Last week at the AAS in Philly, I had an interesting discussion of votebank politics in India and the importance of spatial variation. My contention was that most politics are local, and that electoral dynamics such as Muslim votebanks (i.e. Muslims voting for certain parties) and the extent of ethnic coordination (i.e. Muslims voting for Muslim candidates) depend on largely local factors. Some people disagreed, many agreed - but it remained a gut feeling. Until, on the flight back, I got an idea how to prove my point. This brief post thus explains at which level votebanks form and operate in India (well, in one instance at least)...

I look at this at the example of the last assembly elections in Uttar Pradesh and Muslim / non-Muslim votebanks (based on earlier papers here and here, and this blog post) - but I assume the picture will be largely similar in the current national elections (more on this once results are out in a couple of weeks). The idea is pretty straightforward: I took the last version of my votebank model and ran it on various levels (across UP, district by district, parliamentary constituency by parliamentary constituency, assembly constituency by assembly constituency) - and compared the R squared of the results as a proxy for explanatory power.1

Here are the results of this little experiment: across UP, the model explains 10% of the vote pattern; district by district, the explanatory power rises to a median 16%; parliamentary constituency by parliamentary constituency it stands at a median of 12% (an interesting one, that one - any idea why votebanks by district are stronger than votebank by PC? Comment below, please...); but the highest R squared can be found if one runs the model assembly constituency by assembly constituency - on this level, it explains a median of 24% of the variation in election results (with a mean of 27% and a sd of 18%).

This means three things. First: even a rather complex model of (religious) votebanks does not explain the vote pattern very well at any level - the electoral pattern largely depends on factors other than religion. I feel this needs to be reiterated these days in particular (on the question of what else might matter to voters, watch out for future posts - or see the analyses of Devesh Kapur, Milan Vaishnav and Neelanjan Sircar at CASI's India in Transition blog). Secondly, however, votebanks - to the extent that they matter - explain electoral outcomes better the deeper one disaggregates: all politics, at least all votebank politics, are local; votebanks form and operate on the level of a constituency much more than on higher levels of abstraction. Thirdly and lastly, explanatory power varies across constituencies a lot (as does, incidentally, substantial votebank preferences - but that's another story) - on assembly constituency level, for instance, the model explains 4% of the vote pattern in the constituency where it worked least well and up to 83% of the pattern where it works really well. In short: all politics are local!

  • 1. Very briefly, the model is an SUR (Semi Unrelated Regression) model regressing voteshare differences from station mean of the four major parties within polling stations on difference in Muslim electorate share from station mean within polling stations, a rural/urban dummy and demographic factors (age and gender) - see this post for a justification of why the "within stations" bit works around the ecological fallacy...