Model o Need a model of regulator behavior in terms of inspection as well as how firms with different cost structures respond to the various

Size: px
Start display at page:

Download "Model o Need a model of regulator behavior in terms of inspection as well as how firms with different cost structures respond to the various"

Transcription

1 Model o Need a model of regulator behavior in terms of inspection as well as how firms with different cost structures respond to the various incentives o Model should provide a basis for estimating direct and indirect effects of certification. o Two key empirical facts on which we build model Table 1: High inspection intensity in sectors in which compliance is more difficult Table 2: Non compliance is uncorrelated with inspection intensity before and after certification but certification is.

2 Table 1 Select Sectors with low and high Inspection and certification Intensity Low Inspection/Low Certification Natural fibers Wine Shoes Printing Wooden Furniture Office Supplies Paper Coffee/Tea Industry Chocolates Wooden Construction Supplies High Inspection/High Certification Synthetic fibers Beer Explosives Ink for Printing Paint Cleaning Products Glue Pharmaceuticals Edible Oil Cement

3

4 Structure o Sectors are observable and average costs by sector of compliance ( C ) are known to regulator but firm level variation ( d ij ) in compliance cost is not known. Pr( d d) F( d); f ( d) F'( d) ij o Firms choose whether to certify (if available) or comply given probability of inspection P j and fines M. c1 Compliance without certification C C d Non compliance Certification C P M nc ij C C d c2 ij j ij j ij j ij Latter expression reflects possible benefits to certification as well as costs of private audit j

5 Cost Schedule 1. 1, 1.

6 Cost schedule 2. 1, 1.

7 Cost schedule 3. 1, 0.

8 No certification o Compliant fraction L L Fc () FMP ( C) j j j o Benefit A from having firms in compliance and cost B per inspection so regulators maximize given regime k Objective function k k S N AL ( P, C ) PN B j j j j j j j j k L j FOC A B 0 Pj So MPj C 0 and L j is constant

9

10 Objective function k k k S N AL ( P, C ) N BP(1 D ) So FOC j j j j j j j j j k k Lj D k j A B(1 Dj) BPj 0 P P j j

11

12

13

14 OLS Regression Results Fraction Certified defined from SIEM data Dependent variable: Difference in AOD between 2000 and 2006 Fraction Certified (SIEM) [0.032]*** [0.033]** [0.034]** [0.035]** Medium*Log Cert. Intensity wt [0.034]** [0.034]** [0.035]** Log Size* Log Cert. Intensity wt [0.008] [0.008] [0.008] Log Cert. Intensity wt [0.008] [0.013]** [0.013]** [0.013]* Log Size wt [0.003]*** [0.008] [0.008] [0.008]* % Medium [0.034]* [0.034] [0.033] % Exporting [0.024]*** [0.031]* % Importing [0.023] [0.029]* Exporting*Log Cert. Intensity [0.024] Importing. * Log Cert. Intensity [0.022]** AOD [0.011]*** [0.011]*** [0.011]*** [0.011]*** Constant [0.027]*** [0.028]*** [0.028]*** [0.028]*** Weather Controls Yes Yes Yes Yes Month fixed Effects Yes Yes Yes Yes Observations R squared Robust standard errors clustered at the zip code level in brackets e * significant at 10%; ** significant at 5%; *** significant at 1%

15 Figure 2: Irrigation Cost as Percentage of Gross Farm Income for Electric Pumps (High Share of Fixed Cost in Total Groundwater Irrigation Cost) 70 Share in Gross Income % Marginal Small Medium Large All Fixed Cost Motor Burnout Pump Maintenance Tariff Source: World Bank: India Power Supply to Agriculture (2001)

16 Land scale cut off for private wells under public provision (a * P) C Land Holding (a) Minimum land owned A (a m ) B Intensive Margin Land scale cut off for private wells without public provision a * * NP Intensive Margin Extensive Margin Aquifer Depth (d) Figure 4: Characterization of Differential Water Usage under Public Provision of Groundwater for Irrigation

17 30 treated comparison Depth Below Ground Level A. Investment in Private wells in Period 1 (number of private wells) 50 treated comparison B. Investment in Private Wells in Period 2 (number of private wells) Figure 6: Average Number of Private Wells per Village by Depth of Groundwater: Panel A shows that the average number of wells per village did not differ across treated and comparison villages in low or high cost categories although low cost villages (to the left of the dashed line) had more wells. Panel B shows that the comparison villages systematically had more wells in high cost category than treated villages whereas this is not the case in low cost category. (Vertical dotted line is at the practical cutoff at which surface pumps become infeasible.)

