Mobility Interrupted: A New Framework for Understanding Anti-Left Sentiment Among Brazil’s “Once-Rising Poor”

ABSTRACT How do sequences of upward and downward socioeconomic mobility influence political views among those who have “risen” or “fallen” during periods of leftist governance? While existing studies identify a range of factors, long-term mobility trajectories have been largely unexplored. The question has particular salience in contemporary Brazil, where, after a decade of extraordinary poverty reduction on the watch of the leftist Workers’ Party (PT), a subsequent period of economic and political crises intensified anti-PT sentiment. This article uses original data from the 2016 Brazil’s Once-Rising Poor (BORP) Survey, using a 3-city sample of 822 poor and working-class Brazilians to analyze the relationship between retrospective assessments of prior socioeconomic mobility and anti-PT sentiment. The study found that people who reported a “stalled” mobility sequence (upward mobility followed by static or downward mobility) were more likely to harbor anti-left sentiment than other groups, as measured by this study’s anti-PT index.

government with a social-democratic orientation, a subsequent period of intertwined economic and political crises intensified distrust for democratic politics, disdain for the left, and nostalgia for earlier periods of authoritarian rule-a confluence of processes that helped set the stage for the 2018 presidential election of far-right former military captain Jair Bolsonaro. This study uses original data from the Brazil's Once-Rising Poor (BORP) 2016 Survey (Junge et al. 2022) conducted among a 3-city sample of 822 poor and working-class Brazilians to analyze the relationship between retrospective assessments of prior socioeconomic mobility and anti-leftist sentiment. The key finding is that people who remembered their economic condition as having gotten better between 2003 and 2011 and then worse or no change between 2012 and 2016 (what we refer to as a stalled mobility sequence) were, on average, more inclined toward anti-left sentiment compared to respondents reporting other mobility patterns.
The recent period of economic growth and poverty reduction occurred when Brazil was governed by the left-leaning Workers' Party (Partido dos Trabalhadores or PT). During the two-term presidency of PT cofounder Luiz Inácio Lula da Silva (2003-10), some 30 million people rose above the poverty line, with mean per capita income continuing to rise during the first term of Lula's PT successor, Dilma Rousseff (2010-14). During the PT's first decade in office , Brazil's Gini coefficient, measuring inequality of income distribution, fell by 13 percent, coupled with an overall 55 percent reduction of poverty (Neri 2014;cf. Medeiros et al. 2015, 3-4) and moreover, seeming immunity to the 2008 global financial crisis (Weisbrot et al. 2014). Economic transformation on this scale inspired prominent economists and policymakers to celebrate the expansion of Brazil's middle class as it absorbed millions of rising poor, christening the recent arrivals a "new middle class" (Klein et al. 2018).
The macroeconomic backdrop for this extraordinary and unprecedented reduction of poverty in a country long known for wealth inequality consisted in the 1990s consolidation of the market globalization paradigm, the stabilization of Brazil's national economy, and the global integration of China's massive and fastgrowing economy, coupled with a voracious demand for imported goods. Fortuitous macroeconomic conditions notwithstanding, the PT's prioritization of social-democratic initiatives to reduce poverty and inequality was crucial. These included conditional cash transfer programs (most famously the Bolsa Família, or Family Allowance Program); expanded access to higher education for nonwhite and poor applicants through affirmative action initiatives; increased access to affordable housing for low-income households; expansion of labor formalization and rights, in particular, to domestic workers; and steady increases in the minimum wage, which raised the benchmark for wages and government transfers throughout the economy (Barros et al. 2010).
For millions of poor and working-class Brazilians, the PT years are remembered nostalgically as a time when aspirations for a better life were awakened and finally realized (DataFolha 2021; Genial Investimentos and Quaest Consultoria 2022). Low-income families found themselves with some extra cash at the end of the month, enough perhaps to purchase a flat-screen television, a smartphone, or even a car. The mobility that seemed to be taking off in these years found a symbolic analogue in the figure of Lula himself: a metalworker turned union organizer from Brazil's poorest region (the Northeast), who, with little formal education other than training as a machinist, rose to become president. Despite corruption scandals during Lula's presidency, his popularity continued to grow, particularly among Brazilians who identified with a man who, through his own experiences of poverty, suffering, loss, and perseverance, understood both the plight and the potential of the poor. Sociologist Jessé da Souza has argued that among Brazil's upwardly mobile poor, sympathy for and identification with Lula fostered adherence to his political project of combating social injustice through redistributive state policy (2010,251). During Lula's two terms and the first term of his successor, Dilma Rousseff, electoral support for the PT shifted both demographically and regionally -from middle-class support in the southern regions to support from poor voters in the poorer northeastern region (Hunter and Power 2019;Amaral 2020).
