Introduction

Since the 1990s, Latin American countries have moved toward greater transparency and participation in politics (Reference FisherFisher 1998; Reference Hagopian and MainwaringHagopian and Mainwaring 2005; Reference MunckMunck 2004). As these states have become more democratic, they have devoted more attention to one of their citizens’ most urgent demands: the oversight of administrative and legislative activity (Reference AngélicoAngélico 2012; Reference Berliner and ErlichBerliner and Erlich 2015; Reference MendezMendez 2015; Reference MichenerMichener 2010). However, citizens can only assess the quality of state governance effectively if they are provided with credible and comprehensive information. Hidden knowledge and hidden actions undermine the principal’s ability to monitor the agent’s behavior, and this problem is particularly acute in the public sphere (Reference Downs and RockeDowns and Rocke 1994; Reference MillerMiller 2005; Reference MoeMoe 1984; Reference NiskanenNiskanen 1971). Information asymmetries between policymakers and voters not only tilt electoral results in favor of incumbents but also lead to suboptimal outcomes in the provision of public goods. For instance, political actors can influence macroeconomic business cycles in ways unbeknownst to voters to increase their chances in future elections (Reference NordhausNordhaus 1975). Representatives may provide rent-seeking opportunities to special groups by introducing regulations that limit competition yet draw little attention from the general public (Reference TullockTullock 1967). Similarly, civil servants can impose significant welfare losses to taxpayers by maximizing their agencies’ budgets, as they have expert knowledge of the state cost functions and public finances (Reference Migué, Belanger and A. NiskanenMigué, Belanger, and Niskanen 1974; Reference NiskanenNiskanen 1971; Reference TullockTullock 1965).

Brazil has implemented a number of measures designed to reduce such practices. One notable example is the creation of orçamento participativo, participatory budgeting, which has fostered democratic control over public spending by encouraging citizens to engage in local fiscal administration (Reference Baiocchi, Heller and K. SilvaBaiocchi, Heller, and Silva 2008; Reference KooningsKoonings 2004). The Brazilian government has also improved and reformed accountability institutions such as the Comptroller General (Controladoria-Geral da União, CGU) and the Accounting Tribunal (Tribunal de Contas da União, TCU), which have increased the accountability of elected officials in all levels of government (Reference Praça and TaylorPraça and Taylor 2014; Reference SouzaSouza 2001). A more recent development, however, is the use of modern application programming interfaces (APIs) by Brazilian public agencies and government bodies. APIs are broadly defined as software protocols that allow machines to communicate with each other,Footnote ¹ and they have been largely responsible for the recent rise in digital ecosystems and the “internet of things” (Economist 2014; Wired 2013).

Although such initiatives deserve praise, the task of collecting and managing Brazilian administrative data remains beyond most users’ abilities. Groups interested in public records—such as journalists, social scientists, lawyers, or members of nongovernmental organizations—often do not have the computational skills required to interact with APIs. Even those who are familiar with that technology may find data cleaning a time-consuming task, and manual procedures for data preparation are generally burdensome and error prone (Reference Sandve, Nekrutenko, Taylor and HovigSandve et al. 2013). Consequently, many of the benefits of providing public information to Brazilian citizens may be lost if users do not have access to the data in a timely and convenient manner. Indeed, in this article we show how researchers, in a few minutes, can replicate a study that took the authors many months of arduous data collecting.

In this research note, we present congressbr, a package for the R statistical programming language (R Core Team 2015) that enables users to download data from the APIs of the Brazilian Federal Senate and Chamber of Deputies.Footnote ² With congressbr, we aim to fill some of this software gap in the social sciences and to lessen the workload normally necessary to collect such data. Although the same methods could be implemented in other languages, such as Python, Stata or C, we chose R because of its popularity in political science and its status as the de facto language of data analysis. There are currently over 12,500 user-contributed packages available through the R network,Footnote ³ and methods commonly used in political science have been available in R for many years (e.g., Reference Poole, Lewis, Lo and CarrollPoole et al. 2008; Reference Stuart, King, Imai and HoStuart et al. 2011; Reference Zeileis, Kleiber and JackmanZeileis, Kleiber, and Jackman 2008). It is also free software, has comprehensive documentation, and easily facilitates replication (Reference Baumer and UdwinBaumer and Udwin 2015; Reference TippmannTippmann 2015). For newcomers to R, we have provided the code necessary to replicate the analyses herein in the online appendix.

