In the United States, the affluent are much more likely to participate in politics than those who are not (Leighley and Nagler Reference Leighley and Nagler2013; Ojeda Reference Ojeda2018; Verba, Schlozmann, and Brady Reference Verba, Schlozman and Brady1995; Schlozman, Brady, Verba, Reference Schlozman, Brady and Verba2018; Smets and Van Ham Reference Smets and Van Ham2013; Verba and Nie Reference Verba and Nie1987). Although the presence of a participatory gap between high- and low-income individuals is well established, scholars know little about whether income is the driving force behind these gaps, or instead, income gradients in voting reflect some other unobserved social or contextual factors. As a result, we do not know whether increasing household income would actually improve overall levels of civic engagement and narrow gaps in voting behavior. This question is inherently difficult to answer, as incomes are (typically) not exogenously distributed. Moreover, there has been little research into how income interacts with the life course and whether children’s propensities to vote are affected by the family environment, by family income, or both (for a notable exception see Ojeda Reference Ojeda2018).

In this letter, we explore whether income has an effect on civic participation across two generations. To do so, we investigate the effects of a quasi-experimental unconditional cash transfer. We examine a unique longitudinal dataset from the Great Smoky Mountains Study (GSMS): a study of children in rural western North Carolina, which began in 1993 and consisted of both Native American and non–Native American families in the area. Halfway through the initial 8-year survey time frame, a casino opened on the Eastern Cherokee reservation located in this region. On its opening, a portion of the profits were distributed to all adult tribal members independent of employment status, income, or other characteristics relevant to political engagement. This exogenous unconditional income transfer, along with the unique longitudinal nature of the data, allows us to use various panel techniques to explore the effects of positive changes in household incomes on the political participation of parents and children from the same household.

We first test whether a positive change in income has an effect on parents’ voting. As we describe in the following paragraphs, the Resource Model of Voting (RMV)—and other similar voting models—suggests that positive income shocks should have a noticeable effect on voting (Verba, Schlozman, and Brady Reference Verba, Schlozman and Brady1995). However, we find a precisely estimated null effect across parents. Second, we test whether there is any impact of the additional household income on the household children. We find that average annual unconditional transfers of approximately $4,700 (in 2000 dollars) increase the initially poorer group’s voter turnout by about 8–20 percentage points, depending on the age of the recipients and the type of measures of voting considered. The effect closes the participatory gap between individuals of high- and low-income backgrounds in this rising generation.

Our work makes several important contributions. First, our study helps answer the question whether income contributes to underlying levels of voter participation. Our quasi-experiment, although (perhaps) not allowing us to draw conclusions across all space and time, does provide us with a unique circumstance in which we can make an unusually credible (local) causal estimate of the effect of income on voting. In doing so, it adds nuance to the foundational resource model of voting developed by Verba, Schlozman, and Brady (Reference Verba, Schlozman and Brady1995). Our findings are consistent with a more complex voter turnout model that allows for childhood resources to affect future civic participation in a manner consistent with the human capital formation literature from economics (e.g., Almond and Currie Reference Almond and Currie2011; Becker and Tomes Reference Becker and Tomes1986; Currie Reference Currie2009).

Second, given the intergenerational element to our analyses, the results also contribute to our understanding of political socialization. In seeking to understand why some people develop into active citizens, whereas others do not, social scientists have tended to focus almost exclusively on adult experiences—when citizens are just coming of age or are already eligible to vote—rather than on those that occur in childhood or early adolescence. Political socialization research once focused on childhood in hopes of discovering the roots of political participation (e.g., Dawson and Prewitt Reference Dawson and Prewitt1968; Greenstein Reference Greenstein1965; Langton Reference Langton1969; Miller and Saunders Reference Miller and Saunders2016; Niemi and Hepburn Reference Niemi and Hepburn1995; Sapiro Reference Sapiro2004; Searing, Schwartz, and Lind Reference Searing, Schwartz and Lind1973). Although various models have postulated that resources allocated earlier in the life course may matter more than those delivered later, little to no contemporary research has explored this possibility. The research that has focused on this topic has generally struggled to elicit causal estimates. Our work provides compelling evidence that early life experiences—in this case, the receipt of additional income—have a greater effect on participation than the same experiences among members of the same family later in life. This implies that voting propensities are not a heritable trait that is transmitted from one generation to the next, but, instead, can be influenced by malleable aspects of the household environment during the early years.

Finally, our results have implications for public policy. Discussions about the merits of various income distribution schemes are at the heart of a multitude of policy reforms. Our results suggest that income augmentation programs might be an effective means of closing the income gap in voting, as long as they are targeted early in the life course. This finding is of critical importance in a context of soaring levels of income inequality in the United States and abroad (Piketty Reference Piketty2014).

