An alternative model to predict electoral outcomes

Print edition : June 04, 2021

Opinion polls and exit polls are the usual exercises during elections that kindle the voters’ and the politicians’ curiosity equally. Usually, a sample survey is conducted across a State’s geographical space, considering the demographic and social composition. These opinion polls are criticised by the politicians when an adverse outcome is predicted for the party they belong to. Also, these polls do not correctly predict the outcome at all times.

In recent times, multiple agencies have been conducting these surveys and giving different numbers, so it becomes difficult to validate the efficacy of these predictions. Despite all the disadvantages stemming from sample bias and statistical modelling errors, these survey-based opinion polls are popular owing to the belief that they capture the current mood of the people.

As an alternative, in 1978 the economist Ray C. Fair developed an economic model using regression analysis to predict the result of the Presidential elections in the United States and validated it for a long time.

Initially, he built a model to predict the incumbent vote share with variables such as economic growth rate, inflation, good news quarter, President running, and duration of the incumbent party.

The basic argument is that the people’s mood depends on the economic well-being that prevailed during a government’s tenure. Of course, such models also have disadvantages in terms of problems associated with the regression analysis, but the most significant advantage they enjoy is that they give a future direction to the incumbent government to win back elections, which is not possible in survey-based opinion polls. On the other hand, conducting opinion polls is costly and time-consuming.

In this essay we build two simple models to predict the electoral outcome of Tamil Nadu elections. The first is a regression model similar to Ray C. Fair predicts the incumbent’s vote share, and the second, a mathematical model mapping the seat share with the derived vote share.

Finally, we compare these predictions with the electoral outcome of the just-completed election in Tamil Nadu and discuss what it takes for the incumbent government to win back a majority in the next election.

Estimating vote share

U.S. Presidential elections are different from State Assembly elections in India, but the idea of using economic variables to predict electoral outcomes is utilised with the available data.

We took data from 1967 to 2016 for the predictions of both models. We used three independent variables to calculate the incumbent vote share: the growth rate of per capita Net State Domestic Product (NSDP), fuel inflation calculated from the Wholesale Price Index, and the continuation of the incumbent government (Table 1). The average of these two economic variables during the tenure is considered.

The third variable gets the value 1 if the incumbent government continues to hold on to power and 0 if not. The expected sign of coefficients for the average annual growth rate of per capita NSDP is positive, and the average fuel inflation is negative. Since the per-capita NSDP reflects economic well-being, the incumbent government that increases it should be rewarded with more vote share in the next election.

Similarly, any inflation is a tax on disposable income, and the government that allows inflation to rise will face the public’s wrath with a reduced vote share. There is no expectation for the third variable as we do not know that being an incumbent government is advantageous. The estimated parameters from the model provide very interesting inferences.

As expected, fuel inflation hurts the incumbent vote share. The per capita NSDP also has a negative impact indicating the fact that there is inequality in income distribution.

On the other hand, being an incumbent government has some advantages, as indicated by the estimated positive coefficient for that variable. Further, this simple model explains almost 80 per cent of the variation in the incumbent vote share.

Finally, by plugging the values for 2016 to 2020, the average annual growth rate of NSDP of 7.53 per cent, the average fuel inflation of 4.43 per cent with the incumbent government variables taking value 1, the model estimates the incumbent All India Anna Dravida Munnetra Kazhagam (AIADMK) government’s vote share equals 34.30 per cent. Thus, it is very close to the actual number of 33.29 per cent, with a 1 percentage point error.

We categorise the elections into two-cornered contests and three-cornered contests to map vote share and seats. Except for the election years of 1977, 1980, and 1989, all the other elections are considered two-cornered contests in Tamil Nadu. The 1977 and 1980 elections were contested by the newly formed AIADMK, the DMK, and the Indian National Congress (INC). In 1989, the contest was between the DMK and two factions of the AIADMK led by Jayalalithaa and Janaki Ramachandran, respectively.

All the other elections were two-cornered contests as the DMK and AIADMK formed alliances and faced the polls, so we used only the two-cornered contest data as 2021 also fought with alliances formed by the AIADMK and the DMK. A quadratic equation is fitted to the data (Figure 1). For the vote share of 34.30 per cent estimated from the regression analysis, the quadratic equation predicts 74 seats for AIADMK, but the actual seats won by it is 66. To validate this quadratic equation, we calculate the seats corresponding to the actual 33.29 per cent vote share gained by the AIADMK as 69. This model suggests that any government with a minimum of 42 per cent vote share gets a simple majority in the Assembly.

Three essential insights are derived from this analysis for the present DMK government. First, removing income inequality rather than increasing the average per capita NSDP should be the policy concern as higher average per capita NSDP with income inequality might hurt its vote share in the future.

Second, although controlling fuel inflation is not in its hands, managing it by adjusting the fuel tax levied by the State government will improve the party’s vote share.

Third, and most important, it has to obtain a minimum of 42 per cent vote share to get a majority in the Assembly.

S. Raja Sethu Durai and R. Srinivasan

This article is closed for comments.
Please Email the Editor