Labor Informality and the Vote in Latin America: A Meta-analysis

Abstract

Conventional wisdom among scholars of Latin American politics holds that informal workers are less participatory and less left-leaning than formal workers. Relevant empirical findings, however, are mixed and in need of synthesis. This article provides that synthesis by conducting meta-analyses on the universe of previous quantitative studies of informality and the vote. It finds that informal workers are indeed less likely to vote than formal workers, but the effect of informality is small—just four to seven percentage points. It further finds that informal workers are more likely to vote for the left, not the right, but here the effect size is even smaller. Meta-regression analyses reveal that in countries where organized professional activity among informal workers is high, gaps in turnout between the two sectors are minimal. The article concludes that the conventional wisdom over-states the individual-level political consequences of labor informality in Latin America.

An important conventional wisdom among scholars of Latin American politics holds that the region’s informal workers are less participatory and less left-leaning than its formal workers (Nelson 1970; Roberts 2002). Some experts claim that these two behavioral patterns, along with Latin America’s relatively high rates of informality, explain why the region has historically had weaker leftist parties and less progressive welfare states than Western Europe (Huber and Stephens 2012; Schneider and Soskice 2009). Other scholars contest this conventional wisdom, pointing to instances of mobilization by informal workers or porous borders between the two occupational sectors (Blofield 2012; Maloney 1999). There is now a critical mass of survey-based studies on informality and the vote, but their results are mixed, and no useful summary of their overall findings exists.

This study systematically describes the nature of past statistical findings on informality and voting behavior. We conduct meta-analyses on the universe of previous quantitative studies (plus some new estimates of our own) to estimate the average effects of informality on voter turnout and vote choice in Latin America. We find that informal workers are indeed less likely to vote than formal workers, but the effect size is small. Informality lowers turnout by about four to seven percentage points. We also find that informal workers are more likely to vote for the left, not the right, but here the effect is even smaller. Thus our primary conclusion is that the two sectors do not participate or (especially) vote all that differently from one another.

Moreover, our meta-analyses reveal which countries and years feature larger differences in voting behavior between the two sectors. Countries in which a relatively large share of informal workers belong to professional organizations have the smallest differences in turnout rates between formal and informal workers. We also find that compulsory voting rules widen differences in turnout between the two sectors and that the small gap in turnout between the two has been growing over the past two decades, counter to some arguments. Overall, this systematic review and quantitative summary of the literature’s findings yields important conclusions about informality and politics in Latin America.

INFORMALITY AND VOTING BEHAVIOR IN LATIN AMERICA

A longstanding assertion among scholars of Latin American elections contends that the voting behavior of labor outsiders—a term we use synonymously with “informal workers”—differs from the voting behavior of labor insiders (also “formal workers”) on the practice’s two central dimensions: turnout and vote choice.¹ On the dimension of turnout, informal workers are, according to this conventional wisdom, less likely to vote than formal workers (Portes et al. 1989). As for vote choice, informal workers are allegedly more likely to favor right-leaning parties than are formal workers (Cameron 1991). Both arguments are grounded in a dualistic notion of labor markets, a notion that sees the formal and informal sectors as segmented from one another (Lindbeck and Snower 1988). The two arguments are deeply intertwined, so we discuss their theoretical underpinnings together.

The Conventional Theoretical Wisdom

The conventional scholarly wisdom starts from the perception that informal workers are less likely than formal workers to be organized and unionized. The vast majority of informal workers either own or work in small, extralegal firms. As a result, they are diffusely spread across markets and spaces, often wanting their firms and jobs to remain undetected by the state (Nelson 1970). This reality of social atomization and marginalization makes collective action difficult (Roberts 2002). Partly for this reason, informal workers have rarely been incorporated into corporatist and postcorporatist structures of interest representation. In contrast, formal workers have been and still are more likely to be organized labor union members.

Union membership politicizes formal workers, and according to the conventional wisdom, it politicizes formal workers to lean leftward (Kurtz 2004). Unions persuade and mobilize their members to support labor-based and thus typically left-of-center parties that pursue the policy goals of the organized working class. These efforts, according to the conventional wisdom, imbue most union members and (thus) many formal workers with a left-leaning partisan bias that motivates them not only to turn out on election day but also to cast ballots for programmatic labor-based parties. Lacking in organizational linkages to leftist parties, by contrast, labor outsiders are less attuned to politics, voting less frequently and generally “pacified” in the political sphere (Milner and Rudra 2015, 669; Altamirano 2019). When they do bother to vote, the absence of organizational linkages to labor-based parties allegedly makes informal workers more likely to choose right-of-center parties. The fact that informal workers are, on average, poorer than formal workers compounds this tendency, since rightist or nonprogrammatic clientelistic parties can easily afford to buy the votes of poor informal workers during election season (Levitsky 2003).