Our package is part of a larger movement that is bringing the data of Brazilian public institutions to citizens. For example, electionsBR (Reference Meireles, Silva and CostaMeireles, Silva, and Costa 2016) makes Tribunal Superior Eleitoral data available for users of R; GetHFData (Reference Perlin and RamosPerlin and Ramos 2016) downloads and prepares financial data from the São Paulo stock exchange; while Julio Trecenti has a number of R packages that interact with Brazilian government APIs, such as sabesp, which downloads and plots data from the São Paulo Water Management Company (Reference TrecentiTrecenti 2015), cnpjReceita, which retrieves information from the Brazilian Internal Revenue Service (Reference TrecentiTrecenti 2016), and spgtfs/sptrans, two packages that collect data from the São Paulo City Bus Management Service (Reference TrecentiTrecenti 2017). With specific regard to political science, there are various data sets that have been introduced and offered to the research community; good examples are Alvarez et al. (Reference Alvarez, Cheibub, Limongi and Przeworski1996), Linzer and Staton (Reference Linzer and Staton2015), and Lindberg et al. (Reference Lindberg, Coppedge, Gerring and Teorell2014). The main contribution of this work is to present an easily understood framework for downloading Brazilian legislative data directly into R. This same framework can also be useful as a guide to other researchers wishing to disseminate data from other countries in a similar manner, using publicly available data from APIs such as those utilized with congressbr, which can help foster replication and reproducibility in comparative politics.

While the use of congressbr does require some basic knowledge of R, its functions are, we hope, simple and intuitive for researchers of all levels of programming experience. Moreover, the returned data are in a “tidy” format (Reference WickhamWickham 2014); that is, all data are organized with variables as columns and cases as rows, no encoding incompatibilities, resulting in a final data set that is as “humanly readable” as possible. This means that users can easily export the results and analyze them within R itself or with other software and spreadsheet applications.

Exploring the Brazilian Houses of Legislature

Congressbr has a series of functions that search for the details of votes, individual legislators and commissions from the websites of the Brazilian Congress.Footnote ⁴ Our goal is to simplify the process of obtaining online information that may be used in both qualitative and quantitative analyses. To make the functions easy to memorize, we have adopted a consistent naming pattern for the package. Every function for the Senate starts with sen_ and all functions related to the Chamber of Deputies have the prefix cham_. As of version 0.1.3, there are over forty functions in congressbr, and all of them are described in the package manual. In order to make this section more concise, here we present the functions we believe researchers will utilize more often.

The Brazilian Congress is the source of the data contained in congressbr. It is composed of two legislative houses: the Federal Senate and the Chamber of Deputies. The Senate has eighty-one members, three for each state, elected for eight-year terms by majoritarian election. The Chamber has 513 members, proportionally elected according to the size of each state, and both houses play similar roles in the legislative process. A legislator typically participates in commissions and in the plenary, proposing, discussing and voting on different issues and the national budget. Depending on the type of the issue, in order to be approved, it must be discussed successively in both houses. Collecting data on the inner workings of such legislative bodies is a complex process and justifies reliance on official data.

