Data and Methods

Our data come from the nine waves of Brazilian Legislative Surveys (BLS), which have been carried out in each of Brazil’s postauthoritarian legislatures. The BLS has sampled senators and federal deputies beginning with the 48th Legislature, which was elected in 1986 and interviewed in 1990, through the 56th, elected in 2018 and interviewed in 2021.

In order to estimate the position of each party we combine respondents’ answers with questions asking them to place themselves and each of the main parties in Congress at the time of the survey. The question wording and answer scale has been kept exactly the same in all waves.

The data generated by these questions consist of 17,974 ratings of 24 parties spread over 127 party/years, made by 1,031 different federal legislators, and also cover those individuals’ ideological self-placements.Footnote ¹ We combine these data following the same procedure discussed in previous work employing these same BLS questions, which builds on Aldrich and McKelvey (Reference Aldrich and McKelvey1977). We assume that the parties’ true positions are observed by our survey respondents, with random error. We also assume that each respondent possesses idiosyncratic “stretch” and “scaling” factors that imply respondent-specific uses of the answer scale. We treat these factors as fixed for each individual, even if they participate in BLS more than once.Footnote ²

Formally, let ${P_{ijt}} \in \left[ {1,10} \right]$ be the placement of party $j$ ( $j = 1, \ldots ,J$ ) by legislator $i$ ( $i = 1, \ldots ,I$ ), in year $t$ ( $t = 1, \ldots ,T$ ). Next, define ${P_{iit}} \in \left[ {1,10} \right]$ as the self-placement of each legislator $i$ in each year $t$ . The spatial model we use is simply:

$$\matrix{ \hfill {{P_{ijt}} = } \hfill\ {{\alpha _i} + {\beta _i}{\pi _{jt}} + {\epsilon _i}} \cr \hfill {{P_{ii}} = } \hfill {{\alpha _i} + {\beta _i}{\pi _{it}} + \nu } \cr } $$

where ${\pi _j}$ is the “true” position of each party in each year, ${\pi _{it}}$ is the true position of each respondent in each year, ${\alpha _i}$ and ${\beta _i}$ are legislator-specific “shift” and “stretch” rescaling factors, ϵ_t and $\nu $ are well-behaved disturbance terms.

We estimate all parameters of the model simultaneously in a Bayesian framework.Footnote ³ The estimates of the positions of all parties over the whole period are roughly centered at zero and distributed between approximately $ - $ 1 and 1. Negative values correspond to the ideological left and positive values to the right.

The Ideological Positions of Parties

Figure 1 reports a visual summary the ideological position of the main Brazilian parties in the period since the country’s return to democracy. This scaling ignores possible changes in the meaning of the traditional left–right dimension over the life span of the BLS. In past work, we have addressed this issue by comparing our estimates with answers given to “substantive” BLS survey questions—that is, those concerning specific policies and issues. Throughout the period covered by BLS, our left–right placements map well onto questions regarding state intervention in the economy. This association weakened in the late 1990s, during the heyday of neoliberalism, before strengthening again in the most recent waves. Yet however we slice the data, Brazilian politics at the elite level has always looked remarkably unidimensional (Power and Zucco Reference Power and Zucco2012; Zucco and Power Reference Zucco and Power2021).

Figure 1. Ideological Positions of the Leading Brazilian Parties (1990–2021)

Notes: Figure shows estimates for the main Brazilian parties for all years that are available. For each party, estimates are ordered chronologically from top to bottom. Arrows summarize variations over time. Estimates employ only data from the BLS project. See text for details.

In figure 1, locations in a common left–right space for each party are arrayed chronologically from top to bottom, with a maximum of nine observations for long-standing parties. As of 2023, these 14 leading parties represent less than half of the parties holding seats in the Chamber of Deputies, although this group of parties has accounted for much more than that over the entire period under analysis. While microparties are excluded from the questionnaires by design, we do have estimates for a number of parties that are not shown in the figure in order to keep it visually digestible. All estimates are reported in the online appendix C and are available from the data repository for this research note.

For most of the 30 years under analysis there were some very conspicuous ideological movements toward the right by parties such as CID (Cidadania, formerly PPS), and also by the much larger Party of Brazilian Social Democracy (PSDB) and Brazilian Democratic Movement (MDB). Other than that, the story is mainly one of a slight centrist drift both by left-wing parties but also by most minor right-leaning parties, contributing to a long phase of gradual depolarization that lasted until roughly 2013. Since then—in Brazil’s tumultuous recent period of impeachment, populism, and polarization—we have witnessed substantial changes. Left-wing parties have slowed and sometimes reversed their migration toward the center, and there have been remarkable recent changes on the right of the spectrum.

