Should Pollsters be banned?

There is much discussion in the Indian media on whether opinion polls should be banned. Of course, our view is biased. In our analysis of the upcoming Lok Sabha elections, our predictions will be based on the opinion polls conducted by groups such as the CSDS. Without these opinion polls, we will not be able to aggregate and predict outcomes.

There is, however, little evidence that opinion polls have a significant effect on voters. During the recent assembly elections, the CNN-IBN polls were able to predict victories in Rajasthan, Chhatisgarh and Madhya Pradesh. Although AAP support was underestimated, the opinion poll still predicted the order of the winning parties correctly. Even so, it was not able to take the next step in predicting alliances. Even though the BJP had the largest number of seats, it was unable to form the government in Delhi. Ultimately, the AAP and Congress joined hands to form the Government. For those, like ourselves, wanting to predict the Lok Sabha elections, this is the uniqueness of India’s multiparty system that makes it hard to predict who will ultimately form the government. As former Lok Sabha elections have shown, alliances can change post-election . Predicting seats is one thing. Determining who will ultimately form the Government can be a wholly different exercise.

During the 2004 Lok Sabha election most opinion polls had predicted an NDA landslide. However, it was the UPA that formed the Government. The opinion polls had been wholly inaccurate in their overall predictions. However, looking at their predictions on the State level showed some accuracy. Opinion polls for Rajasthan, Haryana, Karnataka, Tamil Nadu, Odisha, West Bengal, Uttar Pradesh, Bihar, Jharkhand, Maharashtra, and Punjab were mostly accurate. However, for Gujarat, Uttarakhand, Himachal Pradesh, Andhra Pradesh, Assam and Delhi, the polls were mixed. For Gujarat, it could be equated to the Modi effect – even though everyone said they would vote BJP, potentially out of a misconceived fear, on voting day many still voted for other parties and the BJP captured 47.78 percent of the vote. In Assam, where conducting surveys is difficult, a BJP win was overstated in the opinion polls and Congress carried more seats than the other parties with 35% of the votes.

Based on our analysis of opinion polls from 1998-2009, we would state that certain pollsters show a high level of by-state accuracy. There is little evidence that opinion polls have influenced voters greatly.

Rather than banning opinion polls, we should demand more transparency from those conducting them. At its basics, what is the sample size, methodology and margin of error for each poll? In fact, the pollsters should all belong to trade bodies which demand quality and transparency, as has happened in other countries. The solution is not banning pollster. Rather, there should be more transparency in how each poll is conducted.

What do the state elections tell us about the Lok Sabha Elections?

Did the state assembly elections indicate any Modi or even AAP effect?

Within Delhi, Congress suffered a 15% drop in its vote share. Interesting, the BJP also lost in its vote share by 3%. The AAP, choosing to debut in Delhi, gained 30% of the vote share.

Rajasthan, which regularly swings between Congress and BJP, followed a similar pattern. The BJP gained 12% in voter share, while Congress dropped 3%.

In Madhya Pradesh, both the BJP and Congress gained in voter share, by 8% and 4%, respectively. It does not seem that the BJPs gain was at the expense of Congress.

In Chhattisgarh, both the BJP and Congress gained in voter share, both by 2% each.

In Mizoram, where Congress retained power and gained a seat, the party saw a similar voter share to 2008. To note, one district will be repolling on 12Dec2013.

It seems that the smaller players were the ones who decreased their voter share in the states. It may be that the voters are moving towards a two party preference, as the regional parties seemed to have suffered more. However, the regional parties in West Bengal, Odisha, Bihar, UP, Tamil Nadu, Kerala and Maharashtra are much stronger and these state elections are no indication for what will occur there during the Lok Sabha elections.

While there is some evidence of a Modi effect in Rajasthan and Madhya Pradesh, it does not seem significant considering the voter increase that Congress also experienced in MP.

Finally, with the AAP having had a strong showing and hinting at contesting more seats in the upcoming Lok Sabha election, it will be interesting to see how the voter share is affected. Will voter opinion change now that the AAP will likely be in opposition in Delhi for five months until the Lok Sabha elections?



FiveFortyThree aims, through rigorous statistical analysis, to predict the 2014 General Election results for the 16th Lok Sabha in India. Our aim will be to predict the outcomes for each of the 543 parliamentary constituencies.

Our statistical team of NRIs will remain apolitical.

As amateur poll aggregators, we wish to see whether a system can be developed, similar to those performed for elections in other countries, for predicting the General Election winners.

Presently, we have databased the following,

a) Pollster accuracy from previous general elections (1999-2009). This data will be used to present a pollster rating. Our database includes details for AC Nielson, GFK Mode, C-Voter, MDRA, TNS and CSDS.

b) All candidate results from 1998 to 2009. This includes the voter turnout and winner details for each election. This database has also taken into account delimitation. Consequently, some predictions have been made for newly named constituencies on likely winners in older elections, based on the Vidhan Sabha constituencies.

We will also look at surveys conducted on political opinions and attitudes to determine whether these have any predictive abilities in voter decisions.

There are several methodologies that can be taken to perform our predictive analysis. Some of the more famous ones include those of Nate Silver and The Princeton Election Consortium. As we develop our methodology, we will present it on this blog. One of the fundamental differences to account for in our analysis, compared to other countries such as the USA, is in the fact that India has a multiparty system. Also, governments are typically formed through alliances. The political jostling that occurs post-election can be difficult to account for, so our aim will be to predict the outcomes for each of the 543 constituencies and then present scenarios for the most likely alliances.