PARTY DISCIPLINE IN THE BRAZILIAN CONSTITUTIONAL CONGRESS

The question of party discipline in Brazil is examined with reference to the Constitutional Congress of 1987-88. The authors do not find evidence of a strong party discipline and examine the consequences of this. More studies such as this are urgently needed, they argue.

This paper analyzes party discipline in the Brazilian constitutional congress of 1987-88, focusing on roll call votes in 1988.Because of the large number (1,021) of roll call votes during the constitutional congress and the availability of an excellent data base (Ames and Power 1990), the Brazilian constitutional congress offers an opportunity for one of the most detailed studies that has been conducted of party discipline in a Third World legislature.
Very little systematic work has been done on party discipline in Latin America. 1 The lack of information on party discipline in Latin America is a product in part of the fact that many countries in the region have little history of democracy and therefore have usually had at best a subordinate legislature that was not necessarily worth studying.In addition, most Latin American countries have little tradition of taking roll call votes even under democratic conditions.But few analysts have examined party discipline even where it is possible to do so.We help address that lacuna.
We begin with a discussion of how we have calculated discipline scores, given some distinctive features of the Brazilian party system and the constitutional congress.These measurement issues are important; discipline scores vary according to how one measures, and the comparability of lack thereof across cases and over time hinges on measurement.We argue that with multiparty systems it is better to use a fixed opposition threshold in deciding what roll calls to include in the analysis.We indicate how we have adjusted for legislators who switched parties or went on leave, new legislators who replaced those who took leaves, and parties with one legislator.
Substantively, we show that the biggest Brazilian parties of this period were comparatively undisciplined, and we also show that the leftist parties were a powerful exception to this general tendency.We demonstrate that legislators who switched parties during the constitutional congress were more likely than others to be undisciplined before switching and that their discipline increased markedly after their move to new parties.Finally, we attempt to explain why discipline was low in all but the leftist parties.
By party discipline or party unity-we use the terms interchangeably-we mean the extent to which members of the same legislative party vote together in highly contested roll call votes.(We define highly contested votes as those in which a minimum of 25% of the legislators present vote against the winning side.)A disciplined party is one in which members of congress vote together in highly contested roll calls.Even in undisciplined parties, legislators of the same party usually vote together, but this is because many legislative matters are consensual across and within party lines.On issues that cleave the legislature as a whole, undisciplined parties are often divided.Understood in this fashion, discipline does not always imply that legislators of a party act as a cohesive unit except when an issue comes to a floor vote.Some legislators may 1 An important exception is Figueiredo and Limongi (1995).
ultimately vote with the majority only after many concessions, whether in the form of patronage or policy.
Party discipline is an important issue in the study of parties and legislatures.In the electoral setting, as well as within legislatures, parties can be seen as unitary actors with welldefined goals (Downs 1957), as complex organizations with different factions (Panebianco 1988, Kitschelt 1989), or as an agglomeration of individuals.The usefulness of such assumptions is conditioned by the degree of party discipline; it makes sense to conceptualize parties as unitary actors only if the legislative party is disciplined.
Party discipline influences how legislatures function and how executives and legislatures interact.The importance of party discipline has long been recognized for parliamentary systems (MacRae 1967;Sartori 1994, 189-94).Party discipline also affects presidential systems, though in more equivocal ways.With comparatively undisciplined parties, support for the government can become less predictable and stable, and presidents may have difficulties forming stable bases of support.Party leaders can negotiate a deal only to have the party's legislative members back out of it.Presidents are sometimes forced to rely on ad hoc bases of support rather than counting on party leaders who can deliver the votes of their fellow legislators.On the other hand, weaker discipline makes it more likely that the president will be able to win the support of individual defectors from opposition parties for a particular bill.Conversely, with disciplined parties, presidents can negotiate primarily with party leaders, thereby reducing the number of actors involved in negotiations.They can count on party leaders usually being able to deliver the votes of most representatives of their parties, so there is greater predictability in the political process.
Party discipline is also connected to patterns of democratic representation.With disciplined parties, the party per se is the primary vehicle of representation.With less disciplined parties, individual politicians or factions more than national party organizations represent citizens and organized interests (Converse and Pierce 1979).In November 1986 Brazil had elections for 487 federal deputies (the entire lower chamber) and 49 senators (i.e., two of the three senators in 23 states and all three in the Federal District, which elected senators for the first time).Along with 23 incumbent senators (one from each state) who had served four of their eight-year terms, these newly elected members of congress were encharged with creating a new constitution.Table 1 gives the acronyms, names, and thumbnail sketches of the parties in the constitutional congress.Source: Kinzo (1990, 108).

