Wednesday, December 26, 2012

Irrelevance of exit polls

Election exit polls are done with insignificant sample size and hence don’t have more than 50 % chance of being right

A record 71 per cent of Gujarat voters exercised their franchise in the assembly election. Exit poll surveys are predicting a third straight thumping victory for Chief Minister Narendra Modi, strengtheninghisprospectsofbecomingthe prime minister of India. The forecastsrangefromaroundtwothirdof the 182 seats to a landslide for the BJP, relegating the opposition Congress to fewer seats than its tally of 59 in the 2007 polls. 

    An election exit poll is a poll of voters taken immediately after they have exited the polling stations. Such predictions with insignificant sample size, has not more than 50 per cent chance of getting it correct. It’sliketossingacoinwithtwosides.
    Exit polls have a long history, but they actually started in the US sometime in 1936, with the magazine named L i t e r a r y D i g e s tholding a poll among its readers on the US presidential election and predicted a loss for the incumbent Roosevelt. The result went against the prediction but the concept became popular. After the arrival of TV and internet, opinion and exit polls generated enough interest and spreadtoothercountries,including India.
    Exit polls are being presented as if they are the political cousin of quantum physics. Pollsters would have us believe that they can measure the impact of political energy over the masses by collecting the voices of voters as they poured out oftheboothsandconverttheminto seats. Exit poll figures are largely an exercise in fiction by the learned psephologists. Such polls are also frivolousgimmickstoengageacurious and impatient audience.
    Exit polls in India are a spectacle of the absurdities which are almost never close to being correct. The minute the last EVM button was pressed in the Gujarat assembly poll, the pollsters started claiming tohaveintheirpossessionthenumbers of an omnipresent survey. The omniscient news anchors started painting an omnipotent chief ministerandpredictedalandslidevictory for him. Their worst hour was when the results of general elections of 2004 came out when they predicted an NDA majority.
    One would have thought that since then the pollsters would have re-examined the coordinates and the calculus to get the prediction matrix right. One would assume that they would use every opportunity and every election to make their models more robust to survive the vicious turbulence of politics. But with every passing exit poll their predictions resemble the report of the meteorological department of the 1980s.
    In 2007, the pollsters miserably failed to map Mayawati’s magic in Uttar Pradesh. They missed the outcome by the proverbial mile. Even the best projection placed BSP at 152-168.MayasweptUPwithanabsolute majority, winning 205 seats out of 408.
    Conclusions about caste and community vote patterns in Indian elections are all based on erroneous data. Early this year, exit polls predicted a hung assembly in Uttar Pradesh. Star News/Nielson gave 183 seats to the Samajwadi Party. India TV said the SP would get 137-145. News 24 predicted 185 seats for the SP. Nobody came even close to the SP’s absolute majority tally of 224.
    In this Gujarat election, the voting was based on a very complex caste basis which differed geographically.Theexitpoll,therefore, is bound to be faulty. Any prediction of a larger ensemble on the basis of sampling within a smaller ensemble will depend on the law of averages. This, in turn, will depend on the choice of the smaller ensemble and to what extent this would represent the behaviour of the larger. If there are several factors that can govern the behaviour of the smaller ensemble then the law of average behaviour will fail and the prediction based on data of the smallerensemblewillnotreflectthe true picture.
    The exit poll predictions which are based on votes given by a small number of voters in a few constituencies may reflect the average behaviour of all constituencies in case the voters have voted based on class factor,eg,poorversusrich.Butifthe voting is on caste basis, the pattern will wildly vary with the caste composition changing in each constituency and also candidates of each class being put up in the election.