Over the past turbulent decade, the Democrats (DEM), successor to the regime-founding Party of the Liberal Front (PFL)—which had served as the reliable “anchor” of the right during the first eight waves of the BLS—drifted toward the center. The DEM was quickly replaced by the Social Liberal Party (PSL; Jair Bolsonaro’s party in the 2018 elections) on the extreme right of the spectrum. After Bolsonaro moved the ideological goalposts, the DEM acquired a new reputation as a mainstream, “establishment” conservative party. At the new rightmost pole, Bolsonaro’s PSL was accompanied by the New Party (NOVO), a tiny newer entrant (not shown), while the Liberal Party (PL), Brazilian Labor Party (PTB), and Republicanos began to row back toward the conservative horizon. Meanwhile, the PSDB and MDB (which at the time of BLS 9 fieldwork were engaged in awkward efforts to distance themselves from Bolsonaro) drifted back toward the center. Complicating the already considerable realignment of rightist forces, just as survey fieldwork was drawing to an end in 2021, the PSL and DEM merged into a newly formed União Brasil (UNIÃO). At the same time, Bolsonaro—after a period in which he had no formal party affiliation—joined the PL in order to launch his 2022 reelection bid.

The Ideological Positions of Presidents

In this note, for the first time, we also report the estimated ideological positions for Brazil’s eight presidents throughout the post-1985 democratic regime. These estimates are produced from questions in the survey that asked respondents to place presidents on the same left–right scale as they placed themselves and the parties. Our data on the rating of presidents amount to 1,929 placements by participants in the last three BLS waves. These data are pooled with the placements of parties and processed as if they were ratings of different parties. That is, presidents are treated as additional ${\pi _{ijt}}$ parameters in the model and estimated simultaneously with all other parameters.

There are some limitations to our data on the placements of presidents. The first wave in which the questionnaire included the rating of presidents was BLS 7, fielded in 2013. In that year, we asked respondents to rate not only the incumbent president (Dilma Rousseff), who was in the middle of her first term, but also all of the previous presidents during their terms.Footnote ⁴ In BLS 8, carried out in 2017, respondents were again asked to rate all presidents during their term, but the list was expanded to include president Michel Temer, who had replaced Dilma after she was impeached and suspended from office in May 2016.Footnote ⁵ Finally, in BLS 9, respondents rated only Jair Bolsonaro.

The structure of these questions introduces some heterogeneity into our measurement. First of all, some chief executives were rated while they were in office or only shortly thereafter, while others were evaluated years after leaving the Palácio do Planalto. Ratings of presidents in the first situation can be reasonably be treated as accurate. Retrospective ratings, however, are probably influenced by postpresidential careers, which vary considerably. All five presidents who were rated retrospectively were assessed twice, once in 2013 and once in 2017. However, Itamar Franco was rated nearly 20 years after leaving office and two years after his death in 2011. José Sarney and Fernando Collor were also rated more than two decades after leaving the presidential palace and while holding Senate seats; Fernando Henrique Cardoso was rated just over a decade after the end of his second term; and Luiz Inácio Lula da Silva was rated three years after leaving the presidency.Footnote ⁶ Neither Lula nor Cardoso held electoral office at the time, but both were active in politics; and Lula, in particular, clearly harbored future electoral ambitions (which came to fruition in 2022).

Hence, we have contextually informed ratings for Bolsonaro (BLS 9), Temer (BLS 8), Dilma in her first term (BLS 7), and Dilma less than a year after being forced from office (BLS 8). These are our main contemporaneous estimates and should be noncontroversial. We also present our estimates for the other presidents, which are retrospective and should be treated with a grain of salt. For these five presidents, we report the first available rating (from BLS 7) as it was computed closer to their time in office.

Figure 2 shows estimates for all presidents, as well as the positions of each president’s party at the time of their mandate (or for two technically unaffiliated presidents, the parties most closely associated with them; see online appendix C for all estimates).Footnote ⁷ This figure suggests that even our noncontemporaneous estimates of the positions of presidents are quite plausible. Sarney is the only president substantially more to the right of his own party. This seems reasonable because Sarney had been a core supporter of the authoritarian regime, led the defection that created the PFL, and only joined the Party of the Brazilian Democratic Movement (PMDB) for legal reasons to become Tancredo Neves’s running mate in the 1985 indirect election. Although formally a member of the PMDB, Sarney was initially perceived as a stalking horse for the PFL, and therefore represented a political group that was considerably more conservative than the rest of the PMDB.

Figure 2. Ideological Positions of Brazilian Presidents and Their Parties

Notes: Figure shows estimates of the positions of presidents and the party more closely associated with them. Sarney, Itamar, and Temer are shown relative to the PMDB/MDB, FHC to the PSDB, Lula and Dilma to the PT, Collor to the PRN, and Bolsonaro to the PSL. *Indicates that presidents left office prior to the end of their term, and were replaced by the other president rated in the same year.