A related set of arguments asserts that informal workers, because of their class and occupational positions, hold genuine commitments to the capitalist ideals of the partisan right (Portes 1971) or, at the very least, resentment toward a state that is decidedly absent from their working lives (Ronconi 2019). Many labor outsiders are small business owners and thus part of a petty bourgeoisie that may recoil against heavy state regulation and taxation (de Soto 1989). Similarly, many informal workers seem to eschew certain civic duties, including political participation, because the state and politicians offer them so little (Ronconi and Zarazaga 2015).

In addition, informal workers (usually by definition) do not benefit from state-sponsored social security regimes that provide benefits, such as retirement pensions and unemployment insurance, only to workers who contribute through their state-monitored salaries. These social security regimes, which are vehemently defended by leftist parties, disproportionately benefit formal workers in Latin America’s truncated welfare states (Haggard and Kaufman 2008; Holland 2018). Overall, the conventional wisdom sees compelling reasons for informal workers to be less statist in their policy attitudes and less civic minded, potentially inclining many of them to support right-of-center parties or to sit elections out entirely.

This conventional wisdom about individual-level differences in voting behavior between labor outsiders and insiders is so strong that many scholars use it to explain fundamental macrolevel features of Latin America’s economies and political systems. Some authors suggest, for example, that the presence of large informal sectors throughout Latin America explains why the region has often lacked class voting cleavages (i.e., cleavages wherein the lower class votes left and the middle and upper classes vote right) (Dix 1989; Roberts 2002). Others make the even more ambitious claim that high rates of informality reduce the popularity and size of leftist parties, which, in turn, would explain why the region has smaller, less progressive welfare states and thus higher inequality than Western Europe (Huber and Stephens 2012; Schneider and Soskice 2009).

Countering the Conventional Wisdom

Numerous scholars have provided logic and findings to critique the conventional wisdom, leaving the theoretical foundations of research on informality and voting behavior in an ambivalent state. Reasons to question the gap in political participation and voter turnout come from a variety of sources. As early as the 1970s, Janice Perlman’s Myth of Marginality (1976) challenged the notion that residents of Brazil’s urban informal settlements were detached from politics and organizational life. More recently, the rise of new social movements, poor people’s movements, and informal sector unions has demonstrated the impressive mobilizational energies of informal workers (Cross 1998; Hummel 2017). Similarly, informal workers often organize to influence whether and how legal prohibitions against their activities are enforced (Holland 2017).

Some research even finds informal workers to be more involved than formal workers in religious and neighborhood organizations. These organizations sometimes assume a political hue, but in any case, point to the vibrancy of social capital in the informal sector (Temkin Yedwab and Penagos Va´squez 2013; Smith 2019). On the other side of the coin, the rate of unionization among labor insiders has steadily declined in the region in recent decades (Roberts 2014, 100), so unions today are highly compromised instruments of political mobilization in the formal sector.

Scholars have also provided multiple reasons to doubt the conventional wisdom that informal workers lean rightward. First, many of Latin America’s labor-based parties, such as the Peronists in Argentina and the PRI in Mexico, have not necessarily or persistently been on the political left. Unions affiliated with these parties, in other words, sometimes mobilize their formal sector members on behalf of centrist candidates and moderate policy goals (Murillo 2001). Similarly, the region features some conservative labor unions, such as Brazil’s Confederação Geral dos Trabalhadores and Força Sindical, that regularly endorse nonleftist parties.

Second, in recent decades, parties and politicians from across the political spectrum have implemented noncontributory social programs that disproportionately benefit labor outsiders (Brooks 2015; Fairfield and Garay 2017), so the leftward or rightward lean of informal workers in a given country can be idio-syncratic to which party introduced such programs. Since 1990, governments throughout Latin America have expanded their welfare states by introducing noncontributory programs that overwhelmingly benefit those who do not make payroll tax contributions to (and thus are not eligible for) the formal sector social security regimes. These include health insurance and pension plans for informal workers, as well as conditional cash transfers (CCT). Labor outsiders seem to reward the incumbents who implement these programs, which has reshaped political cleavages (Dion 2010; Hunter 2010). In countries where a leftist or centrist party introduced a new benefit to informal workers, for example, newer voting patterns and coalitions may show informal workers leaning leftward and no longer (if ever) rightward (Zucco 2008).

