2. Theory and prior evidence

The present paper analyzes whether a UBI disbursed to all citizens without a means-test affects the turnout rate of an entire electorate. It integrates into research studying the link between income and political participation, finding that more affluent citizens have a higher probability to turn out to vote than poorer citizens (Solt, Reference Solt2008; Leighley and Nagler, Reference Leighley and Nagler2014; Schlozman et al., Reference Schlozman, Brady and Verba2018)—or that with a “diminishing marginal utility of income” (Ojeda, Reference Ojeda2018, 814) the income effect on turnout is larger for those who are poor (Rosenstone, Reference Rosenstone1982).

Navigating the voting process is inherently costly (Akee et al., Reference Akee, Copeland, Costello, Holbein and Simeonova2018, 5); the ability to deal with these costs—to overcome time constraints, navigate registration requirements, or acquire information in order to participate in a political system—depends on the individual's endowment of resources (Knack and White, Reference Knack and White2000; Sondheimer and Green, Reference Sondheimer and Green2010; Markovich and White, Reference Markovich and White2022). Amongst others, the resources time, money, civic skills, health, and education have been identified “that make it possible to take part” (Verba and Nie, Reference Verba and Nie1972; Brady et al., Reference Brady, Verba and Schlozman1995; Schlozman et al., Reference Schlozman, Brady and Verba2018, 73). Accordingly, income allows for the development and acquisition of abilities necessary for participation, having a job or being in school further allows building social connections and civic skills that enhance participation (Schlozman et al., Reference Schlozman, Brady and Verba2018). An increase in income based on the introduction of a UBI may hence affect turnout as a resource effect materializes—allowing recipients to deal with the act of voting.

Research has identified the existence of an “income gradient” in turnout, a clear causal mechanism could not be verified (Akee et al., Reference Akee, Copeland, Costello, Holbein and Simeonova2018). Smets and van Ham (Reference Smets and van Ham2013) demonstrate that in the majority of turnout studies included in their meta-analysis, income has a significant turnout-enhancing effect. It has however also been shown that only the interaction of income with factors such as education, race, neighborhood affluence, or employment status significantly influences voter turnout (Wolfinger and Rosenstone, Reference Wolfinger and Rosenstone1980; Leighley and Nagler, Reference Leighley and Nagler1992; Estrada-Correa and Johnson, Reference Estrada-Correa and Johnson2012; Burden and Wichowsky, Reference Burden and Wichowsky2014; Cebula, Reference Cebula2017).

Moving toward transfer programs, the assignment to CCT programs provide mixed evidence on a turnout-enhancing effect. On the one hand, studies found turnout and the intent to vote to increase (Baez et al., Reference Baez, Camacho, Conover and Zárate2012; Pop-Eleches and Pop-Eleches, Reference Pop-Eleches and Pop-Eleches2012; De La O, Reference De La O2013), yet only when the payment is relatively large (Galiani et al., Reference Galiani, Ibarraran and McEwan2016). On the other hand, Imai et al. (Reference Imai, King and Rivera2020) reevaluate two CCT programs in Mexico whereby they are unable to detect a significant effect, thereby refuting De La O's (Reference De La O2013) findings. After all, the programs being available only to population subgroups combined with a conditionality clause aggravate the reliability of the measurements. Programs were tied to recipients attending community meetings and health checkups, children's school attendance or enrolment, or were earmarked for the purchase of certain goods (Pop-Eleches and Pop-Eleches, Reference Pop-Eleches and Pop-Eleches2012; De La O, Reference De La O2013; Galiani et al., Reference Galiani, Ibarraran and McEwan2016; Imai et al., Reference Imai, King and Rivera2020). Health, education, and civil skills are all resources that are associated with participatory behavior, the effect of additional economic resources on turnout based on CCT programs may thereby not be isolated from the adjusted behavior (Akee et al., Reference Akee, Copeland, Costello, Holbein and Simeonova2018). There hence exists a research gap with regard to turnout responses based on an unconditional income, making the causal identification strategy of this paper more valuable.