18 Table 1 : Differences in Differences Estimates of Public Provision on Local Water Table Depth by Categories of Fixed Cost Coefficients of interactions between Dummies indicating whether the village was in Public Tubewell Program or not, Dummies indicating whether the Water Table depth was measured at the time public wells were put into operation or 7 years later, and Dummies indicating the fixed cost category Panel A Dependent Variable : Depth of Water Table Below Ground Level (i) Public Tube Well * Post *Low Cost 0.35 (1.17) Public Tube Well * Post * High Cost 5.14 (2.13) Observations R Squared Panel B Heterogeneity in Impact of Public Tube Well Program between High Cost and Low Cost Categories (i) Difference Between Point estimates from Panel A 4.79 F statistic (testing if the difference is 0) 3.86 Significance level Notes: std errors are reported in parenthese and are clustered at village level. Regression (i) in Panel A is based on the baseline groundwater depth common support sample described in Data Appendix. Low Cost category is charaterized by the depth below ground level upto which low cost surface pumps are physically feasible.

19 Table 2 : Robustness Check for the Validity of Discontinuity in Cost between High and Low Cost Categories < 3metres <2 metres <1 metre Break >1 metres >2 metres >3 metres >4 metres >5 metres (9 feet) (6 feet) (3 feet) Point (3 feet) (6 feet) (9 feet) (12 feet) (15 feet) Public Tube Well * Post * Low Cost (1) (.8) (.74) (1.17) (1.16) (1.13) (1.13) (1.22) (1.11) Public Tube Well * Post * High Cost (1.8) (2.57) (3.1) (2.13) (2.8) (2.43) (2.45) (2.54) (2.63) Difference F statistic significane level Notes: std errors are reported in parentheses and are clustered at village level. All Regressions are based on the baseline groundwater depth common support sample described in Data Appendix. Low Cost category is charaterized by the depth below ground level upto which low cost surface pumps are physically feasible. 'Break Point' indicates a depth of 25 feet below ground level which is the actual cutoff at which the fixed cost to access groundwater changes substantially. Each Column corresponds to a separate regression where the cutoff of water table depth at which the change in fixed cost occurs is arbitrarily shifted from 25 feet in increments of 3 feet.

20 Table 3 : Differences -in Differences Estimates of Public Provision on Local Water Table Depth by Categories of Fixed Cost Coefficients of interactions between Dummies indicating whether the village was in Public Tubewell Program or not, Dummies indicating whether the Water Table depth was measured at the time public wells were put into operation or 7 years later, and Dummies indicating the fixed cost category Dependent Variable :Depth of Water Table Below Ground Level (i) (ii) (iii) Public Tube Well * Post *Low Cost (1.2) (1.20) (1.19) Public Tube Well * Post *High Cost (2.25) (2.25) (2.28) Demographic and Economic time varying controls No Yes Yes Geographical time varying controls No No Yes Observations R Squared Panel B Heterogeneity in Impact of Public Tube Well Program between High Cost and Low Cost Categories (i) (ii) (iii) Difference Between Point estimates from Panel A F statistic (testing if the difference is 0) Significance level Note: std errors are reported in parentheses and are clustered at village level. Regressions (i), (ii), and (iii) in Panel A are based on the baseline groundwater depth common support sample matched to various other data sources as described in Data Appendix. Low cost category is charaterized by the depth below ground level upto which low cost surface pumps are physically feasible. Demographic and economic controls include number of households, percentage of scheduled caste population, percentage of literate population, and percentage of workers in the population. Geographical variables include rainfall, first lag of rainfall, and average monthly temperature.

