Economic change and political shifts: the role of unions in Brazil’s deindustrialization era (2002-2022)

ABSTRACT

Studies increasingly demonstrate that deindustrialization impacts electoral performance, affecting left - and right - wing parties differently. However, there is limited analysis of the mechanisms linking deindustrialization to voting patterns. This paper tests the hypothesis that union weakening is a key mediating factor in this relationship. Analyzing Brazilian presidential elections (2002-2022), the study finds that municipalities experiencing both deindustrialization and union decline exhibit stronger effects of industrial loss on voting. The results suggest that the erosion of organized labor, driven by deindustrialization, is crucial to understanding current electoral dynamics, highlighting the importance of this mechanism in shaping political outcomes.

KEYWORDS:
Brazilian economy; elections; deindustrialization; union weakening.

INTRODUCTION

How do structural changes in the economic-productive configuration of a nation affect the decisions of elites and voters? Deindustrialization has been identified as a key explanatory variable for the decision of voters and candidates (McQuarrie, 2017; Santos et al., 2024a; Santos et al., 2024b). Our aim in this article is to contribute to this research agenda by testing whether the decline of trade unions can be thought as a mechanism of the relationship between deindustrialization and voting in Brazil’s presidential elections from 2002 to 2022.

The central hypothesis of the research is that the effects of deindustrialization on the vote, when combined with union weakening, are higher. This argument is consistent with other studies that emphasize the relevance of trade unions to political processes in democracies. Trade unions are institutions historically linked to left-wing parties. Nevertheless, in this literature, few studies have attempted to understand the role of union weakening, which several countries are experiencing, as a mediating variable in the causal chain linking deindustrialization and electoral performance.

In order to test the hypothesis, we adopted a four-step empirical strategy: (1) we check the relationship between deindustrialization and electoral performance; (2) we test whether deindustrialization has contributed to the weakening of unions; (3) we investigate whether the weakening of unions influences voting; and (4) we analyze whether the effects of deindustrialization differ between municipalities where unions lost strength vis-à-vis where they did not. In steps (1), (3) and (4), we estimate linear regression models for panel data, with fixed effects at the municipal level; in step (2), in the model with the binary dependent variable, we used logistic regression. The units of analysis are Brazilian municipalities.

The results confirm the hypothesis, showing a substantial negative effect of deindustrialization on the PT’s performance and an increase in the vote for opponents on the right of the spectrum, and that this effect is mediated by variation in the strength of the unions in the municipalities examined. Furthermore, the analyses reveal that the effects are heterogeneous with regard to the levels of deindustrialization and union weakening. A severe deindustrialization shock has twice the impact on electoral performance compared to a moderate one. In addition, the difference in effects between models with and without union weakening, which attests to the mediating role of unions, is strongly driven by cases in which the loss of union strength was high.

In the next section, we advance the argument regarding the relevance of union weakening as a mechanism that contributes to understanding how deindustrialization affects the vote. In the third section, we outline our empirical strategy, detailing the analytical steps and the models and data used. In the fourth section, we present the results of the analysis. In the fifth section, we perform robustness tests. In the sixth section, we present our concluding remarks.

THE ELECTORAL CONSEQUENCES OF THE LOSS OF UNION STRENGTH

In recent years, some studies have found that deindustrialization is a relevant variable in explaining the loss of votes for left-wing parties and candidates and the strengthening of the right. McQuarrie (2017) argues that deindustrialization is central to understanding Donald Trump’s electoral victory in the United States in 2016. According to the author, there was a kind of “revolt” by residents of “Rust Belt” areas against the Democratic Party. Santos et al. (2024a) found that deindustrialization, modelled as an exogenous shock, negatively affected the electoral performance of the Workers’ Party (PT) in presidential elections in Brazil between 2006 and 2018. On the supply side, Santos et al. (2024b) show that the loss of strength of the industrial sector in Brazil is an important phenomenon in explaining the growth of businesspersons’ candidacies in the country. In this article, we suggest that the weakening of trade unions, because of deindustrialization, is one of the mechanisms through which the latter affects voting behavior and thus contributes to explaining Brazil’s electoral results.

As the literature indicates, trade unions are relevant institutions for electoral outcomes because they engage members in collective and political organization (Bryson et al., 2014; Leighley and Nagler, 2007; Kerrissey and Schofer, 2013; Radcliff and Davis, 2000); influence members’ preferences through communication work (Freeman, 2003; Kerrissey and Schofer, 2013; Kim and Margalit, 2016; Zullo, 2008); provide political and financial resources to candidates and parties (Fouirnaies, 2021; Freeman, 2003; Kerrissey and Schofer, 2013); and they impact the political behavior of non-members by indirectly influencing the interpersonal network of their members or by running broader mobilization campaigns and disseminating information to society, which can mobilize other people with similar interests and/or who have positions favorable to the unions (Freeman, 2003; Kim, 2021; Leighley and Nagler, 2007). It is important to note that the influence of trade unions is not neutral, given the historical relationship of these organizations with left-wing parties. Thus, the growth of unions is associated with the strengthening of left-wing candidates and parties (Freeman, 2003; Radcliff and Davis, 2000; Zullo, 2008).

It is reasonable to expect, therefore, that the reverse effect will also occur: the weakening of unions reduces the voting potential of left-wing candidates and parties, favoring the right. McQuarrie (2017) shows that the dismantling of unions helps explain the loss of strength of the Democratic Party in the Rust Belt. Along these lines, considering the Brazilian case, the study closest to ours is that carried out by Ogeda et al. (2021), which considers union weakening as an institutional channel through which trade shocks produce political effects in Brazil. However, our work has some fundamental differences from that of these authors. Firstly, we model union weakening as a mediating variable for deindustrialization. Deindustrialization is measured as an “electoral industrial balance”, which, following Santos et al. (2024), captures the variation in the share of industry from one electoral cycle to the next. Below, in the section describing the variables considered in our models, we present details of the construction of this indicator. Secondly, our units of analysis are municipalities, while Ogeda et al. (2021) use micro-regions. These differences have important consequences in terms of the identification strategy, which will be detailed below.