Besides these empirical limitations, further, theoretical limitations become apparent. Scholars have focused on the actual increase in monetary resources and their participatory effects, yet an increase in income may also be a “meaningful political symbol” to beneficiaries of the payments and can result in interpretive effects (Markovich and White, Reference Markovich and White2022). Selected studies on means-tested government programs utilize the same theoretical mechanism and hypothesize both monetary resources and positive psychological effects to account for a turnout increase (Clinton and Sances, Reference Clinton and Sances2018; Baicker and Finkelstein, Reference Baicker and Finkelstein2019). I hence assume a UBI to also affect the “cognitive processes of social actors” (Pierson, Reference Pierson1993, 610). Attitudes and awareness toward political institutions and their activities may change as policies can convey “messages of inclusion and empowerment”, increasing participatory behavior (Baicker and Finkelstein, Reference Baicker and Finkelstein2019, 387). A policy tends to be interpreted affirmatively as it is not means-tested and individual claims become unconditionally legitimate. This may lead to a different understanding of the role an administration plays in the lives of citizens and of their own role in the political system (Campbell, Reference Campbell2012, 342; Clinton and Sances, Reference Clinton and Sances2018, 17; Markovich and White, Reference Markovich and White2022, 4). Payments may hence cause intrinsic reciprocity as citizens derive utility from “giving back” by turning out to vote (Galiani et al., Reference Galiani, Ibarraran and McEwan2016, 5).

Based on these considerations I theorize both mechanisms—resource and interpretive effects—to occur as a response to the introduction of a UBI and thereby result in overall turnout to increase. The unconditional nature of the UBI payments allows me to overcome empirical limitations of available research as I am able to isolate the effect of income on political participation but also tackle theoretical limitations of studies that solely focus on an increase in monetary resources and leave out psychological or positive interpretive effects.

3. Background: the Alaska permanent fund dividend

To test the effect of a UBI on voter turnout I use data from a natural experiment: the Alaska Permanent Fund Dividend program (dividend or Alaskan UBI thereafter). It was introduced in 1982, disbursing a share of Alaska's Permanent Fund evenly among all recipients each year just one month prior to the November general elections. Every resident that has lived in Alaska during the previous calendar year is eligible for the payments (Alaska Department of Revenue, 2020), which is why the program is considered “essentially universal” (Jones and Marinescu, Reference Jones and Marinescu2018, 7).

The question remains whether the dividend is indeed a case of a UBI. While most UBI models or experiments are funded through a redistributive mechanism, the program disburses a dividend from a sovereign wealth fund that is based on the state's royalty income from oil production. The Alaskan UBI was and is not communicated to be a welfare program; intention however does not stand in contrast to the definitional requirements of a UBI (Zelleke, Reference Zelleke, Widerquist and Howard2016; Mahdavi, Reference Mahdavi2020). The dividend payments are guaranteed and paid out on a regular basis yet given the variable nature of the fund's performance and therefore the variable nature of the dividend payments—its amount is not. Over time, the dividend is neither able to account for a considerable share of the annual personal income in Alaska nor for a social minimum measured in relation to the US poverty thresholds (see Table A4), leading some authors to consider the dividends a “partial basic income” (Goldsmith, Reference Goldsmith, Widerquist and Howard2016, 49–50; Zelleke, Reference Zelleke, Widerquist and Howard2016). On the other hand, an annual survey by the FED shows, that almost 40 percent of Americans are unable to immediately access USD 400 to cover an emergency expense—with the dividend payments fluctuating between USD 300 and USD 2,000 in the analysis period one can still consider the payment size substantial (The Federal Reserve, 2020).

Albeit these limitations, the Alaskan dividend proves to indeed reflect the features of a UBI as the five definitional characteristics that were formulated by the Basic Income Earth Network are met (BIEN Network, 2020). The dividend is paid out in cash, disbursed annually in early October, it is available to all residents to the state of Alaska irrespective of their age or citizenship, applicants neither have to prove that they are in need nor have to demonstrate a willingness to work, and the dividend is guaranteed as it is based on a constitutional amendment, which underlines that the dividend program was designed to last (Alaska Department of Revenue, 2020). Using data from the Alaskan UBI therefore allows me to empirically test whether an exogenous variation in income through the dividend introduction affects turnout over time. It not only fulfills the definitional requirements of a UBI, it is available to all Alaskan citizens and it has been in place for almost 40 years—guaranteed.

