Predicting electoral outcomes in India is almost an occult science that involves a lot of mysteries like imagined caste-vote tilts and representative sample sizes that try to fathom voter perceptions regarding government’s performance et al. For a nation of 1.2 billion people with a million castes/religions/languages/ethnic groups, even after 65 years of independence, we are totally clueless about the percentage of population of different subgroups. In fact, for more than 50 years all electoral pundits have been quoting lefty-CSDS percentages for all sorts of population based permutations and combinations. What is even more heart-breaking is that most of the CSDS-type data is based on weighted representations of sample-sizes as low as interviewing 700 people in a state!

To add to this huge gap in our databases is the total opaqueness of the election commission of India which has never released polling booth level voting numbers to the wider public scrutiny. Polling booth level voting data is released to the political parties and contesting candidates by the election commission and it is a herculean task to get access to that data.

Although this author has succeeded in collating some polling booth level data in some states, that database has a lot of gaps and is by no means exhaustive. In any case, another stumbling block for such a polling-booth level database is the fact that it is available only from the EVM era of 2004 onwards as there was no practical way of maintaining such a database in the paper-ballot era.

The upcoming Gujarat election is the first opportunity for us at ECRI to try and plug some of these gaps in the electoral data by collating and superimposing different databases from different sources. This is our humble effort to analyse opinion polls and past electoral data on one single platform to get a better understanding of Indian electoral calculus and eventually produce better predictive models.

District-wise electoral map of Gujarat

One of the primary flaws in our understanding of the Indian electoral scene is the lack of a visual geographic electoral-topography. As we have seen in the electoral-college methodology of the US presidential elections, understanding the dominance of a political party in each district in a similar fashion in the Indian context can give us a better visual understanding of electoral politics.

Representative electoral map of Gujarat after the 2007 assembly elections. “Saffron” represents BJP dominant areas whereas “Blue” represents Congress dominant areas.


Representative electoral map of Gujarat after the 2009 LS elections. “Saffron” represents BJP dominant areas whereas “Blue” represents Congress dominant areas.

What is clear from the above maps is that BJP’s strength in the Saurashtra region has been considerably eroded in the 2009 LS polls as compared to its performance in the 2007 assembly polls. Narendra Modi and BJP should be reasonably worried about the party’s prospects in Saurashtra this time because of the presence of GPP and Keshubhai Patel.

For instance, take the case of Bhavnagar district (in the above maps), where BJP had swept the polls in the 2007 assembly elections but had to share the spoils in 2009 because of the presence of Gordhanbhai Zadafia as an MJP candidate – he cornered almost 30% of the votes in the Bhavnagar LS polls. This time, in Bhavnagar area there is not only GPP to contend with, but also Kanubhai Kalsaria’s Sadbhavana Manch, both of which can potentially dent BJP’s vote-share.

Despite a poor performance in the LS polls, BJP was still ahead in 106 assembly segments (to the Congress lead positions of 76) even in the 2009 elections. In more than 30 years, since 1991, the only time BJP had been leading in less than a 100 seats (in both LS as well as assembly polls) was in 2004. Only a repeat of that 2004 LS performance by the Congress party can bring it any closer to power.

LS Polls v/s Assembly polls


[Data Source: Election Commission of India]

Having said that the BJP should be weary of its performance in Saurashtra, it must be stressed that the voting behaviour of Gujaratis is distinctly different in assembly polls as compared to LS polls. The one clear visible feature of this difference in voting pattern is the 12 to 15% gap in turnouts between both the set of polls (as seen in the chart above). It seems as though the Gujarati voter has lesser incentive in turning out to vote for national polls vis-à-vis state elections. Also logical analysis suggests that it is the BJP voter who is less enthusiastic about Lok Sabha elections than the Congress voter – this aspect might change in the event of Narendra Modi being declared as the prime-ministerial candidate of the NDA/BJP.

Understanding the caste-calculus and voting patterns of the past polls


[Data Source: Election Commission of India]

  • In each assembly election from 1995 onwards, BJP has maintained a 10% gap over its closest rival Congress
  • A 10% gap over a 4 election-cycle is almost impossible to breach under normal electoral conditions in India
  • Over the years as “others” and smaller parties started becoming irrelevant in the Gujarat electoral scene, the “others” vote started accruing to the two main players.
  • In the Indian first-past-the-post electoral system having a 50% vote-share has a huge advantage in the electoral- arithmetic.
  • Any increase of even 1% from the present base of close to 50% vote-share will give an exponential number of seats for the BJP, but a drop of even 2-3% of vote-share will only mean a marginal drop in the corresponding number of seats in the assembly
  • This time, after a gap of almost 2 decades, there are a lot of “others” in the fray and that will have a huge bearing on the vote-share percentages of both the principal players.



[Data Source: BJP internal database + leaked census data + CSDS + NES]

Even though Gujarat as a state has evolved from being a purely caste-electoral-matrix into a development and progress oriented society, the caste & identity pull on the electoral outcomes cannot be completely ruled out. Thus it becomes important to understand and analyse the voting patterns of different subgroups of the Gujarati populace. There are again three major stumbling blocks in this process of understanding the caste-based voter preferences.

