New Delhi: As the seven-phase Lok Sabha elections enter their last leg, all eyes are on psephologists — the people who predict elections by conducting pre-poll surveys and exit polls. Be it stars like Prannoy Roy, Dorab R. Sopariwala and Yogendra Yadav or an army of nameless politics and data enthusiasts, they have all been as busy as politicians and poll officials on the election trail.
For them, Sunday 19 May is both their biggest festival and their day of reckoning, for that’s the day exit polls will be released, locking in their predictions for the final results on 23 May.
But even as popular interest in their work surges and methodologies become more sophisticated with big data mining and artificial intelligence, there is no formal course for aspiring psephologists in India to pursue — only a handful of them are trained professionals.
Young graduates from various fields like business administration and engineering, who are interested in politics and data interpretation, have taken to joining this stream in recent years to study elections and voter behaviour.
Also read: Why media ground reports, opinion & exit polls don’t agree with each other on elections
‘Psephology should be about big data, not anecdotes’
“An understanding of politics and a passion for data is what one needs to become a good psephologist. There is no formal training or course that is available, and that is the sad part,” said Sanjay Kumar, political analyst and psephologist, who serves as director at the Centre for the Study of Developing Societies (CSDS).
“Some universities introduced a paper called ‘understanding Indian elections’ in political science around five years ago. This is because the subject has been gaining popularity in the last few years.”
The number of trained psephologists in India is minuscule.
“In the last five years, there has been a mushrooming growth in the field of election analysis, and people trying to predict election results are being clubbed in the list of psephologists, even if they don’t deal with big data. But I would hesitate in calling them that. It should be more about big data and research rather than just anecdotes.”
What one needs to be a pollster
Shashi Singh, a political science graduate who holds a master’s in international relations from JNU, entered the profession in 2014 purely because of his interest in politics and a knack for data.
“I have the caste data for each constituency that helps me understand the voting pattern in a better way. Understanding of caste dynamics, a keen interest in politics and data is all it takes to become a pollster,” Singh said.
Partha Das, an IIM Bangalore graduate, said “a general MBA is good enough if you want to become a psephologist in India”, adding that people should also be trained in market research. While making poll predictions, one has to be able to analyse market performance, what is trending on social media, and what investors are looking for, he said.
“If people are searching for a particular political party or a candidate around a certain time, it indicates a trend,” he said.
In this Lok Sabha election, one of the buzz-creators has been Anthro.ai, a platform that uses artificial intelligence to analyse publicly available data sets from published news, demographic and census data, historical data and social media posts. It is run by a team of anthropologists, data-scientists, mathematicians and market specialists.
The team started setting up custom data signals in September 2018 using news, blog posts, tweets, public instant-messaging groups and photos being posted across the states, past electoral data and census data. It claims to have analysed nearly 150,000 booth level records across multiple levels, and between 300,000 and 500,000 geo-located tweets at the constituency level to sample things like personality and attitudes.
The team also analysed more than 500 news articles a day from a curated list of sources to constantly track shifts in opinion about key political figures and issues.
Also read: Indian markets are worried of election ‘bolt from the blue’ on 23 May
Absence of data culture
Psephologists have been able to predict elections correctly several times, but they have also been horribly wrong.
“In 2004, most people predicted a large number of seats for the BJP, but that did not happen. In the 1998 Madhya Pradesh elections, we predicted that Digvijaya Singh would not come back to power again, but he got a second term. We also went wrong with predicting the Aam Aadmi Party’s seats in Punjab in 2017,” said Yashwant Deshmukh, psephologist and founder of CVoter.
“The biggest problem with psephology in India is the absence of data culture. The maximum number of people who are in the business do not have an understanding of data to be able to predict an election result.”
Rahul Verma, fellow at the Centre for Policy Research, wrote in a paper titled ‘Elections, Exit Polls and the Electronic Media’ that exit polls fail to collect a sample representative of the local population.
“Those who get interviewed in an exit poll tend to be more forthcoming, male, from the upper strata of Indian society, among many other such characteristics that skew the representatives of samples,” Verma said.
Also read: For Congress-JD(S), this election is more than just about a good Lok Sabha result
Changing methods
The entry of multinational market-research agencies such as Nielsen, TNS and GfK Mode in the 1990s gave an impetus to India’s polling industry, something Verma also wrote about in his paper.
Many market-research firms in India now prefer direct recruitment and training, use GPS trackers to identify sampled locations, and have moved from the PAPI (paper and pencil interviewing) model to CAPI (computer-assisted personal interviewing).
“Switching to computer-assisted personal training and using things like GPS have been very helpful in conducting market research and policy evaluation studies. However, this model does not look economically viable for election surveys, which have to be conducted in a very short time-frame on a large scale,” Verma told ThePrint.
“The total engagement between a polling firm and investigators for a typical exit poll project is not more than 72 hours. Transporting electronic devices for CAPI is expensive, and wear-and-tear is high. Some pollsters have turned to CATI (computer-assisted telephonic interviewing) as an alternative.”