Given the extraordinary economic transformation achieved during the PT years, Lula's easy reelection to a second term in office in 2006 and the resounding victory of his PT successor, Dilma Rousseff, in 2010 might appear to have followed a straightforward logic: People whose lives get better during periods of sustained rule under one political party will, all other things being equal, incline toward voting for the incumbent party with which those improvements are correlated (Lewis-Beck and Paldam 2000;De La O 2013;Pop-Eleches and Pop-Eleches 2012). Applied to the Brazilian context, one might imagine that awareness of personal and family mobility facilitated through PT social-democratic initiatives would incline beneficiaries toward a leftist orientation rather than toward conservative ideologies, historically associated with Brazilian elites' interest in maintaining the status quo (Pribble 2013, 16;Meltzer and Richard 1981).
This logic, however, never fully accounted for people's experiences-even before poverty reduction came to a grinding halt and Brazil fell into the deep crisis for which the 2010s will long be remembered. Despite their "rise out of poverty," the newest arrivals to the middle class-now paying taxes and electricity bills and increasingly indebted-remained economically vulnerable and constrained in their participation in public consumptive life. Facing grueling commutes, poor public services, and violence disproportionately impacting their neighborhoods, the upwardly mobile poor, moreover, faced disdain and resentment from members of Brazil's established middle and upper classes (Cardoso 2020). Their economic mobility was always precarious and at risk of derailing, due to health problems, shifts in the labor market, policy changes, or simple bad luck.
It is also worth noting that the PT's signature poverty reduction initiative, Bolsa Família, has minimal conditions for continued family support after initial qualification (proof that children are attending school and have up-to-date vaccinations) and operates through market mechanisms within a technocratic framework, leading some critics to argue that the program incites identification with the consumerist individualism of elites, rather than the citizen consciousness of the working class (Singer 2018;Lavinas 2013;Fundação Perseu Abramo 2017;cf. Richmond 2020). Empirical studies of Bolsa Família, however, mostly show increased electoral support for the PT among prior beneficiaries (Simoni Junior 2021;Amaral 2020;Layton et. al 2017;Corrêa 2015;Zucco 2013;Peixoto and Rennó 2011;Licio et al. 2009;Nicolau and Peixoto 2007;cf. Pereira 2015;Bohn 2011). 1 We would not, however, expect this pattern to have survived by the time of our data collection in 2016, when Rousseff was in the final stages of impeachment and many previously ascendant Brazilians were slipping below the poverty line.
However one remembers the PT years, few observers question the reversal of fortune-both in objective economic terms and with regard to subjective aspirational horizons-that unfolded during the second decade of the twenty-first century. The optimistic moment began to wane in 2013, when demonstrators across the country protested bus fare increases and the stark contrast between bloated federal spending for the upcoming World Cup and Olympic Games, on the one hand, and the poor quality of available public services, on the other (Magalhães 2018;Purdy 2019). As a severe economic crisis took hold in 2014, itself fueled in part (Paula and Moura 2021) by the ideologically motivated anticorruption campaign Operation Car Wash (Fishman et al. 2019), the conservative media conglomerate Globo incited a channeling of popular discontent in the form of rage against Lula, against Dilma Rousseff, and against the PT (Van Dijk 2017; De Albuquerque 2019). Amid unprecedented polarization between PT supporters and detractors, Rousseff was impeached in 2016, in a move many regard as a soft coup, and was replaced for the next two years by her center-right vice president, Michel Temer. Stemming from the Car Wash investigation, Lula was imprisoned on a politically motivated personal enrichment corruption charge (Fishman et al. 2019) and prevented from running in the 2018 presidential elections, for which he was the front-runner. In 2021, the convictions against Lula were reversed by Brazil's Supreme Court.
Against this backdrop, former excitement about poverty reduction evaporated, and many of Brazil's "once-rising poor" fell below the poverty line in what we conceptualize as a stalled mobility sequence (i.e., an upward trend followed by a static or downward trend). Structures of affect accompanying this reversal of fortune seem reminiscent of the "cruel optimism" Berlant (2011) has associated with late capitalism, whereby dreams of a better life and a sense of forward momentum erode as the social institutions that once offered upward mobility themselves fall apart. For previously ascendant Brazilians, the capsizing of an earlier, subjective sense that things were getting better has left a distinctive mark on political views, perhaps conferring its own variety of resentment toward, and alienation from, leftist politics and democratic political institutions more generally. Ethnographic studies from this period emphasize the "cruel pessimism" (Mitchell 2021), "despairing hopes" (Rojas et al. 2021), and "withering dreams" (Kopper 2021) that replaced an earlier period of optimism as social mobility trajectories reversed.
In 2016, authors Junge, Klein, and Mitchell began a three-year anthropological investigation of the political values, attitudes, behaviors, and broader lifeways of poor and working-class Brazilians whose material conditions and aspirational horizons were shaped by experiences and expectations of economic ascent during the PT years. We have conceptualized this group as Brazil's "once-rising poor" (Klein et al. 2018). Combining survey methods with a range of ethnographic interviewing and observational techniques, we conducted fieldwork in the cities of Recife, Rio de Janeiro, and São Paulo. The year of Dilma Rousseff 's impeachment, 2016, was a moment of cynicism and despair about all things political, propelled by the unfolding, intertwined economic and political crises described here. As such, we launched our investigation at a moment when most forms of upward mobility had been halted or reversed, presenting us with formidable challenges-but also opportunities-for understanding the political sentiments of "once-rising poor" Brazilians which may have contributed to Bolsonaro's election in 2018.
