Criminally Accused Politicians in India: Economic Outcomes (Chapter 4)

TABLE 9 Effect of Electing Criminally Accused (Financial Charge) Politicians on Roads Dependent Variable Length of Roads Built (in kms) (1) (2) (3) (4) PANEL A: All States Financial Charge -7.26 -6.76 -6.09 -2.73 (5.19) (5.45) (7.19) (3.97) Bandwidth Size No. of observations Financial Charge Bandwidth Size No. of observations Financial Charge Bandwidth Size No. of observations Serious Charge Bandwidth Size No. of observations State and Year Fixed Effects Bandwidth Type Polynomial order of control function 9.03 7.61 4.51 18.06 226 198 120 326 PANEL B: BIMAROU States -13.91* -11.95 -9.00 -5.91 (7.13) (8.18) (10.41) (5.34) 9.09 7.25 4.55 18.19 133 109 77 167 PANEL C: Least Developed States -12.01 -11.17 -13.61 -7.27 (7.79) (8.24) (11.45) (5.29) 7.73 6.74 3.86 15.45 124 110 65 178 PANEL D: High Corruption States -2.46 -0.73 -2.20 0.61 (7.87) (9.05) (11.64) (5.57) 8.61 105 YES IK (h) 6.62 4.3 85 60 YES YES CCT h/2 Local Linear 17.22 145 YES 2h NOTES: Standard errors are clustered at the constituency level and given in parentheses. The dependent variable is the residual from the regression of state and year dummies. Results displayed in each column come year from a separate regression. Panel A includes all the states; Panel B includes the BIMAROU states: Bihar, Chattisgarh, Jharkhand, Orissa, Uttar Pradesh, and Uttarakhand; Panel C includes the Least Developed states as ranked by Ministry of Finance: Arunachal Pradesh, Assam, Bihar, Jharkhand, Odisha and Uttar Pradesh; and Panel D includes the High Corruption states as ranked by Transparency International India (TII) on index of corruption: Tamil Nadu, Haryana, Jharkhand, Assam, and Bihar. The RD estimates in columns (1)–(4) are on a local linear regression using a triangular kernel. Asterisks denote significance levels (*=.10, **=.05, ***=.01) TABLE 10 Effect of Electing Criminally Accused Politicians on Constituency GDP Growth Growth of Night Lights Dependent Variable Baseline Estimate Financial Charge Serious Charge (1) (2) (3) PANEL A RD Estimate -21.74** -39.39*** -33.19*** Polynomial order of control function Bandwidth type Local Linear Imbens-Kalyanaraman PANEL B Estimated Effect on GDP Growth Rate (in percentage points) Using global average (Henderson et al. 2014) Using India-specific average (Bickenback et al. 2014) -6.5 -2.3 -11.8 -4.2 -10.0 -3.6 PANEL C Assuming 6% growth - what would growth look like? Upper Bound Lower Bound 5.61 5.86 5.29 5.75 5.40 5.78 NOTES: The definition of the main explanatory variable changes across the columns: criminally accused, financial criminal accusation, and serious criminal accusation. Column (1) reports the RD estimate for criminally accused from Table 2, Column (1). In Column (2), we report the RD estimate for any financial charge from Column (1) of Table 3 in Panel A, while we report the RD estimate for any serious charge from Column (1) of Table 3 in Panel B in Column (3) of this table. The upper-bound uses an elasticity of 0.3. The lower bound uses an elasticity of 0.107. 0 .01 .02 .03 .04 .05 Figure 1 Continuity of the Victory Margin between Criminally Accused and Non-Accused -100 -50 0 50 100 0 .01 Density .02 .03 .04 (a) McCrary Test -50 0 Margin of Victory 50 (b) Density of the Victory Margin The forcing variable is the margin of victory of a criminally-accused candidate. Negative values are the difference in the vote shares of a criminally-accused runners-up and a non-accused winner. Positive values are the differences in the vote shares of a criminally-accused winner and a non-accused runners-up. The estimated size of discontinuity in margin of victory (log difference in height) is −0.061 (se = 0.2). -300 -100 -200 Growth of Light t-2 -100 Growth of Light t-1 0 100 0 200 