21 Table 4 : Differences-in Differences Estimates of Public Provision on Groundwater Irrigation by Categories of Fixed Cost Coefficients of interactions between Dummies indicating whether the village was in Public Tubewell Program or not, Dummies indicating whether the Water Table depth was measured at the time public wells were put into operation or 7 years later, and Dummies indicating the fixed cost category Panel A Dependent Variable : Ratio of Ground water Irrigated area to Sown Area (i) (ii) (iii) Public Tube Well * Post * Low Cost (.034) (.034) (.034) Public Tube Well * Post *High Cost (.09) (.09) (.09) Demographic and Economic time varying controls NO Yes Yes Geographical Time Varying Controls NO NO Yes Observations R Squared Panel B: Heterogeneity in Impact of Public Tubewell Program on Groundwater Irrigation between High and Low Cost Categories (i) (ii) (iii) Difference Between Point estimates from Panel A F statistic Significance Note: std errors are reported in parentheses and are clustered at village level. Regressions (i), (ii), and (iii) in Panel A are based on the baseline groundwater depth common support sample matched to various other data sources as described in Data Appendix. Low cost category is charaterized by the depth below ground level upto which low cost surface pumps are physically feasible. Demographic and economic controls include number of households, percentage of scheduled caste population, percentage of literate population, and percentage of workers in the population. Geographical variables include rainfall, first lag of rainfall, and average monthly temperature.

22 Figure 1: Rural Water Project (RWP) study region and sample springs 56

23 Figure 2: Rural Water Project (RWP) Timeline Identified universe of springs; June-July 2004 (N springs =562) Conducted initial site visits and water quality tests, Further site visits with Ministry of Water and NGO technical staff, Selection of sample; July-November 2004 (N springs =200) Spring user lists compiled; July 2004-January 2005 (N springs =200) Randomization of springs into year of treatment (N springs =200) Random selection of 7-8 households per spring (N hh =1500) Year 1 Treatment (N springs =50; N hh =371) Sample after nonviable springs eliminated (N springs =47; N hh =350) Year 2 Treatment (N springs =50; N hh =378) Sample after nonviable springs eliminated (N springs =46; N hh =349) Years 3 and 4 Treatment (N springs =100; N hh =751) Sample after nonviable springs eliminated (N springs =91; N hh =685) Household baseline surveys, water quality testing; August 2004-February 2005 (N springs = 184, N hh =1384 in viable sample) Year 1 spring protection; January-April 2005 (N springs =47) Household surveys, water testing; April-August 2005 (N springs =175, N hh =1,250) Year 2 spring protection; August-November 2005 (N springs =46) Household surveys, water testing; August-November 2006 (N springs =183, N hh =1,283) Household surveys, water testing; January-March 2007 (N springs =184, N hh =1,231) 57

24 Table 2: Spring protection source water quality impacts ( ) Dependent variable: ln(spring water E. coli MPN) (1) (2) (3) (4) Treatment (protected) indicator (0.28) *** (0.27) *** (0.23) *** (0.24) *** Baseline ln(spring water E. coli MPN) (0.04) *** (0.04) *** (0.05) *** Baseline ln(spring water E. coli MPN) (0.13) * Treatment indicator (0.12) Baseline latrine density (0.61) Baseline latrine density * Treatment indicator 0.86 (1.75) Baseline diarrhea prevention score (0.07) Baseline diarrhea prevention score *Treatment indicator (0.25) Baseline boiled water yesterday density 0.42 (0.66) Baseline boiled water yesterday density *Treatment indicator 0.82 (1.53) Baseline mother s years of education density (0.04) Baseline mother s years of education density *Treatment indicator 0.07 (0.14) Treatment group 1 (phased in early 2005) (0.30) (0.20) * (0.17) ** (0.20) Treatment group 2 (phased in late 2005) (0.25) (0.17) (0.15) * (0.18) R Observations Mean (s.d.) of dependent variable 3.64 (1.94) 3.64 (1.94) 3.64 (1.94) 3.64 (1.94) Notes: Estimated using OLS. Huber-White robust standard errors are presented (clustered at the spring level), significantly different than zero at * 90% ** 95% *** 99% confidence. There are 184 spring clusters with data for some of the four survey rounds (2004, 2005, 2006, 2007). MPN stands for most probable number coliform forming units (CFU) per 100ml. Average diarrhea prevention knowledge calculated as average of demeaned sum of number of correct responses given to the open ended question to your knowledge, what can be done to prevent diarrhea? All variables that are interacted with the treatment indicator are de-meaned. Time (survey round and wave) fixed effects are included in all regressions but not reported. When interactions included, baseline variables are interacted with time indicators and treatment group indicators in addition to the treatment indicator. These coefficients not reported. Baseline iron roof density and its interaction with the treatment indicator are included as additional control variables (not shown in the table). The -108 log point effect in column 1 is equivalent to a 66% reduction in E. Coli fecal coliform units per 100ml. 46