Choosing a Threshold for Controverted Votes
In analyzing party discipline, most scholars have limited the analysis to issues on which a legislature as a whole is meaningfully divided.Many decisions in most legislatures are highly consensual.In the ANC, for instance, 309 of the 1,011 roll calls for which we had information had an opposition level under 3.0%.Including such bills would skew the results because unity extends across party lines.When consensual votes are considered, parties appear to be highly united because the assembly as a whole is united.
The question is how to select which roll calls to include in assessing party discipline.This issue is important because the criteria for inclusion affect the levels of party discipline as measured.In our view three criteria are important in multiparty systems: 1) It is important to exclude consensual votes but without establishing an unreasonably high threshold, which would excessively limit the number of roll call votes subject to analysis and tend to indicate deceptively low intraparty unity.By threshold, we mean a level of opposition by which we regard a vote as highly contested and thereby include it in the analysis. 3 We would have preferred to analyze the entire post-1985 period, but data are not publicly available.
2) If possible, the data should be somewhat comparable to those of other countries.The advantages of being able to compare are significant; only through comparison can we confidently assert that certain parties are relatively undisciplined.
Several different criteria have been used for inclusion in studies of roll call voting.In Britain a common procedure has been to use 'party votes' in which 90% or more of the members of one major party oppose at least 90% of the members of the other (Lowell 1912, Vol. II, 71-100;Cox 1987, 21-27).This criterion is useful in understanding how often a legislature divides along party lines, but it is not helpful in assessing party discipline.By including only roll calls in which at least 90% of both parties vote together, this criterion eliminates roll calls in which party discipline sags and thereby eliminates most variance in the dependent variable.
The country for which most work has been done on party discipline is the United States, where one common approach has been to consider roll call votes in which at least 50% of one party opposes at least 50% of the other (Brady 1973;Anderson, Watts, and Wilcox 1966). 4Roll calls that meet this criterion are called party unity votes.This criterion is useful both for assessing how frequently the legislature is divided and for measuring party discipline. 5 3) For a multiparty context, we preferred a fixed threshold, that is, an opposition threshold beyond which a vote would be considered highly contested and would be included in our analysis.Unfortunately, this criterion conflicts with the second one.The criteria for a US party unity vote are most applicable in cases of legislative two-or two-and-one-half-party systems.In multiparty situations, the roll call votes that meet the criteria for inclusion would shift according to which pair of parties is being analyzed.This problem is illustrated in Table 3 below, based on a hypothetical distribution of votes on a given roll call.Parties A and B meet the US criterion for a party unity vote, but parties A and C do not since the majorities of both parties are on the same side.As a result, the discipline score for a given party depends on what party it is paired with.Without a fixed threshold, there is no unique score for party discipline in multiparty contexts.This is also the criterion used by the Congressional Quarterly Almanac.

5
There is some debate about how to measure discipline in the US congress.This debate goes beyond our present purposes.For a refreshing recent challenge to the often-used US criteria of a party unity vote, see Cox and McCubbins (1993).
With a fixed opposition threshold, a roll call vote is either included or excluded depending on whether a certain percentage of the congress opposes a bill.This makes individual legislators' loyalty scores and party discipline scores fixed rather than contingent on which pair of parties one takes.
We ultimately opted for a 25% threshold because a lower one would have reduced comparability with the US and a higher one would unduly limit the number of roll calls.That is, we analyzed only votes in which a minimum of 25% of those present opposed the winning side.Some comparability with the US case is advantageous because far more work has been done on party discipline in the American legislature than any other.Having one case with which to compare our data gave us a reference point for judging how disciplined the parties were.Moreover, since Brazilian parties are supposedly undisciplined, we wanted to compare this case with another one known for less disciplined parties.Compared to parties in the other advanced industrial democracies, US parties are undisciplined.Therefore, if Brazilian parties are less disciplined than their American counterparts, it supports the argument that they are undisciplined.
To assess the comparability of our 25% fixed threshold with US party unity votes we measured discipline in roll calls in which at least 50% of the PMDB opposed at least 50% of the PFL.Using these two parties, the party unity vote proved to be a more demanding criterion than the 25% threshold; exactly 200 roll calls qualified as party unity votes, compared to 363 that met the 25% minimum.Nevertheless, we empirically verified (though we do not show the evidence here) that discipline scores using a 25% threshold did not differ substantially from those using the party unity vote criteria.

Individual Loyalty and Party Discipline: Defining the Dependent Variables
Two units of analysis have commonly been used to assess party discipline.Some studies have analyzed discipline for each roll call that is included and compute a party discipline score by taking the mean of these scores for the roll calls. 6Other studies have used individual legislators as the unit of analysis (Brady 1973, chapter 3;Anderson et al. 1966, 54).This approach takes the percentage of times that a legislator voted with the majority of his/her party in highly contested roll 6 Several different indicators have drawn on this basic approach.The Rice (1925) index of party cohesion takes the percentage of a party that supports the majority position within the party on a given roll call and subtracts the percentage who oppose the majority.The index of absolute party cohesion (Derge 1959) gives the percentage of legislators who vote with the majority position in the party; it counts absences as voting against the party.The index of relative cohesion (Anderson et al. 1966, 36) also gives the percentage of party members who vote with the majority of the party, but it uses only those legislators who actually voted.The party's discipline score then equals the mean for all relevant roll calls.Cox and McCubbins (1993, 139-61) focus on party leadership votes in which Democratic and Republican House leaders are each united but oppose one another.calls (or in party unity votes).Party discipline scores are then the mean of the individual loyalty values for all legislators of the party.
Both approaches are useful for different purposes.Focusing on roll calls allows the researcher to analyze the issues on which parties are more disciplined.However, we were more interested in results based on individual legislators, so we calculated loyalty scores for individual members of congress rather than for roll calls.Using legislators as the unit of analysis enabled us to trace some trends that we could otherwise not have, such as the relationship between discipline and party switching.We were also interested in variations in discipline across states and in the effect of federalism on discipline.For these purposes, using individual legislators as the unit of analysis was essential.