In the case of Brazil, it is important to note that the Workers’ Party (PT), which has reached the second round in every presidential election since the country’s re-democratization in 1988, is a union-based organization. Luiz Inácio Lula da Silva, the PT’s main leader, who is in his third term as president of Brazil, was president of the Metalworkers’ Union of São Bernardo and Diadema, two cities in the state of São Paulo (Kekc, 2010; Lucca, 2011). This connection between the party and the unions has electoral effects. Data from the “A Cara da Democracia” survey shows, as can be seen in Table 1, that in the last two elections, the PT had more votes among unionized voters than Jair Bolsonaro, its right-wing opponent. Figure 1 shows, however, that trade unions are losing strength in the country.

Table 1
Left and right votes among union members and non-union members (2018-2022)

Unions	Fernando Haddad	Jair Bolsonaro
	2018
Did not participate	40%	60%
Participated	54%	46%
Unions	Lula	Jair Bolsonaro
	2022
Did not participate	56%	44%
Participated	75%	25%

Figure 1
Number of unionized and General Industry workers by total formal Jobs in Brazil

Given what has been discussed above, it is reasonable to assume that deindustrialization and union weakening are part of the same causal chain leading to electoral performance. Our hypothesis, therefore, is that the effects of deindustrialization on voting, when accompanied by union weakening, are more strongly negative for the left and more strongly positive for the right.

EMPIRICAL STRATEGY

Steps and models

In order to test the relevant hypothesis, this paper follows a four-step empirical strategy, with Brazilian municipalities as units of analysis: (1) we check the relationship between deindustrialization and electoral performance; (2) we test whether deindustrialization contributed to the weakening of unions; (3) we investigate whether the weakening of unions influences voting; and (4) we analyze whether the effect of deindustrialization differs between municipalities where unions were undermined and those where it did not, to corroborate the hypothesis that declining union strength may be considered as an mechanism.

The identification strategy used for steps (1), (3) and (4) were fixed effects models. The choice of this type of model is due to the fact that, as we noted in the previous section, despite some similarities, our research design has notable differences from the one adopted by Ogeda et al. (2021). The tariff reduction they used as a treatment variable was a policy adopted centrally by the federal government as a single, permanent shock between 1990 and 1995. It is therefore possible to identify more clearly before and after the application of the treatment, in order to capture the temporal differences. In the case of this study, both deindustrialization and union weakening are not single, permanent shocks that affected all the units of analysis (municipalities) at once. Because of this, it is more difficult to establish a theory, as Ogeda et al. (2021) has done on tariffs, able to explain the temporal variation of the effects.

The use of fixed effects models with “pooled” data, in this case, is a more efficient way of exploring the temporal variation of our sample (2002-2022), taking into account that this approach to estimating causal effects only considers changes in the same unit over time. In this way, the counterfactual outcomes (controls) that are compared with the treatment observations in a given time period are estimated using outcomes observed in time periods of the same unit (Imai and Kim, 2019; Mummolo and Peterson, 2018). Furthermore, in this process, with the inclusion of fixed effects, the intercept varies between the units and the models control for unobserved variables that are fixed in time, mitigating omitted variable biases (Békés and Kézdi, 2021; Cunningham, 2021).

In order to do this, we estimate linear regression models according to equation (1) below:

(1) $$Yit=\beta xit+\beta Wit+ci+\epsilon it$$

where Yit is the outcome variable (electoral performance or union weakening), xit is the treatment variable (deindustrialization or union weakening), Wit is a set of control variables, ci represents municipality fixed effects.

In step (2), as one of the models uses a binary version of the outcome variable (union weakening), we estimate a logistic regression model in the shape of equation (2) below:

(2) $$Yit=\mathrm{logit}\left(p_{it}\right)=\ln \frac{p_{it}}{1-p_{it}}=\beta 0+\beta xit+\beta Wit+ϵ\mathrm{it}$$

where is the probability of the event occurring (union weakening), ln is the natural logarithm function, is the intercept, is the treatment variable (deindustrialization), is a set of control variables and is the error term. For this model, we interpret the results as odds ratios, indicating the chances of union weakening given a one-unit change in , ceteris paribus.

VARIABLES AND DATA

Electoral performance

In this article, the main outcome variables are the electoral performance of the left and the right in the presidential elections between 2006 and 2022. As a proxy for the performance of the left, we used the percentage of votes for the Workers’ Party (PT), which reached the second round in every presidential election in the country since 1989. For the performance of the right, between 2006 and 2014 we used the vote of the Party of the Brazilian Social Democracy (PSDB). In the 2018 and 2022 elections, the variable for the performance of the right is operationalized by the percentage of votes for Bolsonaro, who ran in 2018 for the Social Liberal Party (PSL) and in 2022 for the Liberal Party (PL). The electoral data was collected on the website of the Superior Electoral Court (TSE).

Deindustrialization

The variable that measures deindustrialization was based on data from the Brazilian Institute of Geography and Statistics (IBGE) on the economic activities that make up the municipalities’ Gross Domestic Product (GDP). First, following Santos et al. (2024a), we calculated the industrial share for the municipalities based on the percentage of industrial gross value added (GVA) in relation to GDP. We then used a delta variable to measure the variation in the industrial share of the municipalities’ GDP over time, which is the difference in the industrial share from one electoral year to the next. For 2006, for example, the electoral industrial balance is the industrial share of 2006 minus that of 2002.

