Adam Michael Auerbach and Tarig Thachil First,clients may prefer brokers with incumbent par- spects.16 First,prior fieldwork informed our selection tisan connectivity,namely ties to the incumbent party of a forced-choice design.Such frameworks better ap- Clients might anticipate such connectivity will help proximate the competitive and voluntary processes brokers access state resources.Indeed,to the degree that define slum leader selection in India.Irrespective top-down models expect clients to have any prefer- of whether they selected leaders through community ences,such incumbent connectivity should head the list. meetings or everyday decisions,slum residents made If client support is a mechanical response to targeted defined choices about whom to seek help from.consis- handouts entirely controlled by party elites,then a bro- tent with our experimental setup. ker's efficacy is largely a function of their ties to these Second,ethnography provided us with a context- elites.However.broker connectivity can also be estab- sensitive way in which to operationalize this se- op//s lished through nonpartisan channels.For example,vot- lection procedure.We presented respondents with ers might prefer brokers with bureaucratic connectivity two hypothetical slum residents running to be pres- to government departments responsible for public ser- ident of a vikas samiti (development committee). vice delivery (Stokes 1995). These neighborhood associations are common orga- Our client-centric perspective also highlights a bro- nizations through which Indian slum dwellers make ker's individual claim-making capabilities in informing claims.7 We leverage the structure of the development their efficacy in lobbying for their clients (Auvero 2000: committee-headed by a president-to ground our ex- Krishna 2002).Prior studies have noted education.in periment in a process of leadership selection familiar to particular,can improve a broker's ability to effectively respondents.Third,our ethnography helped us opera- petition for state services.These studies range from ru- tionalize core concepts into simple,contextually mean- ral India(Manor 2000:Krishna 2002)to Peruvian and ingful candidate attributes.Respondents were given Venezuelan slums (Ray 1969;Stokes 1995).In Indian five pieces of randomized information about each can- slums,Jha,Rao,and Woolcock(2007)also describe in- didate,and then asked to select which would make a formal leaders as well educated.Yet these studies do better leader (full question wording in SI Section S.1) not examine whether this descriptive fact is fueled by Below,we describe how each concept was operational- client preferences for educated leaders,or assess the ized(SI Section S.3 provides the list of treatments). weight of education vis-a-vis other concerns in shaping broker selection. Ethnicity India houses several forms of ethnic categorization. RESEARCH DESIGN Prior studies sometimes use the term "ethnic"to re- fer to single dimensions of ethnicity,notably caste How can we precisely identify the relative weight of (Chandra 2004)or religion (Wilkinson 2004).Here, coethnicity,copartisanship,capability,and incumbent we compare multiple dimensions of ethnicity in slum partisan and bureaucratic connectivity within client leader selection.First,we examine the salience of jatis preferences for brokers?We address this question endogamous subcastes that denote traditional occupa- through a forced-choice conjoint survey experiment.In tions.are highly localized.and number in the hundreds this setup,respondents are presented with information across India.Jatis are nested within broader caste status S5.501g regarding randomized attributes of two slum leaders. groups,indicating whether a jati is considered high or Respondents were then asked which of the two they low caste. prefer. Our treatments varied a leader's name,which indi- This approach has become increasingly popular in cates their subcaste.Respondents were assigned(with the study of political behavior,because it enables re- equal probability)to evaluate a potential slum leader searchers to estimate the causal effects of several treat- from their own jati,8 one of three well-known upper ment components simultaneously (Hainmueller and caste Hindu jatis,one of three well-known lower caste Hopkins 2015).This design also allows us to disentan- Hindu jatis,or one of three well-known Muslim jatis.19 gle the effects of observationally correlated attributes This created a jati match or mismatch between the re- such as caste and party preference.Furthermore,con- spondent and leaders.These names also identified a joint experiments have the potential to reduce socia candidate as Hindu or Muslim.This treatment allowed desirability concerns because they offer respondents us to classify respondents as ethnic matches or mis- the confidentiality of several potential justifications for matches on the broader dimension of religion a decision. Finally,given the multiregional nature of Indian Despite these advantages,we are cognizant of con- slums,we assess the salience of region-of-origin differ- cerns with increasingly complex survey experiments ences by randomizing each leader's home state.Slum These concerns often stem from boilerplate designs that prioritize a researcher's theoretical interest at the 16On ethnographically informed surveys,see Thachil(forthcoming). expense of contextual resonance.Such construct va- 17 See Auerbach(2017). lidity concerns are especially high when working with 18 The respondent's jati was asked at the beginning of the survey.The poorly understood communities. instrument ensured a gap of at least 20 questions between this ques- To improve the validity of our design,we draw tion and the conjoint experiment. We include several jatis within each status level to ensure esti on a combined three years of fieldwork among In- mated effects were not driven by comparisons with any one particu- dia's urban poor to enhance our design in three re- lar jati. 780