Relative and Absolute Loyalty
Relative discipline scores are based only on roll calls in which a legislator is present.If s/he took part in 93 roll calls, and voted with the majority of the party 90 times, the score is 97% ((90/93) x 100).In contrast, absolute loyalty scores include absences; the index is calculated over the total number of highly contested roll calls regardless of whether the person was present.A failure to vote then lowers individual loyalty scores.The absolute loyalty score has one advantage over the relative score: in a disciplined party legislators not only vote with the party majority when they vote (the information tapped by the relative loyalty score), they also show up to vote.Nevertheless, we computed both scores because both pieces of information are valuable.
In computing relative loyalty scores we considered abstentions as absences (that is, we simply ignored them) if the majority of the party voted no but as voting against the majority if it voted yes.The reason for this differential treatment has to do with a procedural rule in the ANC.
The practical effect of an abstention during the amendment phase was exactly the same as a 'No' vote because an absolute majority of the entire constitutional congress (i.e., 280 votes) rather than a majority of some quorum was needed to approve an amendment to the draft that was in place in January 1988.Therefore, when most of the party voted for a measure, an abstention meant effectively voting against the majority.
When most of the party voted against a measure, an abstention effectively supported the majority.The reason for not including abstentions in such cases is that it is not clear whether the abstention means supporting or breaking rank with the majority.If a legislator abstains while most of his/her colleagues vote no, the intent and perception may be closer to voting against the majority rather than voting with it.However, counting all abstentions as votes against the majority of the party might overstate the extent to which an abstainer breaks rank with the majority of his/her party.Most often, an abstention was somewhat ambiguous; it neither indicated full support for nor full opposition to the majority position.In any case, the number of abstentions was low-the mean abstention rate was 2.7% in the first period and 1.7% in the second-so a different coding would have had little effect.Of course, for absolute loyalty scores, an abstention always counted as a failure to vote with the party.
When a party was perfectly divided, we counted the 'nays' as the majority and the 'yeas' as the minority.Individual loyalty scores are slightly affected by this random procedure, but only under extraordinary circumstances would it affect the party score more than marginally. 7

Leaves, Absences, and New Members
A Brazilian legislator may take a leave from congress to assume some other position, usually in the state or federal executive branch, as the head of a major public enterprise, or as mayor.It would prejudice results to consider someone on leave as absent and therefore as failing to vote with his/her party.In order for the absolute loyalty scores to be useful, we needed to deal with legislators who took leaves or died and with suplentes, substitutes who take a legislators' place when the latter take a leave or die.We needed information on leaves that began between 28 January 1988 and 2 September 1988 and on suplentes who took these seats during this time.
We uncovered information about 31 individuals who voted during part of but not the entire amendment phase.We were able to get the exact arrival and departure dates for 19 of the 31 cases and approximate dates for the rest. 8In the latter cases we examined when a legislator had stopped voting or started to vote and used those dates in calculating loyalty scores.It was unreasonable to expect that new suplentes would immediately start voting in the ANC because these individuals were holding other jobs and were not necessarily residing in Brasília.For this reason, rather than using the date that they officially joined the ANC, in computing absolute loyalty scores we used the date when the new suplente actually started to vote. 9

Adjusting for Party Switching and Single Member Parties
We treated the periods from 28 January to 24 June and from 25 June to 2 September separately because of extensive party switching after 25 June.In late June 1988 some fundamental changes in the composition of the legislature occurred (compare columns 2 and 4 of Table 2).Disgruntled with their party, some three dozen PMDB legislators, mostly from the Center-Left faction of the party, defected along with a handful of other legislators and created the 7 It would significantly affect the party score only if 1) one or more legislators participated in very few votes, hence his/her/their individual loyalty score would be substantially altered according to how a tie vote was tabulated, and 2) the party had few legislators, such that one member's loyalty score would substantially affect the party's discipline score.

8
We used three sources: Coelho and Oliveira (1989); Departamento Intersindical de Assessoria Parlamentar (1988); and unpublished data from the Chamber of Deputies.9 With one exception, the longest delay we discovered in voting in the ANC after assuming office was eleven days.The exception was Fadah Gattas, who never voted in the ANC even though our sources reported him as becoming a member on 6 August.
Center-Left PSDB (Party of Brazilian Social Democracy).The party's foundational meeting took place on 25 June 1988.Before that date the incidence of party switching was insignificant.
Therefore, until 24 June, we took the member's initial party affiliation in computing loyalty scores.
After 25 June it was inappropriate to use the member's initial party affiliation because the creation of the PSDB began a process of extensive party switching.By the end of the ANC a net 70 of the 305 initial PMDB legislators had flocked to different parties (see Table 2).Including a member of another party in the discipline score for the PMDB would not make sense.Moreover, we wanted to be able to derive a discipline score for the PSDB since it was the third largest congressional party from the time of its inception.Because we had no means of finding precise dates when all 81 cases of party switching occurred, we assumed that 25 June 1988 was the date for all switches during the ANC. 10   We eliminated single-member parties because unless the single member were absent or abstained in every single highly contested roll call, the only feasible relative loyalty score would be 100.The same problem vitiates absolute loyalty scores.Dropping the members of singlemember parties for the first round eliminated three individuals and left us with 561 individual loyalty scores.For the second round we did not compute loyalty scores for two members who had no party affiliation and three others who belonged to parties with only one congressional.

Party Discipline in the ANC
Table 4 gives an array of information on party discipline in the ANC.There are two sets of scores for each party: the top one for Period I and the bottom one for Period II.Of the 1,011 roll calls in the Ames/Power data base, 363 or 35.9% of the total met our criterion for a highly contested vote; 246 occurred by 24 June (Period I) and 117 took place afterwards (Period II).The table is ordered from the most to the least disciplined party, using the relative loyalty mean for Period I.The relative and absolute loyalty scores are highly correlated (r = .97and .95for the two periods).The number of party members (column 2) does not always match the number of seats in Table 2 because if a member went on leave or died, a suplente would replace him/her.In this situation, one seat would produce scores for two members.