4. Data and empirical strategy

The dataset covers an analysis period from 1978 to 2000 and is determined by both the introduction of the dividend payments in 1982, and the data availability of the dependent variable: US-wide turnout data at the state-level as well as individual-level voting behavior on November general elections (U.S. Census Bureau, Current Population Survey, 2019a, 2019b). I also include a vector of control variables including socioeconomic and demographic factors, GDP, and unemployment rate as well as registration requirements (Rosenstone and Wolfinger, Reference Rosenstone and Wolfinger1978; Barone and de Blasio, Reference Barone and de Blasio2013; Burden et al., Reference Burden, Canon, Mayer and Moynihan2014; Burden and Wichowsky, Reference Burden and Wichowsky2014).

The empirical analyses rely on a differences-in-differences (DiD) estimation methodology comparing the pre- to post-treatment change in both turnout and individual-level voting behavior for the state of Alaska—where the dividend program is introduced in 1982—relative to differences in turnout in all other US states. As no natural control group exists within the state of Alaska, the control group therefore consists of 49 US states plus Washington D.C. that proxy for the counterfactual outcomes in the absence of the dividend payments. Analogous to Bechtel and Hainmueller (Reference Bechtel and Hainmueller2011) I consider i = {1,…, 51} states where biennial statewide general elections took place in the years t = {1978, 1980,…, 2000}. Let ${\cal D}_{it}$ denote the binary coded dividend indicator that is coded 1 for the treatment group (Alaska) after the introduction of the Alaskan UBI and 0 otherwise. The indicator is an interaction of a dummy for the state of Alaska and a dummy for the treatment period starting in 1982. Let $\Upsilon _{it}$ equal the electoral turnout in state i at time t when exposed to the treatment or the control group. I estimate α ₁ which is the causal effect of the dividend payment using a standard fixed-effects model (Equation 1):

(1)$$\Upsilon _{it} = \eta _i + \delta _t + \alpha _1D_{it} + {\rm {\rm X}}_{it}^{\rm ^{\prime}} \beta + \varepsilon _{it}\;, \;$$

where η _i is the state-level fixed effect which allows to control for any time-invariant unobserved factors and δ _t is a period-fixed effect included to control for common trends, capturing factors that would have had an effect on turnout even in the absence of the dividend payments, and ${\rm {\rm X}}_{it}^{\rm ^{\prime}}$ is a vector of socioeconomic time-varying covariates. As Bertrand et al. (Reference Bertrand, Duflo and Mullainathan2004, 249) point out, conventional DiD standard errors may “severely understate” the coefficients standard deviation. The standard errors are therefore clustered by state to account for potential serial correlation and heteroscedasticity. In order to increase the reliability of the results, I run several DiD analyses, corresponding to elections from 1978 to 1982 (short-term), 1978 to 1990 (medium-term), and 1978 to 2000 (long-term). This not only allows to estimate whether the introduction of the dividend had an effect on turnout in general but if the effect has persisted and changed over time.

In order to test for the marginal effect of the dividend size, I also estimate a generalized DiD model where I replace the binary coded dividend indicator with the continuous d_it treatment indicator. This corresponds to the dividend amount in USD divided by 1,000, adjusted for inflation. The variable measures, whether an increase in the dividend amount increases turnout and reports the marginal effect of the payments. I estimate α ₂ which is the causal effect of interest being the dividend payments using a standard fixed-effects model (Equation 2):

(2)$$\Upsilon _{it} = \eta _i + \delta _t + \alpha _2d_{it} + {\rm {\rm X}}_{it}^{\rm ^{\prime}} \beta + \varepsilon _{it}.$$

The other parameters of the equation remain the same.

6. Why does a UBI affect turnout

The presented findings on the introduction of the dividend program provide compelling evidence that an entire electorate responds to the payments with an increased turnout rate, with the effect persisting over time. This runs counter to existing research that finds electoral rewards as a response to policies to deteriorate swiftly (Gerber and Green, Reference Gerber and Green1998; Duch and Stevenson, Reference Duch and Stevenson2006). It thereby refutes a popular argument against a UBI—that especially its introduction would only prompt short-term effects inciting reelection seeking incumbents (Gentilini et al., Reference Gentilini, Grosh, Rigolini and Yemtsov2019). Knowledge with regards to political effects of a UBI is fortified, as my findings suggest that cash transfers which are universally and unconditionally available are able bring about lasting turnout change.

