There was a time in the 1990s when political commentators joked about Samajwadi Party (SP) supremo Mulayam Singh Yadav and his attempts to woo Muslim voters by calling him a "Maulana", an Islamic scholar. With the recent assembly elections in Uttar Pradesh, the talk of "Muslim vote banks" returned with vehemence - and many political observers attributed the clear SP victory to a "return" of Muslims to the party. In their surveys, political scientists did, however, not find any evidence for such a return (see here and here).

In today's blog - based on a draft paper I presented last week at Aligarh Muslim University - I argue that this contradiction might be resolved through closer attention to the local level. Based on my work with electoral roll data (see here, here and there), I unpack the "Muslim vote" hypothesis at the example of urban Lucknow. There, Muslims indeed voted more for SP - but not more than they always did (which might be different at different times, in other parts of Uttar Pradesh, or even in rural parts of Lucknow). This localized perspective nicely complements, but also complicates, existing assesments of the "Muslim vote". Since I am currently broadening the analysis to the whole of Uttar Pradesh, today's post is thus also an appetizer for a larger argument in the making.

Rather than boring you with the pros and cons of both census- and survey-based approaches to Muslim vote pattern with which my paper begins, let me jump directly to slides 8 and 9 in the Prezi above. What you can see there is first a map of SP voteshare in Lucknow (green areas are those where SP did better than average, red where it did worse) - and secondly a modified version of my Muslim map (darker areas indicating a stronger presence of electors with "Muslim names" on the electoral rolls of this area). If you switch back and forth between these two maps, you will see quite some intuitive support for the hypothesis that Muslims voted for SP more. Firstly, the old city West of the river has both higher-than-average votes for SP - and more Muslims. You can even see that, within the old city, the horizontal strip between Chowk and Husainabad has both fewer Muslims and fewer SP votes. But once you study the maps closer, you also find many areas where the calculus does not work that way - for example in the North-East of Lucknow. On first impression, there seems to be some relation between Muslims and the Samajwadi Party - but only in some areas, and only to a certain extent. In other words: the picture is more complicated

In order to get to the bottom of this, I thus propose to statistically exploit a "natural experiment" kind of situation: rather than correlating Muslim electors percentage with SP voteshare (the immediate strategy that comes to mind), I suggest to correlate the difference of either variable within a given polling station: do polling booths with more Muslims on their electoral rolls also have more votes for SP than other booths within the same polling station?

The answer is: indeed they do. The correlation coefficient stands at r=.35 and is highly significant; if I calculate a regression and control for both gender and age, I still arrive at a significant coefficient of .25 (with an R² of .14, which is not a big surprise considering that the voters' decisions depend on many other factors apart from the three captured in this regression). This figure basically means: Muslims did vote more for the Samajwadi Party than non-Muslims in the last assembly elections, at least in urban Lucknow.

Moreover: looking at the problem this way - by exploiting the differences within each polling station - has a number of advantages over census- or survey-based assessments (some of them outlined in the presentation). Above all, it implicitly controls (by means of a spatial bracket, and due to the fact that voters are assigned to a specific booth within their station on quasi-random basis) for many potential intercorrelations: we can be fairly certain that it is indeed Muslims who prefer SP and not, for instance, poor people.

Another key advantage of this method is that we can now decipher more closely in which areas this voting pattern holds - and where it breaks. The map below (larger version) for instance visualizes the covariation between SP votes and Muslim elector percentage (spatially bracketed within each station, as described above). You can see that polling stations with only one booth (the crossed areas) tend to begin towards the outskirts and rural Lucknow. And you can see where Muslims voted more (green) or less (red) for SP than non-Muslims (in white areas, the covariation goes neither way). To pick just one example: Muslims in the posher areas of Hazratganj and Butler colony voted less for SP, as did Muslims around Kashmiri Mohalla, a Shia-concentrated area:1

Now an interesting follow-up question is of course whether Muslims not only vote more for the Samajwadi Party than non-Muslims - but whether they were also responsible for the swing in votes which brought SP back to power in 2012. This is more difficult to assess - primarily because delimitation in 2008 changed the layout of constituencies and polling stations to an extent which makes it difficult to compare election results from 2007 and 2012. But at least a tentative answer to this question is possible, and this answer is: no, Muslims were not responsible for the swing (with a number of statistical caveats, and without being generalizable as of now - see the presentation for details).

So is it indeed Maulana Singh Yadav after all? In urban Lucknow and for this election I would generally say yes. We have fairly robust evidence that Muslims voted indeed more for SP than other communities. But: this is nothing new; the swing which brought the Samajwadi Party back to power came from elsewhere. Now: we already know all of this from CSDS survey data. But the big advantage of my approach is that we can now check for spatial variation: it will be very interesting to see how this plays out across the whole of Uttar Pradesh - do we find a similar pattern, and if so where? My strong suspicion is still that the "Muslim vote" operates on a much more localized level. That urban Lucknow confirms survey data need not imply that other areas don't differ. With the new dataset based on electoral roll data at hand, it is now possible to dwelve deep into this local level - and I will keep you updated once I find out more...

I would like to acknowledge the use of the Oxford Supercomputing Centre (OSC) in carrying out this work.

  • 1. In case any politician reads this: I should probably clarify that this is not at all an argument to benefit only those - I firmly believe that a government should care for all people, irrespective of whether they voted for this particular government or not...