Restricted Candidacy and Political Competition:

Transcription

1 Restricted Candidacy and Political Competition: Evidence from a Policy Change in the Minimum Education Requirement for Legislators in Pakistan* Madiha Afzal, Yale University Abstract This paper looks at the e ect of a unique policy change in the minimum education requirement for legislators on political competition in Pakistan. I assume that political competition is increasing in voter turnout and number of candidates contesting election, and decreasing in the fraction of votes of the winning candidate and the vote margin of the winning candidate. Using panel data, I nd no evidence of any overall e ect of restricted candidacy on political competition. However, the signs on the coe cients of the political competition outcomes are consistent with reduced political competition in the constituencies where the incumbent was disquali ed as a result of the education requirement. In addition, closer examination reveals evidence of heterogeneity in the e ect of disquali cation on political competition: there was a sharp decline in political competition, in terms of a higher vote winning fraction and vote winning margin, as well as a decrease in the number of candidates contesting election and a decrease in the Her ndahl-based political competition index, in those areas where the incumbent was disquali ed and there was a smaller pool of substitute candidates. *Preliminary draft: Please do not cite. 1

2 "The aim of every political Constitution, is or ought to be, rst to obtain for rulers men who possess most wisdom to discern, and most virtue to pursue, the common good of society; and in the next place, to take the most e ectual precautions for keeping them virtuous whilst they continue to hold the public trust." James Madison (1788 [1961]), in the Federalist Papers (#57) 1 1 Introduction Thomas Je erson wrote of the need to form a "natural aristocracy" to hold the o ces of government 2, referring to a political class that was a cut above the public in terms of ability and talent. It is the same idea which underlies the rst part of the quotation above by James Madison, the primary author of the US constitution. And this was claimed to be the thought behind the constitutional amendment invoked by General Musharraf, President of Pakistan, just two months before the October 2002 general elections: in his Chief Executive Order No.7/2002, he imposed a minimum education requirement on all candidates running for the o ce of Member of National Assembly, the lower house of Parliament in Pakistan 3. In order to contest elections in 2002, all candidates for Member of National Assembly (MNA) had to le proof of graduating with a Bachelors degree or higher with their candidacy papers. This policy change had the e ect of disqualifying 60 of the 207 MNAs elected in the 1997 election, 1 From Tim Besley, "Political Selection", typescript, p.2. 2 From Tim Besley, "Political Selection", typescript, p Pakistan s political system is described in detail in Appendix 1. 2

3 29% of the National Assembly incumbents. Opponents in Pakistan claimed that this policy was engineered to reduce political competition in the country, and to bring to power proponents of the Musharraf regime. In particular, this policy restricted 97 percent of the country from running for Parliament, leaving only 3 percent of the population who were college graduates as eligible to contest national elections. Barriers to entry to the political process are in general controversial and typically discredited, in part because they are seen to reduce political competition, which is considered bene cial in terms of greater alignment of politician and public preferences and greater government responsiveness. The rst e ect, if any, of this change in the rules for political selection should be on political competition. The second e ect, if any, takes place once the winning candidates are in power and can in uence policy. However, we also know that changes in political competition a ect policy directly 4. Therefore, in the case of this unique policy experiment, if we can establish that there infact is no e ect on political competition, we can subsequently argue that any e ect on policy comes directly through the di erence in legislator education (or characteristics associated with education). Such evidence has never before been established because legislators education typically cannot be disentangled from voter preferences; this policy experiment allows for exactly that. In this paper, I look at the e ect of Musharraf s policy change on political competition, as measured by voter turnout, winning vote margin and fraction, the Her ndahl-based 4 See Besley and Burgess [2002]. 3

4 political competition index, and number of candidates standing for election, and leave policy outcomes as a topic to be examined in my future research. The empirical strategy employed in this paper is essentially a di erence-indi erence approach. The coe cient of interest measures the e ect of disquali - cation on a number of political competition outcomes, controlling for year and constituency xed e ects. Most obviously, one would expect this policy change to lead to a reduction in political competition, as in a standard barriers-toentry argument. However, in Pakistan there is reason to believe the opposite, albeit slightly convoluted, argument as well. If this policy change disquali ed long standing uneducated incumbent legislators whose presence had traditionally discouraged entry of educated candidates, political competition could increase with the in ux of many educated candidates into the political arena. Finally, there could conceivably be no change in political competition if there is perfectly smooth adjustment to the equilibrium level of political competition. The latter result would perhaps be most surprising of all, especially in a developing country with an unstable political system such as Pakistan s. However, we should note here that the welfare e ects of reduced political competition are ambiguous 5. The main empirical speci cation of this paper does not provide evidence of a signi cant e ect of disquali cation on political competition. However, the signs on all the political competition variables are consistent with reduced political competition in the constituencies where the incumbent was disquali ed as a 5 See Besley and Burgess [2002] and Bardhan and Yang [2004]. 4

