5.1 Alternative Specifications of α

Table 5.1: RQ index with varying degrees of polarization (α)

1Table 5.1 above summarizes the coefficients on the Religious Polarization index for varying degrees of polarization measured by the parametric value of α. I use the normalized Polarization index as defined in Montalvo and Reynal-Querol which can be written as follows:

2Here π_i is the share of group i (i = 1.., N ) in sub district j for some constant K > 0 (normalized to K = 1) and polarization sensitivity α ∈ (0, 1.6]. I use four values of polarization sensitivity – 0, 0.4, 1.2 and 1.6 – in the analysis. I compare the results for various parametric values with my RQ index calculated at the median value of 0.8.

3I observe the extreme values of α at 0 and 1.6 to have a negative impact on the supply of public goods but the magnitude of the results is smaller compared to the estimates from the median value of α. Interestingly, the estimates lose their significance with the extreme values of α. For the 25th and 75th percentile values of α corresponding to 0.4 and 1.2, my estimates are statistically significant and larger than median value estimates. As earlier, locally supplied goods such as toilets, taps, wells and tube wells exhibit mixed results, which should be interpreted cautiously. My findings are consistent with those of Alesina et al. (2003) who find that the RQ performs better closer to the median values relative to the extreme values.

5.2 Robustness to City Size

4Banerjee et al. (2007) focus on the concentration of public goods in specific geographies in India. These include large cities and the areas around them. Interestingly, large Indian cities tend to be spatially fragmented along religious lines (Singh and Sethi, 2018).

5I test for the robustness of my results by focusing on a sub-sample of areas by systematically removing big cities and areas mentioned in Banerjee et al. (2007) to ensure that my results are not systematically biased by the presence of densely populated and ethnically fragmented areas in India. These results are summarized in Table 22 to Table 24. I observe a loss in the statistical significance because some variables which were statistically significant in my full sample are no longer significant, while remaining variables are now significant at 10% compared to 5% in my main estimates. I also observe a reduction in the estimated coefficients for my sub-sample which is likely due to the omission of larger cities with greater fragmentation and a higher concentration of public goods. However, generally my religious polarization index shows a negative effect on the supply of public goods provided by national/provincial government institutions. As discussed previously, I am unable to draw any significant conclusions for locally supplied public goods and their effect remains ambiguous.

5.3 Robustness to Larger Share of Minority Groups

6Horowitz (1985) hypothesizes that ethnic antagonisms are central in places where a sizeable ethnic minority faces an ethnic majority. In the sample I have a sizeable Hindu majority followed by smaller group sizes of other religious communities. To further understand the role of social antagonisms in the supply of public goods, I restrict my analysis to Hindu majority areas where another community forms a sizeable minority, i.e., has a population share greater than 40%. This is in contrast to the average effect estimated from the full sample fixed effects regressions.

7I observe the average effect of religious polarization to be statistically significant and negative for my sub-sample and my estimated coefficients are significantly higher than the coefficient obtained from my baseline fixed effects model. These results are summarized in Table 25 to Table 27. Strikingly, all the public goods supplied by national/provincial government institutions have a strong negative relationship with the religious polarization index. I am unable to draw any tangible conclusions for the locally provided public goods.

8I attribute the larger negative effect of religious polarization on public goods supply to increased social antagonisms in areas with two large and demographically comparable ethnic groups. My findings are consistent with the existing literature and leave further scope to understand the differences in the role of antagonisms across varying minority group sizes.