Both the Brazilian Chamber of Deputies and Senate maintain APIs for the dissemination of data on bills, legislators, commissions, and the budget, among other topics. For those unfamiliar with the concept, APIs are protocols to facilitate the communication between different software programs. In the present case, Brazilian legislative houses provide documentation and protocols for downloading data from a structured server. For instance, one could simply navigate to a certain URL, using the browser, to receive the data in common formats. Instead of having to download these data directly from the API, congressbr implements methods to connect directly to the APIs, collect the data and load it into the local R environment. The Chamber has two API services: one located at http://www2.camara.leg.br/transparencia/dados-abertos/dados-abertos-legislativo, and a newer API (https://dadosabertos.camara.leg.br/) released in 2017. Congressbr connects to the older API only, as the latter is still undergoing construction and has not yet reached a stable state. We do, however, intend to utilize the newer API once it is fully developed. Regarding the API of the Federal Senate, located at https://www12.senado.leg.br/dados-abertos, we have implemented all the methods available on the API. Both the APIs for the Chamber and Senate return data in XML and/or JSON format, and congressbr transforms the data requested to the data frame format, the common R data structure more suitable for handling tables. Since the package is hosted on the standard R network, CRAN (https://cran.r-project.org), to install and load it, we need only the following code:

Users may familiarize themselves quickly with a certain house or legislature by using congressbr. For those new to Brazilian politics, typing statesBR into the R console will return a table of Brazilian states by name and two-letter acronym, which is useful as many requests to the API may be filtered using these same two-letter acronyms. Likewise, sen_parties() will return a data frame of the parties in the Senate, including those that have come and gone.Footnote ⁵ A quick look at the resulting table tells us that a total of forty-seven parties have been present at one time or another in the Federal Senate. Given the bewilderingly large number of Brazilian parties, we suggest newcomers start here. Another useful function is sen_senator_list(). By default, it returns a data frame of the senators currently serving in the Senate. This table may be requested for a certain state, for periods other than the present legislature, and by whether the senator is titular or a suplente (stand-in deputy). This table contains the gender of the senator, meaning that an analysis of the breakdown of gender in the Federal Senate is as easy as typing sen_senator_list() and then using an appropriate summary function, for example with the table() function natively available in R, which will print the following to the R console (here we also include the R commands necessary to reproduce this table):

The function sen_senator_list() also returns information on senators’ party membership, mandate term, and webpage and email information. For example, a journalist or an NGO member looking to quickly gather all the emails of the currently serving senators would only have to use this function to extract this data in a matter of seconds. Congressbr also contains a number of other functions that allow users to quickly familiarize themselves with the data. The function sen_bills_list(), for example, will return a table of the types of bills possible in the Senate, along with their numeric IDs and acronyms, whereas sen_bill_sponsors() produces a table of the sponsors of bills in the Federal Senate, showing that Senator Paulo Paim currently leads the way in the number of bills sponsored with a total of 310.Footnote ⁶ The rich detail of the API of the Federal Senate means that we have been able to make a plethora of such functions available in congressbr: information on senators’ mandates may be had with sen_senator_mandates(); sen_plenary_leaderships() returns data on leadership status in the plenary; mandate data may be had with sen_senator_mandates(), and sen_budget() returns a data frame of information on the budget proposals that have passed through the House. For the full list of functions, we refer readers to the package documentation.

Another important aspect of the package is voting data. The voting behavior of legislators is an area of great interest both inside and outside of academia (e.g., Reference AmesAmes 1995; Reference AttinaAttina 1990; Reference Poole and RosenthalPoole and Rosenthal 2000; Reference SnyderSnyder and Groseclose 2000). Congressbr has two functions that describes voting patterns: cham_votes(), which returns a data frame of votes in the Chamber of Deputies; and sen_votes(), which does the same for the Senate. We should note, however, that these are not necessarily nominal votes, as some may be secret. In this case, the API simply records whether the legislator voted or not.

Cham_votes() returns values such as the summary of the decision (returned as the variable named decision_summary), the guidelines given by the government and the opposition to the members of their respective coalitions (GOV_orientation and Minoria_orientation) and how political parties directed their deputies to vote. A researcher could choose to analyze, for instance, how the PSDB (Partido da Social Democracia Brasileira), the PSOL (Partido Socialismo e Liberdade), and other parties oriented their members on certain votes, using variables such as PSDB_orientation and PSOL_orientation, respectively.