5 result of this policy change. However, closer examination reveals evidence of heterogeneity in the e ect of disquali cation on political competition: there was a sharp decline in political competition in those areas where the incumbent was disquali ed and it was harder to nd a substitute candidate. The remainder of this paper is organized as follows. In the next section, I discuss the related literature. Section 3 outlines the empirical strategy. Section 4 discusses the delimitation of constituencies and GIS matching methods. Section 5 describes the data and provides basic summary statistics, and section 6 presents the econometric results. Section 7 concludes. 2 Related Literature Much of the theoretical political economy literature has typically focussed on the role of incentives in shaping politician behavior. Through recent work by Besley [2005] and Besley, Pande, and Rao [2005], the literature has started to look at political selection, a line of thought which argues that the characteristics of politicians determine how they perform in o ce. This is in line with the views of political thinkers and politicians alike, including James Madison and Thomas Je erson. Indeed, political selection seems to be the exact motivation behind Musharraf s education requirement. Infact, one of the ndings of empirical work by Besley, Pande, and Rao [2005] in India is that better educated politicians exhibit less political opportunism. Besley [2005] discusses how the design of political institutions determines politician selection. He also points 5

6 out that by restricting the domain over which candidates are selected, political reservation could make the pool of good candidates smaller; on the other hand, reservation may limit the extent of policy disagreement between competing candidates, making it more likely that political competence is a deciding factor in determining policy. The same argument can be applied to a discussion of restrictions on legislator candidacy. Recent evidence that politician identity in uences policy also shows that selection matters: Pande [2003] nds that reservation for scheduled castes and tribes positively a ects policy targeting towards these groups in India. Also in India, Chattopadhyay and Du o [2004] nd that women s issues get more attention when women politicians are elected due to reservation. In the sense that this policy change served as a shock to incumbency, it also relates to the growing literature on incumbency. Besley and Case [1995] showed that in the US, governors facing binding term limits increase taxes and expenditures. However, in our case, the shock to incumbency arising from disquali cation is unexpected, and took place when the term of the incumbents had already ended. In in uential work, David Lee [2003] showed a large causal electoral advantage to incumbency in the US in a regression discontinuity framework. Using a similar method, Linden [2004] provides evidence of an incumbency disadvantage in India. In Pakistan, at the national level at least, parties have been subject to an incumbency disadvantage as well, with the majority party switching in each of the four elections between This paper looks at the impact of Musharraf s policy change, which induced 6

7 restricted candidacy, on political competition, but does not comment on how altered political competition impacts welfare. Recent empirical work by Besley and Burgess [2002] and theoretical work by Bardhan and Yang [2004] relates political competition to economic and policy outcomes. Besley and Burgess show that greater electoral turnout and higher political competition is associated with better government responsiveness in the form of greater public food distribution and calamity relief in India. Bardhan and Yang show that greater political competition leads to tradeo s of economic costs and bene ts. In one of their de nitions of political competition which is most closely related to the one I use, they show that it can pose a threat to long term investments which feature uncertainty that extends across election cycles. In general, they argue that information asymmetries, distributional con icts, and the characteristics of public investment opportunities can play a role in mapping electoral competition into good or bad economic outcomes. Hence higher political competition does not necessarily translate into higher welfare. 3 Empirical Strategy and Identi cation The empirical strategy employed in this paper is essentially a di erence-indi erence approach. The coe cient of interest measures the e ect of disquali cation on a number of political competition outcomes, controlling for year and constituency xed e ects. In particular, the empirical speci cation is (the coe cient of interest is ): 7

8 P ct = c + t + D ct + " ct where: P ct : Political competition outcome in constituency c at time t. I look at four di erent political competition outcomes, namely, the fraction of votes of the winning candidate; the number of candidates; the vote margin of the winning candidate as a proportion of total votes; the Her ndahl-based political competition index; and a proxy for voter turnout. These variables are discussed in greater detail in Section 5. c : Constituency xed e ects. t : Year e ects. D ct : Disquali ed is de ned as a dummy equal to 1 for a constituency in 2002, if the MNA elected from that constituency in 1997 did not have a Bachelors degree or higher; it equals 0 for all constituencies in t = 1997, 2002 (election years). 3.1 Identi cation With this empirical approach, any level di erences between constituencies where the incumbent was disquali ed (where the law binds) and those where the incumbent was not disquali ed (where the law does not bind) are taken care of by constituency xed e ects. However, di erence-in-di erences does assume that in the absence of the policy intervention, political competition in the constituencies where the incumbent was disquali ed would have grown at the same rate as in constituencies where the incumbent was not disquali ed. I indirectly tested 8

9 this before the policy intervention, by comparing political competition growth rates 6 between 1993 and 1997 in constituencies hit by disquali cation in 2002 with constituencies not hit by disquali cation in I cannot reject equality of means; therefore, there were no strong pre-existing di erential trends in political competition between these two types of constituencies 7. 4 Delimitation of Constituencies Pakistan had 207 electoral districts, referred to as constituencies, for the four elections which took place between The nationwide population census, which took place in 1998 after 17 years, called for the need for delimitation of constituencies in order to account for the large increases in population which took place in the intervening period. Therefore, in 2002, the Election Commission of Pakistan (ECP) delimited the constituencies again and increased their number to 272 in accordance with the Delimitation of Constituencies Act, In order to use panel data, and at an even more basic level, in order to determine which constituencies in 2002 were a ected by the education requirement, I had to match the 1997 constituencies to the 2002 ones. The Election Commission does not have the maps in GIS formats required for such matching. Therefore, I obtained the maps in graphic formats from the Election Commission for Note that I look at political competition growth rates for the vote fraction, vote margin and candidate variables, but not for the turnout proxy since I do not have constituency population levels for 1993 (and the turnout proxy is not directly comparable to actual voter turnout). 7 The di erence in means t-test for vote fraction growth has a value of: jpj>t=0.19; for number of candidates growth the value is: jpj>t=0.39; and for vote margin growth the value is: jpj>t=