10
The reality was more complex; some legislators switched parties before that date and others after.Although our assumption does not correspond perfectly to reality, of Kinzo's net 83 party switchers, 48 joined the PSDB.Forty-five of them did so around the time of the party's inaugural meeting (Kinzo 1993, 49).We relied on two sources of information on party switching: Coelho and Oliveira (1989) and Departamento Intersindical de Assessoria Parlamentar (1988).These sources identified 81 individuals who switched parties.This is not an exhaustive list, but it comes close to matching the net 83 cases suggested by the aggregate number of members in different parties (see Table 2).The mean first period relative loyalty score was 71.9%, while the average absolute loyalty score was 53.1%.The former figure indicates that the average member of the ANC voted against the majority of his/her party 28.1% of the time in highly contested votes.The absolute loyalty score indicates that the average member of the ANC failed to vote with the majority of his/her party almost half the time.For Period II the mean relative score was 73.1% and the average absolute score was 50.9%.
For most parties discipline differed only marginally in the second period compared to the first.The most notable change from the first to the second period is an increase in the PDT's relative score, caused by the defection of three conservatives who brought down the party's scores in the first period.We had hypothesized that discipline would be slightly higher in the second period, expecting that the parties would be more disciplined after some dissatisfied members departed.But the evidence does not support this hypothesis.
For both periods the interparty differences in discipline are stunning.Whereas the average member of the PT voted against the majority of his/her party only 2% of the time, the average member of the PMDB did so 35% of the time.
The last column of Table 4 shows the absentee rates of the parties, i.e., the mean percentage of times that its legislators failed to vote.Here we include abstentions as voting.
Politicians of leftist parties were much more likely to vote than politicians of other parties, suggesting that leftist legislators were more committed to legislative affairs.(Power 1993 has shown that the Right was especially likely to have high absentee rates.)Again, the magnitude of the differences is noteworthy.During the second period, members of the Center-Right PDC (Christian Democratic Party) were 20 times more likely to be absent than PC do B (Communist Party of Brazil) members.
Table 4 also reports the standard deviations for the loyalty scores.This information is not particularly relevant for the very small parties (since few members could deviate from the mean) or for the very disciplined ones (since by definition, legislators could not deviate much from consistent loyalty).For the relative scores the PDT easily had the highest standard deviation for the first period-the three conservative members who more often than not voted against the majority switched to conservative parties before the conclusion of the ANC, with two joining the PFL and one the PTB.The PDS, PFL, and PL also had high standard deviations.In the second round, the PDT's standard deviation dropped noticeably because of the defections of the conservatives.The other parties did not experience sharp fluctuations in standard deviations from period 1 to period 2. Because of large individual variations in absentee rates, standard deviations were consistently higher for absolute discipline scores than for relative discipline scores.
The largest three catch-all parties (PMDB, PFL, and PDS) had wide internal variations in loyalty levels.Before the creation of the PSDB, the relative loyalty scores of PMDB members ranged from 42% to 100%; those of the PFL from 29% to 93.7%; and of the PDS from 29.5% to 89.8%.In contrast, the lowest loyalty score for a PT member was 94%.For the second period the relative loyalty scores of PMDB members ranged from 38.5% to 100%; of PFL members, from 25% to 91.7%; and of PDS members, from 24% to 93.9%. 11The lowest PT score was 92.7%.

Ideology and Party Discipline
The four leftist parties had very high discipline scores.There is a large gap between them and the catch-all parties (PMDB, PFL, PL, PTB, PDC, and PDS).The PMDB's position at the bottom of the scale was not surprising.The PMDB entered the ANC as a highly heterogeneous party, with prominent factions ranging from the Left to the Right of the ideological spectrum.Its internal squabbles prompted the resignations from the party of 152 of the 305 members of the congress by the time of the October 1990 congressional elections.In between the highly disciplined leftist parties and the undisciplined catch-all parties was the PDT.Since it is a Center-Left party, we expected that its discipline score would fall somewhere between the leftist parties and the rest.The absolute loyalty scores show the same general pattern: highly disciplined leftist parties, weakly disciplined catch-all parties, and the PDT in between.
In Table 5, we group parties by clusters to show the relationship between ideology and party discipline.The evidence again shows that the leftist parties were far more disciplined than the others.
Based on Kinzo's (1990) data, we constructed a Left-Right scale, ranging from -10.0 to 10.0, where -10.0 is the farthest Left position and 10.0 is the farthest Right.Kinzo (1990, 128-32) created several scales based on voting behavior in the ANC in order to assess parties' programmatic positions.Two of those scales are particularly useful for our purposes.Her scale of conservatism (ranging from 0 for the Left to 10 for the Right) distinguished between "more conservative and more reformist positions on social rights, property distribution and economic order."Another scale (ranging from 0 for the most conservative to 10 for the furthest Left) measured opposition to private banking.Our ideological score for each party equals its value on the conservatism scale minus its value on the financial scale.The farthest Right position (with a 10.0 conservatism score and a 0.0 antibanking score) would get a 10.0 value, while the farthest Left position would have a -10.0 value.The results, shown in Table 6, are consistent with the 11 For the PMDB we excluded Ulysses Guimarães who, as president of the ANC, almost always abstained as a matter of principle.Abstentions drove his absolute scores down to 0.8 and 3.8 for the two periods.ordinal scale in Table 5, with two minor oddities: most observers considered the PDS to the Right of the PFL and the PSB (Brazilian Socialist Party) to the Right of the PT.
N = number of congresspersons (cases) for each cluster.a 256 cases for relative scores (two members never voted).
Note: Does not include data for three legislators who were the only members of their party for period I.For period II does not include three legislators who were the only members of their party and two who had no party.The data in Table 6 reinforce two conclusions.First, the Left was markedly more united than the Right: the Left showed loyalty means near 100%, while the centrist and rightist parties are all under 80%.This finding supports the classic thesis (e.g., Duverger 1954;MacRae 1967, 55-57) that leftist parties are more disciplined.A correlation between ideology and unity scores shows that parties to the Left are more disciplined (r = -.79;p.< 0.005).Second, once again the scale is not linear.Discipline scores drop sharply after the PT and again somewhat after the PDT, but the lowest score is found in the centrist PMDB.

