6(10), 1386-1397.\\u003c/li\\u003e\\n\\u003cli\\u003eR\\u0026uuml;\\u0026uuml;tsalu, K., Reinson, H., Tillmann, T., Uusberg, A. \\u003cem\\u003eShort- and Long-Term Effects of SMS Reminders on Men`s Participation in Colorectal Cancer Screening.\\u003c/em\\u003e [Manuscript in preparation]\\u003c/li\\u003e\\n\\u003cli\\u003eSaccardo, S., Dai, H., Han, M. A., Vangala, S., Hoo, J., \\u0026amp; Fujimoto, J. (2024). Field testing the transferability of behavioural science knowledge on promoting vaccinations. \\u003cem\\u003eNature Human Behaviour, 8\\u003c/em\\u003e(5), 878\\u0026ndash;890. https://doi.org/10.1038/s41562-023-01813-4 \\u003c/li\\u003e\\n\\u003cli\\u003eSaez, E. (2009). Details Matter: The Impact of Presentation and Information on the Take-Up of Financial Incentives for Retirement Saving. \\u003cem\\u003eAmerican Economic Journal: Economic Policy\\u003c/em\\u003e, 1(1): 204\\u0026ndash;28. https://doi.org/10.1257/pol.1.1.204 \\u003c/li\\u003e\\n\\u003cli\\u003eShah, A. M., Osborne, M., Lefkowitz Kalter, J., Fertig, A., Fishbane, A., \\u0026amp; Soman, D. (2023). Identifying heterogeneity using recursive partitioning: Evidence from SMS nudges encouraging voluntary retirement savings in Mexico. \\u003cem\\u003ePNAS Nexus\\u003c/em\\u003e, \\u003cem\\u003e2\\u003c/em\\u003e(5), pgad058. https://doi.org/10.1093/pnasnexus/pgad058\\u003c/li\\u003e\\n\\u003cli\\u003eSunstein, C. R. (2022). The distributional effects of nudges. \\u003cem\\u003eNature Human Behaviour, 6\\u003c/em\\u003e(1), 9\\u0026ndash;10. https://doi.org/10.1038/s41562-021-01236-z \\u003c/li\\u003e\\n\\u003cli\\u003eThaler, R. H. (2013, April 6). Shifting Our Retirement Savings Into Automatic. \\u003cem\\u003eThe New York Times.\\u003c/em\\u003e https://www.nytimes.com/2013/04/07/business/an-automatic-solution-for-the-retirement-savings-problem.html\\u003c/li\\u003e\\n\\u003cli\\u003eTIAA-CREF (2014). TIAA-CREF Survey finds Americans spend less time planning their IRA investment than choosing a restaurant. Retrieved from https://www.tiaa.org/public/about-tiaa/news-press/press-releases/pressrelease495.html \\u003c/li\\u003e\\n\\u003cli\\u003eTomar, S., Kent Baker, H., Kumar, S., \\u0026amp; Hoffmann, A. O. I. (2021). Psychological determinants of retirement financial planning behavior. \\u003cem\\u003eJournal of Business Research\\u003c/em\\u003e, \\u003cem\\u003e133\\u003c/em\\u003e, 432\\u0026ndash;449. https://doi.org/10.1016/j.jbusres.2021.05.007\\u003c/li\\u003e\\n\\u003cli\\u003eVan Den Akker, M., \\u0026amp; Sunstein, C. R. (2023). Do People Like Financial Nudges? \\u003cem\\u003eSSRN Electronic Journal\\u003c/em\\u003e. https://doi.org/10.2139/ssrn.4636905\\u003c/li\\u003e\\n\\u003cli\\u003evan Rooij, M. C. J., Lusardi, A., \\u0026amp; Alessie, R. J. M. (2011). Financial literacy and retirement planning in the Netherlands. \\u003cem\\u003eJournal of Economic Psychology\\u003c/em\\u003e, \\u003cem\\u003e32\\u003c/em\\u003e(4), 593\\u0026ndash;608. https://doi.org/10.1016/j.joep.2011.02.004\\u003c/li\\u003e\\n\\u003cli\\u003eVervloet, M., Linn, A. J., van Weert, J. C. M., de Bakker, D. H., Bouvy, M., \\u0026amp; van Dijk, L. (2012). The effectiveness of interventions using