  1. The exact percentage of any subgroup or caste community is not known and all we have are derivatives based on sample-sizes & voting patterns in certain caste-dominated areas. Although certain leaked 2011-census reports more-or-less correspond to our existing databases, one cannot vouch for the veracity or integrity of either the leaked census data points or the sources.
  2. When the base (of the actual percentages of the population) itself is shaky, it is quite difficult to construct an edifice of voting pattern on that base. Yet, by mixing the sample data from past surveys and polling booth level voting patterns (where available) we are trying to build a reasonable edifice of caste-voting-matrix.
  3. Even if certain caste groups have voted in a certain way in the past, there is no way to predict human behaviour with reasonable accuracy about their future voting patterns. This is where the pre-poll surveys come into picture, but that is for later.

[Data Source: BJP internal database + CSDS + NES + Polling booth data]

The above chart represents collated data from two LS polls (2004 & 2009) and two assembly polls (2002 & 2007). The remaining vote-share percentage, if any; other than that of Congress & BJP; has gone to “others”.

  • BJP’s mainstay comes from Upper castes, Patels & OBCs, whereas Congress’ mainstay comes from SC, ST and Muslim votes.
  • BJP has a larger pool of core vote base – 65%, so it can afford to lose some of the vote-share and still manage to emerge victorious.
  • Congress has a smaller pool of core vote base – 35%, so it cannot afford to lose any vote-share if it wants to remain in contention
  • Compared to LS polls with assembly polls, BJP’s biggest drop in vote-share has come from two communities – Patels & Kolis
  • Compared to LS polls with assembly polls, Congress’ biggest gain in vote-share has been from three communities – Patels, Thakors & other upper castes.
  • Almost double the number of Muslims tend to vote for the BJP in assembly polls vis-à-vis LS polls, but that hasn’t impacted the Congress’ Muslim vote-share much till now (it is the non-Congress Muslim vote that has been accruing to the BJP in assembly polls).
  • Unlike other states, in Gujarat, voters tend to vote for “others” more in the LS polls than in assembly polls – suggests a deeper sense of satisfaction with the state politics than the central politics.
  • In the LS polls, the highest percentage of “others” vote-share is in the SC community – 19% (the BSP factor?), whereas in assembly polls, the highest percentage of “others” vote-share is in the Patel community – 15%
  • In the LS polls, the lowest percentage of “others” vote-share is among the other upper castes, Thakors & STs – 5% each, whereas in assembly polls, the lowest vote-share percentage for “others” is among the Kolis & other OBCs.
  • This time the presence of stronger “others” in the electoral arena can potentially take a bigger chunk of the vote-share; the communities to be watched are – Patels, Kolis & Muslim – for their susceptibility to flux-voting.

The urban v/s rural divide


[Data Source: BJP internal database + CSDS + NES + Polling booth data]

The above chart represents collated data from two LS polls (2004 & 2009) and two assembly polls (2002 & 2007). The remaining vote-share percentage; other than that of Congress & BJP, if any; has gone to “others”.

  • 42% of Gujarati voters can be considered as urban voters, which is quite high a percentage in the Indian electoral scenario
  • Rural vote-share of both the parties has remained intact in both LS as well as assembly polls over the decade.
  • Urban vote-share keeps fluctuating between state and national elections
  • This time due to various reasons, including drought, the urban-rural divide is expected to be sharper and also potentially more anti-incumbent in nature

Assembly strongholds

There are basically 56 BJP stronghold seats where the party hasn’t lost an election since two decades (i.e.) since 1995. Among those 56, there are 31 constituencies where BJP hasn’t lost an election since 1990! Congress on the other hand has only 4-6 such stronghold seats in an assembly of 182. This kind of electoral strength is unheard of in the post emergency era of Indian elections.

[Data Source: Election Commission of India] Seats not lost since 90, 95 & 98

For Congress to have any realistic chance of forming a government in Gandhinagar, the party has to win at least half of these BJP stronghold seats. Conversely, for the BJP it is much easier to once again emerge victorious in the Gujarat polls, because all it has to do is retain a majority of its stronghold constituencies and just try and win a couple of dozen more battleground seats.


  • Electoral map suggests that the BJP’s spread in Gujarat is secular across all regions, whereas Congress has pockets of influence here and there
  • Higher voter turnouts in assembly polls compared to LS polls seem to be favouring the BJP more than the Congress – the plus vote goes to BJP
  • In 20 years and 8 election cycles, only once has the BJP tally of assembly segment lead positions been less than 100 – in the 2004 LS polls
  • BJP has consistently maintained a whopping 10% gap in the assembly elections since 1995
  • BJP’s vote-share is almost touching the 50% mark, which is a significant achievement in terms of Indian electoral arithmetic
  • BJP has a larger pool of (caste-community based) core-voters – 65%, while Congress has a smaller pool of core-voters – 35%
  • Surprisingly, BJP’s rural vote remains intact and more loyal to the party in both LS as well as assembly polls, whereas the urban vote-base of the BJP is more susceptible to a flux
  • In the 2012 elections, the “others” vote percentage might play a crucial role in the electoral arena
  • BJP has some 56 assembly segments where it has never lost an election since 1995, while Congress only has 6 such seats.