In the models we built, in addition to using the electoral industrial balance as a continuous variable, we used two other versions, binary and categorical. In the binary variable, category 1 was assigned to municipalities that had industrial balances lower than zero (deindustrialized); category 0 includes municipalities that had industrial balances greater than or equal to 0 (industrialized or remained at the same level). For the categorical version, we classified the cases in the sample into five categories: high deindustrialization, moderate deindustrialization, stability, moderate industrialization and high industrialization. These categories were operationalized based on the distribution of the electoral industrial balance variable in quartiles. The “stability” category included cases where the balance was very close to zero, between -0.5 and 0.5. As can be seen in Figures 2 and 3, while electoral performance on the left increases in municipalities with lower levels of deindustrialization, on the right the result is the opposite.

Figure 2
Average percentage of PT votes in presidential elections by level of industrialization/deindustrialization of municipalities

Figure 3
Average percentage of right-wing votes in presidential elections by level of industrialization/deindustrialization of municipalities

Figure 4
Time series of the variables “number of unions”

UNION STRENGTH

The most traditional way of assessing union strength is through union density (Crouch, 2017). However, there is no data available on union density at the municipal level in Brazil. We therefore used data for the number of unions and the number of people employed in unions by municipality. The data was extracted from the Annual Social Information Report (RAIS) for the period 2002 to 2021, using the “basedosdados” package in the R software (Cavalcante et al., 2023). To operationalize the variables for 2022, we used the information for 2021.

The Figures 4 and 5 show the time series of the two variables. The number of unions began to rise sharply in 2005 and then fell from 2018 onwards. The number of people employed in unions also grew strongly until the early 2010s but began a deeper downward trend in 2012. This data suggests that many unions went through weakening processes, losing employees, before they closed. Based on this picture, the number of people employed in unions is used in this article as a proxy to measure union strength.

Figure 5
Time series of the variables “number of people employed in unions”

Figure 6
Expected effects of deindustrialization on the electoral performance of the left and the right

To capture the variation in union strength between electoral periods, we used the same procedure employed to transform the industrial share of GDP into a delta. As the size of the municipality influences the number of unions and people employed in unions, we used the variation in the proportion of people employed by unions as an indicator of union strength. In the same way, we created a binary variable and a categorical variable. For the binary, category 1 represents the municipalities that saw a decrease in the number of people employed in unions and category 0 classifies those in which this did not occur. For the categorical version, we classified the cases in the sample into five categories: high weakening, moderate weakening, stability, moderate strengthening and high strengthening. The aim is to additionally produce more detailed analyses for step 4.

Controls

A relatively small set of control variables is used in the analyses to avoid the risk of including “bad controls” that could produce unintended discrepancies between the regression coefficients and the expected effects (Angrist and Pischke, 2009, 2014; Cinelli et al., 2022). We used GDP, population size, the percentage of evangelical inhabitants, the share of the population with complete primary education and the percentage of self-declared white people.

GDP and population size are controlled to ensure that the analyses are not biased by the wealth and size of the municipalities. The percentage of evangelicals is used considering the rise of the “evangelical vote” (Nicolau, 2020). “Race” is an important variable for electoral behavior in Brazil (Rodrigues and Pereira, 2022; Zanotti et al., 2023). For this reason, we included the share of the self-declared white population as a proxy. The share of the population with completed primary education was inserted as a proxy for educational level, and we know that the PT has been losing support among more educated voters since 2006 (Nicolau, 2020; Singer, 2012). Table 2 provides descriptive statistics for the variables used in the analysis.

Table 2
Association between electoral performance and deindustrialization (continuous, binary and categorical treatment variable)

PANEL A1
Dep. var.: PT votes	Continuous treatment variable (1)	Binary treatment variable (2)	High. deindus. (3)	Moderate. deindus. (1)	Moderate. Indust. (2)	High. Indust. (3)
Electoral industrial balance	0.12*** (0.021)	-2.16*** (0.161)	-2.18*** (0.271)	-1.09*** (0.285)	1.29*** (0.285)	0.71** (0.262)
Obs.	14605	14605	14605	14605	14605	14605
Adj. R2	0.777	0.779	0.780	0.780	0.780	0.780
Within R2	0.116	0.123	0.127	0.127	0.127	0.127
PANEL A2
Dep. var.: Right votes	Continuous treatment variable (1)	Binary treatment variable (2)	High. deindus. (3)	Moderate. deindus. (1)	Moderate. Indust. (2)	High. Indust. (3)
Electoral industrial balance	-0.05*** (0.012)	0.31* (0.127)	-0.07 (0.262)	-0.29 (0.217)	-0.30 (0.240)	-0.85*** (0.220)
Obs.	14605	14605	14605	14605	14605	14605
Adj. R2	0.823	0.822	0.823	0.823	0.823	0.823
Within R2	0.187	0.186	0.187	0.187	0.187	0.187

• Source: Elaborated by the authors. Notes: For the model with the categorical variable, “Stability” was used as the reference category; p < 0.1,
• ^*
  p < 0.05,
• ^**
  p < 0.01,
• ^***
  p < 0.001.

Table 3
Association between union weakening and deindustrialization (continuous and binary treatment variable)

PANEL B1
Dep. var.: Union Weakening (binary)	Continuous treatment variable (1)	Binary treatment variable (2)
Electoral industrial balance	-0.011** (0.0036)	0.084* (0.0415)
Obs.	9836	9836
AIC	13183.2	13188.9
BIC	13233.5	13239.3
RMSE	0.49	0.49

• Source: Elaborated by the authors. Note: p < 0.1,
• ^*
  p < 0.05,
• ^**
  p < 0.01,
• ^***
  p < 0.001.

ANALYSIS

Deindustrialization and electoral performance

Our first analytical step is to estimate the direct association between deindustrialization and the electoral performance of the left and the right in presidential elections in Brazil. Figure 6 illustrates the expected impact of deindustrialization on the two ideological groups. The minus (-) and plus (+) signs show the expected effects: that deindustrialization affects the left negatively and the right positively.