The short-term turnout estimates in response to the UBI introduction are large, yet when comparing them to existing research, some differences become apparent. The turnout increase is measured with an exogenous variation in income that with its introduction jumps to USD 1,000 as the UBI is disbursed for the first time, just one month prior to the November election of 1982. Akee et al. (Reference Akee, Copeland, Costello, Holbein and Simeonova2018) model one such variation, finding turnout to increase by around 9 percentage points among children of recipients of an exogenously varied income based on a casino opening. That income was however not available on an individual or universal basis and the income stemmed from the operations of a private business. Other research attempts to analyze income effects on turnout, among them studies analyzing Latin American CCT programs estimated to result in a turnout effect of around 2–7 percentage points (Baez et al., Reference Baez, Camacho, Conover and Zárate2012; De La O, Reference De La O2013) and the expansion of Medicaid resulting in an increase of 3–7 pp (Clinton and Sances, Reference Clinton and Sances2018; Baicker and Finkelstein, Reference Baicker and Finkelstein2019). Turning toward more general wealth effects shows that a higher minimum wage increased turnout by 2.5 pp among affected employees (Markovich and White, Reference Markovich and White2022) and homeownership has a 4–8 pp turnout effect (Hall and Yoder, Reference Hall and Yoder2022). Every two-point change in the unemployment rate was matched by a 1 pp change in turnout (Burden and Wichowsky, Reference Burden and Wichowsky2014; Cebula, Reference Cebula2017), yet with a turnout-depressing effect reported to be at around 7 pp among the unemployed (Aytaç et al., Reference Aytaç, Rau and Stokes2020). Neither of those studies was however able to isolate a direct or indirect income effect from other factors that could affect turnout.

The Alaskan UBI was first disbursed prior to a midterm election, an election type generally reflecting lower turnout rates compared to presidential elections, which has been explained with an increased media salience and more sources for mobilization in the run up to presidential elections (Tolbert et al., Reference Tolbert, Grummel and Smith2001; Solt, Reference Solt2010). Midterms may therefore have higher turnout potential and we might expect larger turnout effects as a response to the Alaskan UBI especially during those elections. Diminishing marginal returns of voting would imply that as presidential elections already provide “a large incentive to vote”, the effects of a turnout-enhancing policy are weaker during presidential election years (Smith, Reference Smith2001). Finally, the turnout effect was estimated to be at 12.8 pp when using the Elections Project data which adjust for citizenship and people having lost their right to vote (Holbrook and Heidbreder, Reference Holbrook and Heidbreder2010) (see Appendix).

6.1 Turnout-enhancing mechanisms

There are several possible explanations for the turnout increase in Alaska, the first one being that it may be traced back to the payment amount as such: the resource effect. To navigate the voting process successfully, individuals require resources such as income, but also education, time, health, and civic kills. Income further allows for the development of skills that are necessary for political participation (Schlozman et al., Reference Schlozman, Brady and Verba2018). Contextualizing the dividend payments shows that the Alaskan UBI is not able to account for a considerable share of the per capita personal income (see Appendix) in my analysis period (Zelleke, Reference Zelleke, Widerquist and Howard2016, 150)—the resources necessary for political participation can probably not be expected to increase vastly for the average population.

In order to corroborate the theoretical mechanisms proposed, I analyze heterogeneous effects by socioeconomic background (Table 4). As the CPS voter supplement only publishes family income starting in 1982, I proxy for household resources by using individual educational attainment,Footnote ¹ two variables that are highly correlated (Wolfinger and Rosenstone, Reference Wolfinger and Rosenstone1980).

Table 4. Heterogeneous effects—educational attainment

Notes: Regression coefficients shown with robust standard errors in parentheses that are clustered by state. Coefficients for the fixed effects are not reported. ***p < 0.01, **p < 0.05, *p < 0.1.

The negative estimates of the interaction of dividend and educational attainment mean that the turnout increasing effect of the dividend decreases as educational attainment increases. Figure 2 displays the marginal effect of the dividend on turnout at different levels of educational attainment as well as the distribution of educational attainment, showing that the UBI has a significant turnout-enhancing effect when educational attainment is low.

Figure 2. Marginal effect of dividend on turnout in Alaska with 95 percent CI.

