tag:blogger.com,1999:blog-7435391049679382734.post4648688809331449360..comments2019-05-22T07:49:11.293-04:00Comments on Operations Research and Analytics: Predicting the Indian Elections - A Win for Data ScienceShivahttp://www.blogger.com/profile/05571015480979394896noreply@blogger.comBlogger3125tag:blogger.com,1999:blog-7435391049679382734.post-21788876249351397712016-01-26T22:39:28.067-05:002016-01-26T22:39:28.067-05:00This comment has been removed by a blog administrator.Thy Nguyenhttps://www.blogger.com/profile/08010217919250824694noreply@blogger.comtag:blogger.com,1999:blog-7435391049679382734.post-43997256674675303092014-05-26T16:24:52.346-04:002014-05-26T16:24:52.346-04:00fascinating and insightful observations. Will re-r...fascinating and insightful observations. Will re-read and return via a separate post.Shivahttps://www.blogger.com/profile/05571015480979394896noreply@blogger.comtag:blogger.com,1999:blog-7435391049679382734.post-8285733697828710992014-05-24T02:11:21.503-04:002014-05-24T02:11:21.503-04:00On a related topic, here is an interesting questio...On a related topic, here is an interesting question on statistics, which got triggered off when I was watching the election results unfold on May 16th:<br />Counting was underway, and partial vote counts were available for practically all the constituencies. At some point, NDA had leads in about 305 constituencies out of all 542. At this point, Prannoy Roy, with all his experience and wisdom remarked: "As counting gets completed, this lead is likely to swell and NDA could reach 320 seats". As it turned out, NDA ended up with 336. I wondered if it was a mere co-incidence, or is there a probabilistic basis for Prannoy's observation. I think I figured the answer.<br /><br /><br />This phenomenon is the result of the "first past the post" system, where the winner takes all. The NDA won 62% of the seats but only 38.5% of the votes. During vote counting, early leads are influenced more by the vote share: for instance, if only one vote is counted in each constituency, the expected value of NDA leads would only be 38.5% of the seats (after making the simplified assumption that the 38.5% vote share is uniform across constituencies). As more votes get counted, the probability of the dominant party taking the lead starts increasing and the expected value of leads starts moving towards the final result. In other words, if there is a dominant party, their domination gets established more and more as a larger portions of the votes are counted.<br />Another simple corner case would make this even clearer. Let's say one party gets around 50%-60% of votes in every single constituencies (God forbid that happens). Then they would end up getting all the seats. But leads based on counting one a fraction of votes would be significantly lesser (with just one vote being counted, they would get about 50-60% of the seats). As more votes are counted, the lead would inch towards 100% of the seats.Raghavan Subramaniyanhttps://www.blogger.com/profile/04976660774950657638noreply@blogger.com