This article examines the relationship between retrospective assessments of prior socioeconomic mobility sequences and anti-PT sentiment-the latter referred to hereafter using the Portuguese term antipetismo-among once-rising poor Brazilians. 2 To be clear, our goal in this article is not to explain what drove Bolsonaro's victory. 3 Instead, it is to generate new insight on a cultural formation-antipetismo-that scholars, pundits, and ordinary citizens alike agree was crucial (Rennó et al. 2021;Amaral 2020) and that grew precipitously during the research period. 4 During this period, anti-PT sentiment spread through mainstream print and broadcast media, through organized protest movements against the PT (Telles 2016), and, in the run-up to Bolsonaro's 2018 election, through WhatsApp message groups (Cesarino 2020;Davis and Straubhaar 2020;Pinheiro-Machado and Scalco 2018). In Nicolau's account, these processes incited with everincreasing efficacy the understanding of the PT as "a party of the corrupt, that threatens traditional families, and wants to transform the country into an enormous Venezuela" (2020, 82). In Cardoso's analysis (2020), antipetismo has roots in the most conservative sectors of the middle classes, including among recent arrivals from the poor to the middle class (2020, 272), and it draws on fears of communism, an association of corruption with the PT, and the allegation, circulated widely in the media, that "the PT broke Brazil" economically (2020, 271; see also Messenberg 2017). Missing from these accounts, however, is empirical analysis of how the abrupt reversal of fortune for once-rising Brazilians may have fostered anti-PT sentiment. In this sense, our approach fills an important gap with its focus on retrospective evaluations of socioeconomic mobility trends in different moments of time, which may bring greater analytic purchase than existing approaches.
The next section reviews scholarly conversations about the relationship between socioeconomic mobility and political views. Based on this review, we delineate our core hypothesis for how a range of subjectively recalled mobility trajectories (spanning the PT years and subsequent crisis period) have influenced anti-PT sentiment. Then the data sources, variables, and plan of analysis are presented, followed by empirical evidence relevant to the core hypothesis. The concluding section reflects on the implications of the findings for understanding the era in Brazil and elsewhere in Latin America that followed the Pink Tide of the early twentyfirst century.

SOCIOECONOMIC MOBILITY AND POLITICAL VIEWS
To examine configurations of antipetismo associated with retrospective assessments of mobility during the PT years and subsequent crisis period, this study uses data from a structured household survey, conducted in 2016 in neighborhoods where we could reasonably expect to find large numbers of "once-rising poor" Brazilians who had experienced a range of mobility forms and trajectories, including but not limited to economic mobility "out of poverty." The utilization of mobility as an explanatory variable is informed by existing studies showing that economic status (being "poor," for example) is not a consistently reliable predictor of political attitudes and electoral behavior. Piketty (1995), for example, has argued that European voters with the exact same incomes but different social origins will vote differently on redistribution policies. Focusing on Latin America, Kaufman (2009) has similarly argued that poor people do not necessarily vote for higher taxes on the rich or for redistribution (cf. Meltzer and Richard 1981). Instead of absolute individual or family income at any given instant, relative economic trajectories over time-sequences of mobility-have strong potential to explain emergent political values, attitudes, and behaviors. This is particularly so as economic decline has been publicly seized on and politicized by the PT's opposition and an often hostile press (e.g., Braga and Kracovics 2015).
In survey-based research, mobility can be gauged using both objective measures (e.g., comparing income, occupation, or educational attainment at different points in time) and subjective measures using questions that ask respondents to consider whether their lives have, in some sense, gotten better or worse over time. The predictive value of subjective mobility measures for political attitudes has been argued persuasively by several contemporary scholars, including for voting patterns (Graham and Pettinato 1999;Matějů 1999;Weakliem 1992), for attitudes about redistributive policies (Kaufman 2009;Benabou and Ok 2001), and for adopting political behaviors of the "destination class" (Clifford and Heath 1993). 5 The approach in this study combines both objective and subjective mobility measures, as well as assessments of both intra-and intergenerational mobility. Consistent with observations made by Amaral and Ribeiro (2015, 119), we are not aware of any existing studies that empirically examine the impact of the stalledmobility pattern on political views (Amaral and Ribeiro 2015, 119). There is, however, a wide-ranging literature on the political effects of negative economic shocks. Margalit (2019) has thoroughly reviewed this literature and identifies two recurrent patterns: negative economic shocks (especially job loss) increase support for redistributive policy and expansive social policy more generally, and erode trust in the political system (279). Redistributive sentiment as an effect of negative economic shock is well documented as a historical pattern in the United States (Owens and Pedulla 2014) and in Europe (Martén 2019). 6 However, studies from Brazil and elsewhere in Latin America seem to reveal a different pattern. Drawing on research in Peru, Graham and Pettinato (1999) found that downward mobility correlated with lower support for redistributive policies (however, in accordance with Margalit, lower confidence in the healthy functioning of democracy or "political happiness"). In the Brazilian context, Rennó et al. (2021) found that long-term downward mobility preceding Bolsonaro's inauguration in 2019 correlated with continued support for him during the first year of his presidency.