In the second column, it can be seen that an increase in the electoral industrial balance is associated with a higher electoral performance by the PT in presidential elections. The coefficient of 0.12, significant at the 0.1 percent level, indicates that an increase of one percentage point in the industrial balance is associated with an increase, on average, of 0.12 percentage points in the PT vote. A decrease in the balance therefore has the opposite effect. To the right, the coefficient is -0.05, showing a negative association with the industrial strengthening of municipalities. The coefficients in column 3 show more directly the average effect of deindustrialization on the vote. Deindustrialization is associated with a loss of 2.16 percentage points for the PT and a gain of 0.31 for the right. In competitive elections, this effect can be decisive. In addition, the results for the model in which the treatment is the categorical variable show that the levels of deindustrialization matter. For the left, when compared to the reference category, stability, high deindustrialization has twice the impact of moderate deindustrialization. To the right, only the coefficient for high industrialization is significant.

The results suggest that, in fact, the change in Brazil’s productive structure in recent decades, in the sense of the country’s decline in industrial strength, has produced substantial electoral effects, favorable to the right.

Deindustrialization and union weakening

Our theory suggests that there is a positive association between the increase in the industrial sector and the growth in union strength; consequently, the opposite is true of deindustrialization, given that the industrial sector is the one with the highest rate of unionization of workers. Figure 7 highlights this expected relationship between deindustrialization and union strength.

Figure 7
Expected effects of deindustrialization on union strength

In the second column, the negative coefficient indicates that the chances of union weakening decrease as the electoral industrial balance increases. For every one-percentage point increase in the industrial balance, the expectation is that, on average, the chances of union weakening will decrease by 1.1 percent. It is therefore possible to interpret that deindustrialization has the opposite effect, increasing the chances of union weakening, in line with predictions. In the third column, the positive coefficient when we use a binary treatment variable reinforces that the chances of union weakening increase when there is deindustrialization. In this case, deindustrialization is associated with 8.4 percent higher chances of union weakening.

The third analytical step in this article is based on the premise that the union weakening results in a loss of votes for parties historically linked to unions, such as the PT. The union weakening affects the connection between workers and left-wing parties, thus favoring the right. Figure 8 shows the expected effects of union weakening on the dynamics of the vote: negative for the left and positive for the right.

Figure 8
Expected effects of union weakening on the electoral performance of the left and the right Union weakening and the dynamics of voting

In Table 4, the results show, in general, that union weakening is associated with a loss of electoral performance for the PT and an improvement for the right. In Panel C1, the coefficient of 0.020 indicates that for each percentage point increase in union strength, an average increase of 0.020 percentage points in the PT vote is expected. To the right, the expected result is a decrease of 0.01 percentage points with a one-unit increase in union strength. In the model with the binary treatment variable, the expected result is that the weakening of the unions causes the PT to lose, on average, -1.02 percentage points in its electoral performance. For the right, on the other hand, the expected result is an improvement in electoral performance of 0.82 percentage points.

Table 4
Association between electoral performance and union weakening (continuous and binary treatment variable)

PANEL C1
Dep. var.: PT votes	Continuous treatment variable (1)	Binary treatment variable (2)
Union weakening	0.020*** (0.006)	-1.02*** (0.208)
Obs.	9812	9812
Adj. R2	0.747	0.747
Within R2	0.151	0.152
	PANEL C2
Dep. var.: Right votes	Continuous variable (1)	Binary variable (2)
Electoral Industrial balance	-0.01* (0.004)	0.82*** (0.162)
Obs.	9812	9812
Adj. R2	0.810	0.811
Within R2	0.228	0.230

• Source: Elaborated by de authors. Note: p < 0.1,
• ^*
  p < 0.05,
• ^**
  p < 0.01,
• ^***
  p < 0.001.

Union weakening as a mechanism

We have seen that deindustrialization impacts voting dynamics and union strength, and that the union weakening affects the electoral performance of the left and the right. Now, for our purposes, we need to estimate whether and to what extent deindustrialization influences presidential elections in Brazil through union weakening. To do this, we divided our sample into seven different data sets. First, with the binary variable, we created two groups: one contains only the cases in which unions weakened and the other those in which they did not. Secondly, with the categorical variable, we created five groups, dividing the cases according to the levels of variation in union strength: high weakening, moderate weakening, stability, moderate strengthening and high strengthening. The aim is to investigate whether the effects of deindustrialization on electoral dynamics are different in municipalities with these profiles. Figure 9 illustrates the expected relationships between deindustrialization and electoral performance with union weakening as the mechanism.

Figure 9
Expected relationships between deindustrialization and electoral performance with union weakening as the mechanism

Table 5 presents the results of the analyses. In Panel D1, models (1) and (2) show that the effect of deindustrialization on the PT’s electoral performance is substantially greater among municipalities that have experienced union weakening. Among these, the expected average effect is -3.54 percentage points on the party’s vote; among those that have not suffered union weakening, the coefficient is -2.69. To the right, in Panel D2, the same is true: while in municipalities where unions have weakened the average expected effect is 1.04 percentage points, in those where this has not happened the coefficient is 0.41, significant only at the 10 percent level. These results provide evidence that confirms our hypothesis that union weakening plays an important mediating role in the relationship between deindustrialization and voting in Brazil’s presidential elections.