10 For 1997, I got the maps from a private consulting company, ECIL, which had been hired as a consultant rm by the ECP. I then manually constructed the shape les for both the 1997 and the 2002 maps in GIS by rst georeferencing the maps to have the same projection, and then digitizing the maps to have constituency boundaries (by drawing polygon lines through the maps) 8. I used GIS to intersect the 1997 and 2002 boundaries, and then calculated the exact area weights of these intersected areas relative to the 1997 constituency and 2002 constituency to which they belonged. I constructed a population-weighted average of the 2002 data corresponding to each 1997 constituency, as well as an area-weighted sum matching and an area-weighted average matching, and ran the regressions corresponding to each matching. Details of the construction steps in GIS are in Appendix 2. The preferred matching method for this paper is the population-weighted average, but below I discuss the exact matching formulas for all three methods. Appendix 3 contains the regression results corresponding to the two area-weighted matching methods. For illustration of the di erent matching methods, consider the following hypothetical example based on Figure 1: constituency D is a constituency in 1997, and constituencies A, B, and C are constituencies formed in 2002 after delimitation of constituencies. Constituency D consists of (intersected) areas d1, d2, and d3, part of the new constituencies A, B, and C respectively. 8 These maps are in Appendix 5. 10

11 4.0.1 Population-weighted average 9 : Data_D_in_2002 = P op_d3 P op_d_in_2002 Data_C10 where: P op_d1 P op_d_in_2002 Data_A + P op_d2 P op_d_in_2002 Data_B + 9 In general, population-weighted measures are preferred to area-weighted measures since the former give a larger weight to areas with the highest population (as is more logical), whereas the latter give a greater weight to regions with the largest area. In addition, weighted averages typically are preferred to weighted sums since the former account for the fact that there are multiple units. However, the weighted sum might be a better measure in cases such as the delimitation of constituencies where a unit (1997 constituency) is divided into smaller units (2002 constituencies), if these smaller units do not have as much political competition as a result of the fact that the area has become smaller - that is, although the population of the area has gone up, the number of candidates contesting election goes down since the pool of candidates remains more or less the same across elections in a given region. This does not appear to be the case in Pakistan: see the Summary Statistics section. Speci cally, note that the mean number of candidates contesting election was 8.07 in 1997, and 7.54 in 2002, a very slight reduction. 10 In the following analysis, Data_A, Data_B, Data_C, Pop_A, Pop_B, Pop_C refer to the political competition outcomes and population for these constituencies in

12 P op_d1 = Area_d1 Area_A P op_a P op_d2 = Area_d2 Area_B P op_b P op_d3 = Area_d3 Area_C P op_c P op_d_in_2002 = Area_d1 Area_A P op_a + P op_c Area_d2 Area_B P op_b + Area_d3 Area_C That is, I weighted each intersected area by its population in 2002 (assumed to equal the fraction of the intersected area relative to the 2002 constituency, multiplied by the population of the 2002 constituency) and then multiplied by the 2002 data for the 2002 constituency to which the intersected area belonged. I then added this weighted data for 2002 to correspond to the 1997 constituency, and then divided by population of 1997 constituency in 2002 (calculated with the same assumption by intersected area as above; the 2002 population of the 1997 constituency is used to account for population growth) Area-weighted average: Data_D_in_2002 = Area_d1 Area_D Data_A + Data_C Area_d2 Area_D Data_B + Area_d3 Area_D That is, I weighted each intersected area by the fraction of its area relative to the 1997 constituency, and then multiplied by the 2002 data for the 2002 constituency to which the intersected area belonged. I then added this weighted data for 2002 to correspond to the 1997 constituency. 12

13 4.0.3 Area-weighted sum: Data_D_in_2002 = Area_d1 Area_A Data_A + Data_C Area_d2 Area_B Data_B + Area_d3 Area_C That is, I weighted each intersected area by the fraction of its area relative to the 2002 constituency, and then multiplied by the 2002 data for the 2002 constituency to which the intersected area belonged. I then added this weighted data for 2002 to correspond to the 1997 constituency. The fact that delimitation and the education requirement both occurred at the same time raises possible concerns that these changes were not independent of one another, and more speci cally, politically motivated to bring into power proponents of Musharraf s regime. Despite the fact that I match the 1997 data to the 2002 data, this can be a problem since it could imply endogeneity of my independent variable, disquali cation. To test for orthogonality of disquali cation and delimitation, I run a simple correlation between the disquali ed dummy and a variable that signi es the extent of delimitation in that constituency: this yields a correlation coe cient of 0.01, and a signi cance level of 0.85, implying that these two variables are not correlated, thus validating my analysis The variable that signi es the extent of delimitation in the 1997 constituency is constructed as follows: for each intersected area, I calculated the fraction of its area relative to the 2002 constituency to which it belonged, and then added these percentages to correspond to the 1997 constituency. This variable signi es how many di erent constituencies the 1997 constituency was split into: when it is equal to 1, it signi es no change in the constituency; when it takes on a value greater than 1, it signi es that the constituency is split, and when it is less than 1, it signi es that the constituency now forms party of another constituency. In my data, the mean of this variable is 1.30, and the standard deviation is