100 Figure 2 Balance Test for Constituency Characteristics -20 0 Margin of Victory 20 -20 20 -.4 -.4 Electorate Size t-1 -.2 0 Number Voted t-1 -.2 0 .2 (b) Growth of Light t-2 .2 (a) Growth of Light t-1 0 Margin of Victory -20 0 Margin of Victory 20 -20 20 -.5 -1 -.5 Ruling Party t-1 0 SC Constituencies 0 .5 .5 1 (d) Number Voted t-1 1 (c) Electorate Size t-1 0 Margin of Victory -20 0 Margin of Victory 20 -20 20 (f) SC Constituencies -.5 ST Constituencies 0 .5 1 (e) Ruling Party t-1 0 Margin of Victory -20 0 Margin of Victory 20 (g) ST Constituencies The forcing variable is the margin of victory of a criminally-accused candidate. Negative values are the difference in the vote shares of a criminally-accused runners-up and a non-accused winner. Positive values are the differences in the vote shares of a criminally-accused winner and a non-accused runners-up. Each variable on the y-axis is net of state and year fixed effects. The dots in the scatter plot depict the averages over each successive interval of 0.5% of margin of victory. The curves are local linear regressions fit separately for positive and negative margins of victory using a triangular kernel and an optimal bandwidth calculator as suggested in Imbens and Kalayanaraman (2012). The confidence intervals are the 95% confidence intervals plotted using standard errors that are clustered at the constituency level. -1 -.4 -.2 -.5 Incumbent t-1 0 Winner Gender t-1 0 .2 .5 .4 1 Figure 3 Balance Test for Candidate Characteristics -20 0 Margin of Victory 20 -20 20 (b) Winners Gender t-1 -.5 Runnerup Gender t-1 0 .5 1 (a) Incumbent in t-1 0 Margin of Victory -20 0 Margin of Victory 20 (c) Runners-Up Gender t-1 The forcing variable is the margin of victory of a criminally-accused candidate. Negative values are the difference in the vote shares of a criminally-accused runners-up and a non-accused winner. Positive values are the differences in the vote shares of a criminally-accused winner and a non-accused runners-up. Each variable on the y-axis is net of state and year fixed effects. The dots in the scatter plot depict the averages over each successive interval of 0.5% of margin of victory. The curves are local linear regressions fit separately for positive and negative margins of victory using a triangular kernel and an optimal bandwidth calculator as suggested in Imbens and Kalayanaraman (2012). The confidence intervals are the 95% confidence intervals plotted using standard errors that are clustered at the constituency level. -50 Average yearly Growth in Lights 0 50 100 Figure 4 Effect of Electing Criminally Accused Politicians on Growth of Night Lights -40 -20 0 Margin of Victory 20 40 The forcing variable is the margin of victory of a criminally-accused candidate. Negative values are the difference in the vote shares of a criminally-accused runners-up and a non-accused winner. Positive values are the differences in the vote shares of a criminally-accused winner and a non-accused runners-up. The variable on the y-axis is the growth of night lights net of state and year fixed effects. The dots in the scatter plot depict the average of growth of night lights over each successive interval of 0.5% of margin of victory. The curves are local linear regressions fit separately for positive and negative margins of victory using a triangular kernel and an optimal bandwidth calculator as suggested in Imbens and Kalayanaraman (2012). The confidence intervals are the 95% confidence intervals plotted using standard errors that are clustered at the constituency level. -300 -100 -50 -200 Growth in Lights in t+2 -100 0 Growth in Lights in t+1 0 50 100 100 150 Figure 5 Effect of Electing Criminally Accused Politicians by Year in Power -40 -20 0 Margin of Victory 20 40 -40 0 Margin of Victory 20 40 Growth