All other discrepancies are much more modest. Collor was rated as slightly more moderate than his own party (the defunct Party of National Reconstruction [PRN]), but still very far to the right. Itamar, originally Collor’s vice president, was rated as slightly more right wing than his longtime party (PMDB) and placed in a very similar position to that of his successor, Cardoso. This also makes sense given that Itamar is best remembered for having appointed Cardoso as minister for the economy in 1993 and having sponsored the latter’s successful presidential bid a year later. In Dilma’s second term she was rated as somewhat more centrist than the PT, but both the president and her party moved to the left in the legislature that was marked by her controversial impeachment. All other presidents were placed very close to where we estimated their own party to be during their respective mandates. Jair Bolsonaro is revealed, unsurprisingly, as the most extreme president that Brazil has had in the democratic era.

Position estimates from the Political Representation, Executives, and Political Parties Survey (PREPPS) project provide a reasonable benchmark for the validity of our estimates. PREPPS is an expert survey that measures positions of both parties and presidents in 18 Latin American countries (Wiesehomeier, Singer, and Ruth-Lovell Reference Wiesehomeier, Singer and Ruth-Lovell2021). It has a wider geographical coverage than BLS but covers a shorter time span. For the subset of Brazilian parties and presidents rated by both PREPPS and BLS, ideological estimates correlate almost perfectly ( $r = 0.97$ ). We provide additional details of this comparison in online appendix D.

Coalitional Presidentialism Under Stress

The Brazilian political system is a well-known case of coalitional presidentialism, though clearly not the only such case. While scholars disagree on the normative desirability of such a system, there is considerable agreement that it requires malleable politicians and/or a robust center to function.

We computed three simple indicators from the data presented in this note which, together, suggest that this system might be under severe and increasing stress. The first is simply the share of legislators belonging to clearly centrist parties, defined here as parties whose position estimates lie between $ - 0.25$ and $0.25$ .Footnote ⁸ The second indicator is the weighted average distance between all pairs of parties, which is a measure of polarization in Congress (higher values indicate greater polarization).Footnote ⁹ The third indicator is the weighted average distance between parties and the president, which is computed analogously to the previous one.

We report all three indicators in figure 3, for the nine legislatures for which we have survey data. We also report projected values of the three indicators for the current legislature (57th, 2023–26), for which we do not yet have BLS survey data. Hence, in this exercise we used the electoral results from October 2022 as the basis for the composition of Congress, and utilized the left–right estimates of the parties as recorded in BLS 9, which corresponds to the 56th Legislature (2019–22). We made two modifications to these estimates: we created an estimate for the newcomer party UNIÃO by averaging the positions of the DEM and the PSL in BLS 9, weighting each party by its size at the time of the merger. We also modified the estimated position of the PL by averaging it with the position of Bolsonaro, as he himself and many of his close associates joined the PL after BLS 9. Finally, we used Lula’s retrospective evaluation from BLS 8 (the more recent estimate recorded for Lula) as his ideal point, which was based on the perception of Lula at the time of his second term.

Figure 3. Polarization in the Brazilian Congress (1990–2023)

Notes: Figure shows the share of legislators that belong to centrist parties—measured on the right vertical axis—and the average weighted distances between parties and the average weighted distance between the president and parties—both measured on the left vertical axis. Estimates for the first nine quadrennial legislatures (beginning with the 48th, which ran from 1987–90) were computed by the authors from BLS 1–9. Estimates for the 57th Legislature (2023–26) rely on extrapolation of previous data. See text for details.

Figure 3 shows that the run-up to Dilma’s impeachment in 2016 marks the onset of a period of severe institutional stress. The three directly elected presidents in this period have been, by far, the ones most ideologically distant from Congress. Temer, who replaced Dilma, was the only president in the period to have substantial support in Congress, despite being extremely and consistently unpopular with the public. In lockstep with the increasing distances between presidents and the median legislator, we also see a steep rise in polarization and an emptying of the political center.

While figure 3 is sobering, we note that political polarization is not unambiguously bad. Polarization implies greater differentiation among parties and, as such, improved identifiability by voters (Lau et al. Reference Lau, Patel, Fahmy and Kaufman2014). However, the increasing abandonment of the center creates particular cause for concern because it suggests that compromise might be harder to achieve. Presidents may attempt to counteract this centrifugal tendency via clientelism and the exchange of resources—thereby co-opting opposition parties on an ad hoc basis—but this strategy may prove insufficient if preference differences are too pronounced. The consequences of this new configuration might not have been felt during Bolsonaro’s presidency, as he lacked clear legislative ambitions. However, polarization casts a long shadow on Lula’s third government.

Decades ago, a similar atmosphere of polarization was the key ingredient for the democratic breakdowns in Brazil and in the Southern Cone. Facing a choice between democratic compromise or ideological maximalism, political elites opted for the latter. The silver lining, today, is that as long as all relevant actors are committed to democracy, the current scenario of polarization might provide the incentives for politicians to cool the political temperature and to seek to attenuate conflict and accommodate competing interests. The test is indeed stressful, but the outcome is not a foregone conclusion.