Poll predictions not weather forecasts

Voting is fundamentally a strategy game. Every voter who turns up to cast her vote knows that her individual choice cannot be pivotal in an election to, say, the Indian Parliament. But she also knows that her individual vote in relation to the rest of the votes cast affects the electoral outcome. A voter expecting the majority to vote with her has a reason to show up at the polling station. Another voter anticipating to be in a minority might as well stay at home. A third could wait for more clarity before making a choice. Individual expectation of collective behaviour is not visible to outsiders in secret ballots, except in outcomes. Which makes predicting elections so complex.

That is not to say election managers need a functional familiarity with Nash equilibria. They can, and do, focus their energies on manipulating individual expectations of collective behaviour till campaigning comes to a close. Electioneering is designed to shape the narrative to inform both personal and group choices. This adds another layer of complexity to predicting outcomes before ballots are cast. The entire process is dynamic till polling ends. This narrows the window for effective forecasting from the time the last vote is cast till the time official results are declared. Exit polls, thus, have a higher probability of getting it right than most pre-election forecasts.

Of course, the usual caveats on methodology apply. Samples have to be randomised for individuals and groups. Statistical modelling needs to incorporate the major strands of the electoral narratives. Error margins must allow pollsters to make calls that are not too close. Designed well, exit polls offer the first glimpse of collective behaviour, and do serve a purpose as an advance indicator. Before this stage, most political forecasting is more art than science, and best left to practising politicians who have their ear to the ground and ambitions in the air. The process of manufacturing consent is pretty well-established, and can lead to fairly predictable political outcomes.