After its Bihar exit poll fiasco in 2015, polling agency Today’s Chanakya had explained, ‘A simple computer template coding marking the alliances got interchanged at our end. Due to this our seat numbers remained the same, but respective alliances got interchanged.’ And the reason given was that the day this template was made, a senior member of the staff was on leave. Oops.Is this the level of seriousness of some of our leading pollsters? Now that counting for the four states and one Union territory is scheduled for Sunday, huge variations persist within the exit poll predictions by major agencies. For example, in West Bengal, if TMC gets 191 of the 292 seats (65.4%), some pollster could well claim to have predicted correctly. Some other agency may claim credit if TMC gets routed with a meagre 64 seats (21.9%).Similarly, there will be some pollster or the other who will claim to have predicted DMK+ bagging 133-195 seats in the 234-member Tamil Nadu legislative assembly; BJP+ 58-85 seats among 126 seats in Assam; and LDF 71-120 among 140 seats in Kerala. The actual numbers may even fall outside these broad ranges.It may be worth revisiting the 2016 exit polls in these states. The predictions for Tamil Nadu were disastrous, many polls being unable to predict an AIADMK victory at all. While AIADMK bagged 134 seats, a pollster even predicted 89 seats for the party. While BJP+ got 86 seats in Assam, a pollster predicted 57 seats for it. In West Bengal, most pollsters couldn’t foresee a massive TMC victory with 211 seats — some even predicted only 153 seats. In Kerala, LDF’s victory margin was grossly underestimated in most exit poll predictions.In 2016, while CVoter (Centre for Voting Opinion & Trends in Election Research) read voters’ sentiments in Tamil Nadu and Kerala correctly (by predicting 139 seats for AIADMK and 78 seats for LDF), it utterly failed in West Bengal (where it predicted 167 seats for TMC) and Assam (57 seats for BJP+). Today’s Chanakya was remarkably accurate in Bengal (predicting 210 seats for TMC) and Assam (90 seats for BJP+). But its performance was not great in Kerala (predicting 75±9 seats for LDF), and was miserable in Tamil Nadu (predicting 90 seats for AIADMK). So, can we rely on any single pollster this time around?While the variations for exit polls in Kerala and Tamil Nadu would keep the eventual predicted winners unchanged this time around, a wide range of 64-191 seats for TMC in Bengal predicted by various pollsters would actually take us nowhere. So, what’s the utility of so many poll predictions? Informed guesstimates may sometimes serve better.Is a ‘poll of polls’ — the average of some major exit poll results — more reliable then? Perhaps — unless, of course, one includes some really ‘bad’ predictions into the fray. Some of these exit polls are certainly ‘outliers’, and they should be given much less weight while finding the ‘average’. A sophisticated statistical approach — that of ‘robust averaging’ of several poll results — may yield a more trustworthy prediction.Alternately, some polling organisations may pool their data for possible increased accuracy of their results. This worked successfully when the BBC and ITV merged their data during the 2005 British general election, and when Sky News, Channel 7 and Auspoll collaborated in the 2007 Australian federal election. However, it’s not always easy, as the method of predicting seat numbers by various pollsters may often be different.Often, the methodology, sample size, etc, are not disclosed by the pollsters, thus making their predictions look like prophecies. With so many polling organisations in the predicting business these days, and with such widely varying predictions, it’s hard to know whom to believe. Every political dream may be realised ‘theoretically’ within the domain of exit polls. As a result, the thrill of counting day remains intact.(The writer is professor of statistics, Indian Statistical Institute, Kolkata)