Margalit's summary conclusion that negative economic shocks tend to erode trust in the political system-especially toward incumbent political parties-is well supported. For example, Achen and Bartels (2017) argue that voters take out their frustrations on the incumbents and vote for out-parties. From this perspective, when poor economic performance or other society-level crises occur, voters tend to choose the opposition when elections arrive-which, circa 2016 (i.e., after 14 years of PT governments), would translate into a rightward shift in voting. The apparent exception to Margalit's first finding posed by the Brazilian case may be partly explained by his second finding. Negative economic shocks both increase support for welfare and foster antisystemic sentiment. These two effects themselves stood at odds in the Brazil of 2016, which had been governed by the socialdemocratic PT since 2003; and, we speculate, the latter effect-the erosion of trust in "the system"-probably trumped the former. That is, "the system" became equated with the long-term incumbent (the PT), which contaminated the possibility of a positive interpretation of the PT's signature initiatives in redistributive policy.
Some studies find that negative economic shocks foster rightward political movement. For example, Stevenson (2001) found, from an analysis of 14 Western democracies, that national economic contraction moves voter preferences to the right. However, these are only aggregate results, revealing little about the effects of negative economic shocks in people's lives. Somewhat closer to our own analysis, Dehdari 2021, based on research in Sweden, found that a job layoff led "lowskilled native-born workers" to increase their support for the far right (191). Similarly, Ballard-Rosa et al. (2021) examined the frustration-aggression mechanism whereby economic shocks hindered British citizens' expected attainment of their goals; this, they argue, can dispose them to embrace authoritarian values. We contend, however, that longer-term mobility sequences have political consequences that are not perceptible by studies that focus only on a short-term economic shock.
Based on these insights from theoretical conversations linked especially to Berlant's work about forms of affect stemming from "dashed hopes;" on existing empirical studies about the tendency of economic downturn to erode trust in democratic functioning and to ignite antisystemic and anti-incumbent sentiment; on the historical specificities of contemporary Brazilian political and economic conditions; and on the origins of antipetismo, this study's core hypothesis is: People who report a stalled-mobility sequence (upward during the PT's boom years and then static or downward during the subsequent crisis period) will, on average, harbor more antipetismo than people reporting other mobility sequences.

ANALYSIS
Data for this analysis come from the 2016 Brazil's Once-Rising Poor (BORP) survey, conducted in the three cities of Recife, Rio de Janeiro, and São Paulo. While the survey methodology is described in detail elsewhere (Junge et al. 2022), an abbreviated summary is provided here. Based on a 5 percent margin of error, a 0.25 standard deviation, and a 95 percent confidence interval, and contemplating a conservative design effect to account for the multistage sample design and expected completion rate of 90 percent, we determined that a minimum sample size of 384 households in each city (rounded up to 400) would be necessary to produce statistically generalizable claims. Accordingly, we aimed for, and ultimately obtained, a total sample size of 1,204 respondents, representing 400 households in each of the three cities. In each city, we constructed 4 sampling areas for recruitment using 2010 census data to identify tracts with robust representation of our target population. Each sampling area consisted of 10 to 14 qualifying census tracts; in each tract, we conducted interviews with adult residents from 10 households, following a standardized recruitment protocol. 7 So that respondents would be old enough to have adult memory of the boom years, this analysis is limited to the subsample of respondents aged 36 or older (n = 822).

Measures
The primary outcome of interest for this study is anti-PT sentiment, antipetismo. It used principal components analysis to construct an antipetismo index, based on nine survey questions that, in different ways, gauged (dis)approval of the PT and its leaders. One question asked respondents whom they voted for in the second round of the 2014 presidential elections (with a vote for overtly anti-PT finalist candidate Aécio Neves taken as indicative of anti-PT sentiment). Six of the questions asked respondents to rate their agreement with the following statements on a scale of 1 to 5: 1. "Corruption was the main reason for the impeachment of Dilma Rousseff." 2. "Dilma's impeachment was a coup." 3. "You feel represented by the Temer government." 4. "You felt represented by the Dilma government." 5. "You felt represented by the Lula government." 6. "The PT's government improved life for people such as yourself." Two of the questions included in the index measured respondents' assessments of two well-known and anti-PT movements, Brasil Livre (Free Brazil) and Vem Pra Rua (Come to the Streets). In the subsequent analyses-in the construction of the index, when the constitutive index items appeared as individual dependent variables, and with respect to the index itself-these measures were all coded such that higher values indicated more anti-PT sentiment, or antipetismo. (Items 2, 4, 5, and 6, which all reflect pro-PT sentiment, were all reverse-coded.) As noted, retrospective assessments of prior socioeconomic mobility experiences have been shown to be powerful predictors of political opinions. Accordingly, for our main explanatory variables, we drew from survey questions that asked respondents to compare their household's current situation (i.e., at the moment of the 2016 interview) to two earlier points in time, 2003 and 2011. We compared individuals' responses to these questions to identify their reported mobility trends (i.e., upward, downward, or static/no change) for the periods 2003-2011 (conceptualized as the "precrisis" period of PT governance) and 2011-2016 (understood as the period of political and economic crisis still unfolding at the time of data collection).