Third, the nature of Latin American labor markets may blur the divide in partisan preferences between those who are informal and those who are formal at a particular point in time. The boundaries between the informal and formal sectors are more porous than dualistic portrayals recognize. Latin American workers move between formal and informal jobs over the course of their working lives, often multiple times (Maloney 1999; Perry et al. 2007). In addition, incomes for many households originate in both sectors. Discordant marriages—one formal spouse and one informal spouse—are common in the region, sometimes composing 30 percent of all marriages (Galiani and Wein-schelbaum 2012).

This partial integration of the two sectors can diminish their political differences in multiple ways. Informal workers can realistically aspire to formal sector jobs, so they may support the parties that wish to bolster or expand formal sector social security programs. Meanwhile, labor insiders may support social programs for outsiders because they themselves may, at some future point, transition to informality and be eligible for them (Carnes and Mares 2016). Either way, “the frequent movement of workers between the formal and informal sectors attenuates the insider/outsider cleavage” (Schneider 2013, 108). Additionally, discordant marriages can homogenize partisan preferences across an informal-formal pairing through shared economic policy interests and interpersonal influence (Baker et al. 2020a). In short, political preferences may not differ greatly between formal and informal workers because the two sectors are somewhat integrated.

Wrestling with the Evidence

Recently, scholars have begun to weigh the empirical accuracy of the conventional wisdom and its critics, but their collective findings have also been ambivalent. Various studies now exist, with political scientists taking advantage of significant improvements to measuring informality in surveys. But results are mixed and at times mutually contradictory. For instance, Ronconi and Zarazaga (2015) find that informal workers are less likely to vote, while Thornton (2000) finds that Mexico’s informal workers are more likely to vote than their formal compatriots. Similarly, Singer (2016) alone finds that the effect of informality on voting for the left can be positive or negative, depending on workers’ attitudes toward the macroeconomy.

Given these mixed results, how does one arrive at summary conclusions about the relationship between labor informality and voting behavior? Collecting new survey data is always valuable, but this (expensive) option fails to use the evidence that has come before and yields no conclusions about central tendencies in the overall body of evidence. Moreover, new data would shed light on just a narrow period; it is possible, if not likely, that the nature of the relationship between informal workers and their governments has evolved with shifting partisan appeals and economic realities (Centeno and Portes 2006; Cross and Peña 2006). In addition, measurement of the concept of informality has improved in recent years. And the relationship between informality and voting behavior surely varies across countries in systematic ways. Given these issues and needs, we propose a systematic means of summarizing and analyzing the evidence that scholars have compiled thus far: the meta-analysis.

META-ANALYSIS

A meta-analysis of the literature brings clarity to what previous studies have found. Meta-analysis is, in essence, a systematic and quantitative review of the literature, one that yields a numerical summary of the literature’s overall findings. We conduct meta-analyses to estimate the average effect of informality on turnout and on vote choice in all relevant and extant empirical studies. Further-more, we conduct meta-regression analyses to discern what cross-national and temporal factors shape these effect sizes.

Data and Methods

The raw materials of a meta-analysis are the coefficients and standard errors on the independent variable of interest (informality) in all regressions from existing studies with the appropriate dependent variable (turnout or vote choice). We first establish inclusion criteria to define the universe of relevant studies and regressions. To be included in our pool, a study must report a statistical model that uses a measure of Labor informality as an independent variable and either

Turnout (or Abstention) or directioned Vote choice as a dependent variable.² We use only models that operate at the individual (i.e., survey respondent) level of analysis. Furthermore, we use studies that sample from a single Latin American country, several Latin American countries, or several less developed countries (LDCs) that include multiple Latin American countries. To maximize information and avoid publication bias, we include both published and unpublished studies.³

Table 1 lists the studies and gives some notes about each one. Our search yielded 41 independent estimates of the effect of informality on turnout from 6 different studies.⁴ For vote choice, we found only 3 studies with relevant quantitative estimates, which is shocking, considering how widespread is the perception that informal workers are right-leaning. We urge scholars to become more attentive to informality when specifying statistical models of voting behavior, but for now we partially address this shortfall by supplementing these 3 studies with our own original model estimates, generated from recently released data, the 2018–19 Latin American Public Opinion Project (LAPOP) surveys. These model results are reported in the online appendix (part A). Altogether, previous studies plus these new estimates yielded 69 estimates of informality’s effect on vote choice.⁵ The original studies do not all use identical measures and functional forms—a challenge common to all meta-analyses—so the estimates they report are not always directly comparable to one another. Some models of turnout use logit, for example, while others use probit. We address this by converting the original estimates to adjusted estimates that share a common metric within each meta-analysis. The need to have common metrics in any single meta-analysis necessitates that we conduct three meta-analyses: one for turnout and two for vote choice.

