Will BJP Really Lose Forty Seats In Uttar Pradesh?

“If you torture data long enough, it will confess to anything,” observed Ronald Coase, the Nobel-winning British economist. This article, however, is a tale of a couple of economists who did not even try.

Of the most ludicrous pieces that came out in this election cycle, a series of articles on election predictions published by an online news portal deserves a special place. The authors of these delusive pieces initially predicted that in Uttar Pradesh, the Bharatiya Janata Party (BJP) could lose up to 40 seats. Probably buoyed by their cheerleaders, the authors then brazenly extended their “model” to some of the so-called ‘Hindi-heartland’ states—UP, Rajasthan, Madhya Pradesh and Chhattisgarh, and West Bengal—ceteris paribus, this “model” pegged the BJP’s losses to over 75 parliamentary seats in the region.

After the exit polls deluge, which predicted a thumping victory to the BJP, one can’t but help analysing this farcical contrarian “model” for whatever it is worth.

Any decent forecasting model that should work in an uncertain future ought to have a probability distribution of numerous scenarios. Theoretically, every scenario is a possibility. For example, consider a parliamentary seat in Uttar Pradesh where a BJP candidate is facing a Samajwadi Party candidate (Mahaghatbandan). At one extreme of the probability distribution all the SP-BSP-RLD votes cumulate and transfer to the SP candidate; and at the other extreme, none of the BSP and RLD votes transfer to the SP candidate. Being extreme possibilities, both these scenarios have very less likelihood of becoming realities. And, in between these two scenarios lies an array of scenarios, each with a certain probability of occurrence.

The murder of modelling begins when the authors predict that in Uttar Pradesh all of the votes of SP-BSP-RLD would add up cumulatively for each and every Mahaghatbandan (consensus) candidate. They do this by taking readily available vote share data from the Uttar Pradesh (UP) state elections and general elections, provided by the Election Commission of India (ECI), and add them. In the process, the authors, determined to churn out a precogitated result, consciously decide to shun all complexities resulting from voters’ choices. This becomes a classic case of applying the statistics of univariate data to a multivariate scenario. The univariate fallacy is in pretending that all other variables are invalid when the one variable that suits the narrative is present—in this case, they have discarded all the other scenarios of the probability distribution. This injudicious adding up of data in an excel sheet, which the authors have tried to pass off as a "model", could have been easily carried out on a hand calculator.

Any economist worth his salt would have been aware of these issues, and certainly, the authors, who teach economics as a profession, ought to know this. Modern econometrics has bestowed us with extremely sophisticated models of estimation, which is the result of the sweat of thousands of economists, statisticians, and experts from other disciplines. These models have enabled us not only to comprehend large macroeconomic phenomenon but have applications in wide-ranging subjects like microbiology.

But the authors' “model” is far from any of this. It has no backing in existing literature in psephology nor has it been tested on any past data. In fact, a basic analysis shows that their model fails its own model.

Yes, you read it right. And here is how.

They have used 2014 Lok Sabha and 2017 assembly election vote share data to predict 2019 Lok Sabha outcome. But the 2014 Lok Sabha vote share model fails in predicting the 2017 election results—where SP + INC vote share went down. Not surprising considering that they are using the hand calculator model! This model does not work for any of the previous election data, nor will it work for the current elections. The only success it has achieved is in generating a false sense of euphoria amongst the flying-on-pigs binge consumers of the online portal’s content.

The intellectual masquerading, however, continues. The authors get into extending this hand-calculator “model” to West Bengal. They say,

What challenge is this to an economist? Taking a previous vote share and adding it to the current one. They look at no historical precedents, they ignore gains of the BJP elsewhere (for example, the increased 43 per cent vote share in Tripura), and most appalling of all, they expect that 2011-2014 would be same as 2014 to 2019. And even if they had used the proxy of the number of rallies (and turnout) held in West Bengal by Prime Minister Modi (10) and Amit Shah (11) they could have obtained more nuanced numbers.

At this point, if we refrain from questioning the competence of these authors, we are forced to question their motivation. Overwhelmed by their eagerness to cater to the anti-BJP echo chamber, they have successfully risked their personal credibility.

Do we believe they were unaware of the gaping holes in their “model” or does it speak of the audience they are catering to? Alas, some economists are willing to sell their disciplines to forward their ideology. This reminds me of C.G. Jung’s wisdom: “People don't have ideas. Ideas have people.”

The problem is not in the numbers being unacceptable, it is in the fact that the authors did not even try to accommodate any sort of realistic indicators in their forecasting: no sampling or construction of probability distribution, no controls for national political climate, no account of BJP’s dedicated focus on Bengal, no testing of the model (and fixing the errors), and no information on the candidates. Just a 2+2 = 4.

Economists are supposed to raise the technicality of the generic politically-dominated discourse. Instead, these authors have lazily given into the lure of sensationalism.