Party Switching and Party Loyalty
We hypothesized that legislators with low loyalty levels in Period I would be more likely to switch parties, and conversely that those who switched would be more likely to have low loyalty levels.The mean Period I scores of the 81 members who switched parties were 65.1% (relative) and 50% (absolute) compared to 73% and 53.6% for the 480 members who did not switch parties.However, bivariate analysis is inadequate for testing this proposition because members of leftist parties were simultaneously more loyal and more likely to stay with their parties.Because the members of leftist parties had a much lower propensity to switch out of their original parties-none did so-the lower mean loyalty of party switchers could be related to the characteristics of their parties rather than of individual politicians.To deal with this situation, we introduced a control for the original party affiliation to see whether party switching was related to low loyalty.Table 7 shows the results.The data generally support the hypothesis that those who switched parties would have lower initial loyalty levels, but with sharp variations by party.PFL, PDS, and PDT defectors had much lower first-round scores than party loyalists.The five PDT members who switched parties had a mean relative score of 52.6% in the first period, compared to a mean of 91% for the 21 PDT members who did not switch parties.In the PMDB, however, the differences were negligible (and in the wrong direction).
Two hypotheses might explain the motivations of party switchers.One is that party switching was related to policy conflict: legislators might leave their party because of policy differences.In this case, their Period II loyalty scores should be higher than their Period I scores.
The other possibility was that some politicians might be inclined to vote as individual agents regardless of what party they were in.In this case, switching from one party to another would represent an attempt to gain personal benefits (better access to patronage, better committee assignments, etc.) with no predictable increase in their loyalty scores after moving to a new party.
Therefore, if the loyalty scores of party switchers did not increase (compared to the whole ANC-for which there was virtually no change from Period I to Period II), it would support a view that party switching was related more toward obtaining individual benefits than toward ideological issues.The mean loyalty scores of legislators who switched parties increased substantially during Period II when they were in their new parties, to 79.9% (relative) and 62.1% (absolute).This suggests that for a substantial number of cases, policy differences were an important motivating factor for party switching.Because Brazilian politicians are patronage oriented (Ames 1995; Hagopian 1996), we were surprised at how robust this finding was.
These period II scores were markedly higher than the mean for the 480 politicians who did not switch parties.The five politicians who switched to the PSB, the five who joined the PDT, and the five who left the PDT showed particularly large increases.

Interstate Differences in Party Discipline
Brazil is a country of huge interstate differences and inequalities.In 1985, for example, the state of São Paulo had a per capita income 6.8 times greater than that of the poorest state (Piauí), and the south and southeast are much more developed than the north, northeast, and center-west.Historically and to this day Brazil's poorer states have been known for more clientelistic, patrimonial political practices and for weaker parties (Soares 1973).Given these differences, we hypothesized that levels of party discipline would be lower in the poor states.(1985).b Significant at a .05level.c Significant at a .005level.
The evidence proved to be mixed.We first looked at the bivariate relationship between a state's per capita income and its discipline scores.For period I there was no statistically significant correlation between a state's per capita income and the mean relative loyalty scores of its legislators.For absolute loyalty greater discipline was positively correlated (.35) to higher per capita income, but the correlation was significant only at the .10level.In Period II states with higher per capita income had more disciplined legislators.The correlation between the mean relative loyalty score of a state's legislators and its per capita income was .60,significant at the .005level.The correlation between per capita income and absolute scores by state increased to .53 (significant at the .01level) for Period II.
Because legislators of leftist parties were more disciplined and because the leftist parties were electorally stronger in some states (e.g., São Paulo) than others, the bivariate analysis, while interesting, is not an adequate means to examine the relationship between a state's per capita income and party discipline.We therefore ran a regression adding dummy variables for parties, so that the independent variables are per capita income in the state of the legislator and a dummy for each party except the PFL.(A member of the PFL would have the constant discipline score.) Table 8 shows the results.For Period I the relationship between per capita income and discipline is statistically insignificant.For Period II the results fit our expectation that states with higher per capita incomes would have more disciplined legislators.Both the relative and absolute scores are statistically significant at the .05level, but the coefficients are low.To make a one point difference in expected relative discipline scores for Period II a state needed, on average, a 33,333 increase in per capita income, measured in 1985 cruzeiros.The mean per capita income at the time was 76,468 cruzeiros, so a doubling of income would l have resulted in only a 2.3 point increase in expected relative discipline.
Party Discipline, 1989-94 Figueiredo and Limongi (1995) show that party discipline increased in the post-1989 period.
They analyzed 221 roll call votes in which at least 10% of the deputies opposed the winning side.They focused on roll calls rather than legislators as the unit of analysis.Although this reverses our procedure, the effect on party discipline scores should be marginal.Translating their Rice scores to our party discipline scores yields the results shown in Table 9. Party discipline increased substantially in the PFL, PMDB, PDS, and PTB.Figueiredo and Limongi correctly assert that these discipline scores paint a picture of greater discipline than some scholars had previously suggested.
What brought about this significant increase?We agree with Figueiredo and Limongi (1995) that internal congressional rules played an important part.In 1989 the Chamber of Deputies adopted internal regulations that led to greater centralization, and party leaders assumed greater control over legislative matters.This change gave party leaders more power over rank-and-file members.In addition, whereas the president had little formal capacity to set the agenda of the ANC, the 1988 constitution granted presidents considerable power over the legislative agenda.As a result the governing parties tend to bring issues to a vote mostly when they are confident of approval.Whereas the ANC needed to vote on issues regardless of the prior degree of consensus, the regular congress usually votes only after a consensus has been forged.Thus, the move to greater discipline does not necessarily imply that the traditional factionalism of Brazilian parties has faded; it is possible that the locus of disunity has shifted from being expressed on the floor to negotiations that precede the voting.Source: Calculated on the basis of Figueiredo and Limongi (1995, 506).