Table 1. Sources and Notes for Meta-analyses

Note: Productive definition defines informality as self-employment. Benefits definition defines informality as having a job that makes no payroll tax contributions to social security.

For the meta-analysis of turnout, we convert (when necessary) all estimates to logit coefficients.⁶ Each adjusted estimate, in other words, is the estimated effect of informality (relative to formality) on the logit of a respondent voting:

$$\log \left( {{{\Pr (Y) = 1} \over {\Pr (Y) = 0}}} \right)$$

where 1 is a valid vote and 0 is an abstention.

For vote choice, we must set up two different meta-analyses. Many of the original estimates are based on binary or nominal measures of vote choice (N=35; e.g., Singer 2016), so for one meta-analysis of vote choice we analyze logit coefficients:

$$\log \left( {{{\Pr (Y) = 1} \over {\Pr (Y) = 0}}} \right)$$

where 1 is a vote for the leading leftist candidate and 0 is a vote for the leading nonleftist candidate. Other estimates from the literature are based on models that use a continuous measure of vote choice (N = 34). This measure gives each respondent the ideology score (ranging from far right [1] to far left [20]) of the candidate for whom the respondent voted (e.g., Castan˜eda and Doyle 2019).⁷ We thus run a second meta-analysis of vote choice based on these ordinary least squares (OLS) models. For these instances, each adjusted estimate (i.e., the original OLS coefficient) is the average difference between informal and formal workers in the ideological scores of their vote choices. For most of the survey samples, we were able to obtain adjusted estimates using both the logit specification and the OLS specification, so there is heavy overlap between the two vote-choice meta-analyses.⁸

Differences other than how the dependent variable is specified, such as how informality is measured, also exist across the studies, but the effects (if any) of these other sources of variation are easily calculated and adjusted for with meta-regression analyses (MRAs). We conduct MRAs in the next section.

Once the adjusted estimates are in place, we can calculate the primary statistic of interest for each of our three meta-analyses, the average effect (⁻ε) of informality:

$$\overline {\rm{\varepsilon }} = {{\sum {w_i}{{\rm{\varepsilon }}_i}} \over {\sum {w_i}}}$$

where ε _i is the vector of adjusted estimates (each denoted by i) and w _i is a vector of weights. There are different philosophies on what weights to use, but the general logic is that estimates derived from large samples weigh more heavily (because of their greater precision) than do estimates derived from small samples.⁹ Each of the three average effects is calculated with a seemingly small sample size (N = 41, 35, 34), but recall that each observation is actually an estimate based on hundreds or even thousands of respondents. In other words, the three average effects are grounded in ample information, and their standard errors (formula not shown) incorporate the overall amount of uncertainty based on the underlying sample sizes.

The Average Effects of Informality on Voting Behavior

Figure 1 summarizes with a funnel graph the results of the meta-analysis for turnout. A funnel graph plots each adjusted estimate (x axis) with its precision (y axis). Precision is the inverse of the estimate’s standard error. To reiterate, each estimate is a logit coefficient. (Below we describe these effects in more intuitive terms than as logits.) For ease of exposition, we will also refer to each estimate or effect in figure 1 as a “turnout gap,” since each one represents the gap in turnout between informal and formal workers.

Figure 1. Meta-analysis of the Impact of Informality on Voter Turnout in Latin America: Funnel Graph of Logit Coefficients and Their Precisions

Note: The vertical solid line is the average effect (⁻ε), and the 95 percent confidence interval is shaded in gray. Points with black labels were statistically significant in the original study, while those with gray labels were not.

Each point is labeled with its corresponding country (using two-letter ISO codes) and year. For instance, UY6/8 is an estimate derived from a single model conducted on surveys of Uruguayans in both 2006 and 2008, and LA18 is an estimate derived from a model that pools 2018 surveys from all available Latin American countries. A black label indicates that the estimate was statistically significant in the original study (at the 95 percent confidence level), and a gray label indicates that it was not.

The solid vertical line is the average effect (ε), and the shaded region captures its 95 percent confidence interval. More precise estimates (higher on the y axis) receive greater weighting in calculating this weighted average and thus tend to cluster near it. Less precise estimates (lower on the y axis) tend to have greater variance and do not weigh as heavily in calculating the overall average. Consequently, the point cloud assumes the shape of an inverted funnel centered on the average effect.