25 Table 3: Spring protection household water quality impacts ( ) Dependent variable: ln(home water E. coli MPN) (1) (2) (3) Treatment (protected) indicator (0.15) * (0.19) (0.27) ** Baseline ln(spring water E. coli MPN) (0.02) *** (0.02) *** (0.03) *** Baseline multi-source user (0.17) * (0.17) Baseline multi-source user * Treatment indicator (0.25) (0.25) Baseline latrine density (0.33) ** (0.32) *** (0.59) Baseline latrine density * Treatment indicator 1.43 (1.01) Baseline diarrhea prevention score (0.02) (0.02) (0.04) Baseline diarrhea prevention score * Treatment indicator (0.06) Baseline boiled water yesterday indicator (0.08) ** (0.08) * (0.16) * Baseline boiled water yesterday indicator * Treatment indicator 0.51 (0.28) * Baseline mother s years of education (0.01) (0.01) (0.02) Baseline mother s years of education * Treatment indicator 0.02 (0.04) Treatment group 1 (phased in early 2005) (0.14) (0.18) (0.27) Treatment group 2 (phased in late 2005) (0.12) (0.16) (0.28) R Observations (spring clusters) 4341 (184) 4341 (184) 4341 (184) Mean (s.d.) of dependent variable in comparison group 3.25 (2.15) 3.25 (2.15) 3.25 (2.15) Notes: Estimated using OLS. Huber-White robust standard errors (clustered at the spring level) are presented, significantly different than zero at * 90% ** 95% *** 99% confidence. MPN stands for most probable number coliform forming units (CFU) per 100ml. Additional control variables included are: season fixed effects, number of children under 12 living in the home, home has iron roof indicator, iron roof density within spring community. When differential treatment effects are reported in column 3, we also include interactions with all of these control variables and the treatment indicator (not shown in the table). Baseline spring water quality, latrine density, and diarrhea prevention score are de-meaned. Time (survey round and wave) fixed effects included in all regressions but not reported. When interactions are included, baseline variables are interacted with time effects and treatment group indicators, in addition to interactions with treatment (protected) indicator. These coefficients not reported in the table. The -26 log point effect in column 1 is equivalent to a 23% reduction in E. Coli fecal coliform units per 100ml. 47

26 Table 4: Health outcomes for children under age three at baseline or born since 2004 ( data) Dependent variable Dependent variable: Diarrhea in past week Weight (kg) Dependent variable Body mass index, BMI (kg/m 2 ) (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Probit Treatment (protected) indicator *** *** * ** *** (0.012) (0.019) (0.023) (0.023) (0.029) (0.039) (0.073) (0.094) (0.13) (0.16) Treatment (protected) indicator * Male ** (0.040) (0.123) (0.186) Treatment (protected) indicator * Baseline latrine density (0.123) Treatment (protected) indicator * Baseline diarrhea prevention score Treatment (protected) indicator * Baseline mother s years of education (0.0073) (0.0044) Child fixed effects No No Yes Yes Yes Yes Yes Yes Yes Yes Treatment group fixed effects No Yes Yes Yes Yes Yes Yes Yes Yes Yes Month of year controls No Yes Yes Yes Yes Yes Yes Yes Yes Yes Gender-age controls No No No Yes Yes Yes Yes Yes Yes Yes R Child-year observations Mean (s.d.) of the dependent variable in the comparison group 0.19 (0.39) 0.19 (0. 39) 0.19 (0. 39) Notes: Column 2 estimated using probit (marginal effects presented), columns 1 and 3-10 estimated using OLS. Huber-White robust standard errors (clustered at the spring level) are presented, significantly different than zero at * 90% ** 95% *** 99% confidence. Data from all four survey rounds (2004, 2005, 2006, 2007), sample restricted to children under age three at baseline (in 2004) and children born since Diarrhea defined as three or more looser than normal stools within 24 hours at any time in the past week. The gender-age controls include linear and quadratic current age (by month), and these terms interacted with a gender indicator. Columns 2-10 also contain survey round controls. In column 6, additional control variables are number of children under 12 living in the home, home has iron roof indicator, iron roof density within spring community, and the boiled water yesterday indicator (all measured at baseline), all interacted with the treatment indicator (0. 39) 0.19 (0. 39) 0.19 (0. 39) (3.52) (3.52) 17.0 (2.2) 17.0 (2.2) 48