The function requires that users provide the type, number, and year of the bill in question. The type parameter accepts four entries: “PL” for law proposal (projeto de lei), “PEC” for constitutional amendments (projeto de emenda constitucional), “PDC” for legislative decree (decreto legislativo), and “PLP” for supplementary laws (projeto de lei complementar). Unfortunately, a particular bill can have more than one roll call, and the API does not provide a way to readily identify these repeated roll calls. We have therefore provided the variable rollcall id in the returned data frame, which is a unique identity number (ID) for each roll call. For instance, we can retrieve information about proposition 1992/2007 with the command cham_votes(type = “PL”, number = 1992, year = “2007”). This will return a table of 2056 observations on 31 variables. The second column, decision_summary, contains a useful summary of the vote. We can access it with some simple R code:Footnote ⁷

This will print out the following: [1] “Aprovada a Subemenda Substitutiva Global oferecida pelo Relator da Comissao de Seguridade Social e Familia, ressalvados os destaques. Sim: 318; nao: 134; abstencao: 02; total: 454.”; this tells us that bill 1992/2007 was approved by 318 votes to 134. The table returned also shows that the government directed deputies to vote yea, while several parties, from both the right and left sides of the political spectrum, ordered their members to vote nay.

We could also see how many legislators voted against their party on this vote. Taking the PT as an example, the following R code tells us eight PT (Partido dos Trabalhadores) deputies voted against their party (note we use the filter() function from the dplyr package [Reference Wickham, François, Henry and MüllerWickham et al. 2017]). We exploit the unique roll call ID and simply filter the data:

The result gives us eight against the party and sixty-eight with. The function that downloads votes from the Senate, sen_votes(), works similarly. It provides variables that pertain to the time of the vote, its number, ID, year, description and the result of the roll call. Information on individual senators (their party, name, ID, gender, and the state they represent) is also returned. This function has a binary argument that if TRUE transforms the recorded (nominal) votes from yea to 1 and nay to 0, which is useful for any following quantitative analysis. Please note that dates are in yyyymmdd format, that is, to query the API for votes on September 8, 2016, we type: sen_votes(“20160809”). This returns a table of 405 rows and 16 columns. Supposing we named this object sen, typing the line unique(sen$vote_round) into the R console shows us that, on this day, bills went through up to five separate rounds of voting.

The package also contains some ready data sets. The data set sen_nominal_votes(), for instance, returns a data frame of votes in the Federal Senate. Cham_nominal_votes() returns all the votes, by legislator, for the Chamber of Deputies from 1991 to 2017 with a few other columns—attributes such as party, state, and bill ID, among others. We here use this data set to calculate some simple statistics. Figure 1 shows the number of parties in the Chamber of Deputies by year.Footnote ⁸ Readers may note the high level of party fragmentation in Brazil and that this fragmentation has grown worse in the last fifteen years.

Figure 1 Number of parties by year in the Chamber of Deputies, 1991–2017.

Analysts can also use congressbr for collecting details on individual legislators and on commissions, both rich sources of information. Details on individual senators can be obtained with the sen_senator() function. For example, sen_senator(id = 391)Footnote ⁹ will return details on Senator Aécio Neves, and will show us that he was born in March 1960 in Belo Horizonte and that he joined the PMDB (Partido do Movimento Democrático Brasileiro) in 1988. Similar information for other senators is easily available, and is useful not only for qualitative analyses but also for adding description to quantitative work.

The Senate API also contains information on coalitions and commissions. The function sen_commissions() will return a table of details on the commissions in the Senate. (For reasons of space, only columns 3 and 4 are shown here).

We can also see which senators serve on certain commissions. For example, the function sen_commissions_senators(code = “CCJ”) will return the senators who serve on the Commission on Citizenship and Justice (Comissão de Cidadania e Justiça). Temporary coalitions may also be of interest. Sen_coalitions() will return a table of the coalitions on the Senate, including specific ID numbers for each coalition. These ID numbers can then be used to get more detailed information on that particular group. For example, 200 is the ID of the bloco moderador.Footnote ¹⁰ With sen_coalition_info(code = 200), we can get more complete information on the bloc, for example, on its date of formation, January 2, 2015. Table 1 shows an example output of the function.