METHODS

To test the effect of exogenously increased household incomes on parents and children, we use several estimation strategies.^{Footnote 2} We first show in Appendix Figure A.6 that there is a sizeable and statistically significant change in unearned household incomes because of the casino transfer payments that accrue to households with enrolled tribal citizens after the start of casino operations. We base our analysis on this exogenous increase in household incomes for Native American households.

Given that there are strong reasons to expect heterogeneities in the estimated effects by income levels in our analyses, we test for these formally across both the adults and the children in our sample. To do so, we take two approaches: first, we stratify our regressions at the median household income level measured at baseline (before the casino opened) and second, we interact our treatment variable with the baseline household income variable.

Methods: Parents

We conduct a standard difference-in-difference analysis for the parents because they are eligible to vote before the income intervention (in 1992 and 1994), where we are able to use the same household’s voting record overtime. The comparison in the case of parents is across the same household before and after the transfers began and across AIs (eligible) and non-AI households (not eligible). Because the treatment is on the household level, we run the analysis at that level as well.^{Footnote 3}

Equation (1) formalizes the model that we run for parents, with γ ₁ being the coefficient of interest. In this case, the treatment is an indicator for household exposure to the casino transfer in the time period after the start of the casino intervention. We include a control for Native American household status and a binary variable for whether the observation is drawn from the time period after the intervention. The variable NativeAmerican _i × AfterCasino _i is simply the interaction between those two binary indicator variables. We also include an individual household fixed-effect α _i because we observe the same household over multiple periods in our strongly-balanced panel; note that this implies that we will not be able to separately identify the level effect of Native American household in the regression equation as it will be captured in the individual household fixed-effect. Finally, we include election year fixed-effects to account for potentially different average voter turnout for presidential versus congressional-only elections θ _t and an error term. Our estimation equation is given as follows:

(1)$$\scale 85%\eqalign{ & {Y_{it}} = {\alpha _i} + {\gamma _1}\;NativeAmerica{n_i} \,\times\, AfterCasin{o_t} + {\delta _1}\;NativeAmerica{n_i} + {\lambda _1}\;AfterCasin{o_t} + {\theta _t} + {\varepsilon _{it}}. \cr}$$

Identification in equation (1) is based on the assumption that the parallel trend assumption holds. We show pretreatment trends based on the parents’ voting records across age cohorts by race and year in Figure 1. Voting data for the years 1992 and 1994 serve as the pretreatment observations. We find no substantial differences between the two groups. This standard check provides supportive evidence of the internal validity of our quasi-experimental design—among other things, it provides evidence that before the income transfers began, the treated and untreated groups voted at indistinguishable rates.

FIGURE 1.

Effect of Casino Transfers on Parental Voting by Initial Household Income Status around the Start of Casino Operations

Notes: Figure displays coefficients from event analysis model for parents’ voter turnout in the 1992-2014elections. The estimates are split by median family income levels at baseline. Standard errors are clustered at the individual level. Top panel N = 15,984 (1,332 GSMS individuals); Middle panel N =8,172 (681 GSMS individuals); Bottom panel N = 7,812 (651 GSMS individuals).

Methods: Children

For children, there is no “before” period as they are not eligible to vote as minors. Thus, we conduct a difference-in-difference analysis where we compare the voting records across cohorts of children who are exposed to the income intervention for different lengths of time. Younger children live in households with exogenously increased unearned incomes for a longer time period than their older counterparts because the income intervention started at a single point in time in this community. This approach leverages two differences: the first difference is between Native American (eligible for the transfer) and non–Native American children (not eligible) and the second difference is across age cohorts of AI children who were exposed to the income transfers at different points in the life course. Our hypothesis, consistent with Ojeda’s model (Reference Ojeda2018) and with the human capital model from economics, is that income transfers will have larger effects on the younger children in the survey.

Equation (2) formalizes the difference-in-difference model that we estimate using data on the children in the GSMS sample:

(2)$$\eqalignb{ {Y_i} = \alpha + {\beta _1}\;AgeCohort{1_i} + {\beta _2}\;AgeCohort{2_i} + {\delta _1}\;NativeAmerica{n_i} + {\gamma _1}\;AgeCohort{1_i} \times \,NativeAmerica{n_i} + {\gamma _2}\;AgeCohort2 \times\, NativeAmerica{n_i} + X\prime \theta + {\varepsilon _i}. \cr}$$