This indicates that the Alaskan citizenry did not react uniformly to the introduction of the payments. More specifically, the strong reaction of individuals with a lower socioeconomic status indicates a resource effect being at play for those very individuals. A heterogeneous treatment effect analysis by race that is included in the Appendix increases the robustness of this finding, showing that especially American Indians/Aleuts reflected large marginal turnout effects, a population subgroup that fares worse than the average population with regard to income but also factors such as financial literacy, education, and health (Paisano, Reference Paisano1995; Denny et al., Reference Denny, Holtzman, Turner Goins and Croft2005; Murphy et al., Reference Murphy, Gourd and Begay2014). This is in line with research showing that albeit more affluent citizens turn out to vote at higher rates, an income effect voting is higher for poorer citizens (Rosenstone, Reference Rosenstone1982; Ojeda, Reference Ojeda2018, 814). The results stipulate that the dividend—which can account for 20 percent of the poverty threshold in Alaska in some years—can substantially increase citizens' resources and therefore their ability to deal with the costs connected to the voting process. As the results indicate a resource effect having materialized especially among Alaskans with a lower socioeconomic background, the average increase in turnout may thereby be explained by a second mechanism that needs to be considered: interpretive effects.

The Alaskan UBI is paid unconditionally to every permanent resident, and citizens do not approach the government as petitioners applying for public transfer payments with the need to prove their indigence, minimizing the risk of a stigmatizing certain population groups (Bidadanure, Reference Bidadanure2019, 485). Alaskans approach the administration as universally legitimate recipients of the dividend, some even call them shareholders in “Alaska Inc.” (O'Brien and Olson, Reference O'Brien and Olson1990). The program's guaranteed nature may lend legitimacy to the citizens “claim[s] to benefits and their worth to society” (Campbell, Reference Campbell2012, 342). Among the electorate, attitudes toward the administration may shift as the government policy of the dividend is not means-tested—possibly causing reciprocal behavior among citizens as they turn out to vote (Campbell, Reference Campbell2012; Galiani et al., Reference Galiani, Ibarraran and McEwan2016; Markovich and White, Reference Markovich and White2022).

A third potential policy feedback mechanism is the mobilization effect, meaning that a UBI could incentivize program recipients that do not want the program benefits revoked to increase political participation (Campbell, Reference Campbell2012). The dividend program was however only introduced after the Alaskan Permanent Fund was founded based on a constitutional amendment following a referendum. This indicates that the dividend payments never became a politically contested subject as the program was designed to last. The main goal of the sovereign wealth fund is to preserve a share of Alaska's “nonrenewable oil and mineral wealth for future generations” by utilizing a diversified investment strategy to generate additional income, and to place a share of Alaska's oil revenues “beyond the reach of day-to-day government spending” (O'Brien and Olson, Reference O'Brien and Olson1990, 139–141). I therefore argue that mobilization is an unlikely mechanism driving turnout in Alaska.

Finally, the turnout increase could have also been based on new voters. Franklin (Reference Franklin2004) states that turnout behavior remains constant during adult life, a turnout increase should accordingly be grounded on the voting habits of new voters that either only reached voting age or moved to the state. When looking at the marginal effect of the dividend on turnout at different age groups (Table A12, Appendix), it can be seen that the dividend has a stronger turnout effect among older citizens, this holds both in the short- and medium-run. The share of young voters aged 18–24 can also be seen to have decreased over time in Alaska. The development of the population growth rate in Alaska further indicates that while it increased from immediately after the introduction of the UBI to 5.4 percent it decreased again after 1984. This does not appear to be an unusual development as Alaska's population growth has historically exhibited fluctuations, growth in the early 1980s was due in particular to a booming oil industry (Williams, Reference Williams1985; SEER Program, National Cancer Institute, 2018). Both this population growth and a general population volatility over time are also mirrored in other oil-producing US states including Texas and Oklahoma. I have plotted these growth rates against each other in the Appendix. In order to account for this population growth, I control for the population size in all of my models and am able to estimate a turnout effect persisting until 2000. I further also use the synthetic control method by Abadie et al. (Reference Abadie, Diamond and Hainmueller2010, Reference Abadie, Diamond and Hainmueller2015) that relaxes the parallel trends assumption with the above-mentioned US states included in the baseline group.