14 Disquali ed Frequency Percent Total Table 1: Incumbents disquali ed in Data and Summary Statistics 5.1 Disquali cation I have data on the exact education levels for the 207 MNAs elected from general seats in 1997, as well as the 272 MNAs elected from general seats in For the MNAs elected in 2002, the information was part of the application for candidacy led with the Election Commission. For each of the MNAs elected in 1997, I asked each of the District Coordination O cers in Pakistan to report the education levels of all MNAs elected from the constituencies which fell within their district in I have de ned "disquali ed" as a dummy which is equal to 1 for a constituency in 2002, if the MNA elected from that constituency in 1997 did not have a Bachelors degree (or higher). As Table 1 shows, 60 out of 207 MNAs in 1997 were disquali ed. 5.2 Political Competition The Election Commission of Pakistan publishes a booklet containing the number of votes received by each contesting candidate in each constituency after 12 An o cer from the DCO s o ce obtained this information either through phone, or by personally going to the ex-mna s house. I am extremely grateful to the Home Department, Government of Punjab, for their invaluable help in getting this information. 14

15 every general election 13. I also obtained the population of each constituency in 1997 and 2002 from the Election Commission. From this information, I de- ned the following variables of interest for each constituency: the number of candidates contesting election; the fraction of votes of the winning candidate (votes received by the winning candidate divided by the total votes polled in the constituency); the vote margin of the winning candidate (votes received by the winning candidate minus votes received by the rst runner-up candidate, divided by the total votes polled in the constituency); a proxy for voter turnout (total votes polled in the constituency divided by the constituency s population) 14 ; and nally, the Her ndahl-based political competition index (de ned as one minus the political Her ndahl index, which equals the sum of squares of the vote shares of each candidate running from election in a constituency). I postulate that political competition (or alternatively, political contest) is increasing in voter turnout, the number of candidates contesting election and the Her ndahl-based political competition index, and decreasing in the fraction of votes of the winning candidate and the vote margin of the winning candidate. 15

16 Variable Observations Mean SD Min Max Vote fraction No. of candidates Vote margin Total votes polled Population Turnout proxy Her ndahl Competition Table 2: 1997 Summary Statistics Variable Observations Mean SD Min Max Votes fraction No. of candidates Vote margin Total votes polled Population Turnout proxy Her ndahl Competition Table 3: 2002 Summary Statistics 5.3 Summary Statistics Firstly, note that constituency 28 (in the Tribal Areas), which had 107 candidates contesting election in 1997, is an outlier 15. Note that I have dropped constituency 28 in all of the subsequent empirical analysis. Excluding constituency 28, the mean number of candidates standing in 1997 was 8.07, the SD 13 I obtained this booklet from the Election Commission directly for 2002 and entered the data. For 1997 and previous years, I am grateful to Ali Cheema and Asim Khwaja for sharing the data. 14 Note that actual turnout is total votes polled divided by the number of registered voters. Since I do not have the number of registered voters in 2002 for each constituency, I constructed this proxy for turnout. This proxy is valid as long as the demographic composition of the population is similar across constituencies. 15 This constituency was an outlier in terms of candidates standing for election, but not in terms of other political competition outcomes. The winning candidate in that constituency had 29% of the total votes polled in the constituency and his winning margin as a proportion of the total votes was 11%. 16

17 was 5.02, the minimum number of candidates running was 2 and the maximum was 35. The average number of candidates standing in 2002 was 7.54, a slight reduction [see Tables 2 and 3]. Similarly, the average of the percent votes of the winning candidate went down from 55% in 1997 to 49% in The mean of the vote margin of the winning candidate also decreased from 0.26 in 1997 to 0.17 in The turnout proxy decreased slightly from 0.23 in 1997 to 0.22 in Finally, the mean of the Her ndahl political competition index increased from 0.57 to 0.62, although the distribution shifted towards zero as both the minimum and the maximum decreased. Overall, therefore, political competition seems to have decreased on some dimensions and increased on others. Total votes polled increased from in 1997 to in 2002, primarily because the populations of the constituencies increased. Note that although the number of constituencies increased from 207 in 1997 to 272 in 2002 to account for the rise in population, the average size of the constituencies still increased from approximately 406,000 in 1997 to 489,000 in Results In the aftermath of this policy change, opponents claimed that relatives replaced the disquali ed members of Parliament, acting as puppets who blindly followed the policy preferences of the incumbents they had replaced. I manually matched the names of the MNAs in 2002 to the names of MNAs of the matched constituencies in 1997, identifying a 2002 MNA as a relative if two of 17

18 Disquali ed (0.065) Fixed e ects province R-squared 0.22 Obs MNA Relative of 1997 MNA Table 4: Relatives Regression Note: OLS regression with robust standard errors clustered by province in parentheses. * denotes signi cance at 10%, ** at 5%, and *** at 1%. his names (either his rst and last name, or last and middle name) matched those of the 1997 MNA 16. Simple cross-tabulations show that 14 relatives of the 60 disquali ed incumbents won elections in 2002, a percentage of Of the 147 incumbents quali ed to run for election in 2002, 29 were re-elected. Of the 118 incumbents who were quali ed to run for election in 2002 but who did not win in that year, 29 of their relatives won in 2002, equalling 19.49%. Therefore, although a higher proportion of disquali ed incumbents were replaced by relatives (23.73% versus 19.49%), this di erence is not signi cant. Running a simple OLS regression of an indicator for a matched relative on whether the incumbent was disquali ed (Table 4) also shows that there is no evidence to support the claim that relatives replaced disquali ed incumbents in Parliament; the sign of the coe cient, however, is consistent with this claim. The size of the coe cient shows that a 2002 MNA is 8 percent more likely to be a relative of a 1997 MNA in areas where the 1997 MNA was disquali ed. 16 This is in accordance with the custom of naming in Pakistan. In a few cases where the last name was uncommon, I matched on the basis of the last name alone. This matching method may not be perfect, but it only causes possible measurement error in the dependent variable. 18

