Exit polls have proven to be notoriously ineffective in predicting the actual result in India & critics point to the 2004 & 2009 elections as examples.
In 2014, most exit polls have predicted a NDA/BJP victory
We have crunched through the numbers specifically for the BJP/NDA seat projections & shown 5 nuggets that point out the hazardous nature of exit poll predictions in India
Vozag.com Key Takeaways:
- The median national exit poll figure for the NDA government is 282 with a high of 354 (Chanakya) & a low of 249 (TimesNow) & a difference of 105 seats between the high & low
Projected Number of Seats for NDA by Polling Firm
- TimesNow’s low estimate is driven by extremely low projections for NDA in Rajasthan & Madhya Pradesh which most experts expect to be swept by BJP
TimesNow Comparison Versus Median Estimate in “Big 6” States for BJP/NDA
- For the 6 biggest states for NDA/BJP, the difference between the worst exit poll & the best exit poll (excluding Chanakya) is 60 seats. Critics can point to this difference & the 2004 & 2009 poor predictions of exit polls to the unreliability of such polls.
- Chanakya may seem like the “crazy” outlier with an aggressive projection of 340 seats for NDA but they were the most accurate exit polling firm in the 2013 State Assembly elections
Estimates of Various Exit Polling Firms to Actuals in State Assembly Elections in November 2013
- CNN-IBN’s estimates are in the mid-range of all exit polling estimates & are considered “typical”. However, a closer examination of polling data shows that in two key states tiny differences in vote share led to big changes in seat projections: 1) A 1% positive change in Maharashtra between the last opinion poll & the current exit poll shows a gain of 8 seats for the NDA & 2) Similarly, a 2% positive change in Karnataka leads to a 4 seat difference for the BJP
Vozag.com is a quantitative research and data analytics service firm doing data analysis in the home remodeling, repair and contracting industry. Now and then they utilise their analytical capabilities to comment on a wide range of serious and frivolous issues alike.
Try deep learning using MATLAB