27 Table 6: Discrete choice models (conditional and mixed logit) of water source choice (2007 surveys) Revealed Preference Stated Ranking --- (1) (2) (3) (4) (5) (6) (7) Treatment (protected) indicator 0.56 *** *** 0.54 *** 0.96 *** (0.05) (0.07) (0.08) (0.08) (0.24) Mixed logit Mean (normal): 0.56 *** (0.05) Mixed logit Std. dev. (normal): 0.11 *** (0.04) ln (source water E. coli MPN) *** (0.01) Water quality at source perceived to be above average 0.95 *** 0.97 *** (0.24) 0.22 (0.19) (0.06) Distance to water source (minutes walking) *** *** *** *** *** *** *** (0.002) (0.002) (0.003) (0.002) (0.002) (0.009) (0009) Distance * Children aged 0-5 with diarrhea last week *** (0.003) Treatment indicator * Children aged 0-5 with diarrhea last week 0.23 *** (0.06) Treatment indicator * Baseline latrine ownership 1.8 *** (0.2) Treatment indicator * Baseline diarrhea prevention score (0.018) Treatment indicator * Baseline mother s years of education ** (0.010) Source type: Borehole/piped * ** * (0.07) (0.07) (0.07) (0.07) (0.25) (0.25) Source type: Well *** *** *** *** * * (0.06) (0.06) (0.06) (0.06) (0.23) (0.24) Source type: Stream/river *** *** *** *** *** *** (0.09) (0.09) (0.09) (0.08) (0.50) (0.51) Source type: Lake/pond (0.19) (0.20) (0.18) (0.19) (1.80) (1.76) Log likelihood at convergence Number of observations Number of households

28 Figure 4: Household revealed preference and stated preference valuations of one year of spring protection (2007) Normal distribution density $5.79 $ Revealed preference (US$) Contingent Valuation (US$) Stated ranking (US$) Notes: The revealed preference estimates are from the mixed logit results in Table 6, regression 5, and the stated preference ranking results are from the mixed.logit results in Table 6, regression 7. The contingent valuation data are presented in Table 7, Panel C. 59

29 Panel A: Revealed preference valuation (from mixed logit Table 6, column 5) Table 7: Valuation of one year of spring protection (2007 survey) One year of spring protection Mean Std. dev. Work days (8 hour days) 12.7 days 6.7 days Assume value of time is 50% Kenyan worker average wage $9.03 $4.75 Assume value of time is 25% Kenyan worker average wage $4.52 $2.38 Equate mean stated preference ranking and contingent valuation $5.79 $3.05 Panel B: Stated preference ranking valuation (from mixed logit Table 6, column 7) Work days (8 hour days) 38.3 days 20.2 days Assume value of time is 50% Kenyan worker average wage $27.52 $15.76 Assume value of time is 25% Kenyan worker average wage $13.76 $7.88 Equate mean stated preference ranking and contingent valuation $17.64 $10.10 Final Wave, emphasizing trade-offs Panel C: Contingent Valuation Full Round Proportion willing to pay this for spring protection: US$3.57 (250 Kenya Shillings) 0.94 [308] 0.80 [98] US$7.14 (500 Kenya Shillings) 0.90 [316] 0.79 [204] US$14.29 (1000 Kenya Shillings) [204] One year of spring protection Mean Std. dev. Sample: Final Wave, emphasizing trade-offs $17.64 $13.09 Subsample with 250 KSH starting value $12.62 $11.06 Subsample with 500 KSH starting value $23.91 $14.28 Notes: The results in Panels A and B all correct for attenuation bias in the coefficient estimate on distance walking to water source, assuming a correction for classical measurement error (the correlation between reported distance walking to the sample spring across survey rounds is 0.38.) Number of observations in brackets in Panel C. The contingent valuation questions were only asked of households in the treatment group, since they have a firsthand sense of what spring protection is worth. In the final wave of the survey, respondents were first asked if they would be willing to pay either 250 or 500 Kenya Shillings, followed by the question that emphasized the expenditure trade-off for their assigned amount, and then were asked if they would be willing to pay the next higher amounts also with emphasis on the expenditure trade-off. 53