19 Vote Number of Vote Turnout Her ndahl Fraction Candidates Margin Proxy Competition Disquali ed ** * (0.006) (0.415) (0.014) (0.006) (0.007) Fixed e ects province province province province province R-squared Obs Table 5: 1997 Cross Section Results Note: OLS regression with robust standard errors clustered by province in parentheses. * denotes signi cance at 10%, ** at 5%, and *** at 1%. For preliminary analysis, I ran both the 1997 and 2002 political competition outcomes in the cross-section on disquali cation. For the 1997 election, this cross-section regression re ects on initial di erences in political competition in constituencies with uneducated versus educated legislators (who would later be disquali ed versus non-disquali ed). For the 2002 election, this regression reveals di erences in political competition between constituencies with disquali- ed versus non-disquali ed incumbents. The 1997 cross-section results (Table 5) show that constituencies where the winner would later be disquali ed displayed high political competition in 1997, in terms of a lower winning vote fraction and a lower winning vote margin. This implies that uneducated candidates had to ght for their seats, and goes against the conventional argument that they exercised unopposed control over these seats. On the other hand, in the 2002 elections (Table 6), consituencies where the incumbent had been disquali ed displayed weakly lower political competition (in terms of a higher winning vote fraction and a lower her ndahl competition index) compared to those where the incumbent had not been disquali ed. 19

20 Vote Number of Vote Turnout Her ndahl Fraction Candidates Margin Proxy Competition Disquali ed 0.020* ** (0.006) (0.720) (0.018) (0.006) (0.007) Fixed e ects province province province province province R-squared Obs Table 6: 2002 Cross Section Results: Population-weighted Average Note: OLS regressions with robust standard errors clustered by province in parentheses. * denotes signi cance at 10%, ** at 5%, and *** at 1%. Note that both the 2002 cross-section and all the panel regression results shown in this section are based on the population-weighted average method of matching constituencies over time. Results based on the area-weighted sum and area-weighted matching methods reveal broadly the same patterns, and are discussed in Appendix 3. The main empirical speci cation of this paper, the basic panel regression (Table 7), shows no signi cant e ect of disquali cation on political competition. However, the signs of the coe cients on all the political competition variables are consistent with reduced political competition in the constituencies where the incumbent was disquali ed as a result of this policy change. In terms of magnitudes, these numbers are moderate. Vote share and vote winning margin increase by 4-5% as a proportion of total votes in constituencies where the incumbent was disquali ed; this implies that the disquali ed incumbent is not replaced by another candidate standing for election. Similarly, the e ect on the number of candidates is sizeable: exactly one less candidate runs from constituencies where the incumbent was disquali ed The e ect on turnout seems to be negligable. Finally, the her ndahl political 20

21 Vote Number of Vote Turnout Her ndahl Fraction Candidates Margin Proxy Competition Disquali ed (0.038) (1.060) (0.047) (0.011) (0.033) Year *** *** *** (0.020) (0.415) (0.025) (0.006) (0.017) Fixed e ects const const const const const R-squared Obs Table 7: Basic Panel Regression: Population-weighted Average Note: OLS regressions with constituency xed e ects and robust standard errors clustered by constituency in parentheses. * denotes signi cance at 10%, ** at 5%, and *** at 1%. competition index decreases by 0.04, a small amount. This is then consistent with the widely held opposition claims; however, we should note again that the welfare e ects of reduced political competition are ambiguous. The fact that we do not nd any signi cant e ect overall motivates us to look for some sort of heterogeneity in the e ect of disquali cation on political competition. One element of heterogeneity can arise from the substitutability of the disquali ed incumbent; that is, a measure of how di cult it is to nd a replacement for him. One would think that in areas where it is harder to nd a substitute for the disquali ed incumbent, political competition should decrease more severely. A proxy for substitutability can be party size. The assumption is that if there is an equal proportion of educated candidates in small and large parties, small parties will be less likely to nd an educated candidate to replace their disquali ed incumbent; this will reduce political competition. In the limit, the smallest party is an independent candidate. I de ne small party based on the number of candidates (and not just the 21

22 winners) put up for election by that party in Table 8 clearly illustrates this de nition 17. I de ne a party as large if it elds 50 or more candidates, and as small if it elds less than 50 candidates for election. Alternatively, this de nition means that a party is small if it elds candidates in approximately less than one-fourth of the total national assembly constituencies, which number 207 for This seems to be a natural cuto in the data, as can be seen in the following table. However, I also used three alternate cuto s for small party to check for robustness - whether the number of candidates elded is less than 10, 20, or 100. The regression results (Table 10) con rm this substitutability hypothesis: constituencies where the winner in 1997 was from a small party or was an independent candidate and was disquali ed face a subsequent reduction in political competition, in terms of winning vote fraction and winning vote margin, as well as the number of candidates and the Her ndahl-based political competition index 18. These results are robust to using the three alternate cuto s for small party. For the 10 and 20 candidate cuto, winning vote fraction and winning vote margin increase signi cantly where the winner in 1997 was from a small party and was disquali ed, but number of candidates do not.decrease signi - cantly. For the 100 candidate cuto, the results are slightly weaker (which may be explained by the fact that even some large parties are coded as small by this 17 The party a liations of the winning candidates in 1997 and 2002 are contained in Tables 20 and 21 in Appendix I have not included small party independently in the regression because the coe cient is just a linear combination of the xed e ects. 22

