Opinion polls conducted without full disclosures are not open to scrutiny and thus not credible
For the last few months, the Indian voter has been inundated with a plethora of opinion polls. The Election Commission is expected to announce an embargo on opinion polls during the election period once the elections to Indian Parliament are announced.
The diversity of India in religion, regions, languages, caste, urban/rural, disparities in income and literacy levels coupled with the system of parliamentary democracy makes the task of predicting results based on opinion polls very complex and difficult — but not impossible.
Over 800 million voters of India will elect candidates from 7 national parties, 55 state parties, 1000 plus unregistered parties and about 5000 independents to 543 seats of Indian Parliament shortly.
Statistical sampling is the foundation on which opinion polls are based. It typically simplifies answers to complex questions such as “Who will form the next government at the Center and how?”
The question may not be very difficult to answer but the scope of the question is vast – over 800 million voters will vote in 543 constituencies. Statistics help in seeking the answer to such questions by a technique called sampling.
Sampling technique helps to maintain the character of the population in a small sample selected from the voters. Smallness of the sample makes the exercise of conducting opinion polls feasible in terms of time, money and effort.
It is essential that the sample be selected at random which means each voter has an equal chance of getting selected for it. Compromise made on the randomness will not make it a representative sample and the results are bound to go awry.
The sample is not random if any bias has influenced its selection. The best way to avoid any bias seeping into the sample is to generate samples out of a complete list of Indian electors using random numbers. If a voter from the remotest part of India is selected in the sample, that voter should be contacted and his/her response collected.
The sample size of at least 16,500 voters (543 multiplied by 30 ) should be selected. The randomness of the sample will ensure that it is selected from different demographics of most of the constituencies across all states of India making it a truly representative sample.
The voters thus selected should be interviewed in their native language and their responses recorded. The questions asked should be direct and unambiguous. The option available for the interview of the selected voter is only face-to-face as the voter data does not offer phone numbers. Their responses should be aggregated.
Indian electorate has almost equal gender ratio and 80% of it happens to be Hindu. If the sample data has 30% women (the sample is not likely to be so skewed), it is not representative of the population. The sample size is to be increased without purging the existing data until the sample data is nearly (not exactly) the same as that of population. Like wise, if Hindus are only 50% of the sample (not likely), the sample size is to be increased.
The results drawn from the sample can thus help in the prediction of winners for each constituency within a range of margin of error by applying appropriate statistical formulae, tables and using extrapolation methods.
Recent opinion polls have not disclosed fully the methodology used for conducting them. This non-partial disclosure makes one skeptical of the veracity of these opinion polls.
It is therefore imperative that a complete disclosure of the methodology is made for each opinion poll. The disclosures should be comprehensive enough to include methods used for the selection of sample, mode of interview, states, constituencies, margin of error, urban/rural and other demographics of the sample.
Some of the pollsters use different sampling methods which makes the task of conducting opinion polls less cumbersome, less expensive and faster. Such efforts do inject bias into samples at the expense of randomness and should be avoided. Some estimate the number of seats from the vote share by induction methods which are not appropriate.
Each opinion poll conducted in the United States gives information about methodology used. It also gives email-address and phone numbers to be contacted for any methodology related questions.
The Election Commission ought to have laid down comprehensive standards/guidelines for methodology and ensure compliance and insisted on complete disclosure of each opinion poll. Instead, it is taking the easy route of banning opinion polls during election time, the time when they really matter.
Opinion polls conducted without full disclosure are not open to scrutiny and hence findings of such polls are not credible. Sadly, such opinion polls have become a tool for abuse by the vested political parties.