electronic reminders to improve adherence to chronic medication: A systematic review. \\u003cem\\u003eJournal of the American Medical Informatics Association, 19\\u003c/em\\u003e(5), 696\\u0026ndash;704. https://doi.org/10.1136/amiajnl-2011-000748 \\u003c/li\\u003e\\n\\u003cli\\u003eWager, S., \\u0026amp; Athey, S. (2018). Estimation and inference of heterogeneous treatment effects using random forests. \\u003cem\\u003eJournal of the American Statistical Association, 113\\u003c/em\\u003e(523), 1228\\u0026ndash;1242. https://doi.org/10.1080/01621459.2017.1319839 \\u003c/li\\u003e\\n\\u003cli\\u003eZaval, L., Markowitz, E. M., \\u0026amp; Weber, E. U. (2015). How will I be remembered? Conserving the environment for the sake of one\\u0026rsquo;s legacy. \\u003cem\\u003ePsychological Science, 26\\u003c/em\\u003e(2), 231\\u0026ndash;236. https://doi.org/10.1177/0956797614561266 \\u003c/li\\u003e\\n\\u003c/ol\\u003e\"},{\"header\":\"Footnotes\",\"content\":\"\\u003cp\\u003e\\u003csup\\u003e\\u003csup\\u003e[1]\\u003c/sup\\u003e\\u003c/sup\\u003e The 3rd pillar offers investment flexibility, allowing contributions to be allocated across a broad range of mutual funds with varying asset classes and management styles, from low-cost index funds to actively managed portfolios, enabling individuals to align their investments with their risk preferences and financial goals.\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003csup\\u003e\\u003csup\\u003e[1]\\u003c/sup\\u003e\\u003c/sup\\u003e The 2nd pillar is an auto-enrolment retirement account that complements the 3rd pillar. While both offer tax advantages, 2nd pillar contributions are automatically set at 6% of gross income, making them a reliable proxy for income levels.\\u003c/p\\u003e\"}],\"fulltextSource\":\"\",\"fullText\":\"\",\"funders\":[],\"hasAdminPriorityOnWorkflow\":false,\"hasManuscriptDocX\":true,\"hasOptedInToPreprint\":true,\"hasPassedJournalQc\":\"\",\"hasAnyPriority\":true,\"hideJournal\":false,\"highlight\":\"\",\"institution\":\"\",\"isAcceptedByJournal\":false,\"isAuthorSuppliedPdf\":false,\"isDeskRejected\":\"\",\"isHiddenFromSearch\":false,\"isInQc\":false,\"isInWorkflow\":false,\"isPdf\":false,\"isPdfUpToDate\":true,\"isWithdrawnOrRetracted\":false,\"journal\":{\"display\":true,\"email\":\"info@researchsquare.com\",\"identity\":\"nature-portfolio\",\"isNatureJournal\":true,\"hasQc\":false,\"allowDirectSubmit\":false,\"externalIdentity\":\"\",\"sideBox\":\"\",\"snPcode\":\"\",\"submissionUrl\":\"\",\"title\":\"Nature Portfolio\",\"twitterHandle\":\"\",\"acdcEnabled\":false,\"dfaEnabled\":false,\"editorialSystem\":\"ejp\",\"reportingPortfolio\":\"\",\"inReviewEnabled\":true,\"inReviewRevisionsEnabled\":false},\"keywords\":\"\",\"lastPublishedDoi\":\"10.21203/rs.3.rs-8008072/v1\",\"lastPublishedDoiUrl\":\"https://doi.org/10.21203/rs.3.rs-8008072/v1\",\"license\":{\"name\":\"CC BY 4.0\",\"url\":\"https://creativecommons.org/licenses/by/4.0/\"},\"manuscriptAbstract\":\"\\u003cp\\u003eInsufficient retirement savings threaten financial security worldwide as individuals increasingly bear responsibility for funding their retirement. We present the first population-level test of behaviourally informed