Election 2019: Where We Went Wrong
Our models predicted two diametrically opposite outcomes — one where the SP+BSP+RLD won the majority of seats, the other was a mirror where the BJP won the majority of seats. Guess which one we bet on.
We got it wrong. We have always maintained that if we were wrong we would be very wrong — that we would see the polar opposite of our projections. We take what little satisfaction is available from being right about how wrong we would be.
We started this journey by talking about how this was the BJP’s election to lose in UP. Having won 73 seats in 2014, they were up against a formidable local combination, blunted only by the Congress’s unwillingness to accept the existential battle they were in.
Today, the BJP gained ~10% vote share but won fewer seats than in 2014. Akhilesh Yadav and Mayawati’s gathbandhan in Uttar Pradesh is the only pocket of resistance against the Modi-Shah juggernaut in North India. Both have won more seats than the last election.
But, we have learnt far more:
We bet on traditional political wisdom when we had to assign weights: that traditional vote bases built around caste identities would consolidate; that the trends we had seen in recent elections would allow for a strong opposition to emerge.
What we have seen instead is consolidation around a majoritarian identity and a single person — Narendra Modi who successfully framed this as an election where he was the only option. The voter was left believing they could either choose him, or reject him and hope for the best.
Mid April, we wrote about the formidable post-caste bloc the BJP has built. Today, that bloc covers half the voting population of Uttar Pradesh. This is FAR more than we expected. We under-indexed weights that predicted a rise of this size, thinking it impossible because of the support we saw for the Gathbandhan. Clearly, we misread this trend.
We found it much harder to measure the anti-Muslim sentiment that dominates a section of the BJP voter. We under-indexed this as well. The BJP knows, and has now proven, that the Muslim vote isn’t needed to win an election.
Anger over an issue (demonetisation, unemployment etc) kept showing up in our data and we kept giving these greater weight. In Uttar Pradesh, we also extrapolated the disenchantment against the state government into a vote against the BJP. What we saw instead was a vote for Narendra Modi. And a clear vote against a Congress-led government. Odisha teaches us an important lesson — the split in outcomes between the BJD in the assembly and the parliament clearly tell us that people are voting for different leaders, not parties, at the state and national level.
In each phase, the excitement we picked up amongst people who hadn’t voted for the BJP validated our choices.
It’s easy to say the Gathbandhan didn’t work, but with THE BJP’S ~10 per cent increase in vote share, they could have won all 80 seats. The fact that they won fewer seats than 2014 points to the power of this Gathbandhan. There are at least 15 more seats which the Gathbandhan would have won if the Congress hadn’t played spoiler.
Our mistakes are clear — we arrived at the right insights about the people we are studying. But we weighed electoral impact based on an outdated political model.
The way forward for us is clear — we believe we have the data we need to strengthen our weight allocations.
We want to acknowledge the work done by traditional pollsters — they deserve both respect and applause for calling UP right — the variance in exit polls shows what an incredibly hard task it was.
Finally, we want to thank all the people who have followed our work these last six weeks — the first time something like this has been attempted at this scale. We are going to give ourselves points for our ambition, and for testing out data gathering and processing systems that will hold us in good stead in the work we will continue to do.
These notes are an experiment in data-driven points of view. We are immersing ourselves in information screens and data patterns and allowing ourselves to connect dots. We emerge to write a note — like the one you’re reading — which is our best understanding at a given moment in time. We believe ourselves to be correct in the moment, but are happy to be proven wrong. In either case we learn and improve.