The survey gauged two varieties of subjectively recalled mobility. First, it asked respondents to evaluate the financial situation of their home at the time of the interview (mid-2016) compared to 2011 and, separately, compared to 2003. Similarly, two questions asked respondents to compare the "quality of life of the people living in this home today" with 2011 and, separately, 2003. With these four questions, using the process described above, we were able to gauge retrospective assessments of respondents' financial mobility trends for the periods 2003-2011 and 2011-2016, and  Thus, the primary outcome variable is the antipetismo index, and the main explanatory variables are the measures of financial situation and quality of life mobility. The analyses also employ control variables commonly used in survey research. These include measures of race, gender, age, marital status, education, religion, employment status, income, social assistance beneficiary status, and how safe respondents feel in their neighborhood. We also control for the city in which a respondent lives (Recife, Rio de Janeiro, or São Paulo) to account for cityspecific characteristics.
Characteristics of the study sample are presented in table 1, showing differences by city. Because each of the three cities is distinctive in its history, its regional positioning in national political economy, and the makeup of its residents and urban geographies, we would expect the intercity variation on characteristics reported here. Accordingly, we do not include in the table results of significance tests we carried out, nor do we devote substantive attention to discussion of observed differences. 8 Median age was 53 years. With respect to race, nearly three-quarters of respondents selected an option other than White (26.1 percent Black, 43.8 percent mixed, and 3.5 percent Asian or Indigenous). 9 Women were notably more represented in our sample than men (59.6 percent vs. 40.4 percent, respectively), and slightly more than half (53.1 percent) of the respondents were married. Nearly a third (31.1 percent) of the sample were high school graduates, and 35.0 It is perhaps surprising that approximately half of the sample reported either no change or a worsening of household conditions (in terms of both financial situation and quality of life) during the period 2003-2011-the historical moment most strongly associated with economic ascendance for poor and working-class Brazilians. We take this as an important reminder that the "boom years" associated with the early PT administrations were, for many Brazilians, not experienced (or at least not remembered) as such. Similarly, we underscore that approximately half the respondents did not report downward mobility between 2012 and 2016, suggesting a more complicated experience of the recent national-level economic and political crises than scholars have often imagined. It is also worth noting that since Brazil's economic crisis began in 2014, it is possible that had we asked the question with that start date, we would have obtained different answers. Additionally, these surveys were carried out as the impeachment of Dilma Rousseff was ongoing and the legacy of PT governments was receiving sustained criticism The survey question gauged work for pay during the past 30 days. b "Safe" = selected "safe" or "very safe" on a 5-point Likert scale question.
in the mass media. It seems likely that such a context affected people's recollections of the PT years. We also assessed the prevalence of all possible mobility sequences for the two periods considered. For both financial situation and quality of life, approximately 40 percent of our sample reported downward or no mobility during both periods, approximately 10 percent reported downward or no mobility during the earlier period and upward mobility thereafter, approximately 37 percent reported upward mobility during both periods, and approximately 13 percent reported stalled mobility in the form of upward mobility between 2003 and 2011, followed by downward or no mobility during the 2011-2016 period.

Empirical Strategy
To test the main hypothesis, we utilized two linear models-using maximum likelihood estimation-that employed the antipetismo index as the dependent variable. This index is standardized, and higher values of the index indicate more anti-PT sentiment. Testing the main hypothesis required estimating the effects of different mobility patterns, in terms of both financial situation and quality of life, on levels of antipetismo. The first model corresponds to financial situation mobility, and the second to quality of life mobility. Both share identical specifications, aside from utilizing these different sets of explanatory variables. For each set of explanatory variables, we first binarized the mobility measures. The positive class (i.e., coded as 1) includes respondents who reported upward mobility for the period in question. The negative class (i.e., coded as 0) includes respondents who reported static or downward mobility for the period in question. This coding scheme facilitates interpretation, balances the positive and negative classes, and fits the conceptual framework of this study.
In each of the two main models, we interacted the two mobility measures-one measuring whether respondents' reported upward mobility for the period 2003-11 and the other for the period 2011-16. This gave four terms of particular interest, associated with a particular mobility pattern sequence: the model intercept, 2003-11 upward mobility, 2011-16 upward mobility, and the interaction of 2003-11 upward mobility with 2011-16 upward mobility. The intercept corresponds to respondents who reported static or downward mobility during both periods. The 2003-11 upward mobility term corresponds to the stalled-mobility patternpeople who were upwardly mobile during that period and who did not experience upward mobility during 2011-16. The 2011-16 upward mobility term corresponds to people who did not experience upward mobility between 2003 and 2011 but who did experience upward mobility between 2011 and 2016. And the interaction term itself corresponds to respondents who were upwardly mobile during both periods-2003-11 and 2011-16. Thus, the tests of our primary hypothesis center on the 2003-11 upward mobility terms (one in the first model for financial situation, and one in the second model for quality of life).