Following the previous practice (Holbein Reference Holbein2017; Sondheimer and Green Reference Sondheimer and Green2010), in equation (2), we specify Y _i in three ways—as a binary variable indicating whether an individual has ever voted, as a continuous variable measuring the proportion of elections that a person voted in, and as a measure for voting in specific elections. In equation (2), YoungestCohort _i is an indicator variable for the child belonging to the youngest cohort (age 9 years at the start of the survey and age 13 years at the first cash transfer). MiddleCohort _i is an indicator that the child belongs to the second youngest cohort (age 11 years at the beginning of the survey and age 15 years at the first cash transfer). The omitted group is the third (oldest) cohort, so all coefficients are interpretable as differences from that cohort. The variable NativeAmerican _i is a dummy equal to one for Native Americans and zero otherwise. The vector X is a set of baseline covariates that include the parents’ voter turnout rate before the casino opened, child age, number of children younger than six years in the household, and child gender. The coefficients of interest are γ ₁ and γ ₂, which capture the difference-in-difference coefficient estimates.

RESULTS

Parents’ Voting Outcomes

We first estimate the effects of the casino transfer on parents’ voting rates. Figure 1 provides the event analysis for parents for the pooled sample and separately by initial household income above and below the median.^{Footnote 4} Note that the first two years (1992 and 1994) serve as a test of the parallel trends assumption. We plot the coefficient on the AI household dummy for all elections starting in 1992, which is before the income transfers began in 1996 (Appendix Table A.9 provides the regression results used to construct this figure.) As the figure shows, we find that the increase in household income has no economically substantive or statistically significant effect on parents’ voting probabilities. This null effect is precise. Using equivalence testing (Hartman and Hidalgo Reference Hartman and Hidalgo2018), our 95% confidence intervals allow us to confidently rule out effects larger than 2.5 percentage points. Regardless of how we specify the model, unconditional cash transfers have little to no effect on parents’ voting.^{Footnote 5}

The figure displays coefficients from the event analysis model for parents’ voter turnout in the 1992–2014 elections. The estimates are split by median family income levels at baseline. Standard errors are clustered at the individual level.

These precisely estimated null effects are vitally important in their own right. They suggest that resource-based models of voter turnout may be oversimplified. Resources (in this case, income) when distributed in adulthood do not uniformly increase adult turnout.

Children’s Voting Outcomes

We next turn to estimating the effect of casino transfers on children’s voting patterns. Figure 2 provides a graphical depiction of the differential effect of income received earlier and later during the life course. Here, we combine cohorts 1 and 2 for readability, so the plotted coefficients represent the average effect of the transfers on the voting propensities of these two cohorts relative to the oldest cohort.^{Footnote 6} In the top panel, we plot the estimated coefficients for observations below the baseline median household income. The effect of the casino transfers on the younger cohorts from poorer households is positive, substantively large, and (in virtually all elections) statistically significant. The effects range from 10.3 percentage points (p < 0.05) in the 2010 election to 20.2 percentage points (p < 0.01) in the 2012 election. In the bottom panel, we provide the same analysis for individuals from above the baseline median household income level. The effect among this group of children is smaller and not statistically significant.^{Footnote 7}

FIGURE 2.

Effect of Casino Transfers on Child Voting by Initial Household Income Status

Note: The Figure displays coefficients from event analysis model for children’s voter turnout in the 2004-2014 elections. The estimates are split by median family income levels at baseline. To make visualization easier, cohorts 1 and 2 are collapsed together and compared to cohort 3. Standard errors are clustered at the individual level. Top panel N = 4,557 (651 GSMS individuals); Bottom panel N = 4,767 (681 GSMS individuals).

The figure displays coefficients from the event analysis model for children’s voter turnout in the 2004–14 elections. The estimates are split by median family income levels at baseline. To make visualization easier, cohorts 1 and 2 are collapsed together and compared with cohort 3. Standard errors are clustered at the individual level.

These results are consistent with the predictions based on a human capital perspective of voting and with the implications of Ojeda’s two-gap conceptual model (Reference Ojeda2018). They suggest that income transfers in relatively early life narrow participatory gaps considerably. For young people who are in their formative years and who have yet to finish high school, household incomes matter a great deal in determining whether they become active voters or fail to do so—with these effects concentrated among those who have the lowest initial household incomes. This suggests that voting rates are not predetermined because of household characteristics or rigidly transferred from one generation to the next. In addition to the other positive effects of elevating families out of poverty, this movement out of poverty also has large effects on children’s levels of civic participation.

We provide suggestive tests for potential mechanisms in the Appendix. Our results suggest that this effect is not likely to be driven by the transmission of voting from parents to children—one of the most commonly cited channels in the literature (e.g., Dawson and Prewitt Reference Dawson and Prewitt1968). Instead, we provide suggestive evidence that the results may, indeed, be driven by increases in educational attainment and social skill acquisition over the life course. These channels may help explain why the timing of the income intervention mattered so much for children from the youngest cohort at receipt.