Adam Michael Auerbach and Tariq Thachil First, clients may prefer brokers with incumbent partisan connectivity, namely ties to the incumbent party. Clients might anticipate such connectivity will help brokers access state resources. Indeed, to the degree top-down models expect clients to have any preferences, such incumbent connectivity should head the list. If client support is a mechanical response to targeted handouts entirely controlled by party elites, then a broker’s efficacy is largely a function of their ties to these elites. However, broker connectivity can also be established through nonpartisan channels. For example, voters might prefer brokers with bureaucratic connectivity to government departments responsible for public service delivery (Stokes 1995). Our client-centric perspective also highlights a broker’s individual claim-making capabilities in informing their efficacy in lobbying for their clients (Auyero 2000; Krishna 2002). Prior studies have noted education, in particular, can improve a broker’s ability to effectively petition for state services. These studies range from rural India (Manor 2000; Krishna 2002) to Peruvian and Venezuelan slums (Ray 1969; Stokes 1995). In Indian slums, Jha, Rao, and Woolcock (2007) also describe informal leaders as well educated. Yet these studies do not examine whether this descriptive fact is fueled by client preferences for educated leaders, or assess the weight of education vis-à-vis other concerns in shaping broker selection. RESEARCH DESIGN How can we precisely identify the relative weight of coethnicity, copartisanship, capability, and incumbent partisan and bureaucratic connectivity within client preferences for brokers? We address this question through a forced-choice conjoint survey experiment. In this setup, respondents are presented with information regarding randomized attributes of two slum leaders. Respondents were then asked which of the two they prefer. This approach has become increasingly popular in the study of political behavior, because it enables researchers to estimate the causal effects of several treatment components simultaneously (Hainmueller and Hopkins 2015). This design also allows us to disentangle the effects of observationally correlated attributes, such as caste and party preference. Furthermore, conjoint experiments have the potential to reduce social desirability concerns because they offer respondents the confidentiality of several potential justifications for a decision. Despite these advantages, we are cognizant of concerns with increasingly complex survey experiments. These concerns often stem from boilerplate designs that prioritize a researcher’s theoretical interest at the expense of contextual resonance. Such construct validity concerns are especially high when working with poorly understood communities. To improve the validity of our design, we draw on a combined three years of fieldwork among India’s urban poor to enhance our design in three respects.16 First, prior fieldwork informed our selection of a forced-choice design. Such frameworks better approximate the competitive and voluntary processes that define slum leader selection in India. Irrespective of whether they selected leaders through community meetings or everyday decisions, slum residents made defined choices about whom to seek help from, consistent with our experimental setup. Second, ethnography provided us with a contextsensitive way in which to operationalize this selection procedure. We presented respondents with two hypothetical slum residents running to be president of a vikas samiti (development committee). These neighborhood associations are common organizations through which Indian slum dwellers make claims.17 We leverage the structure of the development committee—headed by a president—to ground our experiment in a process of leadership selection familiar to respondents. Third, our ethnography helped us operationalize core concepts into simple, contextually meaningful candidate attributes. Respondents were given five pieces of randomized information about each candidate, and then asked to select which would make a better leader (full question wording in SI Section S.1). Below, we describe how each concept was operationalized (SI Section S.3 provides the list of treatments). Ethnicity India houses several forms of ethnic categorization. Prior studies sometimes use the term “ethnic” to refer to single dimensions of ethnicity, notably caste (Chandra 2004) or religion (Wilkinson 2004). Here, we compare multiple dimensions of ethnicity in slum leader selection. First, we examine the salience of jatis, endogamous subcastes that denote traditional occupations, are highly localized, and number in the hundreds across India. Jatis are nested within broader caste status groups, indicating whether a jati is considered high or low caste. Our treatments varied a leader’s name, which indicates their subcaste. Respondents were assigned (with equal probability) to evaluate a potential slum leader from their own jati, 18 one of three well-known upper caste Hindu jatis, one of three well-known lower caste Hindu jatis, or one of three well-known Muslim jatis. 19 This created a jati match or mismatch between the respondent and leaders. These names also identified a candidate as Hindu or Muslim. This treatment allowed us to classify respondents as ethnic matches or mismatches on the broader dimension of religion. Finally, given the multiregional nature of Indian slums, we assess the salience of region-of-origin differences by randomizing each leader’s home state. Slum 16 On ethnographically informed surveys, see Thachil (forthcoming). 17 See Auerbach (2017). 18 The respondent’s jati was asked at the beginning of the survey. The instrument ensured a gap of at least 20 questions between this question and the conjoint experiment. 19 We include several jatis within each status level to ensure estimated effects were not driven by comparisons with any one particular jati. 780 Downloaded from https://www.cambridge.org/core. Shanghai JiaoTong University, on 26 Oct 2018 at 03:53:05, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/S000305541800028X