30 Table 8: Property Rights Institutions: Counterfactual Policy Simulations Proportion of springs protected Average price per water trip (USD) NPV profits, per land owner (USD) NPV household welfare, per spring (USD) Proportion households with lower utility than status quo NPV of public spending on springs (USD) Status quo Social welfare, per spring (USD) Social planner Public investment (including tax deadweight loss) Springs social planner does not protect Springs social planner protects Pure privatization Springs social planner does not protect Springs social planner protects Conditional privatization Springs social planner does not protect Springs social planner protects Conditional privatization, access to unprotected water Springs social planner does not protect Springs social planner protects Notes: The status quo assumes that water prices are zero at all sources, and that all springs remain unprotected. The pure privatization scenario allows land owners complete freedom in charging water prices and in deciding whether or not to protect their springs. The conditional privatization scenario only allows land owners to charge positive prices at protected springs. Conditional privatization access to unprotected water prohibits land owners from charging for this unprotected water. Household utility is normalized to take on a value of zero in the Status Quo case. Net present values are discounted at 5% annually for both households and land owners, over 10 years. Household utility values are converted into USD using the value of time that equates mean stated preference ranking and contingent valuation (as in Table 7 above). 54

31 Figure 3 Model of two well interference

32

33 Theoretical Results Marginal cost of water increasing in own wage but decreasing in neighbors total water.. Profits are decreasing in neighbors total water So strategic complementarity with negative spillovers

34 Table 3 Village FE-IV Estimates: Determinants of the Log of Irrigated Acreage and Tubewell Depth Dependent variable Log total owned land (acres).184 (1.42) th Total owned land>9 acres (20 percentile) (2.84) Log total village land owned by bottom 20% of -3 landowners (x10 ) Log total village land owned by top 40% of -3 landowners (x10 ) Tubewell Depth (2.25) (0.40) a Log of HYV yield on irrigated land in village.151 (.114).198 (1.52) (2.79) Irrigated Acreage.885 (19.1) (0.86) (0.74) (2.35).210 (1.44) a Total village wells per acre (1.70).199 (2.32) N a Endogenous variable. Instruments include whether a rice-growing or wheat growing village in 1971, log of HYV yields on irrigated land in the village in 1971, whether the village had a pond or river, number of village wells per acre in Absolute values of robust asymptotic t-ratios in parentheses. -

35 rate is 50 points for one dollar. After each round each subject is randomly matched with another subject in his or her group for the next round. In part 2 of the experiment the subjects play 10 rounds as in part 1 but the payoffs can be modified at the beginning of this part to the payoffs in Table 1 (Modified Payoffs). The modification of payoffs consists of imposing a tax or fine on unilateral defection. While under the initial payoffs the unique Nash equilibrium is mutual defection, under the modified payoffs both mutual defection and mutual cooperation are Nash equilibria. Table 1: Stage Game Payoffs (in points) Initial Payoffs Other s action Modified Payoffs Other s action Own C D Own C D action C action C D D We chose a prisoners dilemma game as the initial game as the tension between personal incentives and efficiency is not only an important feature of human interaction but also a feature that groups attempt to solve by imposing different kinds of policies. We chose a prisoners dilemma game over other kind of social dilemma games (i.e. public good games) as the former is simpler which allows a simple explanation of the policy. The modified game was chosen to be a coordination game as it is intuitive to think that the incentive to follow policies and regulations may depend on the behavior of others and may result in a multiplicity of equilibria. Whether the payoffs are modified in the policy selection stage is determined as follows. First, subjects vote on whether to modify payoffs. Second, the computer randomly chooses whether to consider the votes in each group. If the computer considers the votes, then the majority wins and in case of a tie the computer breaks the tie. If the computer does not consider the votes in a group, it randomly chooses whether to modify payoffs or not in that group. The voting stage is summarized in Figure 1. The subjects 5

36 Figure 1: Voting Stage consider votes Majority decides to (computer breaks ties) modify Payoffs not modify payoffs (EndoMod) (EndoNot) Vote Computer decides to not consider votes Computer decides to modify payoffs not modify payoffs (ExoMod) (ExoNot)

