Association between statewide financial incentive programs and COVID-19 vaccination rates

To promote COVID-19 vaccination, many states in the US introduced financial incentives ranging from small, guaranteed rewards to lotteries that give vaccinated individuals a chance to win large prizes. There is limited evidence on the effectiveness of these programs and conflicting evidence from survey experiments and studies of individual states’ lotteries. To assess the effectiveness of COVID-19 vaccination incentive programs, we combined information on statewide incentive programs in the US with data on daily vaccine doses administered in each state. Leveraging variation across states in the daily availability of incentives, our difference-in-differences analyses showed that statewide programs were not associated with a significant change in vaccination rates. Furthermore, there was no significant difference in vaccination trends between states with and without incentives in any of the 14 days before or after incentives were introduced. Heterogeneity analyses indicated that neither lotteries nor guaranteed rewards were associated with significant change in vaccination rates.


Background
Despite widespread availability of COVID-19 vaccines, about half of Americans are not fully vaccinated and vaccination rates have declined since their peak in April 2021. To encourage vaccination, many states introduced incentive programs ranging from guaranteed rewards (e.g., small rewards like free beverages or larger rewards like gift cards of up to $100) to lotteries in which vaccinated individuals had a chance to win $1 million or more. Lotteries in particular have been widely used because of evidence that people tend to overweight small probabilities and thus respond to large jackpots more than they would to small cash payments [1]. However, it is unclear whether such incentive programs are effective at increasing COVID vaccination rates given the strong disinterest many of the unvaccinated may have in being vaccinated. a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 Existing studies of incentives for COVID-19 vaccination offer conflicting accounts of their effectiveness. A survey experiment in Germany suggests guaranteed rewards of about $25 or higher can significantly increase vaccine uptake [2]. In contrast, evaluations of the first vaccine lottery in Ohio have yielded mixed results [3,4]. The relatively early introduction of incentives in Ohio may also limit the generalizability of results from there. One other study examining statewide incentive programs introduced prior to early-June 2021 concluded that lottery incentives were effective in 10 of 12 states that were studied [5]. However, methodological limitations stemming from cross-sectional-analysis confounds and unadjusted correlation in errors in daily vaccination rates within a state warrant further evaluation of these programs. Heterogeneity in the success of statewide incentive programs remains underexplored as well. Lotteries and guaranteed rewards may vary in their effectiveness. Given substantial political partisanship in COVID-19 prevention policies [6,7], conservative and liberal states may also have varying success in promoting vaccination with incentives.
We study the effectiveness of statewide incentive programs for COVID-19 vaccination using detailed information on 24 statewide incentive programs and data on daily vaccine doses administered in each state. We also examine heterogeneity in program effectiveness based on key characteristics of incentive programs and the states where they were introduced.

Methods
We compiled information on statewide incentive programs from the National Governors' Association [8] and Google News. For each program, we recorded the start date (when vaccinations were first incentivized), end date, and incentive type (lottery or guaranteed reward). For states with multiple programs, we recorded the dates for the program with the highest expected value per resident. We calculated the number of vaccine doses administered daily per 100,000 individuals in each state with data from the US Centers for Disease Control and Prevention [9], focusing on the period April 1, 2021 (before the first statewide incentive program began) to July 29, 2021. We also classified states based on whether they voted for the Republican or Democratic Party candidate in the 2020 Presidential Election.
For each state-date combination, we created an indicator variable reflecting an active incentive program. Leveraging variation across states in the daily presence of incentives, we used difference-in-differences regressions to examine the association between incentive programs and vaccination rates. Specifically, to determine this association during the entire period when incentives were active, we estimated the following model: Vaccinations jt = Incentive jt +θ j +γ t , where j indexes the state and t indexes the date. Incentive jt is our indicator variable for an active incentive program in state j on date t, and θ j and γ t are fixed effects for each state and day, respectively, in order to adjust for time-invariant differences across states in vaccination rates and for national trends over time. The standard errors we report are clustered to account for arbitrary correlation of error terms at the state level.
To test for heterogeneity in the effectiveness of incentive programs, we performed subgroup analyses that separately examined effects of lotteries vs. guaranteed rewards, incentive programs introduced early vs. late (based on the median date when incentive programs were introduced in our sample), and incentive programs in Republican-vs. Democratic-leaning states.
In contrast to the difference-in-differences analyses that combined data from all states that had incentive programs, we also used a synthetic control approach that examined the impact of each state's incentive program one at a time relative to a synthetic control comprised of the "control group" of states that never had incentive programs. The synthetic control was defined on the basis of the daily vaccination rate in control group states in the 1 day before an incentive program was introduced as well as the 8 days before the program was introduced.
In additional analyses, we assessed whether responsiveness to incentives may have been greatest in the weeks immediately after incentives were introduced. In these analyses we also tested whether pre-program time trends were similar between states with and without incentive programs, a key assumption of the difference-in-differences model. Specifically, we defined indicator variables for each of the 14 days before and after incentive programs were introduced and estimated the following model that compared daily vaccination rates between states with and without incentive programs in the two 14-day periods, while again clustering standard errors at the state level: Vaccinations jt ¼ P 14 t¼À 14 Incentive jt þ y j þ g t . This model allowed us to look at vaccination trends leading up to the date when incentives were introduced and the subsequent effect of incentives on vaccination uptake.