23 Party Candidates Percent Small Party ANP AQP BNM BNP HPG IND JUI(F) JUI(FG) JUI(S) JUI(SG) JUP JWP KJP KT MIP MKP(KBG) MQM(H) NPP NPP(WG) PAP PDP PK-MAP PMI PML(J) PML(N) PML(Q) PPP PPP(SB) PPP(ZAB) PSL PTI SI SNP TI TIP UNA WP One Candidate Parties Total Table 8: Small Party De nition 23

24 Vote Number of Vote Turnout Her ndahl Fraction Candidates Margin Proxy Competition Disquali ed * (0.010) (0.771) (0.027) (0.006) (0.006) Small Party (0.027) (1.999) (0.024) (0.022) (0.030) Small Party * Disquali ed (0.016) (1.372) (0.044) (0.016) (0.014) Fixed e ects province province province province province R-squared Obs Table 9: 2002 Cross Section Results with Small Party Interaction: Populationweighted Average Note: OLS regressions with robust standard errors clustered by province in parentheses. * denotes signi cance at 10%, ** at 5%, and *** at 1%. cuto ): the signs are still the same, but for winning vote fraction, our coe cient of interest becomes marginally insigni cant, whereas for winning vote margin, it is now signi cant at the 10% level 19. This reveals that there was indeed a real e ect of this education requirement on political competition: political competition decreased sharply in those areas where the incumbent was disquali ed and it was harder to nd a substitute candidate. Moreover, these numbers are huge in terms of magnitude; they reveal that winning candidates get an increased 19% of the total votes in these constituencies, gain an astounding 25% over the runner-up candidates, the number of candidates running for election decrease drastically - by 5 candidates, and the Her ndahl-based political competition index drops by I also use this de nition of small party but rede ne the substitutability measure - I use the number of candidates belonging to a small party instead of just 19 Results available upon request. 24

25 Vote Number of Vote Turnout Her ndahl Fraction Candidates Margin Proxy Competition Disquali ed (0.037) (0.853) (0.047) (0.009) (0.032) Year *** *** *** (0.020) (0.416) (0.025) (0.006) (0.017) Small party*disquali ed 0.188** * 0.253*** ** (0.077) (3.121) (0.082) (0.033) (0.069) Fixed e ects const const const const const R-squared Obs Table 10: Panel Regression - Party Interaction: Population-weighted Average Note: OLS regressions with constituency xed e ects and robust standard errors clustered by constituency in parentheses. * denotes signi cance at 10%, ** at 5%, and *** at 1%. the winner belonging to a small party; that is, a disquali ed incumbent is considered less substitutable if more of the candidates who ran in his constituency in 1997 belonged to a small party. This would be a more plausible measure if I had the education levels of the other candidates as well - as it stands, for useful inference it assumes that the other candidates were also disquali ed. The results are similar to results from above, but remain very strong for the 10% cuto 20. To the extent that the education of legislators in 1997 may be misreported, this biases my estimates downward, such that they are a lower bound on the true e ect. It should also be noted here that the GIS matching may not be perfect, since I constructed the GIS maps based on imperfect images 21. However, this can only cause possible measurement error in the dependent political 20 Again, the results are available upon request. 21 However, these are the best images in Pakistan at this point. 25

26 competition variables which were matched for Conclusion Contrary to popular opinion, the empirical analysis in this paper provides no evidence of any overall e ect of restricted legislator candidacy on political competition in Pakistan. It is not entirely clear whether this implies a perfectly smooth adjustment to some political equilibrium or whether reduced political competition is not being captured here because of noisy data. In any case, closer examination does reveal stark heterogeneity in the e ect of disquali cation on political competition: political competition declines sharply in those areas where the incumbent was disquali ed and there was a smaller pool of substitute candidates. There are a number of steps that could possibly be taken to re ne the analysis in this paper, but they are demanding in terms of data requirements. For one, disquali cation can be de ned over the entire pool of candidates as opposed to just the winner in However, this requires knowledge of the education levels of all of the candidates running for election in One should note that using a better measure of disquali cation will infact make the e ect on political competition larger: using my measure of disquali cation only serves to attenuate the e ect. Secondly, I can de ne a more direct measure for substitutability of candidates than the size of the party. I use small party as a proxy for a smaller pool of educated candidates in a party, but I can nd the actual pool of 26

27 educated candidates in a party in the party list, and therefore use this measure for substitutability. These are the next steps I hope to take in this research. I also plan to explore the e ect of this education requirement on policy outcomes in future research. 8 Bibliography Bardhan, Pranab and Tsung-Tao Yang, [2004], "Political Competition in Economic Perspective," mimeo University of California, Berkeley. Besley, Timothy and Anne Case, [1995], "Does Electoral Accountability A ect Economic Policy Choices? Evidence from Gubernatorial Term Limits," Quarterly Journal of Economics, 110(3), Besley, Timothy and Robin Burgess, [2002], "The Political Economy of Government Responsiveness: Theory and Evidence from India," Quarterly Journal of Economics, 117(4), Besley, Timothy, Rohini Pande and Vijayendra Rao, [2005], "Political Selection and the Quality of Government: Evidence from South India," typescript. Besley, Timothy, [2005], "Political Selection", typescript. Chattopadhyay, Raghabendra and Esther Du o, [2004], "Women as Policy Makers: Evidence from a Randomized Policy Experiment in India," Econometrica, 72(5), Lee, David S., [2003], "Randomized Experiments from Non-random Selection 27