messaging for boosting saving in retirement accounts. In collaboration with the Estonian Ministry of Finance, we conducted a preregistered randomized controlled trial among all eligible account holders (N\\u0026thinsp;=\\u0026thinsp;127,974) testing whether email reminders about tax benefits boost voluntary pension contributions. Across nine behaviourally designed reminders versus a no-reminder control, reminders increased contribution likelihood by 10.49% and raised average contributions by 14.75% within one week before the deadline. Most framings increased participation but notably, a family-security message also increased individual contribution size (from \\u0026euro;868 to \\u0026euro;1,009). Exploratory causal forest analysis revealed the baseline message most strongly influenced younger, higher-income individuals with prior contributions, while the family-security message had the largest impact on high-income individuals. The intervention generated ~\\u0026euro;1.2\\u0026nbsp;million in additional retirement savings within one week. These results show that simple, low-cost behavioural reminders can meaningfully amplify tax incentives at scale, offering practical tools for policymakers.\\u003c/p\\u003e\",\"manuscriptTitle\":\"Behavioural messages amplify tax incentives: \\nA nationwide megastudy of retirement savings reminders\",\"msid\":\"\",\"msnumber\":\"\",\"nonDraftVersions\":[{\"code\":1,\"date\":\"2026-01-06 14:53:47\",\"doi\":\"10.21203/rs.3.rs-8008072/v1\",\"editorialEvents\":[],\"status\":\"published\",\"journal\":{\"display\":true,\"email\":\"info@researchsquare.com\",\"identity\":\"nature-communications\",\"isNatureJournal\":true,\"hasQc\":false,\"allowDirectSubmit\":false,\"externalIdentity\":\"NCOMMS\",\"sideBox\":\"Learn more about [Nature Communications](http://www.nature.com/ncomms/)\",\"snPcode\":\"\",\"submissionUrl\":\"https://mts-ncomms.nature.com/\",\"title\":\"Nature Communications\",\"twitterHandle\":\"\",\"acdcEnabled\":true,\"dfaEnabled\":true,\"editorialSystem\":\"ejp\",\"reportingPortfolio\":\"Nature Communications\",\"inReviewEnabled\":true,\"inReviewRevisionsEnabled\":false}}],\"origin\":\"\",\"ownerIdentity\":\"dee71a14-8a5c-4ffd-9de3-be068e4aac64\",\"owner\":[],\"postedDate\":\"January 6th, 2026\",\"published\":true,\"recentEditorialEvents\":[],\"rejectedJournal\":[],\"revision\":\"\",\"amendment\":\"\",\"status\":\"under-review\",\"subjectAreas\":[{\"id\":58524747,\"name\":\"Social science/Psychology/Human behaviour\"},{\"id\":58524748,\"name\":\"Social science/Economics\"}],\"tags\":[],\"updatedAt\":\"2026-01-06T14:53:47+00:00\",\"versionOfRecord\":[],\"versionCreatedAt\":\"2026-01-06 14:53:47\",\"video\":\"\",\"vorDoi\":\"\",\"vorDoiUrl\":\"\",\"workflowStages\":[]},\"version\":\"v1\",\"identity\":\"rs-8008072\",\"journalConfig\":\"researchsquare\"},\"__N_SSP\":true},\"page\":\"/article/[identity]/[[...version]]\",\"query\":{\"redirect\":\"/article/rs-8008072\",\"identity\":\"rs-8008072\",\"version\":[\"v1\"]},\"buildId\":\"XKTyCvWXoU3ODBz1xrDgd\",\"isFallback\":false,\"isExperimentalCompile\":false,\"dynamicIds\":[84888],\"gssp\":true,\"scriptLoader\":[]}","source_license":"CC-BY-4.0","license_restricted":false}