Results
Twenty-four states introduced vaccination incentive programs during the study period ( Table 1). The median (interquartile range) percent of the population that was fully vaccinated when incentives began was 43.8% (39.3%-47.2%). Among these states, vaccination rates declined from a daily average of 486/100,000 individuals in the 14 days pre-incentives to a daily average of 351/100,000 individuals in the 14 days post-incentives. This reflected a national trend, as daily vaccination rates also declined in the 26 states without statewide incentive programs during comparable 14-day periods (from 351/100,000 to 272/100,000 individuals vaccinated daily). Difference-in-differences analysis showed that overall, incentive programs were associated with a non-significant relative decline in daily vaccination rates of 8.9/100,000 individuals (p = 0.75) during the period when incentives were deployed ( Table 2, Column 1). The 95% confidence interval for this main effect suggests we can rule out that the incentive programs increased daily vaccinations by 45/100,000 (a~10% increase in daily vaccination rates given the average daily vaccinations in our dataset). In the Appendix, we show the evolution of vaccination rates in the period before and after incentives were introduced in each of the 24 states with incentive programs as well as that state's unique synthetic control (S1 Appendix). While our main results in Table 2 show that incentive programs did not increase vaccination rates, the results from the synthetic control analyses allow readers to visually inspect each state's incentive program individually for suggestive evidence that incentives may have been effective in a few of the states.
In subgroup analyses, neither lottery incentives (in 18 states) nor the provision of guaranteed incentives (in 6 states) had significant effects on vaccination rates (Table 2, Columns 2 and 3). As columns 4 and 5 show, incentives also did not have a significant effect on vaccination rates in states that began offering incentives before or after the median date when statewide incentive programs began (May 27, 2021). Finally, Columns 6 and 7 of Table 2 provide suggestive evidence that incentives were (marginally) effective in promoting vaccination in states with Republican-leaning electorates, which had considerably lower vaccination rates than Democratic-leaning states. In Republican-leaning states, incentive programs were associated with an increase in daily vaccination rates of 56.8/100,000 individuals (p = 0.073). In contrast, incentives were associated with a decline in daily vaccination rates of 51.1/100,000 individuals (p = 0.095) in states with Democratic-leaning electorates.
Comparing vaccination rates in the 14 days before incentives were introduced, we confirmed that states with and without incentive programs had similar trends in vaccination rates (Fig 1)-a finding that supports the parallel trends assumption in our difference-in-differences analyses. Each point in Fig 1 shows the difference between states with and without incentive programs in the days leading up to and after the date when incentives were introduced. In the 14 days before incentives, there did not appear to be a sharp increase or decrease in the difference in vaccination rates. This was confirmed with an F-test that tested the joint significance of the coefficients. A joint test of the coefficients for the 7 days before incentives found that the coefficients were not significantly different (F(7,49) = 1.21; p = 0.31). Expanding to the 14 days before incentives, we found a marginally significant difference (F(14, 49) = 1.98; p = 0.04), that was driven by slight differences between states with and without incentives in the 7-14 days before incentives were introduced. Furthermore, there was no significant difference in vaccination trends between states with and without incentives in any of the 14 days after incentives were introduced. The latter result provides additional confirmation that the All 24 states with incentive programs n/a n/a n/a n/a n/a 486 351 -135 All 26 states without incentive programs n/a n/a n/a n/a n/a 351 272 -79 Eligibility indicates whether all vaccinated individuals or newly vaccinated were eligible for incentives. Program descriptions focus on incentives that were typically offered for individuals aged �18 years, as individuals aged 12-17 years typically received other incentives such as scholarship funds. Among the 26 states without statewide incentive programs, the average daily vaccine doses administered per 100,000 individuals are calculated over the 14-day periods before (and after) the start dates of incentives in the 24 states with statewide incentive programs.  1. Difference-in-differences analysis of daily vaccination rates in states with and without incentive programs, 14 days before and after start of incentive programs. Coefficients and 95% confidence intervals shown from estimation of a difference-in-difference regression model with indicator variables for each of the 14 days before and after statewide incentive programs are introduced. For each day, the y-axis shows the difference in daily vaccine doses administered per 100,000 individuals between states with and without statewide incentive programs, after adjusting for time-invariant differences across states and for national trends with state and date fixed effects, respectively. main difference-in-difference results do not mask short-term increases in vaccination rates after incentives were launched.

Discussion
Lotteries and other incentives offered by 24 states were not associated with a significant change in COVID-19 vaccination rates. Adjusting for national trends in vaccination rates and correlation in daily vaccination rates within states, this study goes beyond existing studies of statewide programs that have focused on individual states or may not have adjusted for confounding factors. Confidence intervals for our analyses indicate we had insufficient statistical power to detect small effects of incentives, but that increases of greater than 10% in daily vaccination rates can be ruled out. Our findings are also consistent with recent evaluations of city-wide lottery incentives like those offered in Philadelphia [10]. Many factors likely explain our findings.
With about 40% of individuals already fully vaccinated when incentives were introduced-and significant resistance to vaccination among many of the unvaccinated-small rewards (e.g. $5-$50) or low-probability lotteries may have been insufficiently persuasive to unvaccinated individuals. Incentives have been effective in other contexts [11], but their impact may be attenuated among those whose vaccine intentions are shaped by misinformation or distrust. In certain contexts in which incentives have been effective at promoting healthy behavior (such as smoking cessation on weight loss), individuals who were offered incentives typically had an underlying desire to change their behavior whereas the desire to get vaccinated may be minimal for many who are now being offered incentives. Low awareness of incentive programs may also reduce effectiveness. A limitation of the study is that employer or local government incentive programs that we did not observe may have dampened the effects of statewide programs. Overall, our findings suggest that more substantial incentives or mandates may be necessary to raise vaccination rates.