While this term should be interpreted in reference to and in conjunction with the other mobility terms, in short, a positive and statistically significant estimate of this coefficient would indicate that respondents reporting a stalled-mobility pattern express more antipetismo, on average. Turning to the several standard control variables included in the models, the measures of income and age are standardized, and the other measures are binarized, with reference categories (the negative class) corresponding to the modal category. This approach increases model parsimony and-given the importance of the interaction terms-facilitates interpretation. The results of the two main models are presented below.

MAIN RESULTS
In table 3, model 1 shows the estimated effects of the four financial situation mobility patterns on the antipetismo index. Including the intercept, the only statistically significant mobility term is 2003-11. This is the term that captures respondents reporting stalled mobility. The term is positive, and the value of 0.52 indicates that this group on average scores approximately half a standard deviation higher on the antipetismo index-compared to each of the groups that experienced other mobility patterns, and to the sample in general. The same pattern appears in model 2. Stalled-mobility respondents, here in terms of quality of life rather than financial situation mobility, also score slightly more than half (0.58) a standard deviation higher on the antipetismo index on average. Together, these results support our hypothesis that individuals who experienced upward mobility between 2003 and 2011 and downward or no upward mobility between 2011 and 2016 express greater anti-PT sentiment.
Interpreting the main results presented in table 3 is straightforward, largely because none of the mobility terms (including the intercept) are statistically significant, aside from the 2003-11 term that corresponds to the stalled-mobility pattern. These results are displayed in figure 1. The four mobility patterns are  the table). The results from model 1 (financial situation mobility) are shown in red, and the results from model 2 (quality of life mobility) are in blue. As described above, the results from these two models closely match one another, and together provide strong support for our main hypothesis. Figure 1 shows clearly that of the four groups, only respondents reporting the stalled-mobility sequence exhibit higher-than-average levels of antipetismo. The groups that experienced mobility patterns other than upward mobility between 2003 and 2011 and stalled mobility thereafter report neither higher nor lower levels of antipetismo on average. (That is, compared to the whole sample-they express lower levels of antipetismo compared to respondents reporting stalledmobility sequences.) As both the table and figure demonstrate, we find the same results when considering mobility in terms of either financial situation or quality of life. 11

Antipetismo Disaggregated
With the initial support for our hypothesis that results described above provide, it is useful to analyze some of the items that constitute the antipetismo index; that is, as separate dependent variables, affording further insight into why individuals who reported different mobility patterns express more or less anti-PT sentiment. More specifically, we can gain sharper understanding of elements driving antipetismo among respondents reporting stalled mobility sequences.
We selected the following six items from the nine-item antipetismo index to analyze as separate dependent variables: whether respondents reported voting for Aécio Neves in the second round of the 2014 presidential elections, feeling  represented by the Temer government, having felt represented by the Dilma government, having felt represented by the Lula government, agreement that the PT's government improved life for people such as they, and agreement that Dilma's impeachment was a coup. We reverse-coded the last four measures in that list, such that higher values indicate more antipetismo, to facilitate interpretation across the models. We did not analyze as separate dependent variables respondents' assessments of the anti-PT social movements (Brasil Livre and Vem Pra Rua), since few respondents were familiar with these movements, or agreement that corruption was the main reason for the impeachment of Dilma, since this sentiment was captured in the question asking to what extent respondents agreed that Dilma's impeachment was a coup.
To analyze these six individual antipetismo outcomes, we employed the same empirical strategy and model specifications used above: two sets of models, one utilizing the financial situation mobility measures as explanatory variables and the other utilizing quality of life mobility measures. The measure of whether respondents reported voting for Neves is binary, and therefore we analyzed it with logistic regression. For the other outcomes, we employed linear maximum likelihood models. Table 4 shows the results of the models with the financial situation mobility measures as the main explanatory variables.
The models in the table show that there is no unique (i.e., statistically distinguishable from zero) estimated effect of any of the different mobility patterns on the outcomes measuring whether respondents voted for Neves, whether they felt represented by Temer, or whether they felt that the impeachment of Dilma was not a coup. Respondents reporting stalled-mobility sequences are less likely on average to have felt represented by the Dilma Rousseff government, but only at a significance level of p < 0.1. At conventional significance levels (p < 0.05), however, respondents reporting stalled mobility on average felt less represented by the Lula government and were less likely to report that the PT's government improved life for people like them. It is noteworthy that stalled mobility is more highly correlated with a lack of perceived representation by Lula than by Dilma. It seems plausible that because Lula is the figure most strongly associated with the socioeconomic ascension of the previously poor, and with the PT in general, he was also the figure most likely to be associated with the disappointment that came when that ascension was stalled or reversed. 12 Table 5 shows the same model estimations, but now using the quality of life mobility measures. The results are similar: we do not observe strong relationships between the different mobility patterns and whether respondents agree that the impeachment of Dilma Rousseff was a coup, whether respondents felt represented by the Dilma Rousseff government, or whether they voted for Neves. At a significance level of p < 0.1, the stalled-mobility subsample is, on average, less likely to report that the PT's government improved life for people like them. Though, again-at conventional significance levels-respondents reporting stalledmobility sequences are, on average, less likely to have felt represented by the Lula government. 13 In addition, two other types of patterns may help to illuminate the recollected mobility experiences of stalled-mobility respondents as these experiences relate to antipetismo. The first is respondents' disapproval of the PT's signature povertyreduction initiative, Bolsa Família. The second is respondents' dissatisfaction with the functioning of Brazilian democracy. The findings presented in table 6 are supplementary to the main results, for several reasons. First, these two outcomes are secondary indicators of antipetismo, measuring attitudes toward the most prominent national redistributive program (which was indeed implemented under PT governance) and toward Brazilian democracy, which was, of course, headed by PT leaders for more than a decade. Second, the question about respondents' satisfaction with democracy was asked only in Recife, and thus those model results are restricted to the sample from that city. Third, the two models in the table include the quality of life mobility measures only as explanatory variables (though they share the same general specification and strategy as the models analyzed earlier). No notable patterns appear when using the financial situation mobility measures as explanatory variables instead. The models employing the quality of life mobility measures offer nonetheless useful insight into the relationship between the subjectively recalled mobility experiences and antipetismo.
