Why numbers differ so much in Exit Polls?

Why numbers differ so much in Exit Polls? 

As a professional who has been doing various surveys for past 10 years,  people often ask me this question. 

One of the main reason for this is faulty analysis. 

It’s all about doing Hasty generalization and rule of small numbers. It includes some sort of statistics and some sort of psychology.

For India, in layman’s terms, this is how exit poll is calculated:-

1. Surveyor is allotted a particular constituency and he is supposed to  ask voters whom they voted.
2. The diversity like language, age, gender and religion have to be take in calculations.
3. They are then counted as representative of that constituency; vote share is taken into account and then prediction is done.

But here comes the villain and that is rule of small numbers. It states that small sample tends to show extreme results.

---

Let me explain in detail how it works:-

Suppose in some constituency, youth consist of 60% of total population but in survey, it consists 80% of total sample. Conclusion is then made up of this survey.

Suppose the surveyor went to just one booth of a constituency and it represents only urban area. Surveyor avoided to go in the interior part of the constituency.

Suppose I do survey in my society where all people are unapologetically Modi supporters and do declare that here BJP will win.

Do you think it will represent the actual scenario? No.

---

So, to know the reliability of some exit poll, most important factors are Sample size and diversity.

But Exit poll predictors only provide us sample size. When I look at some survey then I just look at that sample size of the survey to know how accurate they can be.

Here is sample size of various polls:-

1. India Today-Axis My India poll- 7,42,187[1]
2. News 18 India- 1,21,542[2] [Only 199 constituency]
3. Times Now-VMR poll- 40,000[3]

As per the sample size, India Today- Axis My India poll will have less extreme result as their sample size is largest. [Sample size is 20 times bigger than 2014.]

Again, to generalize the 48 crore voters [60% do vote] by sample of 7,42,187 is faulty. It consists only 0.15% of total voters and conclusions are made up by considering only these much of population.

But mathematically, it is more reliable than other polls which have sample size of 20,000 to 40,000.

Always remember, that when sample size varies, the results will vary. 

We have to wait till 23rd May because we can’t predict nation’s mood just by surveying 0.15% people of nation.

However, these exit polls can give an indication of the way the wind is blowing.