There’s something for everyone in Indian politics this week. The Aam Aadmi Party has won the Delhi municipal elections; the Bharatiya Janata Party has bagged Gujarat; and the Congress will be back in power in Himachal Pradesh.

Then there are those who have fewer reasons to be happy. The pollster Axis My India, for instance.

In Gujarat and Delhi, the Pradeep Gupta-led polling firm broadly put out the correct outcome. But it seems to have seriously underestimated the BJP.

The hitch is the vote share. Its exit poll predicted a 35 percent share for the saffron party in Delhi MCD elections, but the party cornered 39.1 percent of all votes. In Gujarat assembly election, it gave 46 percent to the BJP, though it is hovering around 53 percent.

The pollster, which partners with the India Today group, usually keeps a margin of error of 3 percent in its estimates. That makes its predictions for the Narendra Modi-led party in Gujarat and Delhi fairly behind the actual numbers.

In Delhi, it projected 149 to 171 seats for AAP and only 69 to 91 seats for the BJP. The AAP ended up with 134 of the total 250 seats, eight seats over the 126 majority mark. The incumbent BJP won 104. 

This means that the pollster’s seat predictions for the two parties were off by 25 seats – a tenth of the total number of seats.

These numbers might seem uninspiring because Axis My India is known to be the best in the game. According to a dataset prepared by journalist Samarth Bansal, the pollster’s projections have been vindicated on 18 of the 21 exit polls it has produced between February 2018 and March 2022. This is better than those of its competitors Jan Ki Baat and CVoter.

Delhi and Gujarat will likely not count in these vindications.

In the western state, the pollster’s low vote share and seat projections for the BJP has come at the cost of bumped-up expectations with AAP. Axis My India predicted a whopping 20 percent of votes in Gujarat for the Arvind Kejriwal-led party with 9 to 21 seats. As of 4 pm today, the AAP could only manage about 13 percent of votes with 4 seats.

Its Gujarat seat projection for the BJP was 129 to 151 seats. The party is actually leading in 158.

On December 5, when India Today released its exit polls, Gupta explained that AAP was popular in Gujarat three months ago, especially among Muslim and Dalit voters. But this changed a month before the election, when votes from the two communities shifted back to the Congress. The reason for this, he added, was AAP’s advocacy of Hindu gods on currency notes and the uniform civil code and its “silence on the Bilkis Bano case”.

In Delhi, where AAP and BJP fought a shrill and aggressive contest, the pollster gave the correct prediction for AAP’s vote share (43 percent), but could not convert it accurately to the number of seats.

The sample size of the Axis My India exit poll in Delhi was 34,505 – nearly 0.2 percent of all eligible voters. Its data also brought out nuanced insights on voter preferences. It showed that uneducated voters were more inclined to vote for the AAP in the capital, but degree-holders gravitated towards the BJP. Similarly, younger voters preferred AAP over BJP.

In the exit poll business, pollsters bet on trained surveyors, representative sampling, a sound mathematical model and a good grasp of social behaviour to get the right numbers. An error on either of these fronts can produce poor polls.

Newslaundry reached out to Gupta to understand his analysis of why Axis My India could not capture the scale of pro-BJP sentiment in Delhi and Gujarat.

Gupta said that his Delhi numbers were “fairly in line” with the results. The lack of precision in his predictions, he explained, came because of change in the voting patterns of the Muslim community in the capital. “In Chandni Chowk, Northeast Delhi and in East Delhi, Muslim vote shifted from the AAP to the Congress,” he said.

In Gujarat, where Axis My India had a sample size of 42,156, Gupta said that his exit poll numbers ran into difficulties because he had to reveal the projections while phase two polling was still on. “I told Rajdeep (Sardesai) and Rahul (Kanwal) to correct our numbers the following day,” he told Newslaundry. “I suggested we drop AAP’s vote share from 20 percent to 15 percent and increase BJP’s from 46 percent to 49 percent. This was broadcast on Aaj Tak around 7pm on December 6.”

The pollster Chanakya performed better than Axis My India in Gujarat. In fact, it performed better than all exit polls in the state if we count Jan Ki Baat on NewsX, CVoter on ABP News, ETG on Times Now and P-Marq on Republic TV.

Chanakya’s vote share prediction for the saffron party in Gujarat was 50 percent with a 3 percent margin of error. It extrapolated this to 139 to 161 seats – a bracket that no other pollster projected. 

But Chanakya too overestimated AAP with a 20 percent vote share prediction – seven percent above what the party managed.

Himachal Pradesh was a saving grace for Axis My India, and a marvellous one. It predicted 44 percent vote share for Congress in the hill state and 42 for the BJP. With Congress’s actual vote share at 43.9 percent, this is spot on. The BJP is at 43 percent and that’s within the margin of error.

Gupta told India Today that “Himachal Pradesh was 100 times more difficult to predict than Gujarat”. He has a point. With a tiny gap between vote shares of two parties in a small electorate – Himachal has merely 55 lakh voters – conversion to seat projections is difficult and needs quality data. With a sample size of 16,758, Axis My India seemed to have it. The seats won by BJP and Congress fall within its anticipated brackets. The window for Congress was 30 to 40 seats and for the BJP it was 24 to 34 seats. At 4 pm, the Congress was leading in 39 seats and the BJP in 26.