Table 1 The output of function sen_commissions().

The APIs of the Chamber and in particular the Federal Senate are rich in details such as these, and more detailed information may be found in the package documentation.Footnote ¹¹ We hope that congressbr can help scholars interested in qualitative studies of the Brazilian houses of legislature to more easily access this data, regardless of computer programming experience.

Producing Legislative Statistics

Congressbr also allows researchers to produce ready and direct summaries of legislative data. Much of the political science literature on the Brazilian houses of legislature has frequently utilized certain summaries of behavior (e.g., Reference Figueiredo and LimongiFigueiredo and Limongi 1995, Reference Figueiredo and Limongi1999), and we hope that the package can help such political scientists create these summaries more easily.

For example, during the 1990s, the practice of empirically analyzing the Brazilian legislature began to grow in popularity and sophistication. Argelina Figueiredo and Fernando Limongi, pioneers in this area, studied party cohesion in the Congress and had to collect their data by hand, a time-consuming and labor-intensive process. They then used these data to analyze patterns of legislative votes, discovering that parties in the Brazilian Chamber are quite cohesive, contrary to previous findings in the literature (Reference Figueiredo and LimongiFigueiredo and Limongi 1995). Congressbr allows researchers to replicate these findings in a matter of minutes by collecting data from the API and employing a few R functions. For example, an important statistic, employed by the authors and often used in the political science literature to measure party cohesion, is the Rice Index (Reference RiceRice 1928; Reference DesposatoDesposato 2005).Footnote ¹² Calculating this measure consists of taking the absolute value of the subtraction between the number of yea votes and nay votes, and dividing this by the absolute number of votes. In mathematical terms:

$\mathrm{Rice}= \frac{\left(\mathrm{yea}-\mathrm{nay}\right)}{\mathrm{yea}+\mathrm{nay}}$

To do this in R, one can write a simple function, where votes is a numeric vector of recorded votes (traditionally coded as 1 and 0 for yea and nay, respectively):

This function can then be used to calculate the Rice Index for the vote data that is returned from the functions in congressbr.Footnote ¹³ Figure 2 shows the historical evolution of the Rice Index for the three major parties in the Chamber of Deputies—the PMDB, the PT, and the PSDB—using votes downloaded with congressbr. The result is consistent with the view of Brazilian political history found in the literature. The PT has always been a party whose members are known for being staunch defenders of its ideology, whereas the PMDB are well-known “kingmakers” who excel at building coalitions and are not famed for ideological purity. The PSDB can be considered to be somewhere in-between.

Figure 2 The Rice Index for major parties in the Chamber of Deputies, 1990–2010.

Spatial Models of Voting Behavior

Another important application of legislative data is for use with spatial models of legislative voting (Reference PoolePoole 2005; Reference Clinton, Jackman and RiversClinton, Jackman, and Rivers 2004). Analyses with spatial models usually focus on “ideal points,” that is, the positions legislators take relative to one another on a scale formed by their voting records. Examples of such ideal-point analyses with Brazilian legislative data include Desposato (Reference Desposato2006) and McDonnell (Reference McDonnell2017). These types of analyses can be easily carried out with congressbr. In order to facilitate these types of large-N nominal vote studies, we have included two data sets of nominal votes in the package, one for each legislative house, beginning in 1991 through to early 2017. The following example uses the Senate data set, which may be loaded into R with the command data(“senate nominal votes”). The votes have been coded 1 for yea and 0 for nay and abstentions.