Table 5
Association between electoral performance and deindustrialization with union weakening as mechanism (binary treatment variable)

PANEL D1
Dep. var.: PT votes	Weakened unions (1)	Not weakened unions (2)	High union weakening (3)	Moderate weakening (4)	Stability (5)	Moderate strengthening (6)	High strengthening (7)
Electoral industrial balance	-3.54*** (0.397)	-2.69*** (0.289)	-3.56*** (0.597)	-2.63*** (0.751)	-2.34*** (0.553)	-2.11* (0.852)	-2.21*** (0.631)
Obs.	4264	5548	2454	1772	2181	1405	2000
Adj. R2	0.739	0.761	0.763	0.730	0.756	0.739	0.765
Within R2	0.201	0.154	0.261	0.149	0.084	0.187	0.184
PANEL D2
Dep. var.: Right votes	Weakened unions (1)	Not weakened unions (2)	High union weakening (3)	Moderate weakening (4)	Stability (5)	Moderate strengthening (6)	High strengthening (7)
Electoral industrial balance	1.04*** (0.306)	0.41 . (0.239)	1.35** (0.464)	0.22 (0.595)	0.35 (0.417)	0.08 (0.607)	-0.04 (0.046)
Obs.	4264	5548	2454	1772	2181	1405	2000
Adj. R2	0.807	0.809	0.810	0.812	0.845	0.833	0.771
Within R2	0.242	0.230	0.253	0.219	0.236	0.305	0.262

• Source: Elaborated by the authors. Note: p < 0.1,
• ^*
  p < 0.05,
• ^**
  p < 0.01,
• ^***
  p < 0.001.

In the models for the data divided into categories according to the levels of variation in union strength, it can be seen that the mediating role of union weakening is strongly driven by the cases in which there was strong weakening. To the left, the expected average effect is -3.56 among municipalities with strong union weakening, while to the right the coefficient is 1.35. Although on the left the coefficients of the models with the other levels of union strength are statistically significant, it is possible to see that there is uniformity among them in terms of the expected average impact. To the right, only the coefficient of model (3) is statistically significant.

The results of our last analytical step reveal that deindustrialization, as a comprehensive phenomenon, has substantial impacts on municipalities with different profiles in relation to union strength. Nonetheless, the estimated coefficients also consistently demonstrate that the results for municipalities where there has been union weakening are considerably higher compared to those where there has been no loss of union strength. It is striking to note that among municipalities that have lost union strength the estimated negative impact on the left is -3.54 percentage points. In competitive elections, together with the estimated positive effect on the right of 1.19, this factor can be decisive.

ROBUSTNESS TEST

In this section, we present a series of robustness tests that we conducted to verify the consistency of the models applied in the research. Given its complexity, it is likely that the phenomenon under study is affected by confounding variables not included in our models. To examine the level of exposure of our models to this type of problem, we ran sensitivity analyses to check how strong an unobserved confounding variable must be to invalidate our results. Following a strategy specified by Cinelli and Hazlett (2020), we used partial R² values to assess the strength of association between confounders and treatment and outcome variables. For simplicity, these analyses were only implemented for the models of the steps 1 and 4. The operationalization was carried out in R with the sensemakr package.

Table 6 presents the results for the analyses in step 1. The robustness value (RVq=1) indicates that to make the estimated effect of deindustrialization on left performance exactly zero, omitted confounding variables must explain at least 11.5 percent of the residual variance of both the treatment and the outcome. In an extreme scenario, where unobserved confounding variables are assumed to explain all the remaining variance in the outcome, these measures would need to explain at least 1.47 percent of the residual variance. Furthermore, to eliminate the statistical significance of the estimate, omitted variables would need to explain at least 9.86 of the residual variances (RVq=1, α =0.05). To demonstrate that these scenarios are highly unlikely, since we cannot directly test the effects of unobserved variables, we used an observed variable that is strongly related to the treatment and outcome variables, GDP. It is plausible to assume that hardly any other factor would have the potential to cause omitted variable bias as great as GDP. Table 7 shows the impact on the model of unobserved confounding variables with the same strength as GDP, twice the strength of GDP and three times the strength of GDP. The adjusted estimates show that these variables would not substantially alter the results.

Table 6
Robustness tests for the association between electoral performance and deindustrialization

Outcome: PT votes
Treatment	Estimate	S.E.	t-value	R2 Y~D|X	RV q=1	RV q=1, α = 0.05
Deindustrialization	-2.16	0.161	-13.048	1.47%	11.5%	9.86%
Note: Critical t-value = 1.96
Outcome: Right votes
Treatment	Estimate	S.E.	t-value	R2 Y~D|X	RV q=1	RV q=1, α =0.05
Deindustrialization	0.31	0.135	2.3084	0.05%	2.14%	0.33%
Note: Critical t-value = 1.96

Table 7
Adjusted estimates using GDP as a reference (step 1)

Outcome: PT votes/ Right Votes	Adjusted estimate	Adjusted t-value
1x GDP	-2.09/ 0.24	-12.9085/ 1.8771
2x GDP	-2.02/ 0.18	-12.7695/ 1.4201
3x GDP	-1.95/ 0.11	-12.6309/ 0.9334

Table 8
Robustness tests for the association between electoral performance and deindustrialization among municipalities that weakened unions

Outcome: PT votes
Treatment	Estimate	S.E.	t-value	R2Y~D|X	RVq=1	RVq=1, α = 0.05
Deindustrialization	-3.54	0.394	-8.987	3.6%	17.5%	13.9%
Note: Critical t-value = 1.96
Outcome: Right votes
Treatment	Estimate	S.E.	t-value	R2Y~D|X	RVq=1	RVq=1, α = 0.05
Deindustrialization	1.05	0.304	3.437	0.5%	7.1%	3.1%
Note: Critical t-value = 1.96

When we consider the electoral performance of the right, as expected, the robustness of the effect is much lower. To invalidate the results presented, the unobserved variables need to be at least 2.14 percent. In the extreme scenario, just 0.05 percent would be enough. To suppress statistical significance, 0.33 percent would be enough. Consequently, unobserved variables with the strength of GDP would be sufficient.

Tables 8, 9, 10 and 11 show the results of the sensitivity analyses for step 4. In Tables 8 and 9, with the analyses done only with municipalities that have experienced union weakening, the robustness of the effects is greater, both in terms of the magnitudes of the coefficients and its statistical significance. It is remarkable that when the mediation of union weakening is considered, the estimated positive effect of deindustrialization on right-wing performance is not only greater, but also more robust. Not even unobserved confounding variables with three times the strength of GDP would be enough to eliminate the statistical significance of the coefficient.