28 in US House Elections", mimeo University of California, Berkeley. Linden, Leigh, [2004], "Are Incumbents Always Advantaged? The Preference for Non-Incumbents in India," Job market paper, Columbia University. Pande, Rohini, [2003], "Minority Representation and Policy Choices: The Signi cance of Legislator Identity," American Economic Review, 93(4), Appendix 1: Pakistan s Political System Pakistan has a parliamentary system of government. The legislative branch of government is the Parliament, which consists of the Senate (Upper house), whose 87 members are indirectly elected by the 4 provincial assemblies, and the National Assembly (Lower house), 272 of whose 342 members are directly elected from single seat constituencies using plurality rule. Of the other 70 seats, 60 are reserved for women and 10 for minorities, but women and minorities can stand from the 272 general seats as well 22. Candidates can stand for election from more than one constituency. Each party typically elds one candidate per constituency, and various independent candidates (not aligned with any party) contest elections as well. 22 These numbers are for 2002, post the delimitation of constituencies in response to the population census of For the elections, the National Assembly consisted of 207 general seats. 28

29 10 Appendix 2: Converting Image Files into GIS Shape Files I am extremely grateful to Abraham Parrish for his help and guidance throughout this task, and specially for getting me started. All the following steps were undertaken in ArcMap 9.0 in the Lambert Conformal Conic (LCC) projection. 1. I georeferenced the 2002 constituencies map for each province (originally a pdf le from the Election Commission of Pakistan website which I then converted into a tagged image format (tif) le, which is readable in ArcMap) to the province map (province selected from Pakistan s districts outlined in a shape le available from the ESRI database) using three or more control points and a rst order or second order polynomial transformation. Then I updated the georeferencing, and used rectify to save the georeferenced le. 2. I overlaid the province outline (merged the districts in the province shape le) onto the georeferenced tif le. In Edit Mode, I selected the entire province and cut polygon lines on the province outline le to draw the rst 2002 constituency. Then select the part you want to cut across to make a separate constituency and cut polygon lines; repeat the process until each constituency becomes a separate feature. This part requires you to be very careful because you can slightly move a feature while in edit mode and create spaces between polygons. You should also explode multipart feature and delete any islands. 3. Repeat step 1 for the 1997 georeferencing. Since the 1997 tif map is less precise than the 2002 one, I used the 2002 digitized map as the base, and kept 29

30 all boundaries that seemed the same (this was value judgment because of lack of precision of the 1997 map), cut any new boundaries and then merged old boundaries to form the 1997 constituencies. 4. Make an attribute table for each shape les and name the attributes. 5. Intersect the 1997 constituency boundaries with 2002 constituency boundaries using Arc Toolbox -> Analysis Tools -> Overlay -> Intersect (Wizard Steps). This tends to create extra slivers of areas. 6. Then export intersected le as well as 1997 constituency shape le and 2002 constituency shape le as geodatabases (to get feature/shape areas). Make sure you use the same coordinate system. Save as personal geodatabase feature classes. 7. To deal with the problem of slivers with the intersect command: Advanced Editing -> Explode multipart feature. Use Select by attribute to select the sliver features (those which have shape areas less than the smallest shape area that should be an actual feature) and delete them. 8. Join the 1997 geodatabase with the intersected geodatabase so that the areas of the 1997 constituencies are appended on to the attribute table of the intersected le (both called Shape_Area, need to create a new eld = Shape_Area of the 1997 constituencies). Remove Join. Do the same thing for Divide the intersected shape areas by the 1997 and 2002 shape areas to get the relative percentages. 30

31 Vote Fraction Candidate Vote Margin Turnout Proxy Disquali ed (0.008) (0.600) (0.016) (0.005) Fixed e ects province province province province R-squared Obs Table 11: 2002 Cross Section Results: Area-weighted Average Note: OLS regressions with robust standard errors clustered by province in parentheses. * denotes signi cance at 10%, ** at 5%, and *** at 1%. 11 Appendix 3: Area Matching Results 11.1 Area-weighted average matching results The 2002 cross section results (Table 10) reveal that in the 2002 elections, constituencies where the incumbent had been disquali ed displayed no di erence in political competition from those where the incumbent had not been disquali ed. This is di erent from the population-weighted average matching where constituencies with disquali ed incumbents displayed reduced political competition (in terms of a higher winning vote fraction) in The simple panel results (Table 11) reveal no signi cant e ect of disquali cation on political competition. However, as with the population-weighted average matching, the signs are consistent with reduced political competition in constituencies where the incumbent was disquali ed. Finally, the party interaction results (Table 12) con rm the substitutability hypothesis once again: constituencies where the winner in 1997 was from a small party or was an independent candidate and was disquali ed face a subsequent reduction in political competition. 31