37 Table 2: Summary statistics of sessions Total/Means Std. Dev. Subjects 276 Economics 12.68% Class Political Philosophy SAT Math SAT Verbal Beauty Contest Num Subject Comprehension Vote stage 92.03% Initial Payoffs 89.13% Modified Payoffs 80.43% Earnings Maximum Average Minimum Note: Economics is the percentage of Economics majors in the session; Class is equal to 1 for freshment, 2 for sophomore, etc.; Political Philosophy is equal to 1 for very liberal to 5 for very conservative; Beauty contest num. is the number chosen in the beauty contest game.

38 Table 4: The effect of the democracy - Individual Level Data Panel A: Number of observations by Vote Stage Outcome and Individual Vote Consider Votes Not Consider Votes Vote for Modify Not Modify Modify Not Modify Modify (EndoMod) (EndoNot) (ExoMod) (ExoNot) Total No Yes Total Panel B: Cooperation Percentage in Round 10 Consider Votes Not Consider Votes Vote for Modify Not Modify Modify Not Modify Modify (EndoMod) (EndoNot) (ExoMod) (ExoNot) No 5.88% 3.64% 9.68% 11.54% Yes 5.45% 4.00% 9.09% 8.82% Total 5.56% 3.75% 9.38% 10.00% Panel C: Cooperation Percentage in Round 11 Consider Votes Not Consider Votes Vote for Modify Not Modify Modify Not Modify Modify (EndoMod) (EndoNot) (ExoMod) (ExoNot) No 41.18% 14.55% 41.94% 3.85% Yes 81.82% 24.00% 57.58% 23.53% Total 72.22% 17.50% 50.00% 15.00%

39 Table 5: The effect of the democracy - Individual Level Data Dependent Variable: Individual cooperation in round 11 (1) (2) (3) (4) (5) EndoMod [0.050]*** EndoNot [0.048]*** ExoMod 0.5 [0.053]*** ExoNot 0.15 [0.055]*** EndoModn [0.101]*** [0.102]*** [0.106]*** [0.106]*** EndoNotn [0.056]** [0.067] [0.057]** [0.069] ExoModn [0.075]*** [0.086]*** [0.075]*** [0.086]*** ExoNotn [0.082] [0.084] [0.087] [0.090] EndoMody [0.056]*** [0.063]*** [0.056]*** [0.064]*** EndoNoty [0.083]*** [0.090] [0.087]*** [0.095] ExoMody [0.072]*** [0.082]*** [0.075]*** [0.085]*** ExoNoty [0.071]*** [0.079] [0.074]*** [0.080] Own Part 1 Coop [0.139]*** [0.141]*** Partners' Part 1 Coop [0.179] [0.181] Exclude did not remember vote result No No No Yes Yes Observations R-squared Tests of differences of cooperation rates by mechanism (Endo versus Exo), payoffs (Mod versus Not) and vote (y versus n) p-values EndoNot=ExoNot EndoMod=ExoMod EndoMod=EndoNot ExoMod=ExoNot EndoNotn=ExoNotn EndoModn=ExoModn EndoModn=EndoNotn ExoModn=ExoNotn EndoNoty=ExoNoty EndoMody=ExoMody EndoMody=EndoNoty ExoMody=ExoNoty Note: All results are from OLS regressions. The dependent variable is an indicator variable for whether the subject cooperated in round 11. The explanatory variables in column (1) are indicator variables for the vote stage result. In the rest of the columns the explanatory variable are the interaction of indicator variables for vote stage results with indicator variables for the vote of the subject. EndoMod: endogenous modification, EndoNot: endogenous non-modification, ExoMod: exogenous modification, ExoNot: exogenous non-modification, n and y denote the individual vote of the subject (agains or for modification). Regressions in columns (3) and (5) control for the individuals' and their partner's cooperation rate in the rounds before the voting stage (Part 1). Standard errors in brackets: * significant at 10%; ** significant at 5%; *** significant at 1%. The p-values correspond to Wald tests based on the regression results.