Table 9
Adjusted estimates using GDP as a reference (step 4, - unions)

Outcome: PT votes/ Right Votes	Adjusted estimate	Adjusted t-value
1x GDP	-3.48/ 0.97	-9.0471/ 3.4224
2x GDP	-3.42/ 0.90	-9.1178/ 3.4276
3x GDP	-3.36/ 0.82	-8.1980/ 3.4638

Table 10
Robustness tests for the association between electoral performance and deindustrialization among municipalities that have not weakened unions

Outcome: PT votes
Treatment	Estimate	S.E.	t-value	R2Y~D|X	RVq=1	RVq=1, α = 0.05
Deindustrialization	-2.69	0.298	-9.014	2.5%	14.6%	11.7%
Note: Critical t-value = 1.96
Outcome: Right votes
Treatment	Estimate	S.E.	t-value	R2Y~D|X	RVq=1	RVq=1, α = 0.05
Deindustrialization	0.41	0.248	1.646	0.1%	2.9%	0 %
Note: Critical t-value = 1.96

Table 11
Adjusted estimates using GDP as a reference (step 4, + unions)

Outcome: PT votes/ Right Votes	Adjusted estimate	Adjusted t-value
1x GDP	-2.60/ 0.32	-9.0347/ 1.3687
2x GDP	-2.51/ 0.23	-9.0704/ 1.0627
3x GDP	-2.43/ 0.15	-9.1226/ 0.7190

Tables 10 and 11 contain the results for the analyses in municipalities that have not experienced union weakening. As expected, for the performance on the left, in addition to the decrease in the size of the effect already identified earlier, there is also a drop in the levels of robustness, but nothing that has the potential to affect the results. For the performance on the right, as the statistical significance of the coefficient of the analysis among municipalities that did not weaken unions is low in the original estimates, the robustness test only confirmed this scenario, which is in line with the predictions of our theory.

These sensitivity analyses provide important elements for understanding the phenomenon in question. About the direct negative effect of deindustrialization on the electoral performance of the left, consistent with what the literature has been presenting, there is solid evidence that the effect is robust, because extreme omitted confounding variables would be needed to invalidate the estimates. When we consider the mediation of union weakening, the effects are not only larger but even more robust, which indicates that we are getting closer to reality when we consider this factor as a mechanism. For the performance of the right, the results show that union weakening is even more relevant. The coefficient of the model that tests the direct effect proved to be fragile to possible unobserved variables. Nonetheless, in those municipalities where deindustrialization is accompanied by union weakening, the positive effect on the right-wing vote is robust.

CONCLUSION

In the same way that industrialization has shaped the organization of political forces, the economic shocks caused by deindustrialization have also substantially altered political elites, but in the opposite direction. In this paper, we demonstrate, in line with the literature, that deindustrialization does indeed negatively affect the performance of the left and positively affects that of the right in presidential elections in Brazil. Nevertheless, we also found that union weakening is a key transmission mechanism for understanding how deindustrialization affects the vote.

In general, the results of the analysis suggest that deindustrialization has a direct negative effect on the performance of the left, which is robust to possible unobserved confounding variables. For the right, the positive effect identified is smaller and not robust. However, when we consider union weakening as a mechanism, the coefficients are larger and robust for both political forces, in the expected sense. In addition, we have shown that the levels of deindustrialization and union weakening matter. The effect of high deindustrialization is substantially greater than that of other levels. With regard to union weakening, the analyses indicated that the mediating role of union weakening is strongly driven by cases in which there has been a strong deterioration of unions.

These results therefore reveal, on the one hand, that the deterioration of the organized labor system as we know it in industrial society is having a major impact on the electoral performance of the Brazilian political forces. Understanding these forms of interaction that favor right-wing political groups calls for new studies that attempt to identify other transmission mechanisms for the complex relationship between deindustrialization and voting.

• JEL Classification: A11; C13; C23; D72.

DATA AVAILABILITY

The entire dataset supporting the findings of this study is published within the article and in the “Supplementary Materials” section.

