Votes for Water: Ethnic Service Delivery and Criminality in Karachi, Pakistan

Abstract

How do voters in ethnically polarized settings evaluate coethnic candidates in an environment of hybrid provision of public goods, especially where such hybrid provision includes links to criminal actors? In numerous urban settings around the world, local public goods provision involves a complex mix of private vendors, state services, and criminal actors. This paper explores how voters in Karachi, Pakistan evaluate candidates making distinct claims to water provision. We present findings from a survey experiment of over 2000 Karachi residents surveyed in 2021–2022. We find that while voters generally prefer coethnic candidates regardless of their ability to provide water, a non-coethnic candidate’s access to the state water bureaucracy can decrease the coethnic advantage and increase the credibility of a non-coethnic candidate. This is particularly the case among voters least satisfied with their water supply and most reliant on private sources of water. However, contrary to literature that finds that criminality can signal competence or the likelihood of goods and services being directed to coethnics, ties to the illegal water mafia do not offer either coethnic or non-coethnic candidates any additional advantage.

Appendices

Appendix 1

Table 5

Table 5 Sources of Water for Bathing and Cleaning

Appendix 2

Figure 4

Fig. 4

Reported tanker use as secondary source of household water. n = 2038. Respondents aggregated to 1998 Union Council boundaries

Appendix 3 Sampling Strategy

The household survey was administered among 2038 respondents in Karachi. The survey was carried out by the Pakistan Institute of Public Opinion (an affiliate of Gallup International in Pakistan), a public opinion and research firm. The survey was conducted in Urdu by teams of male and female enumerators.

The survey was randomly administered in five of Karachi’s six districts, with the predominantly rural/suburban district, Malir, excluded. The sample was divided into 205 primary sampling units (PSUs). Within each district, we used the census block population as per Pakistan’s 2017 Census to assign a proportional number of sampling units.

In the absence of official census maps released by the government, the PSUs were mapped onto proximate polling stations. These served as the starting point of the random walk for the household selection for the PSU. Enumerators used the right-hand rule, where the interviewer would enter the first street from the starting point and interview the first house. Within the house, all 18 + individuals were listed, and the Kish Grid method was used to identify a random male or female in the household. Enumerators interviewed every third house in the neighborhood until the sample was completed.

Respondents were randomly assigned to one of the four treatments. Randomization was conducted on ODK, and the survey was read out loud by an enumerator, who recorded the answers on a tablet. Physical show-cards were created (see Appendix 7) and shown to respondents based on a number that appeared on the tablet after a respondent’s ethnicity was entered. Randomization was successful (see balance tests in Appendix 6).

Appendix 4 Complete text of Experiment (Vignettes that a Muhajir Respondent Received)

Treatment 1

Suppose that in the upcoming local government elections, the top two candidates from this area have said that they will prioritize water if they are elected. Suppose that of the two candidates, one belongs to the Muhajir ethnic group and one belongs to the Punjabi ethnic group. Compared to the Punjabi candidate, the Muhajir candidate has stronger ties with the management, engineers, and linemen of the Karachi Water & Sewerage Board, and has highlighted your constituency as an area where he intends to use these links to increase clean water supply and resolve sewerage issues. That is, the Muhajir candidate has linkages to the KWSB.

Treatment 2

Suppose that in the upcoming local government elections, the top two candidates from this area have said that they will prioritize water if they are elected. Suppose that of the two candidates, one belongs to the Muhajir ethnic group and one belongs to the Punjabi ethnic group. Compared to the Muhajir candidate, the Punjabi candidate has stronger ties with the management, engineers, and linemen of the Karachi Water & Sewerage Board, and has highlighted your constituency as an area where he intends to use these links to increase clean water supply and resolve sewerage issues. That is, the Punjabi candidate has linkages to the KWSB.

Treatment 3

Suppose that in the upcoming local government elections, the top two candidates from this area have said that they will prioritize water if they are elected. Suppose that of the two candidates, one belongs to the Muhajir ethnic group and one belongs to the Punjabi ethnic group. Compared to the Punjabi candidate, the Muhajir candidate has stronger ties with the water tanker mafia, and has highlighted your constituency as an area where he intends to use these links to increase clean water supply and resolve sewerage issues. That is, the Muhajir candidate has linkages to the water mafia.

Treatment 4

Suppose that in the upcoming local government elections, the top two candidates from this area have said that they will prioritize water if they are elected. Suppose that of the two candidates, one belongs to the Muhajir ethnic group and one belongs to the Punjabi ethnic group. Compared to the Muhajir candidate, the Punjabi candidate has stronger ties with the water tanker mafia, and has highlighted your constituency as an area where he intends to use these links to increase clean water supply and resolve sewerage issues. That is, the Punjabi candidate has linkages to the water mafia.

Appendix 5

Table 6

Table 6 Descriptive Statistics

Appendix 6 Balance tests

Table 7

Table 7 Difference in means (with t-test p values) between covariate mean for respondents in T1 (Coethnic KWSB treatment) and _______.