Model 1 in table 6 shows that respondents reporting stalled-mobility trajectories-in terms of quality of life mobility-are, on average, more likely to disapprove of Bolsa Família. This is in comparison to the groups who experienced different mobility sequences, and to the sample as a whole. Regarding model 2, though the sample is limited to Recife residents and-once more-only in terms of quality of life mobility, the results indicate that respondents reporting stalledmobility sequences are, on average, more dissatisfied with the functioning of Brazilian democracy. The interaction term is statistically significant at the p < 0.1 level, but here again, the magnitude and direction of this coefficient suggests, if anything, only that respondents who experienced upward mobility during both periods share approximately the same baseline of dissatisfaction with the other groups-with the notable exception of the stalled-mobility subgroup. Together, these results show that all else being equal (including controlling for past reliance on a social assistance initiative), respondents reporting stalled quality of life mobility trajectories are more likely to disapprove of the PT's signature redistributive initiative, and for the Recife subsample, are more dissatisfied with the functioning of Brazilian democracy. In conjunction with the empirical evidence presented, these supplementary findings help to further illuminate the relationship between retrospectively recalled stalled-mobility trajectories and anti-PT sentiment.

CONCLUSIONS
This study's major finding is that, controlling for the effect of other variables, people who report a stalled-mobility sequence (upward during the PT's boom years and then static or downward during the subsequent crisis period) harbor more anti-PT sentiment, on average. More precisely, respondents who reported improvements in household financial situation or quality of life during the period of PT governance most associated with poverty reduction, and worsening or no improvement during the subsequent period, expressed significantly more antipetismo than subgroups reporting other mobility patterns (and more than the overall sample).
The correlation between stalled mobility and antipetismo was strong and consistent across the two measures of mobility. To our knowledge, this is the first large-sample, multicity survey of once-rising poor Brazilians. As such, our survey overcomes some limitations of regionwide omnibus surveys (e.g., LAPOP and Latinobarometer), which, by seeking nationally representative samples, underrepresent poor and working-class respondents. Furthermore, our survey attends to the specificities of Brazil's recent political history by asking about mobility patterns during the key moments of the past two decades (the PT years and the subsequent crisis period). As such, our study brings a powerful new insight into the driving forces behind anti-PT sentiment as it congealed in recent years and influenced the outcome of the 2018 presidential elections.  While this article does not seek to explain voting outcomes, we may nonetheless speculate, in line with economic voting theory, on how retrospective assessments may have entered into anti-PT voting in the 2018 elections. From the premise that when poor economic performance occurs, voters tend to choose the opposition during subsequent elections, we speculate that stalled mobility (as we have conceptualized and measured it) leads to a retrospective blaming of the PT. In this sense, antipetismo becomes a mediator between the experience of stalled mobility and voting behavior. The PT had been out of the presidency for two years by the time of the election of Jair Bolsonaro, but antipetismo was central to Bolsonaro's campaign (Nicolau 2020), and we believe it highly likely that the political effects of stalled mobility were an important factor in that election.
The limitations of this study concern issues common to cross-sectional surveys with questions about the recall of earlier time periods. Responses to questions we used to construct our explanatory variables (retrospective assessments of mobility patterns) may have been influenced by recall bias, insofar as the periods referenced may have been recalled without full accuracy and these recollections may have been influenced by subsequent events and experiences. While the control variables we included in our models ranged widely, we did not include as controls media exposures (e.g., exposure to the conservative and highly anti-PT conglomerate Globo) or personal experiences of violence, which may precipitate a pivot toward a pro-police or pro-military candidate or one critical of human rights discourses.