A popular way to model voting behavior utilizes Bayesian item-response theory (IRT) (Reference Bafumi, Gelman, Park and KaplanBafumi et al. 2005; Reference Clinton, Jackman and RiversClinton, Jackman, and Rivers 2004; Reference Martin and QuinnMartin and Quinn 2002). Bayesian IRT models estimate the probability of a yea vote (y = 1) as a latent regression:

$P\left(y=1\right)=\phi \left({\beta }_j{x}_i-{\alpha }_j\right),$

where x_i is the ideal point of senator i, and β_j and α_j are the discrimination and difficulty parameters of bill j.Footnote ¹⁴ The ideal points of the senators may be estimated in various ways in R; researchers can use probabilistic programming languages such as JAGS (Reference PlummerPlummer 2003) and Stan (Stan Development Team 2016), or specific R functions for ideal-point analysis, such as ideal() from the pscl package (Reference JackmanJackman 2015) or the IRT modeling functions from MCMCpack (Reference Martin, Quinn and ParkMartin, Quinn, and Park 2011).

For longer periods, or for when change over time is of primary interest, a dynamic ideal point model may be more suitable (Reference Martin and QuinnMartin and Quinn 2002). Here we take a subset of the data for speed, convenience, and to avoid complications with modeling the data over time.

Ideal-point analysis requires that the data be in a particular format. We have provided a convenience function, vote_to_rollcall(), for this purpose. By default it returns data suitable for use with ideal(), but it may also be used to structure the data in a format suitable for other R packages and the programming languages mentioned above.Footnote ¹⁵ We can then use this format to run the analysis and plot the results.Footnote ¹⁶ For this example, we used the MCMCirt1d function from MCMCpack.Footnote ¹⁷ This workflow then has three simple stages: (a) load or download the data with congressbr, (b) utilize the vote_to_rollcall() function on the data, and (c) run an analysis using the IRT software the researcher has chosen.

We here present an example of this workflow. Figure 3 shows the changes in ideal points for selected senators from an analysis of the Dilma Rousseff and Michel Temer administrations (we select only a few senators to avoid clutter in the plot). The left-hand side of the figure shows the senators’ ideal points as they were when Dilma Rousseff was president. Left-wing supporters of her administration are to be found on the negative end of the scale at the bottom of the plot, while the other senators (although all except Romario were part of her coalition) can be found some distance away from their left-wing colleagues, clearly showing the disharmony in the Rousseff administration.Footnote ¹⁸

Figure 3 Differences in senators’ ideal points between the Rousseff and Temer administrations.

On the right-hand side of the plot, we see evidence of the strong support some of these senators (Neves, Calheiros, Malta, Romario, and Jereissati) gave to the Temer government, whereas those who were part of the Rousseff coalition (Calheiros, for example), offered no such support to Rousseff. The two senators who opposed the impeachment process and maintained their support for Rousseff, Senators Grazziotin and Farias, display a notably different voting history. Also of note is the increased polarization seen after the impeachment, with the ideal points of each group coalescing separately, typifying closer coalitions.

Ideal points such as these can be produced quickly and easily from the voting records provided by congressbr. Data from the two houses may also be combined as in McDonnell (Reference McDonnell2017) to facilitate interesting comparisons over time and across the institutions. As may be seen from the R code in the online appendix, visualization of the results of this analysis produces the most verbose code—the actual requesting of data and its preprocessing before modeling necessitate comparatively few lines of R code.

Conclusion

We have introduced the congressbr package for R in this short research note. The purpose of making such a package is to put useful and interesting political science data in the hands of researchers. Our goal is to provide a suite of easy-to-use functions that even the novice R user can understand and use to produce analyses of Brazilian politics. This opens up the analysis of such data to more scholars than was previously possible, as studies such as those cited in the text have often been restricted to those with significant programming experience, or to those with the time and resources to collect data by hand.

In future versions of the package, we plan to include functions that download and standardize data from other levels of the Brazilian political structure, such as state and municipal legislatures. We believe that researchers will have their work greatly simplified with such an array of legislative data available with the use of only a few simple functions. We also believe congressbr can act as a useful guide for other researchers who wish to build similar packages for disseminating the data available in other countries. These types of APIs are usually quite similar in design, so that the format used by congressbr can be used for other similar APIs that make social science data available. As the source code of congressbr is freely available, researchers can copy the parts applicable to their case.

We hope users find congressbr useful for their research. Feedback and suggestions are greatly appreciated.