in Lights in t+4 0 200 -200 -100 Growth in Lights in t+3 0 100 400 (b) Growth in Lights in t+2 200 (a) Growth in Lights in t+1 -20 -40 -20 0 Margin of Victory 20 (c) Growth in Lights in t+3 40 -40 -20 0 Margin of Victory 20 40 (d) Growth in Lights in t+4 The forcing variable is the margin of victory of a criminally-accused candidate. Negative values are the difference in the vote shares of a criminally-accused runners-up and a non-accused winner. Positive values are the differences in the vote shares of a criminally-accused winner and a non-accused runners-up. Each variable on the y-axis is net of state and year fixed effects. The dots in the scatter plot depict the averages over each successive interval of 0.5% of margin of victory. The curves are local linear regressions fit separately for positive and negative margins of victory using a triangular kernel and an optimal bandwidth calculator as suggested in Imbens and Kalayanaraman (2012). The confidence intervals are the 95% confidence intervals plotted using standard errors that are clustered at the constituency level. -20 Length of Roads Built under PMGSY (in kms) 0 20 40 60 Figure 6 Effect of Electing Criminally Accused Politicians on Length of Roads Built Annually under PMGSY (in kms) -20 -10 0 Margin of Victory 10 20 10 20 Length of Roads Built under PMGSY (in kms) in t-1 -10 0 10 20 30 (a) RD Effect -20 -10 0 Margin of Victory (b) Balance Test The forcing variable is the margin of victory of a criminally-accused candidate. Negative values are the difference in the vote shares of a criminally-accused runners-up and a non-accused winner. Positive values are the differences in the vote shares of a criminally-accused winner and a non-accused runners-up. The variable on the y-axis is Roads Built Annually under PMGSY (in kms) net of state and year fixed effects. The dots in the scatter plot depict the average number of incomplete road projects over each successive interval of 0.5% of margin of victory. The curves are local linear regressions fit separately for positive and negative margins of victory using a triangular kernel and an optimal bandwidth calculator as suggested in Imbens and Kalayanaraman (2012). The confidence intervals are the 95% confidence intervals plotted using standard errors that are clustered at the constituency level. State Name TABLE A-1 State Name and Year of Elections Number of Constituencies Election Years* Arunachal Pradesh 60 1999, 2004, 2009 Assam 126 2001, 2006, 2011 Bihar 243 2000, 2005, 2010 Goa 40 2002, 2007, 2012 Gujarat 182 2002, 2007, 2012 Haryana 90 2000, 2005, 2009 Himachal Pradesh 68 2003, 2007, 2012 Jharkhand 81 2005, 2009 Kerala 140 2001, 2006, 2011 Maharashtra 288 1999, 2004, 2009 Manipur 60 2002, 2007, 2012 Meghalaya 60 2003, 2008, 2013 Nagaland 60 2003, 2008, 2013 Odisha 147 2000, 2004, 2009 Punjab 117 2002, 2007, 2012 Tamil Nadu 234 2001, 2006, 2011 Tripura 60 2003, 2008, 2013 Uttar Pradesh 403 2002, 2007, 2012 Uttarakhand 70 2002, 2007, 2012 West Bengal 294 2001, 2006, 2011 Total 2823 Notes: Bold years are the first election in each state in which candidates were required to file affidavits detailing criminal and financial background. TABLE A-2 Robustness of Main Results Dependent Variable Growth of Night Lights (1) (2) (3) (4) PANEL A Criminally Accused -22.35** -19.42** -25.65* -12.25** (9.89) (8.69) (14.84) (6.14) Year Fixed Effects Bandwidth Size No. of observations YES 6.15 1,550 Criminally Accused -22.20** (8.86) Year-State Fixed Effects Bandwidth Size No. of observations Bandwidth Type Polynomial order of control function YES 6.31 1,570 IK (h) YES YES YES 7.53 3.08 12.31 1,757 756 2,493 PANEL B -20.82** -24.01* -12.88** (8.17) (13.30) (5.52) YES YES 7.15 3.15 1,708 772 CCT h/2 Local Linear YES 12.62 2,531 2h NOTES: Standard errors are