How Clients Select Brokers leaders were randomly assigned to come from the Education is valued because it is seen to improve a respondent's home state,the state of the study city. leader's practical abilities to engage in written claim- another prominent source state within north India's making.22 Our fieldwork unearthed numerous exam- "Hindi belt,"or a prominent source state from a dif ples of claims made through leader-written applica- ferent linguistic region of India. tions.23 For example: The salience of shared jati,faith,or regional iden- tities is provided by the difference in probability of a Since last year we have been suffering from water scarcity. leader being preferred when they are coethnics with a At times,we have to go to the factories or the cremation respondent on that dimension,compared to when they grounds for water...We are in trouble and request that you are not. take action.24 olop//s Each of these treatments assesses the horizontal con- cept of coethnicity.However,ethnic categories also ver- Education also signals a slum leader's ability to navi- tically partition society into groups of unequal status. gate complex state institutions,interact with public offi- In India,lower caste Hindus and Muslims are socioeco- cials,and stay abreast with government policies for the nomically marginalized,relative to upper caste Hindus. urban poor: Our experiment's design allows us to assess how this vertical hierarchy affects respondent preferences.To do I was educated.So I knew about the policies...I was always 4 so,we include dummy variables identifying lower caste in search for any loans with which people could find em- and Muslim leaders ployment and gain something...There are many policies through which our worker brothers can benefit. Partisanship To vary a leader's claim-making capability,we ma- We randomly assign leaders to be affiliated with one nipulated their level of education.Leaders were ran- of the two major parties in our study cities,the In- domly assigned to have no schooling,an eighth grade 4号 dian National Congress (INC)and Bharatiya Janata education,or a college B.A.Our fieldwork confirmed Party (BJP),or to be nonpartisan independents.Co- that each of these manipulations was realistic:our sur- partisanship was then coded by matching leader parti- vey found 40%of slum residents had at least an eighth san profiles with the partisan preferences expressed by grade education and 8.9%had at least some college residents. education.26 This treatment also indicated whether the leader was affiliated with the local incumbent party.Slum lead- (Bureaucratic)Connectivity ers who enjoy such partisan connectivity might find it easier to have requests met.The BJP was the incum- A final attribute we sought to manipulate was per- bent party at the state and municipal levels in our study ceptions of a candidate's connectedness to urban bu- 是 cities.We therefore coded all hypothetical leaders be- reaucracies.Residents may value leaders whom they longing to the BJP as incumbents,those belonging to perceive to be connected with municipal authorities. the Congress as opposition,and the rest as nonpartisan Bureaucratically connected brokers may be regarded independents. as more likely to be informed about the dizzying array of government benefits residents might be eligible for, Capability and better able to pressure municipal personnel into providing benefits. Our fieldwork revealed that slum residents were con- Our fieldwork revealed occupations to be a useful in- cerned with whether a broker possessed the raw capa- dicator of bureaucratic connectivity.The range of jobs bility to lobby public officials. we found Indian slum leaders engaging in enabled re- alistic experimental variations of each leader's job,and We have chosen them as leaders for a reason-they have hence perceptions of their connectivity.We preferred information and knowledge,and perhaps connections,so they should get our work done.20 this conceptualization to several alternatives.First.us ing a treatment that explicitly specified a level of con- nectivity(Candidate A has a high/medium/low level of In interviews,slum leaders underscored how their educational qualifications were often used as a mea- connectivity)can induce social desirability bias.Such evaluative statements that provide an ordering of can- sure of such capabilities: didates carry strong normative connotations that one See,here in the slum,we have only poor people.Most peo- ple are uneducated.So when there is an issue,they need 22 We do not believe education is valued because it signals a resident as wealthier,and hence perhaps more powerful within the city.In help in writing applications.So they began coming to me, fact,education only weakly correlates with household income in our saying brother fill out this application for me...slowly peo- ple told others I do]this kind of work...that's how I built mple(0.192). Figure S.6 provides example slum development council letterhead my early support base.21 stationery used to make claims. 24 Saraswati petition letter,Jaipur,late 2000s 25 Interview with Pramod,an informal leader in Anna Slum,Bhopal 士 20 Interview with Kamal Nagar.Resident 13.August 2017 (June272016). 21 Interview with Sen,an informal leader in Ganpati Slum,Jaipur 23.35%of urban Indians have finished high school,and 12.18% (June1,2016). have finished college(2011 Census of India). 781