REFERENCES

• Amaral, O. (2011), “Ainda conectado: o PT e seus vínculos com a sociedade”, Opinião Pública, Vol. 17 No. 1, pp.1-44.
• Angrist, J.D. and Pischke, J.-S. (2009), Mostly Harmless Econometrics: An Empiricist’s Companion, Princeton University Press, Princeton.
• Angrist, J. and Pischke, J.-S. (2014,) Mastering ’Metrics: The Path from Cause to Effect, Princeton University Press, Princeton and Oxford.
• Araújo, V.L. and Gentil, D. (2021), “O primeiro governo FHC (1995-1998): o neoliberalismo escancarado”, in Araújo, V.L. and Mattos, F. (Ed.s), A Economia Brasileira de Getúlio a Dilma - novas interpretações, Hucitec, São Paulo, pp.424-447.
• Bastos, C.P. and Ferraz, F. (2021), “A economia brasileira na primeira metade dos anos 1990: inflação, mudança estrutural e estabilização”, in Araújo, V.L. and Mattos, F. (Ed.s), A Economia Brasileira de Getúlio a Dilma - novas interpretações, Hucitec, São Paulo, pp.387-423.
• Békés, G. and Kézdi, G. (2021), Data Analysis for Business, Economics, and Policy, Cambridge University Press, Cambridge.
• Berman, S. and Snegovaya, M. (2019), “Populism and the Decline of Social Democracy”, Journal of Democracy, Vol. 30 No. 3, pp.5-19.
• Bresser-Pereira, L.C. (2014), A construção política do Brasil: sociedade, economia e Estado desde a Independência, Editora 34, São Paulo.
• Bryson, A. et al (2014), “What Accounts for the Union Member Advantage in Voter Turnout? Evidence from the European Union, 2002-2008”, Relations Industrielles / Industrial Relations, Vol. 69 No. 4, pp.732-765. doi: 10.7202/1028110ar.
  » https://doi.org/10.7202/1028110ar.
• Cameron, D. (1978), “The Expansion of the Public Economy: a Comparative Analysis”, American Political Science Review, Vol. 72 No. 4, pp.1243-1261.
• Cameron, D. (1984), “Social Democracy, Corporatism, Labour Quiescence, and the Representation of Economic Interest in Advanced Capitalist Society”, in Goldthorpe, J. (Ed.), Order and Conflict in Contemporary Capitalism, Oxford University Press, Oxford.
• Campello, D. and Urdinez, F. (2021), “Voter and Legislator Responses to Localized Trade Shocks from China in Brazil”, Comparative Political Studies, Vol. 54 No. 7, pp.1131-1162.
• Cano, W. (2012), “América Latina: do desenvolvimentismo ao neoliberalismo”, in Fiori, J.L. (Ed.), Estados e moedas no desenvolvimento das nações, 4. ed., Vozes, Petrópolis, pp. 287-326.
• Cardoso, A. (2014), “Os sindicatos no Brasil”, Nota técnica, mercado de trabalho, 56.
• Cardoso, A. and Gindin, J. (2017), “O movimento sindical na Argentina e no Brasil (2002-2014)”, Revista Sociedade e Estado, Vol. 32 No. 1.
• Castro, L.B. (2011), “Privatização, abertura e desindexação: a primeira metade dos anos 1990 (1990-1994)”, in Giambiagi, F. et al., Economia Brasileira Contemporânea: 1945-2010 2. ed., Elsevier, Rio de Janeiro, pp.131-164.
• Cavalcante, P. et al (2023), Base dos Dados. Available at: https://cran.r-project.org/web/packages/basedosdados/index.html
  » https://cran.r-project.org/web/packages/basedosdados/index.html
• Cinelli, C. et al (2022), “A Crash Course in Good and Bad Controls”, Sociological Methods & Research: 00491241221099552.
• Crouch, C. (1994), Industrial Relations and European State Traditions, Clarendon Press, Oxford.
• Crouch, C. (2017), “Membership Density and Trade Union Power”, Transfer, Vol. 23 No. 1, pp.47-61. doi: 10.1177/1024258916673533
  » https://doi.org/10.1177/1024258916673533
• Cunningham, S. (2021), Causal Inference: The Mixtape, Yale University Press, New Haven and London.
• Diniz, E. and Boschi, R.R. (2004), Empresários, interesses e mercado: dilemas do desenvolvimento no Brasil, Editora UFMG, Belo Horizonte.
• Fetzer, Th. (2020), “Austerity and Brexit”, Intereconomics, Vol. 55 No. 1, pp.27-33.
• Fouirnaies, A. (2021), “Can Interest Groups Influence Elections? Evidence from Unions in Great Britain 1900-2019”. doi: 10.7910/DVN/DVRNMH
  » https://doi.org/10.7910/DVN/DVRNMH
• Freeman, R.B. (2003) “What do Unions do… to Voting?”, NBER Working Paper Series, 9992.
• Gamez, D. (2021), “Are Deindustrialization and European Integration Fostering Populism?”Available at: https://www.populismstudies.org/are-deindustrialization-and-european-integration-fostering-populism/.IBGE (2022a), SCN - Sistema de Contas Nacionais Available at: https://www.ibge.gov.br/estatisticas/economicas/contas-nacionais/9052-sistema-de-contas-nacionais-brasil.html?=&t=resultados
  » https://www.populismstudies.org/are-deindustrialization-and-european-integration-fostering-populism/.IBGE» https://www.ibge.gov.br/estatisticas/economicas/contas-nacionais/9052-sistema-de-contas-nacionais-brasil.html?=&t=resultados
• IBGE (2022b), Quadro Geográfico de Referência para Produção, Análise e Disseminação de Estatísticas Available at: https://www.ibge.gov.br/apps/quadrogeografico
  » https://www.ibge.gov.br/apps/quadrogeografico
• Imai, K. and Kim, I. (2019), “When Should We Use Unit Fixed Effects Regression Models for Causal Inference with Longitudinal Data?”, American Journal of Political Science, Vol. 63 No. 2, pp.467-490. https://doi.org/10.1111/ajps.12417
  » https://doi.org/10.1111/ajps.12417
• Ingham, G. (1974), Strikes and Industrial Conflict: Britain and Scandinavia, Palgrave Macmillan, London.
• Keck, M.E. (2010), PT - A lógica da diferença: o partido dos trabalhadores na construção da democracia brasileira, Centro Edelstein de Pesquisas Sociais, Rio de Janeiro, pp.93-126. doi: 10.7476/9788579820298.
  » https://doi.org/10.7476/9788579820298.
• Kerrissey, J. and Schofer, E. (2013), “Union Membership and Political Participation in the United States”, Social Forces, Vol. 91 No. 3, pp.895-928.