clustered at the constituency level and given in parentheses. The dependent variable is the residual from the regression of year growth of night lights on state and year dummies. Criminally accused is a dummy variable that is 1 if a criminally accused candidate wins against a non-accused candidate and 0 if criminally accused candidate loses against a non-accused candidate. The RD estimates in column (1)–(4) are on a local linear regression using a triangular kernel. Asterisks denote significance levels (*=.10, **=.05, ***=.01) TABLE A-3 Effect of Electing Criminally Accused Politicians by State Characteristics Dependent Variable Growth of Night Lights (1) (2) (3) (4) PANEL A: Non-BIMAROU States Criminally Accused -4.46 -3.95 -7.08 -2.01 (3.99) (3.61) (6.61) (2.28) Bandwidth Size No. of observations Criminally Accused Bandwidth Size No. of observations Criminally Accused 9.44 10.62 4.72 18.88 1,226 1,335 651 1,910 PANEL B: Relatively Developed States -4.44 -4.23 -6.12 -2.38 (3.71) (3.51) (5.95) (2.12) 9.67 10.37 4.84 19.34 1,243 1,306 681 1,934 PANEL C: Low Corruption States -7.83 -6.01 -9.88 -4.21 (4.87) (4.30) (6.78) (3.23) Bandwidth Size 7.28 No. of observations 1,152 State and Year Fixed Effects YES Bandwidth Type IK (h) Polynomial order of control function 9.22 1,358 YES CCT 3.64 611 YES h/2 Local Linear 14.57 1,789 YES 2h NOTES: Standard errors are clustered at the constituency level and given in parentheses. The dependent variable is the residual from the regression of state and year dummies. Results displayed in each column come year from a separate regression. Panel A includes the non-BIMAROU states: Arunachal Pradesh, Assam, Goa, Gujarat, Haryana, Himachal Pradesh, Kerala, Maharashtra, Manipur, Meghalaya, Nagaland, Punjab, Tamil Nadu, Tripura, and West Bengal; Panel B includes the Relatively Developed states as ranked by Ministry of Finance: Goa, Haryana, Kerala, Maharashtra, Punjab, Tamil Nadu, and Uttarakhand; and Panel C includes the Low Corruption states as ranked by Transparency International India (TII) on index of corruption: Kerala, Himachal Pradesh, Gujarat, Maharashtra, Punjab, West Bengal, Orissa, and Uttar Pradesh. The RD estimates in columns (1)–(4) are on a local linear regression using a triangular kernel. Asterisks denote significance levels (*=.10, **=.05, ***=.01) TABLE A-4 Sensitivity Analysis of RD Specification Dependent Variable Growth of Night Lights Type of Accusation Criminally Accused (1) (2) (3) (4) Linear -21.73** -20.07** -23.96* -12.44** (8.83) (8.06) (13.28) (5.50) Quadratic -27.03** -27.32** (13.46) (12.74) -23.97 (18.14) -22.48** (9.26) Cubic -20.69 (17.00) -23.35 (16.10) -17.76 (22.00) -28.53** (12.65) Quartic -22.69 (19.87) -20.2 (18.96) -10.65 (24.03) -26.33* (15.26) Bandwidth Size No. of observations State and Year Fixed Effects Bandwidth Type 6.35 1,581 YES IK (h) 13.27 1,728 YES CCT 3.17 783 YES h/2 12.70 2,547 YES 2h NOTES: Standard errors are clustered at the constituency level and given in parentheses. Results displayed in each panel-column come from a separate regression that also controls for state and year fixed effects. Criminally accused is a dummy variable that is 1 if a criminally accused candidate wins against a non-accused candidate and 0 if criminally accused candidate loses against a non-accused candidate. The RD estimates in columns (1)–(4) are on a local linear regression using a triangular kernel. Asterisks denote significance levels (*=.10, **=.05, ***=.01) TABLE A-5 Effect of Electing Criminally Accused Politicians using Alternate Dependent Variables Dependent Variable Log(Night Lights) Average over Election Term (1) (2) Criminally Accused -0.91** -23.30** (0.37) (9.35) Bandwidth Size No. of observations State and Year Fixed Effects Bandwidth Type Polynomial order of control