4.1 Ordinary Least Squares Results

1I begin by investigating the relationship between religious heterogeneity and public goods using a cross-sectional data set as is common in the literature. I use the OLS specification drawn from Alesina et al. (1999) with slight modifications as illustrated in Section 3.1. I use the sub-district level Census of India data for 2001 and 2011 and district-level data for 1991.

2I report the detailed results in the appendix from Table 1 to Table 9. I find that the public goods provided by the national/provincial government have a negative and statistically significant relationship with the Religious Polarization index. This includes public goods such as primary schools, middle schools, secondary schools, roads and access to electricity. These results are robust to additional controls="true" such as regional GDP proxied with nighttime luminosity, population growth, mean literacy level, area size, mean rainfall and mean temperature.

3My results are ambiguous for goods provided at the local village/town level through local authorities or collective action. I find a positive relationship between religious polarization and availability of public toilets and tube wells while the relationship is negative for public taps and other wells.

4I propose a deep dive into interpreting my coefficients, given the per capita measures as the dependent variables and non-linear elements in the regressions. A decimal unit increase in the Religious Polarization index results in a 0.001 decrease in primary schools per capita (Table 1). The mean primary schools per capita in India are deficient at 0.0011050, with an average of 20.59 schools in every sub-district. Therefore, a unit increase in the Religious Polarization index decreases the average number of schools to 19.65. Notwithstanding the coefficients’ small magnitudes, I believe that these results have considerable influence on the total quantity of public goods supplied by national/provincial government institutions. Further, the share of religious groups enters my model in two ways: 1) RQ index and 2) religious group size and its squared values. This allows me to look at the heterogeneous effects of the presence of different group sizes. I observe that the larger group size of Christian and Muslim communities results in a lower supply of public goods delivered while the relationship is positive for Hindu and Sikh communities. This result stems from the social antagonisms biased toward Muslims and Christians in India (Pandya, 2010).

4.2 Two-way Fixed Effects Results

5The results from the two-way fixed effects model are summarized in Table 4.1 to Table 4.3. I use a two-way specification to account for sub-district-invariant and time-invariant factors and cluster standard errors at the sub-district level.

6My results suggest that religious polarization generally has a negative and significant effect on public goods provision by national/provincial government institutions, while it has a surprisingly positive effect on locally provided goods. I hypothesize this positive effect results from patronage along religious lines at the village/town level by political actors (Bhattacharya et al., 2018). These results are robust to additional controls="true" such as regional GDP proxied with nighttime luminosity, population growth, mean literacy level, area size, mean rainfall and mean temperature.

7Further, my results bring to light the existence of heterogeneity in the effect of population shares for different religious groups, thereby confirming the merits of the chosen identification strategy. I solve for the combined effects of the linear and quadratic terms for the group size of each religious group and find a positive impact of Hindu and Sikh group size and a negative impact of Muslim and Christian group size; however, the magnitude of these estimates is smaller than that obtained from the OLS estimates.

Table 4.1: FE estimation of RQ on Educational Facilities

Table 4.2: FE estimation of RQ on Infrastructure

Table 4.3: FE estimation of RQ on Local Facilities

4.3 Dynamic Spatial Autoregressive Model Results

8I summarize the results from the dynamic spatial autoregressive model in Table 4.4 to Table 4.6. I employ a one-way fixed effects model with spatial invariant effects and cluster at the sub-district level. Religious polarization measured through the RQ index has a negative and statistically significant impact on the availability of primary schools, middle schools, senior schools, colleges, roads, toilets, taps and tube wells. Religious polarization also shares a negative relationship with secondary schools, electricity and wells but this relationship is not statistically significant. My results remain robust for controls="true" such as regional GDP proxied with nighttime luminosity, population growth, mean literacy level, area size, mean rainfall and mean temperature. Further, I find that caste fragmentation positively impacts the availability of education, infrastructure and local facilities. My results corroborate the findings of Adukia et al. (2020), who posit that cities with higher populations of lower caste individuals have better access to schools, doctors and government hospitals.

9I also report the effect of population shares for different religious groups. I only use linear shares of religious groups in the spatial regressions due to computational constraints. On average, I find that higher Hindu and Sikh shares positively impact the availability of these goods, while higher Muslim and Christian shares have a negative impact on the availability of these goods. However, the effect sizes on the religious population shares are smaller and insignificant relative to the OLS and fixed effects panel data analysis coefficients.

10I observe strong path dependence for all public goods considered. My model’s lag term has a positive and statistically significant impact on public goods’ availability in the next period. Further, I observe positive, statistically significant neighborhood effects at the sub-district level for all considered public goods. This finding, as far as I know, is the first to report the relationship between neighborhood effects and access to public goods in the context of India.

Table 4.4: Spatial estimation of RQ on Educational Facilities

Table 4.5: Spatial estimation of RQ on Infrastructure

Table 4.6: Spatial estimation of RQ on Local Facilities

4.4 Estimated Effect Size

11I summarize the results from the OLS, fixed effects and spatial regressions in Table 4.7 below. I observe my effect sizes to be the largest and statistically significant in the spatial regression for primary schools, middle schools, senior schools, secondary schools, colleges and electricity, while it is the largest in the OLS regression for roads, toilets, taps, wells and tube wells. Further, I use the spatial regression results to compute the average estimated effect size in nominal terms for all these variables. With a decimal unit increase in the RQ index, I observe a reduction of 6.4% in the number of primary schools, 1.1% in the number of middle schools, 0.7% in the number of senior schools, 10.1% in the number of colleges, 1.6% in the number of roads, 0.26% in the number of public toilets, 1% in the number of public taps and 0.84% in the number of public tube wells. I do not observe a statistically significant change in the number of secondary schools, electricity supply or public wells.

Table 4.7: Effects sizes of RQ on Public Goods

4.5 Possible Mechanisms

12I discuss the possible mechanisms through which religious polarization can impact public goods’ availability at the sub-district level. These mechanisms are speculative due to the difficulty establishing the exact individual and community behavior using sub-district level aggregates.

4.5.1 Collective Dislike

13I extend the collective model described in The Economics of Discrimination (Becker, 1957). In collective models, groups act collectively against each other and hold a preference for discrimination. Individual and groups observe disutility in residing, interacting and working with the out-group and have strict preferences for the in- group. This in-group preference and out-group disutility have resulted in strong segregation of population settlements in India. Miklian and Sahoo (2016) find that Muslims and members of Scheduled Castes face discriminatory treatment in the form of lack of access to housing, education and health care in urban areas. They find that segregated settlements are formed along caste and religious group boundaries. Their findings are corroborated by ethnographers such as Gist (1957), Gould (1965) and Susewind (2017), who describe the persistence of residential segregation in urban India.

14This has contributed to accelerating the growth of religious identity politics. Colloquially termed identity politics in India, this segregation of communities allows political agents to pitch the dominant and competing groups against each other to stoke fear and disharmony (Hasan, 2010). Political parties and religious groups have used this segregation by promoting identity issues to build their political capital, with a retracted focus on social and economic issues, reducing access to public goods.

4.5.2 Different Preferences

15Alesina et al. (1999) posit that preferences for goods differ across groups. However, due to financial and technological constraints, only a few varieties of goods can be provided. Sub-districts with a higher degree of religious polarization may have communities refraining from making contributions since they will not get their preferred goods. In the Indian context, this situation is applicable in the case of education goods, where different religious groups may have a divergent preference regarding the education goods that are provided publicly due to the Minority Educational Institutions Act, which was enacted to safeguard the educational rights of minorities enshrined in Article 30(1) of the Indian Constitution.