In 30 t-tests, we find one variable where the difference in means exceeds the p < 0.10 level and one where it exceeds the p < 0.05 level, which is consistent with expectations. Given that this is what we would expect by chance, we can be confident that the randomization succeeded in achieving balance on demographic and partisan covariates.

Given the presence of multiple treatment conditions, we fit a multinomial logistic regression model to the following equation:

$$Version=\alpha +{\beta }_{1}Female+{\beta }_{2}Age+{\beta }_{3}Muhajir+{\beta }_{4}Shia+{\beta }_{5}Income+{\beta }_{6}SatisfiedWater+{\beta }_{7}Education+{\beta }_{8}Voter+{\beta }_{9}TankerUser+{\beta }_{10}PTI+\varepsilon$$

The test for joint orthogonality for the model as a whole fails to reject the null (p = 0.99). This further contributes to our confidence that randomization succeeded in achieving balance on covariates.

Appendix 7 Show-Card Examples (Provided to a Punjabi voter)

Urdu translation: Candidate 1 (Muhajir)/Candidate 2 (Punjabi).

Urdu translation: Candidate 1 (Muhajir)/Candidate 2 (Punjabi).

Appendix 8

Table 8

Table 8 Likelihood of Voting for Coethnic Candidate in Each Treatment

Appendix 9

Table 9

Table 9 Difference in means, Likelihood of Voting for Coethnic Candidate Across Treatment Groups

Appendix 10 Assessment of candidate credibility and helpfulness by ethnicity in each treatment

Tables 10 and 11

Table 10 Candidate credibility

Table 11 Candidate Helpfulness

Appendix 11

Table 12

Table 12 Diff-in-means of perceived candidate credibility and helpfulness, across treatment group

Credibility was measured in response to the following question: “On a scale of 1–5, with 1 being not at all credible and 5 being very credible, how credible do you find [the] candidate’s promise that they will improve water in your area?” Helpfulness was measured in response to: “On a scale of 1–5, with 1 being not at all helpful and 5 being very helpful, how helpful do you think [the] candidate will be to you and people like you over others?”.

Appendix 12

Table 13

Table 13 Perceptions of Candidate Credibility and Helpfulness Across Treatments. Baseline Group: Coethnic Candidate with Water Tanker Mafia (T3)

Appendix 13

Table 14

Table 14 Perceptions of candidate credibility and helpfulness across treatments, full model with covariates. Baseline group: coethnic candidate with water tanker mafia ties

Appendix 14

Table 15

Table 15 Correlates of Satisfaction with Water

Appendix 15

Tables 16 and 17

Table 16 Results for Voters Who Reported Being Satisfied with Water Provision

This table is limited to 941 respondents who we code as “satisfied” with their water provision. Respondents were asked to what extent they agreed with the statement, “The water facilities in my area are satisfactory” on a 1–5 scale. Those who selected “somewhat agree” (541), “strongly agree” (541), and “neither agree nor disagree” (92) were coded as “satisfied.”

Table 17 Results for Voters Who Reported Being Unsatisfied with Water Provision

Appendix 16

Table 18

Table 18 Likelihood of Voting for Coethnic Candidate with KWSB Ties Compared to When Non-Coethnic Candidate Has KWSB Ties, Full Interaction Model (DV: 2 = Vote for Coethnic Candidate, 1 = vote for Non-Coethnic Candidate)

Appendix 17

Table 19

Table 19 Likelihood of voting for coethnic candidate with water tanker mafia ties compared to when non-coethnic candidate has water mafia ties, full interaction model (DV: 2 = vote for coethnic candidate, 1 = vote for non-coethnic candidate)

Appendix 18

Tables 20 and 21

Table 20 Results in Ethnically Homogenous Neighborhoods

This table is limited to 1246 respondents who live in sampling clusters that had more than one ethnic group resident. Each primary sampling unit (PSU) was assigned a value based on the number of unique ethnic groups within it. PSU scores take a minimum value of 1 to a maximum of 6; the mean score was 2.57 (SD 0.029). PSUs with more than one ethnic group present were categorized as diverse.

Table 21 Results in Ethnically Heterogenous Neighborhoods

Appendix 19 Deviations From Pre-Analysis plan

We pre-registered our study with EGAP (EGAP # 20211217AA) in December 2020 prior to fielding the experiment. The full version of the PAP is available here: https://doi.org/https://doi.org/10.17605/OSF.IO/Q3DBU.

In the pre-analysis plan (PAP), we include discussion of our theory and potential mechanisms. We also describe our experimental treatment and hypotheses. Our analysis in the manuscript deviates from the PAP in a few ways. We do not include in this manuscript a sub-group hypothesis pertaining to strength of ethnicity in respondent likelihood of voting for coethnic politicians, as well as hypotheses regarding the main treatments that were duplicative of exiting hypotheses. We also made one change to our primary outcome variable after piloting the survey. In the PAP, we asked respondents to rank their likelihood of voting for either candidate presented to them in the experiment. Following piloting, we found that this was confusing and non-intuitive for many respondents. We therefore replaced this outcome variable with a forced choice vote for either Candidate 1 or Candidate 2.