Our meta-analysis finds that, on average, Latin America’s informal workers are indeed less likely than its formal workers to vote. For these 41 estimates, scholars reported a negative coefficient 39 times. The average effect of informality is –.256, and the upper end of the 95 percent confidence interval [–.321, –.190] comes nowhere close to zero.¹⁰ Yet the meta-analysis also reveals that the substantive effect of informality on turnout is small. To convert the average effect (currently in logits) to a more intuitive quantity—namely, differences between the two groups in the probability of voting—we need only assume a level of overall turnout.¹¹ At 65 percent turnout (toward the low end in Latin America), formal workers are about 7 percentage points more likely to vote than their informal counterparts. At 85 percent turnout (toward the high end), the gap is just 4 points. In sum, informal workers are less participatory than formal workers, but the turnout gap is in the single digits.

Figure 2 reports the funnel graphs for the two meta-analyses of vote choice, and they both tell the same story: informal workers are more left-leaning at the ballot box than formal workers, but the overall difference between the two groups, while statistically significant, is small in size. The upper panel plots logit coefficients, while the bottom panel plots OLS coefficients.¹²

Figure 2. Two Meta-analyses of the Impact of Informality on Voting for the Left: Funnel Graphs of Coefficients and Their Precisions

Note: The vertical solid line is the average effect (⁻ε), and the 95 percent confidence interval is shaded in gray. Points with black labels were statistically significant in the original study, while those with gray labels were not.

The average effect in the meta-analysis of logit coefficients (panel A) is .121 with a confidence interval that excludes zero [.018, .223].¹³ This amounts to, at most, a three-percentage-point gap between informal and formal workers in their support for the left, and the figure reveals, at most, five statistically significant coefficients across these model estimates. Moreover, only one of these significant coefficients (Nicaragua 2006) supports the conventional wisdom that informal voters are more right-leaning. The average effect in the meta-analysis of OLS coefficients (panel B) is .120 [.004, .236].¹⁴

On the 1–20-point ideological scale, in other words, informal workers vote one-eighth of a point more leftward than do formal workers, a gap we consider to be quite small. To put this in perspective, the gap between Mexico’s leftist PRD and its rightist PAN in this ideological space is 13 points. To add a final substantive interpretation, if we convert this average effect to a partial correlation, it is barely nonzero at +.013. In short, once inside the polling booth, informal and formal workers are not that different from one another.

To sum up our findings thus far, we find that informality has a statistically significant and negative average effect on voter turnout in Latin America, and it has a statistically significant and positive average effect on voting for the left. But these average effects are substantively small, large enough to sway only the closest of elections. The average effects are particularly small when it comes to the matter of left versus right voting.

META-REGRESSION ANALYSIS

Nailing down these central tendencies is an important contribution to the literature on Latin American politics, especially because assertions about these effect sizes underlie important scholarly arguments about the region. Yet averages fail to describe variation. The horizontal spread of the points in the funnel graphs captures the variation in effect sizes. In figure 1, for example, the negative effect of informality on turnout in some countries and years is large in magnitude; in other countries and years, it is very close to zero. What explains these differences across time and space in turnout gaps? To answer this question, we turn to meta-regression analysis (MRA).

MRA seeks to explain the variation in effect sizes by regressing the effects (ε _i ) on some of their characteristics. These characteristics can be substantive features of the contexts in which the data were collected, such as political traits of the country and year. They can also be the decisions made by researchers when estimating the effects, such as how to measure key concepts or specify models. For our independent variables in the MRAs, in other words, we seek characteristics that could explain variation in the effects of informality, rather than characteristics that could explain variation in the original dependent variables (turnout and vote choice).

Because we are regressing effects obtained from individual-level regressions on country-year traits, our MRAs are akin to the exercise of specifying cross-level interactions in a hierarchical model of cross-national survey attitudes. At the same time, an MRA carries important advantages over a hierarchical modeling approach because it incorporates information from multiple studies, whereas a hierarchical model is constrained to a single survey dataset that has equivalent measures across all countries. For this reason, an MRA not only yields a larger level 2 (i.e., country-year level) sample size, but it can also uncover the impact on effect sizes of researchers’ decisions, such as modeling, publication, and measurement choices.

We report a set of MRAs designed to explain differences across time and space in turnout gaps. We focus on turnout because the variation to be explained in the meta-analyses of vote choice appears to be more noise than signal. Nearly 90 percent of the estimates from vote choice models were statistically indistinguishable from zero (figure 2), whereas nearly 50 percent of the effect sizes from models of turnout were statistically significant (figure 1). One still can (and we did) conduct MRAs when most effects are statistical zeroes, but we describe the central (and mostly null) conclusions in a footnote and report the coefficients in the online appendix (part B).