While our analysis does not prove (and indeed was not designed to prove) that the observed correlation between stalled mobility and antipetismo is causal in nature, our interpretation of the observed empirical associations is consistent with our overarching causal argument. We contend that among poor and working-class Brazilians, the experience of a reversal of fortune has exerted its own distinctive effects on political sentiment, engendering resentment and antisystemic thinking, which, amid Brazil's political, economic, and mass-media conditions circa 2016, could readily intensify disdain for the perceived incumbent and culprit, the PT. While it is conceivable that anti-PT sentiment may influence retrospective assessments of earlier mobility trends, we note that respondents who report upward mobility during the early PT years express greater anti-PT sentiment, suggesting that individuals with greater anti-PT sentiment do not report that their lives got worse during periods of PT governance. This finding points to the political consequences when expectations are raised and then dashed, when material conditions and prospects for material wellbeing improve and then become precarious. Among respondents reporting stalled-mobility sequences, our analysis revealed a consistent element (across both mobility measures) driving the antipetismo index; namely, having felt less represented by the Lula government and having felt less that PT governments improved lives for people like them. There is, of course, no way to know whether respondents indicating these feelings at the time of our survey (2016) actually felt this way when Lula was in power (2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010). Regardless, we see the observed pattern as pointing to the complex relationship between assessments of and sentiment around Lula, on the one hand, and around the PT, on the other hand.
As noted earlier, perceptions of Lula among poor and working-class Brazilians often rest on identification with his story of poverty, suffering, and perseverance, a sentiment that has long exceeded partisan support for the PT (Hunter and Power 2019). Among our sample, it would thus appear that disdain for Lula (even if retroactively projected) may have an especially corrosive effect on support for the PT. It is also noteworthy that in our analysis of secondary indicators of antipetismo, stalled quality of life mobility correlated highly with disapproval of the Bolsa Família program and, for the Recife subsample, with greater dissatisfaction with the functioning of Brazilian democracy. The former result may signal a loss of faith in redistributive policy (which is, after all, founded on the promise of a better life) following reversal of fortune; the latter would appear to underscore the antisystemic sentiment to which stalled mobility can give rise.
While this study was not designed to predict voting outcomes, its key finding that once-rising poor Brazilians reporting stalled mobility harbor greater levels of anti-PT sentiment suggests a gap in how social scientists have explained the origins of antipetismo and its possible reverberations during the 2018 elections. Beyond the explanations reviewed above (i.e., the influence of conservative media, Evangelical Christianity, corruption scandals, the imprisonment of Lula, etc.), our analysis points to the crucial influence that a derailing of hope and plans for the future can have on political affinities-an insight that could help break the deadlocked debates over why people vote against their apparent economic interests (Frank 2004;Cramer 2016). To the extent that other countries in Latin Americacountries emblematic of the region's so-called Pink Tide-have more recently been mired in economic, political, and cultural crises followed by the election of conservative governments, our findings may contribute new insight to understanding the post-pink moment of cynicism and antipolitics-and what may come after it.
Since it did not collect data on expectations for the future, this study is not suited to empirically engaging this hypothesis.
6. Margalit's own empirical work also lends support to this thesis. In a study of the 2007-8 economic crisis in the United States (2013), he found that loss of employment increased support for greater welfare spending by between 22 and 25 percentage points (81), a pattern more pronounced for Republicans (who are traditionally hostile to welfare spending) than for Democrats. 7. While the category "head of household" is often used in community survey research, we found that the complexity of family structure in our sampling areas made such a category unviable. In all cases, we interviewed an adult household member who claimed knowledge of the whole household. 8. As an example, lower income levels reported in the Recife subsample probably reflect the well-known pattern of greater poverty in Brazil's northeast region. At the same time, a cost of living adjustment might account for some of this variation; however, we are reluctant to apply city-level adjustments (i.e., based on each city's overall cost of living), given that our sampling areas were peripheral neighborhoods, which cannot easily be argued to represent citywide trends.
10. To gauge our success in achieving a sample representative of the once-rising poor neighborhoods in which interviews took place (see Junge et al. 2022 for methodological details on sampling criteria), we compared mean incomes for survey respondents from each sampling area to corresponding census data figures for each entire sampling area (data not shown), indicating that the strategy was successful in capturing poor and working-class neighborhoods with robust representation of the target study population.
11. As a robustness check for these main results, we also employed nearest-neighbor propensity score matching. Within cities, we matched respondents on all of the control variables displayed in table 3, with the stalled mobility sequence (upward mobility from 2003 to 2011, followed by downward or no mobility 2011-16) composing the treatment condition. In the control group we included only respondents who reported downward or no mobility during both periods, in part to bias the test against our hypothesis. For both financial situation and quality of life mobility, the results of the matching analysis are statistically significant at p > 0.05 (data not shown) and entirely consistent with the findings presented here.
12. As mentioned, the Lula, PT, Dilma, and Coup outcome variables are reverse-coded such that larger values of these variables reflect more anti-PT sentiment. Thus, the positive associations displayed in tables 4 and 5 with respect to the main explanatory variables indicate that reversal-of-fortune respondents feel less represented by Lula, less that the PT's government improved life, and so on.
13. The interaction term in the Lula model is statistically significant at the p < 0.1 level. If anything, the direction and magnitude of the estimate indicates that people who were upwardly mobile during both periods share approximately the same baseline level of affinity toward the Lula government as did other "mobility groups," with the notable exception of the stalledmobility subgroup. Similarly, the estimated effects of the mobility terms in the Temer model essentially "cancel each other out."