Assessing Party Discipline in Brazil: Some Comparisons
Because every legislature has unique rules and procedures, a comparison of discipline scores must be undertaken with caution.Nevertheless, such comparisons can also generate some useful information.The best case for comparison with Brazil because it is the only case for which reasonably comparable data are readily available, is the United States.Between 1984 and 1994 the mean absolute party discipline score (in which absence/failure to vote counts the same as a disloyal vote) for Senate Democrats was 78%; for Senate Republicans, 77.2%; for House Democrats, 80.4%; and for House Republicans, 75.8%. 12In the ANC only the leftist parties had absolute loyalty scores as high as those for the US parties.The mean for the ANC (53.1% in Period I and 50.9% in Period II) was vastly lower than the mean for the US congress.On the other hand, Brazil's leftist parties had much higher absolute unity levels than the US parties.
Relative loyalty scores considerably narrow the gap between the ANC and the US scores because absentee levels in the US are lower.From information in the Congressional Quarterly

12
Calculation based on data from the Congressional Quarterly Almanac, various years We calculated the mean by taking the figures for each year and dividing by eleven (the number of years) rather than weighting each year by the number of roll call votes.
Almanac we computed relative loyalty scores.The mean relative scores for the 1984-94 period are 81.4% for Senate Democrats, 80.5% for Senate Republicans, 86.4% for House Democrats, and 80.3% for House Republicans, while the mean for the ANC was 71.9% in Period I and 73.1% in Period II.Yet it is revealing that the Brazilian centrist and conservative parties could not promote greater participation at such an important time as the ANC.Even using the relative scores, the US scores are markedly higher than scores for the four major centrist and rightist parties.However, Brazilian discipline scores for 1989-94 were probably slightly higher than those for the US (bear in mind the different thresholds used).
Compared to parties in the other advanced industrial democracies, US parties are undisciplined.The norm in the advanced industrial democracies is very disciplined parties (for summaries, see Collie 1984;Ozbudin 1970).These scores confirm that Brazil's centrist and conservative parties were very undisciplined during the ANC and remain comparatively undisciplined since then.
Although we do not have comparable scores from other Latin American countries, we know that parties in Venezuela and Argentina are much more disciplined than the catch-all Brazilian parties.Coppedge (1994) reports that in Venezuela roll call votes are rarely taken; instead, party leaders cast votes for the entire legislative party.Discipline is ironclad, and party members follow the decision of party leaders.Mustapic and Goretti (1992) and Jones (1997) report that there are few roll call votes in Argentina, but that parties are highly disciplined.In Uruguay, which by reputation has comparatively undisciplined and factionalized parties, roll calls mostly occur when the congress acts to maintain or override a presidential veto.Between 1985 and January 1995 there were 41 roll calls to override presidential vetoes in which at least 25% opposed the winning side.For these 41 roll calls, the mean unity scores were 99% for the leftist Broad Front, 93.1% for the centrist Colorados, and 92.2% for the centrist Blancos (Moraes and Morgenstern 1995).
Roll call votes are likely to be relevant primarily when the individual legislator has sufficient autonomy vis-à-vis the party leadership to vote against it.Where party discipline is ironclad, there is little reason to take roll call votes-and without roll call votes there is little empirical information for studying party discipline.Empirical analyses of roll call votes are most meaningful if discipline is less than austere.
The discipline scores of Brazil's catch-all parties during the ANC were low from yet another perspective.The mean relative score for the members of an assembly cannot be lower than the mean support level on all highly contested votes. 13For the whole ANC, on average, 62.8% of the 13 This is true unless there is a discrepancy in individual loyalty scores between those who often did not vote and those who almost always voted.A significant discrepancy of this kind could marginally alter the relationship between the mean support level on all controverted votes and the lowest possible mean loyalty score.
assembly voted with the majority on the 363 highly contested votes.This is the lowest possible mean relative loyalty score for members of the ANC.Given this mean support level in highly contested votes, at least 62.8% on average had to support the majority of their party.The actual mean relative scores (71.9% in the first period and 73.1% in the second) were not dramatically higher.
Still, the notion of weak party discipline must be contextualized in two ways.First, party labels were still somewhat meaningful predictors of how legislators would vote.If members of congress were randomly assigned to parties, one would expect the mean loyalty score for the entire ANC to be 62.8% since this was the mean support level on highly contested votes.
Second, even in a system characterized on the whole by low discipline, the leftist parties were extremely united.

