Okay, I looked into the sample profile of Assam in the CSDS survey, and now I am beginning to get why they are getting such high vote shares for the Congress.
Total sample size for Assam is 462 (over 14 constituencies, each of which have their own character). The dynamics of upper Assam even within are very different (Tezpur, believe me, is very different from Jorhat, which is a world apart from Lakhimpur). So, it translates into just around 33 people per Lok Sabha constituency, which in Assam, with its very fragmented communities, is not attractive. Now, if this is not very bad enough, there is the sampling part which should raise eyebrows.
The vote shares were decided based on dummy voting, as they have themselves said. 3% of their samples gave no answer/preference.
In Assam, the urban vote is 15.3%, while only 5.9% of their samples are from the urban regions. Second, Muslims constitute 30.98% of Assam (according to 2001 census, now it is probably around 33%), while only 15.8% of their samples are from Muslims. Muslim vote, especially in lower Assam and Barak valley is expected to accrue to AUDF (maybe even in Central Assam), so it is unsurprising that AUDF is getting only 10% of the vote (in fact, it agrees with the rough calculation that the AUDF is taking most Bengali Muslim votes). Finally, tribal and SC population are 7.1% and 12.4% respectively. However, 15% of the the sampling population was Dalit, and 26% was tribal (is this biased or what?). Tribal vote, particularly the tea tribes, are the Congress’ last Hindu (well, some tea tribes are now Christian) votebank left, while even Bodos, and Rabhas tend to vote for their own parties. Only Mishings, Sonowal Kacharis, and a few tribes of Barak valley vote for the BJP. However, it also hides a second problem. 14% of the voters expressed a wish to vote for `Others’ (non Cong, BJP, AUDF, AGP parties). Most of the other vote, I would say, is going to the Bodo-Rabha-Kamtapur parties of lower Assam (this is the only big block not represented in the list of named parties).
In short, 58% of their samples are taken from a population base base that is actively hostile or indifferent to the BJP (it constitutes about 50% of the population), while 15.3% of the total population which would be BJP’s main catchment area, accounts for only 5.9% of their voting block. Even worse, it does not take into account the set of alliances that the BJP has struck up, particularly with the Bodo and the Kamtapur parties.
I am not sure the CSDS vote shares can be trusted here.
Having said that, let me try to re-compute the BJP, and Congress vote shares here.
1) If we assume that 10% of the `Others’ are from the Bodo-Rabha-Kamtapur parties block (it does agree with past vote shares of around 12% for this bunch of parties), then we should expect a split of their votes in the actual polls. With the BPPF, ABSU and a bunch of Kamtapuri parties supporting the BJP, most of the Bodo, and Rajbanshi vote in Mangaldoi, Tezpur and maybe even Barpeta is going to the BJP (which is seen as the best party to stop Bangladeshis), while we can expect a split of the Bodo and Rabha votes in Kokrajhar, along the BPF-BPPF lines. So out of the 10% votes of the others, I would expect about 6% going to BJP+).
2) Out of the 42% of the population that is a good catchment for the BJP, we can see about 15-16% or so voting for the BJP (roughly 40% of the total vote). Now, re-normalising for the 50% state population (instead of the 42% as in the sample), we should expect to see the BJP getting around 20% of the vote on its own. Also, if we account for the bias against the urban population (honestly, I would like to see how much of the urban population voted for the BJP – I would not be surprised if it were around 50% in this Modi wave), we should expect to see the BJP getting a bit more, maybe about 22-23% of the vote (which is much more credible, since BJP has had a vote share of about 25% in Assam since 98 in Lok Sabha elections).
Adding the NDA vote together, we should see BJP+ getting around 28-30% of the vote.
3) Now, if we examine the Muslim voting, we see that the Bengali Muslims have been voting for the AUDF. Two thirds of the 15.9% Muslims (very few non-Muslims will vote for the AUDF, believe me) represented in the survey chose to vote for the AUDF. Technically, this will mean about 20% of the total vote, but this won’t happen. In places like Tezpur, Darrang, Kaliabor, and Gauhati, Muslims will vote more for the Congress than for AUDF. So, we can expect about 15% o the vote to accrue to the AUDF, rather than the 10% shown here.
4) AGP and the others can be left at 9% and 7% respectively.
This results in the Congress getting a vote share of around 38-40%, which is much more credible.
Note that in all this I have assumed zero influence due to the Modi wave (since we have few numbers here).
PS: This analysis is based on January findings, which gave Congress 47% of the vote. Now, they have come down to 44% in March polls (but no increase to the BJP, so where did that 3% go?)