• Kim, G. (2022), “Labor Unions and Voter Turnout in the American States: Direct Versus Indirect Mobilization”, State Politics & Policy Quarterly, Vol. 22 No. 2, pp.161-180. doi:10.1017/spq.2021.33.
  » https://doi.org/10.1017/spq.2021.33.
• Kim, S.E. and Margalit, Y. (2016), “Informed Preferences? The Impact of Unions on Workers’ Policy Views”, American Journal of Political Science, Vol. 61 No. 3 (July 2017), pp. 728-743. (First published on 9 November 2016). doi: 10.1111/ajps.12280
  » https://doi.org/10.1111/ajps.12280
• Leighley, J.E. and Nagler, J. (2007), “Unions, Voter Turnout, and Class Bias in the U.S. Electorate, 1964-2004”, The Journal of Politics, Vol. 69 No. 2, pp.430-441.
• Lucca, J.B. (2011), “Origem e transformação do enraizamento sindical do Partido Justicialista (Argentina) e do Partido dos Trabalhadores (Brasil)”, Revista Brasileira de Ciência Política, No. 5, pp.287-320.
• Margalit, Y. (2019), “Political Responses to Economic Shocks”, Annual Review of Political Science, Vol. 22, pp.277-295.
• McQuarrie, M. (2017), “The Revolt of the Rust Belt: Place and Politics in the Age of Anger”, British Journal of Sociology, Vol. 68, issue S1.
• Morceiro, P. (2012), Desindustrialização na economia brasileira no período 2000-2011: abordagens e indicadores, Cultura Acadêmica, São Paulo.
• Mummolo, J. and Peterson, E. (2018), “Improving the Interpretation of Fixed Effects Regression Results”, Political Science Research and Methods, Vol. 6 No. 4, pp.829-835. doi: 10.1017/psrm.2017.44.
  » https://doi.org/10.1017/psrm.2017.44.
• Nassif, A. et al (2013), “O debate sobre a desindustrialização precoce no Brasil: estamos avançando ou regredindo em relação aos países desenvolvidos?”, in Feijó, C. et al (Ed.s), A Desindustrialização Brasileira, Editora Unisinos São Leopoldo, RS, pp.25-67.
• Nicolau, J. (2020), O Brasil dobrou à direita: uma radiografia da eleição de Bolsonaro em 2018, Editora Schwarcz-Companhia das Letras, São Paulo.
• Ogeda, P. et al (2021), “Labor Unions and the Electoral Consequences of Trade Liberalization”, CESIFO Working Papers, No. 9418, pp.1-61.
• Oreiro, J.L. and Paula, L.F. (2021), Macroeconomia da Estagnação Brasileira, Alta Books, Rio de Janeiro.
• Pesquisa “A Cara da Democracia” (2024), Banco de Dados INCT IDDC. Available at: https://www.institutodademocracia.org/a-cara-da-democracia
  » https://www.institutodademocracia.org/a-cara-da-democracia
• Pontusson, J. (1995), “Explaining the Decline of European Social Democracy: The Role of Structural Economic Change”, World Politics, Vol. 47, issue 4, pp.495-533. doi:10.1017/S0043887100015197.
  » https://doi.org/10.1017/S0043887100015197.
• Radcliff, B. and Davis, P. (2000), “Labor Organization and Electoral Participation in Industrial Democracies”, American Journal of Political Science, Vol. 44 No. 1, January, pp.132-141.
• Rodrigues, C. and Pereira, M. (2022), “Disputando o partido, enfrentando opositores: efeitos políticos dos movimentos negro e LGBT+ nos governos Lula e Dilma (2003-2014)”, Opinião Pública, Vol. 28 No. 3, pp.635-677.
• Rodrigues, L.M. (2009), Mudanças na classe política brasileira, Centro Edelstein de Pesquisas Sociais, Rio de Janeiro.
• Saad Filho, A. and Morais, L. (2018), Brasil: Neoliberalismo versus Democracia, Boitempo, São Paulo.
• Serra, J. (1982), “Ciclos e mudanças estruturais na economia brasileira do após-guerra”, Revista de Economia Política, Vol. 2 No. 6, pp.5-45.
• Silva, J. (2019), “Regional deindustrialization: concepts, causes, effects and the Brazilian case”, Gestão & Produção, Vol. 26 No. 4, pp.1-12.
• Simoni, M. (2013), “The Left and Organized Labor in Low-Inflation Times”, World Politics, Vol. 65, issue 2, pp.314-349.
• Singer, A. (2012), Os sentidos do lulismo: reforma gradual e pacto conservador, Companhia das Letras, São Paulo.
• Soares, G. and Terron, S. (2008), “Dois Lulas: a geografia eleitoral da reeleição (explorando conceitos, métodos e técnicas de análise geoespacial)”, Opinião Pública, Vol. 14 No. 2, pp.269-301.
• Sojourner, A.J. (2013), “Do Unions Promote Members’ Electoral Office Holding? Evidence from Correlates of State Legislatures’ Occupational Shares”, ILR Review, Vol. 66 No. 2.
• Stephens, J. and Wallerstein, M. (1991), “Industrial Concentration, Country Size, and Trade Union Membership”, American Political Science Review, Vol. 85 No. 3, pp.941-953.
• Streeck, W. and Hassel, A. (2003), “Trade unions as political actors”, in Addison, J.T. and Schnabel, C. (Ed.s), International Handbook of Trade Unions, Elgar, Cheltenham, pp.335-365.
• Terron, S. and Soares, G. (2010), “As bases eleitorais de Lula e do PT: do distanciamento ao divórcio”, Opinião Pública, Vol. 16 No. 2, pp.310-337.
• Vlandas, T. and Halikiopoulou, D. (2022), “Welfare state policies and far right party support: moderating ‘insecurity effects’ among different social groups”, West European Politics, Vol. 45 No. 1, pp.24-49.
• World Bank (2022), World Development Indicators Available at: https://databank.worldbank.org/source/world-development-indicators
  » https://databank.worldbank.org/source/world-development-indicators
• Zanotti, L. et al. (2023), “Assessing the fourth wave of the populist radical right: Jair Bolsonaro’s voters in comparative perspective”, Opinião Pública, Vol. 29 No. 1, pp.1-23.
• Zullo, R. (2008), “Union Membership and Political Inclusion”, Industrial and Labor Relations Review, Vol. 62 No. 1.

Edited by

• Editor responsible for the evaluation process:
  Luiz Carlos Bresser-Pereira

Publication Dates

• Publication in this collection
  14 Sept 2026
• Date of issue
  2026

History

• Received
  13 Feb 2025
• Accepted
  29 Oct 2025