Explaining Weak Discipline
Party discipline is largely a function of institutional rules that give legislators incentives to follow party leaders.Six institutional arrangements are of particular importance in shaping party discipline.In Brazil in 1987-88 all six favored comparatively weak discipline.
The first rule is the electoral system by which members of congress win their seats.Some electoral rules, such as closed-list proportional representation (PR), make legislators heavily dependent on party leaders since their election depends first and foremost on the order of the ticket drawn up by the party (Carey and Shugart 1995;Mainwaring and Shugart 1997).With closed list PR, party leaders determine in what order candidates are elected.In Brazil, in contrast, the electoral system encourages individual candidates to cultivate their own constituencies.Brazil uses an open list proportional system for the lower chamber.Citizens vote for one name on lengthy party lists. 14Seats are assigned first to parties, but the intraparty order of seats is entirely determined by the number of preference votes candidates receive (Katz 1986).Party leaders have no formal control over the order of election, and candidates have strong incentives to develop their own political constituencies (Ames 1995;Lamounier 1986;Mainwaring 1991;Samuels forthcoming).Senate races are also highly personalized.Senators are elected by plurality through a direct popular vote, and personal positions, skills, and organization outweigh party in senate races.Again, party leaders have little influence over the electoral outcomes.
Second, rules for candidate selection also affect party discipline.If central party leaders control the selection process, individual legislators owe their candidacy to the party and are more 14 Voters are allowed to choose a party rather than an individual, but except for PT voters, the substantial majority select an individual candidate.See Samuels (forthcoming).likely to toe the party line.Conversely, if politicians secure nomination through their own efforts, as they do with primary elections, they cannot be denied candidacy by party leaders.
In Brazil national party leaders had comparatively little power over candidate selection.
Until the promulgation of a new law in September 1995 that allowed parties to determine their own mechanisms of candidate selection, public law (specifically, some articles of the Organic Law of Political Parties of 1971) governed this process.Local party conventions were held either in municípios (which can only roughly be translated as counties) or, in the large cities, in submunicípio units called districts.These local party conventions were open to all party members, and membership was quite extensive.By 1995 the PMDB had 6 million members, the PFL 3 million, and the PSDB and PT 700,000 each, 15 out of an electorate of 95 million registered voters.
Local party conventions directly chose the list of names for municipal councilors and the candidates for mayor and vice-mayor.This arrangement gave party members complete formal control of candidate selection at the local level.
Local conventions elected delegates to the state party convention.Delegates at the state convention approved the list of candidates for state and federal deputies, senators, governor, and vice-governor.The state party convention included, in addition to delegates chosen in local conventions, all members who were federal or state deputies, senators, and incumbent members of the State Party Directorate.Finally, the state convention elected delegates to the national convention.The national convention formally determined the candidates for president and vice-president.
Even though conventions generally ratified agreements previously negotiated among top party leaders, what took place at local and state conventions was of decisive importance for internal party control.The relative power of party leaders hinged on how many delegates they controlled; as a result battles over local and state politics were intense.Popular input into candidate selection was crucial.By electing delegates at the local level, who in turn chose leaders at the state and ultimately the national level, party members shaped candidate selection at higher levels as well.
Thus far we have described the 'normal' process of congressional candidate selection.
Incumbent deputies bypassed this process because of a provision that further reduced party control over candidate selection and enabled politicians to act as free-wheeling entrepreneurs: the candidato nato (literally birth-right candidate).This was a rule by which federal deputies had the right to be on the ballot for the same position in the next election, unless the party leadership specifically met and vetoed a candidate.The candidato nato rule provided leverage to politicians who wished to operate free of the fetters imposed by party leaders.
Third, party leaders in congress control resources that give them leverage over individual members, this is likely to enhance discipline (Cox 1987;Cox and McCubbins 1993;Figueiredo and Limongi 1995).In Brazil the constitutional congress functioned in a decentralized fashion.
Party leaders in congress controlled only moderate resources vital to the political success of individual members; they had some control over committee assignments but little over patronage.
The increase in discipline after 1988 is, as noted, partly a result of congressional rules that gave party leaders greater leverage.
Fourth, the rules that shape when roll calls are taken also affect discipline.During the constitutional congress individual legislators were allowed to propose amendments, which then went to the floor.This decentralized system made it easy to get measures to the floor, so conflictual issues that had not been resolved by party leaders could reach the floor.
Fifth, ceteris paribus, presidential systems lack one incentive for party discipline found in parliamentary systems.With presidentialism members of congress are free to vote on individual bills without fearing that their refusal to support the government may cause the latter to fall.In a parliamentary system a no confidence vote not only brings down the government, it also affects the legislator directly, subjecting him/her to new elections and to the possibility that even if he/she is reelected, his/her party will not retain power, thereby adversely affecting the MP's access to patronage and influence (however limited) over policy.
Finally, federalism can affect party discipline if legislators are more attuned to local constituencies than to national issues.Brazil has a federal system that truly works as such; local and state politics are very important.Politicians build their careers around local and state politics (Abrúcio 1994;Ames 1995;Hagopian 1996;Nicolau 1996).
Whereas national party organizations, with the exception of the leftist parties, are weak, state organizations are slightly more robust.State governors often hold considerable sway over national legislators of their own party and coalition.This situation poses a question: to what extent do the centrist and conservative national parties lack discipline because of interstate conflicts?Are state-level organizations markedly more disciplined?
It is easy to devise an empirical test to verify the effect of federalism on party discipline in Brazil.We examined state-level discipline for the PMDB and the PFL.Rather than seeing whether a member voted with the majority of his/her national-level party, here we examine whether the member voted with the majority of copartisans from the same state.(The other parties did not have enough representatives from enough states to test the proposition.) As expected, state-level discipline scores are markedly higher than national level scores.
For Period II the relative discipline scores increase by about seven points for both parties: from 64.7% to 71.8% for the PMDB and from 76.5% to 83.1% for the PFL.On the other hand, at least for the ANC, governors did not exercise the level of control over state delegations that is sometimes (e.g., by Abrúcio 1994) asserted.Most of the governors-22 out of 23-at this time were from the PMDB, which was the least disciplined party.If governors controlled their parties, the PMDB would have manifested greater discipline among state-level copartisans.In sum, federalism creates the possibility that some state copartisans will follow local interests and dissent from the national party majority, thereby lowering national discipline.Nevertheless, state delegations were not vastly more disciplined than national parties.
This conclusion about the impact of federalism on party discipline was supported by the results from using an anova model to see how much of the total variance in individual loyalty scores was explained by party alone, then by party plus state, and finally by the interaction of state and party.Adding state to party made a statistical difference, but state explained much less of the variance than party.The results are shown in Table 10. a Values in cells represent the proportion of the total variance explained by the anova model.All models are significant at the 0.0005 level.b The F value for the additional term is not significant at the 0.001 level.