Independent Variables

We consider two broad types of independent variables: those that capture substantive features and those that capture methodological or other decisions made by researchers.

The amount of collective action among labor outsiders lies at the heart of the debate over informality status and political participation. Advocates of the conventional wisdom claim that the informal sector’s alleged atomization and lack of organization underlie its relative political apathy (Oxhorn 1998), but detractors point out countless instances of collective mobilization by informal workers (Bhowmik 2012). As it turns out, rates of organizational activity within the informal sector vary cross-nationally. Recent research shows, for example, that a third of Bolivia’s informal workers are members of a professional association or labor union, in contrast to only a tenth of Argentina’s informal workers (Hummel 2021).

We hypothesize that Organizational density within the informal sector, measured as the proportion of informal workers who attended a meeting of a professional association or labor union in the preceding 12 months (Hummel 2021), is correlated positively with the effect of informality on voter turnout. In other words, countries with high organizational density within their informal sectors will have small turnout gaps, while countries with low organizational density will have large turnout gaps (Boulding and Holzner 2021, chap. 4). We also consider the mirror of this claim: the extent of organization among formal workers. Data to capture this concept are limited, however, so we report our more tentative findings on this front in footnotes.

Party systems may also influence the size of the turnout gap between informal and formal workers. Many party systems in Latin America are elite-based, built by politicians as oligarchic cartels or electoral vehicles with little pressure or input from below. By contrast, mass-based party systems were forged by bottom-up processes in response to mobilization by voters, and typically by workers themselves (Roberts 2014). Today, scholars argue that mass-based party systems have deeper organizational linkages in society and thus mobilize the popular classes more than elite-based party systems. Using Boulding and Holzner’s 2021 coding of Mass party systems (= 1) versus elitist party systems (= 0), we hypothesize that this variable correlates positively with our effect sizes.

We also consider the effect of Compulsory voting on the gap in turnout between informal and formal workers. Previous research shows that by requiring everyone to vote, compulsory voting increases turnout in the Latin American countries where it exists and is binding (Carreras and Castañeda-Angarita 2014). Compulsory voting may, as a result, close turnout gaps between informal and formal workers, effectively inducing to vote those otherwise more likely to abstain (Carlin and Love 2015). We therefore hypothesize that compulsory voting will correlate positively with the effect of informality on turnout. That said, we must also point out that voters can cast a blank or spoiled ballot under compulsory voting rules, which by standard definitions means they have abstained. Compulsory voting rules thus may be irrelevant to differences in rates of turnout between the two sectors. We use the four-point scale developed by the Varieties of Democracy Institute. This scale moves from 0 (no compulsory voting) to 3 (compulsory voting with considerable and enforced sanctions) (Coppedge et al. 2020).

We are also interested in the trend in the effect of informality through time. Opinions vary on this question, and until now, no hard data existed to adjudicate. Some scholars argue that rates of political engagement by informal workers have increased in recent years because of the opening of new democratic spaces and because politicians increasingly see labor outsiders as an electoral and legislative resource (Blofield 2012; Garay 2016). Yet the argument that informal workers do organize and participate has been around for decades (Perlman 1976). Hummel (2021), for example, documents the presence of street vendor associations in Bolivia in the 1950s and even the 1880s. We estimate any linear trend in the size of turnout gaps between the two sectors by including the Year of data collection.

Our MRAs also include a set of variables to gauge factors that scholars of Latin American political behavior often overlook: researchers’ decisions. Methodological factors, such as model specification and measurement error, can have major consequences on estimated effect sizes, and publication bias can also shape collective findings. First, we include Controls for income, a binary measure of whether the original model includes a measure of income or wealth. Some scholars control for respondent’s income when estimating the effect of informality on voting behavior. This practice, we hypothesize, introduces posttreatment bias because income is endogenous to informality status. Posttreatment bias would attenuate the estimated effect of informality on voting behavior and boost the chance of committing a type II error. Since the average effect of informality on turnout is negative, we hypothesize that controlling for income correlates positively with effect sizes.