How Clients Select Brokers leaders were randomly assigned to come from the respondent’s home state, the state of the study city, another prominent source state within north India’s “Hindi belt,” or a prominent source state from a different linguistic region of India. The salience of shared jati, faith, or regional identities is provided by the difference in probability of a leader being preferred when they are coethnics with a respondent on that dimension, compared to when they are not. Each of these treatments assesses the horizontal concept of coethnicity.However, ethnic categories also vertically partition society into groups of unequal status. In India,lower caste Hindus and Muslims are socioeconomically marginalized, relative to upper caste Hindus. Our experiment’s design allows us to assess how this vertical hierarchy affects respondent preferences.To do so, we include dummy variables identifying lower caste and Muslim leaders. Partisanship We randomly assign leaders to be affiliated with one of the two major parties in our study cities, the Indian National Congress (INC) and Bharatiya Janata Party (BJP), or to be nonpartisan independents. Copartisanship was then coded by matching leader partisan profiles with the partisan preferences expressed by residents. This treatment also indicated whether the leader was affiliated with the local incumbent party. Slum leaders who enjoy such partisan connectivity might find it easier to have requests met. The BJP was the incumbent party at the state and municipal levels in our study cities. We therefore coded all hypothetical leaders belonging to the BJP as incumbents, those belonging to the Congress as opposition, and the rest as nonpartisan independents. Capability Our fieldwork revealed that slum residents were concerned with whether a broker possessed the raw capability to lobby public officials. We have chosen them as leaders for a reason—they have information and knowledge, and perhaps connections, so they should get our work done.20 In interviews, slum leaders underscored how their educational qualifications were often used as a measure of such capabilities: See, here in the slum, we have only poor people.Most people are uneducated. So when there is an issue, they need help in writing applications. So they began coming to me, saying brother fill out this application for me…slowly people told others [I do] this kind of work…that’s how I built my early support base.21 20 Interview with Kamal Nagar, Resident 13, August 2017. 21 Interview with Sen, an informal leader in Ganpati Slum, Jaipur (June 1, 2016). Education is valued because it is seen to improve a leader’s practical abilities to engage in written claimmaking.22 Our fieldwork unearthed numerous examples of claims made through leader-written applications.23 For example: Since last year we have been suffering from water scarcity. At times, we have to go to the factories or the cremation grounds for water…We are in trouble and request that you take action.24 Education also signals a slum leader’s ability to navigate complex state institutions,interact with public officials, and stay abreast with government policies for the urban poor: I was educated. So I knew about the policies…I was always in search for any loans with which people could find employment and gain something…There are many policies through which our worker brothers can benefit.25 To vary a leader’s claim-making capability, we manipulated their level of education. Leaders were randomly assigned to have no schooling, an eighth grade education, or a college B.A. Our fieldwork confirmed that each of these manipulations was realistic: our survey found 40% of slum residents had at least an eighth grade education and 8.9% had at least some college education.26 (Bureaucratic) Connectivity A final attribute we sought to manipulate was perceptions of a candidate’s connectedness to urban bureaucracies. Residents may value leaders whom they perceive to be connected with municipal authorities. Bureaucratically connected brokers may be regarded as more likely to be informed about the dizzying array of government benefits residents might be eligible for, and better able to pressure municipal personnel into providing benefits. Our fieldwork revealed occupations to be a useful indicator of bureaucratic connectivity. The range of jobs we found Indian slum leaders engaging in enabled realistic experimental variations of each leader’s job, and hence perceptions of their connectivity. We preferred this conceptualization to several alternatives. First, using a treatment that explicitly specified a level of connectivity (Candidate A has a high/medium/low level of connectivity) can induce social desirability bias. Such evaluative statements that provide an ordering of candidates carry strong normative connotations that one 22 We do not believe education is valued because it signals a resident as wealthier, and hence perhaps more powerful within the city. In fact, education only weakly correlates with household income in our sample (0.192). 23 Figure S.6 provides example slum development council letterhead stationery used to make claims. 24 Saraswati petition letter, Jaipur, late 2000s. 25 Interview with Pramod, an informal leader in Anna Slum, Bhopal (June 27, 2016). 26 23.35% of urban Indians have finished high school, and 12.18% have finished college (2011 Census of India). 781 Downloaded from https://www.cambridge.org/core. Shanghai JiaoTong University, on 26 Oct 2018 at 03:53:05, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/S000305541800028X