40 Table 6: The effect of democracy - Individual Level Data - All Rounds Vote for Modify Round Yes No [0.124]** [0.103] [0.116]** [0.154] [0.144]* [0.120]*** [0.147]** [0.132]** [0.138]* [0.145]** [0.154]* [0.136]*** [0.154]** [0.197] [0.153]** [0.194] [0.157]** [0.171] [0.160]** [0.175] Note: table reports estimated impact of democracy on likelyhood of choosing C by round for groups with modified payoffs following the model in the Appendix. Jackknife standard errors by group: * significant at 10%; ** significant at 5%; *** significant at 1%.

41 Table 7: The effect of democracy - Group Level Data Panel A: Number of groups by Vote Stage Outcome and Vote Share Consider Votes Not Consider Votes Modify Not Modify Modify Not Modify Vote Share (EndoMod) (EndoNot) (ExoMod) (ExoNot) Total 0 X X X X Total Panel B: Cooperation Percentage in Part 1 Consider Votes Not Consider Votes Modify Not Modify Modify Not Modify Vote Share (EndoMod) (EndoNot) (ExoMod) (ExoNot) 0 X 19.17% 1 X 21.39% 31.00% 11.25% % 16.88% 16.50% 16.88% % X 17.92% 19.58% % X 10.00% Panel C: Cooperation Percentage in Part 2 Consider Votes Not Consider Votes Modify Not Modify Modify Not Modify Vote Share (EndoMod) (EndoNot) (ExoMod) (ExoNot) 0 X 21.67% 1 X 11.67% 24.50% 12.50% % 8.44% 43.50% 9.38% % X 32.50% 12.50% % X 7.50%

42 Table 8: The effect of democracy - Group level data - Voteshare=2 Dependent Variable: Group cooperation rate in part 2 (1) (2) (3) (4) EndoMod [0.090]*** [0.117]*** [0.094]*** [0.123]*** EndoNot [0.078] [0.139] [0.071] [0.134] ExoMod [0.099]*** [0.149]* [0.094]*** [0.136] ExoNot [0.111] [0.158] [0.109] [0.155] Part 1 Cooperation [0.687] [0.636] Exclude did not remember vote result No No Yes Yes Observations R-squared Tests of differences of cooperation rates by mechanism (Endo versus Exo) and payoffs (Mod versus Not) p-values EndoNot=ExoNot EndoMod=ExoMod EndoMod=EndoNot ExoMod=ExoNot Note: All results are from OLS regressions. The dependent variable is the cooperation rate by group in the 10 rounds after the voting stage (part 2). The explanatory variables are indicator variables for the vote stage result. EndoMod: endogenous modification, EndoNot: endogenous non-modification, ExoMod: exogenous modification, ExoNot: exogenous non-modification. Regressions in columns (2) and (4) control for the cooperation rate of the group before the voting stage (Part 1). Standard errors in brackets: * significant at 10%; ** significant at 5%; *** significant at 1%. The p-values correspond to Wald tests based on the regression results.

43 Table 10: Summary statistics of additional sessions Total/Means Std. Dev. Subjects 148 Economics 23.65% Class Political Philosophy SAT Math SAT Verbal Beauty Contest Num Subject Comprehension Vote stage 97.30% Initial Payoffs 81.76% Modified Payoffs 91.22% Earnings Maximum Average Minimum Note: Economics is the percentage of Economics majors in the session; Class is equal to 1 for freshment, 2 for sophomore, etc.; Political Philosophy is equal to 1 for very liberal to 5 for very conservative; Beauty contest num. is the number chosen in the beauty contest game.

44 Table 11: The effect of the democracy controling for information - Individual Level Data Panel A: Number of observations Original Sessions Additional Sessions Not Consider Votes Consider Votes Vote Share Vote for Yes No 2 2 Modify (EndoMod) (ExoMod) (ExoModH) (ExoModL) No Yes Total Panel B: Cooperation Percentage in Round 11 Vote for Modify (EndoMod) (ExoMod) (ExoModH) (ExoModL) No 41.18% 41.94% 35.00% 23.68% Yes 81.82% 57.58% 62.50% 64.29% Total 72.22% 50.00% 55.26% 34.62% Panel C: Cooperation Percentage in Part 2 Vote for Modify (EndoMod) (ExoMod) (ExoModH) (ExoModL) No 43.53% 26.45% 22.00% 18.42% Yes 71.82% 40.00% 50.36% 33.57% Total 65.14% 33.44% 42.89% 22.50%