Second, measurement may also matter for effect sizes, since poor measurement of independent variables biases their coefficients toward zero. Labor informality may be acutely susceptible to measurement error because the concept has only recently been included in politically oriented surveys, and its measurement remains inconsistent across studies (Perry et al. 2007). Scholars increasingly agree that the best measure uses the benefits definition (Gasparini and Tornarolli 2009): workers who contribute to the social security system via payroll taxes are formal while others are not. Fortunately, almost all the studies we compiled use this definition, but a few use a version of the so-called productive definition, which amounts to treating self-employment as a proxy for informality (Altamirano and Wibbels 2012). This productive definition is “theoretically weak” because it tallies informal wage earners (i.e., people who work for someone else but without legal recognition) as formal workers (Gasparini and Tornarolli 2009, 20). We created a dummy variable, Benefits definition, that indicates whether the original study uses the benefits definition or not (see table 1 for scores).¹⁵ We expect it to correlate negatively with effect sizes; usage of the more valid benefits measure should yield effect sizes that are further from zero.

In addition, publication bias can yield the opposite effect of measurement error. The literature’s average effect could be biased away from zero (type I error) because of missing statistically insignificant or “wrongly” signed results (i.e., not supportive of some conventional wisdom) in published work. Many of the estimates we compiled have not been published, although our overall average effects seem to show no contamination from the scholarly aversion to reporting null results (see note 10). Still, the published estimates may be larger in magnitude than the unpublished estimates, so we include a dummy variable that indicates if the estimate is Unpublished or not.¹⁶ If publication bias is present, then this variable will correlate positively with effect sizes, meaning unpublished estimates are closer to zero.

Meta-regression Analysis Results

Results from two meta-regression analyses are shown in table 2. Model 1 focuses on the substantive hypotheses while controlling for one researcher decision variable (Controls for income). This limits the sample to single-country estimates because the first three of the four substantive variables are not defined for multicountry estimates. As for the other two researcher decision variables, they do not vary in this sample (i.e., all single-country estimates are published and use the benefits definition). Model 2 contains the full set of 41 estimates, but in doing so cannot include the first three substantive variables. The sample sizes for these regressions are seemingly small, but recall that each observation is derived from hundreds to thousands of observations.¹⁷

Table 2. Meta-regression Analyses: Explaining the Effects of Informality on Turnout

*p <.05

Note: Dependent variable is effect sizes (in logits) of informality on turnout. Entries are weighted least squares regression coefficients with standard errors in parentheses. Cases are weighted by the inverse of the original estimate’s standard error.

According to model 1, country-level factors clearly explain some of the variation in the sizes of turnout gaps between informal and formal workers in Latin America. Most important, high organizational density within the informal sector keeps gaps small to nonexistent. This variable has a positive and statistically significant sign in model 1 and is associated with obvious variation between cases. Notice in figure 1, for example, that Bolivia consistently (in 2008 and 2018) has no meaningful gap in turnout between insiders and outsiders. Bolivia also has the highest rate of organizational density in the region. By comparison, Uruguay repeatedly has some of the largest turnout gaps between the two sectors, while the country’s level of organizational density (at 13 percent) is among the region’s lowest. The difference between the two countries in organizational density explains about half their difference in turnout gaps. (According to results not shown, organizational density among formal workers does not matter, although recall that data for this concept are patchier.¹⁸) In sum, professional organizations for informal workers can mobilize them to vote at the same rates as formal workers.

By contrast, political party systems do not seem to influence turnout gaps; mass-based party systems do not feature smaller turnout gaps than elitist party systems (model 1). Bottom-up, mass-based parties were and are often rooted in organized labor (Collier and Collier 1991), so the finding that mass-based party systems do not mobilize informal workers is not entirely surprising.

Perhaps the most surprising finding is that compulsory voting increases turnout gaps. Model 1 predicts that countries with enforced compulsory voting laws, such as Brazil, Ecuador, Peru, and Uruguay, average larger turnout gaps than countries without compulsory voting, and indeed it is clear from figure 1 that some of the largest (negative) effect sizes exist in these four countries. Why? A convincing answer to this question will require further research, but we propose one premised on the fact that citizens can abstain (by casting a blank or spoiled ballot) while still complying with compulsory voting laws. The statistically significant negative coefficient on compulsory voting in table 2 is compatible with a pattern in which the marginal formal worker—meaning a formal worker who would not show up at the polls in the absence of compulsory voting rules—is less likely to cast a blank or spoiled ballot than the marginal informal worker. While compulsory voting brings extra formal and informal workers to the polls, in other words, these extra formal workers are more likely to cast a directioned vote than the extra informal workers. The underlying reasons for this behavioral difference between the two sectors are the same as those that underlie the overall average turnout gap, but this difference becomes sharper when many formal and informal workers go to the polls simply to avoid mandatory voting sanctions. This is speculation, so we encourage future inquiry on the matter.