function 3.16 779 YES 5.71 371 YES Imbens-Kalyanaraman Local Linear NOTES: Standard errors are clustered at the constituency level and given in parentheses. Results displayed in each column come year from a seperate regression. The dependent variable is the residual from the regression of state and year dummies. Log(Night Lights) is the intensity of night lights in levels; and Average Growth over the Election Term is the growth of night lights averaged over the election term of the candidate. The RD estimates in columns (1)–(2) are on a local linear regression using a triangular kernel. Asterisks denote significance levels (*=.10, **=.05, ***=.01) TABLE A-6 Controlling for Covariates Growth of Night Lights Dependent Variable (2) (3) (1) PANEL A -21.78** -21.73** -21.27** Criminally Accused (9.67) (8.83) (8.77) PANEL B -21.69** -21.97** -20.58** Criminally Accused (9.60) (8.78) (8.07) 6.35 6.35 6.35 Bandwidth Size No. of observations 1,581 1,581 1,581 YES NO NO Year Fixed Effects State and Year Fixed Effects NO YES YES Control for covariates NO NO YES Bandwidth Type Imbens-Kalyanaraman Polynomial order of control function Local Linear NOTES: Standard errors are clustered at the constituency level and given in parentheses. In Panel A, we do not control for baseline value of log night-lights in the year before election, while in Panel B we control for the baseline value of log night-lights in the year before election. Column (1) includes only year fixed effects; column (2) includes state and year fixed effects; and column (3) includes state and year fixed effects, constituency characteristics: growth of night lights in t-1, growth of night lights in t-2, electorate size, number of voters, whether a constituency was aligned with the ruling party in the state, in the previous election, and whether a constituency is reserved for Scheduled Caste (SC) or Scheduled Tribes (ST); and candidate characteristics: MLA’s incumbency status and gender, in the previous election. The RD estimates in columns (1)–(3) are on a local linear regression using a triangular kernel. Asterisks denote significance levels (*=.10, **=.05, ***=.01 Dependent Variable Criminally Accused TABLE A-7 Does Top Coding Matter? Growth of Night Lights PANEL A (1) (2) (3) (4) -21.74** -20.23** -23.78* -12.35** (8.82) (8.11) (13.27) (5.51) Bandwidth Size No. of observations 6.36 1,567 Criminally Accused -21.74** (8.83) Bandwidth Size No. of observations State and Year Fixed Effects Bandwidth Type Polynomial order of control function 6.35 1,581 YES IK (h) 7.25 3.18 12.73 1,710 779 2,522 PANEL B -20.08** -23.95* -12.44** (8.07) (13.28) (5.50) 7.32 3.17 1,728 783 YES YES CCT h/2 Local Linear 12.70 2,543 YES 2h NOTES: Standard errors are clustered at the constituency level and given in parentheses. The dependent variable is the residual from the regression of year growth of night lights on state and year dummies. Criminally accused is a dummy variable that is 1 if a criminally accused candidate wins against a non-accused candidate and 0 if criminally accused candidate loses against a non-accused candidate. Panel A drops any observations where the constituency-year pixel average is 63. Panel B drops any constituency in which the average pixel intensity for any year is 63. The RD estimates in column (1)–(4) are on a local linear regression using a triangular kernel. Asterisks denote significance levels (*=.10, **=.05, ***=.01) Figure A-7 Share of Criminally Accused Candidates in India ¯ Himachal Pradesh Punjab Uttarakhand Haryana Arunachal Pradesh Uttar Pradesh Assam Bihar Meghalaya Jharkhand West Bengal Gujarat Nagaland Manipur Tripura Odisha Maharashtra Goa Kerala Tamilnadu Percent elected MLA 0.00 - 0.10 0.11 - 0.20 0.21 - 0.30 0.31 - 0.40 0.41 - 0.50 0.51 - 0.60 0 250 500 1,000 Kilometers