32 Vote Fraction Candidate Vote Margin Turnout Proxy Disquali ed (0.039) (1.060) (0.046) (0.014) Year *** *** 0.011* (0.021) (0.432) (0.025) (0.007) Fixed e ects constituency constituency constituency constituency R-squared Obs Table 12: Basic Panel Regression: Area-weighted Average Note: OLS regressions with robust standard errors clustered by constituency in parentheses. * denotes signi cance at 10%, ** at 5%, and *** at 1%. Vote Fraction Candidate Vote Margin Turnout Proxy Disquali ed (0.039) (0.852) (0.046) (0.013) Year *** *** * (0.021) (0.433) (0.025) (0.007) Small party*disquali ed 0.208** *** (0.092) (4.515) (0.090) (0.053) Fixed e ects constituency constituency constituency constituency R-squared Obs Table 13: Panel Regression - Party Interaction: Area-weighted Average Note: OLS regressions with robust standard errors clustered by constituency in parentheses. * denotes signi cance at 10%, ** at 5%, and *** at 1%. 32

33 Vote Fraction Candidate Vote Margin Turnout Proxy Disquali ed (0.042) (0.577) (0.022) (0.013) Fixed e ects province province province province R-squared Obs Table 14: 2002 Cross Section Results: Area-weighted Sum Note: OLS regressions with robust standard errors clustered by province in parentheses. * denotes signi cance at 10%, ** at 5%, and *** at 1% Area-weighted sum matching results The area-weighted sum matching results parallel the area-weighted average matching results. The 2002 cross section results (Table 13) reveal that in the 2002 elections, constituencies where the incumbent had been disquali ed displayed no di erence in political competition from those where the incumbent had not been disquali ed. The simple panel results (Table 14) reveal no signi - cant e ect of disquali cation on political competition but the signs are consistent with reduced political competition in constituencies where the incumbent was disquali ed. Finally, the party interaction results (Table 15) con rm the substitutability hypothesis yet again: constituencies where the winner in 1997 was from a small party or was an independent candidate and was disquali ed face a subsequent reduction in political competition. 33

34 Vote Fraction Candidate Vote Margin Turnout Proxy Disquali ed (0.091) (2.211) (0.062) (0.033) Year * 0.042* (0.064) (1.591) (0.038) (0.023) Fixed e ects constituency constituency constituency constituency R-squared Obs Table 15: Basic Panel Regression: Area-weighted Sum Note: OLS regressions with robust standard errors clustered by constituency in parentheses. * denotes signi cance at 10%, ** at 5%, and *** at 1%. Vote Fraction Candidate Vote Margin Turnout Proxy Disquali ed (0.096) (2.220) (0.066) (0.034) Year * 0.042* (0.064) (1.595) (0.038) (0.023) Small party*disquali ed * 0.209** (0.148) (4.355) (0.102) (0.063) Fixed e ects constituency constituency constituency constituency R-squared Obs Table 16: Panel Regression - Party Interaction: Area-weighted Sum Note: OLS regressions with robust standard errors clustered by constituency in parentheses. * denotes signi cance at 10%, ** at 5%, and *** at 1%. Party A liation Frequency Percent Small Party ANP BNP HPG IND JUI (F) JWP NPP PML (N) PPP PPP (SB) Total Table 17: Party A liations of 1997 MNAs 34

35 Party A liation Frequency Percent BNP IND JWP MMA MQM MTQM NA PAT PML (F) PML (J) PML (N) PML (Q) PML (Z) PPMAP PPP (S) PPPP PSPP PTI Total Table 18: Party A liations of 2002 MNAs 12 Appendix 4: National Assembly Composition in 1997 and 2002 Below are the detailed party names for parties of winning candidates in 1997 and 2002: ANP: Awami National Party. BNP: Baluch National Party. HPG: Haq Parast Group, now more commonly called the Muttahida Qaumi Movement. JUI (F): Jamiat Ulema-i-Islam (Fazl-ur-Rehman Group). JWP: Jamhoori Watan Party. 35

36 MMA: Muttahidda Majlis-e-Amal Pakistan MQM: Mohajir Qaumi Movement MTQM: Muttahida Qaumi Movement NA: National Alliance NPP: National People s Party. PML(F): Pakistan Muslim League(F) PML(J): Pakistan Muslim League (Junejo) PML(N): Pakistan Muslim League (Nawaz) PML(Q): Pakistan Muslim League(Quaid), formed by a cuto of the PML(N). PML(Z): Pakistan Muslim League(Zia) PPMAP: Pakistan Pakhtoonkhawa Milli Awami Party PPP(S): Pakistan Peoples Party(Sherpao) PPP: Pakistan People s Party. PPP (SB): Pakistan People s Party (Shaheed Bhutto Group). PPPP: Pakistan Peoples Party Parliamentarians, formed primarily of the former PPPP. PSPP: Pakistan Shia Political Party PTI: Pakistan Tehreek-e-Insaaf IND: Independent Candidates. 36

37 13 Appendix 5: Maps The following are 12 maps I constructed in GIS: one for each of the four provinces in Pakistan, and two for the largest cities, Karachi and Lahore. These maps were constructed separately by province for higher precision. The purpose of displaying these maps is twofold: one, to show how the constituency boundaries evolved from 1997 to 2002 due to delimitation; and secondly, to illustrate how disquali cation hit the constituencies spatially and by province. In Sindh, Baluchistan, and NWFP, the more feudal and tribal areas of the country, disquali cation seems to have hit in clusters, which means that the constituencies with uneducated MNAs in 1997 were clustered spatially. In Punjab, the most urban and richest province, disquali cation seems to be spread out spatially, albeit in small clusters. Finally, no MNA from any of Lahore s 9 constituencies in 1997 was uneducated, whereas only 2 out of the 13 constituencies in Karachi in 1997 elected uneducated MNAs (uneducated here means less than a Bachelors degree). 37

38 38

39 39

40 40

41 41

42 42