The Bharatiya Janata Party and the National Democratic Alliance have had a resounding victory in a landslide election. But, how well did the Exit Polls do?In 2004 and 2009, most polling firms overestimated NDA’s performance, had they now underestimated NDA performance?
Getting straight to the point, here are the rankings of different exit polling firms and the explanation of their methodology.
Please Note: This is as of 4 PM IST on May 16, 2014 so this is based on leads & some results data and we will continue to update this post.
RANKING OF EXIT POLLING FIRMS BASED ON SEAT VARIANCE
Vozag has collected an extensive data set of exit polls by various agencies which we have compared against actual results to see which firms have the best and which have the worst prediction rates. Our methodology doesn’t just take into account the total numbers and see the closest estimates of each firm because that could lead to inaccuracies.
For example, a polling firm can overestimate seats for a party in one state (e.g. Congress +10 in State A) & underestimate in another state (e.g. -10 for Congress in State B). When we look at the total figures, these variances cancel each other out (+10 & -10) making a polling firm with wide variances look good.
Here is the overall national level variance for example:
However, to get the true predictive quality, we go to each individual forecast at the state and party level and calculate variances from the actual result.
For example, if NDTV predicted 22 seats for NDA in Bihar and NDA actually got 29, the variance would be 7 seats. We have given a few more examples of State and Party variances from actuals for each polling firm.
We sum up all the variances of forecasted versus actuals and come up with a consolidated figure wherever only in places where all the firms have a forecast. For example, if Firm A has a forecast in Odisha but Firm B doesn’t, it is not included in our data set.
All the details of the variance data, background methodology, errors & feedback can be provided by contacting Vozag.com (http://vozag.com/indiapolls/ )
Why do firms have variances in exit poll projections? Differences in exit poll projections and actuals come from two big factors:
1. The underlying vote share surveyed is incorrect. For e.g. a firm predicts a 45% share for BJP but it ends up getting only 35%.
2. Each polling firm has a methodology to convert those vote shares to seat projections. For example, two firms both have BJP at 35% but one firm forecasts 7 seats & another firm forecasts 11 seats using different methodologies.
Yeshwanth Deshmukh from cVoter polling firm tweeted the following on
1) Vote share and 2) Vote Share to seat projections
In our follow-up research, we will take a look at which firms got the vote share projections mostly right & which ones had lots of variances.