Another finding from model 1 is that turnout gaps have been growing larger, and model 2 confirms this finding. The statistically significant negative coefficient on the year variable means that turnout gaps tended to be smaller in the early years of data collection (the mid-1990s) than in recent years (2018). Indeed, according to predicted values from model 2, the average effect size was effectively zero in 1995 and grew to almost –0.40 by 2018, roughly equal to a nine-point gap in turnout between the two sectors (in a country with 65 percent turnout). This is well above the overall average effect size of –0.26. In short, the notion that recent decades have witnessed a steady upswing in political participation by labor outsiders (relative to labor insiders) is false.

Furthermore, decisions by researchers have not affected statistical findings on turnout gaps. To be sure, Controls for income is positively signed and statistically significant in model 1, hinting that a control for income introduces posttreatment bias by lowering the magnitude of coefficients on informality. In model 1, however, this dummy variable equals 1 for just a single outlying case, the large and positive MX97 outlier (Thornton 2000; see figure 1). The positive coefficient is thus driven by the particularities of that case and data. In fact, we reran the 17 single-country models from Baker et al. (2020b) while including a measure of wealth, and the coefficients on informality attenuated by a median of just 7 percent.¹⁹ Moreover, Controls for income, along with the two other researcher decision variables, have statistically insignificant coefficients in model 2.

Part B of the online appendix reports MRAs that explore which country-year variables correlate with differences between formal and informal workers in their propensity to vote for the left. We find no statistically significant results. Besides examining the impact of organizational density and mass-based party systems, we considered two new variables. One is a measure of each country’s employment protection legislation (EPL), based on the hypothesis that countries with high barriers to dismissal of formal sector workers (e.g., Argentina) have more segmentation and therefore deeper political cleavages between the two sectors. We also considered a measure of whether a left-leaning party had previously introduced a CCT in the country.²⁰ Coefficients on all these substantive variables, as well as the researcher decision variables, were statistically insignificant, even when using restrictive models with few independent variables (part C). Again, we are not surprised, because so few of the original estimates were themselves statistically significant (figure 2).

Overall, our most important takeaway from the MRAs is that organizational energy within the informal sector can bring the sector’s turnout to parity with the turnout of the formal sector. Turnout gaps are small to nonexistent in countries where professional associations have a relatively high density among labor outsiders. Also, we find that voter turnout among informal workers (relative to turnout among formal workers) has been declining through time.

DISCUSSION AND CONCLUSIONS

Based on a systematic review and summary of previous quantitative findings, we find that informal workers in Latin America are somewhat less likely to vote than their formal sector counterparts, but the difference between the two in turnout rates is just 4 to 7 percentage points. In addition, we find that those informal workers who do turn out are more likely to vote for the left than are formal workers, but here the difference is even smaller at less than 3 percentage points. Despite these small average effects, we do find that political differences between the two sectors vary by country and by timing, with the two most important findings on this front as follows. First, countries in which informal workers are relatively well organized into professionally minded groups have almost no gap in participation between the two sectors. Second, gaps in turnout between informal and formal workers have been growing in recent decades.

These results lead us to a nuanced conclusion about informal workers and voting behavior. On the one hand, conventional wisdom is correct in viewing informal workers as less participatory. On the other hand, this judgment should not be exaggerated, since the small overall turnout gap means that the informal sector is far from pacified and atomized into political apathy. Indeed, in some countries, organizational vigor among informal workers brings their participation rates to parity with those of formal workers. Moreover, the conventional wisdom is wrong in seeing informal workers as more right-leaning. If anything, the opposite is true, but only barely so.

The tendency for scholars to overblow political differences between formal and informal workers lies, we suspect, in the stubbornness of the dualist model. The fact that most Latin American workers are neither entirely formal nor entirely informal over their working lives has yet to take hold in most scholarship on Latin American politics. The political implications of having so many households with both informal and formal workers under the same roof are also poorly understood. In the end, scholars must explore the possibility that policy preferences are more convergent between the two sectors than previously understood (Kanbur et al. 2019).

Perhaps most important, what do our findings say about arguments that attribute some of the region’s macrolevel features and even maladies—high inequality and truncated welfare states, the absence of class voting, the relative weakness of left-labor parties, rampant clientelism—to high rates of informality? We cannot entirely dismiss these arguments because our findings pertain to individual-level behavior. Where informal sectors are large, parties and politicians may still behave differently than they behave where formal sectors dominate, even if voting patterns in the two sectors are similar in the former. The authors of these arguments have been some-what vague about whether informality works through micro- or macrolevel mechanisms. Regardless, we hope our findings inform these and other debates.