Conclusions
Detailing the consequences of weak party discipline goes beyond our objectives in this paper, but it seems obvious that this pattern has had implications for democracy in Brazil.
Presidents cannot consistently rely on national party leaders to deliver the vote of their copartisans.Rather, they must often win the support of factional leaders or governors, both of whom exercise influence over individual legislators, or they must win the backing of individual members of congress.Party labels are not totally reliable indicators of the political preferences of individual members of congress (Novaes 1994).
We close with a plea that analysts start to look more systematically at party discipline in countries where it is less than ironclad.Despite the recent emergence of many fine studies on Latin American legislatures and executive/legislative relations, almost no systematic work has been done on party discipline (for an exception, see Figueiredo and Limongi 1995).Whether or not parties are disciplined has important implications not only for executive/legislative relations, but also for patterns of representation.Detailed studies of party discipline are needed especially when parties do not act as unified blocs, but at present we do not even have good information about where this is the case.
After 18 years of democracy Brazil had a military regime from 1964 to 1985.The military government initiated a slow and gradual process of political liberalization in 1974, and over time the opposition gained ground.In 1984-85 opposition forces coalesced with defectors from the regime to form a coalition that won the indirect presidential election of 1985, thereby ending 21 years of military rule.The 1967 constitution, designed by the military regime and substantially amended in 1969, remained in place even after a new civilian president took office in March 1985.Some of the patently authoritarian measures of the 1967 constitution were revised during the later period of military rule and the first two years of democracy.In November 1985 the congress approved Constitutional Amendment #26, convoking a constitutional congress to be elected one year later.

TABLE 1 Political Parties in the Brazilian Constitutional Congress PMDB
Capitalizing on its image as the party responsible for the transition to democracy and on an economic plan that suppressed (albeit artificially) inflation in the months preceding the 1986 elections, the centrist PMDB (Party of the Brazilian Democratic Movement) won an overwhelming victory and ended up with 305 of the 559 seats (54.6%) in the constitutional congress (ANC, Assembléia Nacional Constituinte).The rightist PFL (Party of the Liberal Front), the junior partner in the governing coalition, won 134 seats (24.0%) and was easily the second largest party.The remaining 120 seats were divided among eleven parties, with none winning more than the rightist PDS's (Democratic Social Party) 37 (6.6%).2Thepartisancomposition of the congress after the November 1986 elections is shown in Table2.Because the constitutional congress operated as a unicameral body of 559 members, we have collapsed the number of seats for the two chambers.
Party of the Brazilian Democratic Movement.Its precursor, the MDB (Brazilian Democratic Movement) was created in 1966 as the official party of opposition to the military regime.Renamed the PMDB in 1979.Since its inception, the PMDB has been a heterogeneous party, but its hegemonic group is centrist.Many conservatives flocked to the party after 1982.PFLParty of the Liberal Front.A conservative party created in 1984 by dissidents of the PDS.PDS Democratic Social Party.Despite its name, a conservative party.The PDS and its predecessor, Arena (1966-79) provided the partisan support for the military regime.PDT Democratic Labor Party.A populist party with predominantly social democratic tendencies.Created in 1979.PTB Brazilian Labor Party.A predominantly Center-Right party.Created in 1979.PT The Workers' Party.The most important leftist party, ranging from some revolutionary groups to social democrats.Created in 1979.PL The Liberal Party.A conservative party known for its antistatist discourse.Created in 1985.PDC Christian Democratic Party.Created in 1985.Unlike Christian Democracy parties in many countries, the Brazilian party is on the center-right of the political spectrum.Merged with the PDS in 1993 to form the PPR.PC do B Communist Party of Brazil.The most Leninist of the leftist parties in Brazil.Created in 1962 as a schism of the Brazilian Communist Party.

TABLE 2 Party Composition of the Brazilian Congress, Feb. 1987 and Sept. 1988 F
e b r u a r y 1 9 8 7 S e p t e m b e r 1 9 8 8 a Number of party members gained or lost during the constitutional congress.
The constitutional congress began its deliberations on 1 February 1987.It was expected to produce a new constitution by the end of 1987, but debates over many issues were highly contentious.It functioned in a decentralized manner with extensive space for public input.This procedure generated a sense of openness, but it contributed to extensive delays in the process.

TABLE 4 Party Discipline in the ANC
b Includes one legislator each from the PMB, PJ, and PRT, and two legislators who had no party.c Relative scores calculated over N = 562 (two members never voted).