Abstract
We analyse the Bihar assembly elections of 2020, and find that poverty was the key driving factor, over and above female voters as determinants. The results show that the poor were more likely to support the National Democratic Alliance (NDA). The relevance of this result for an election held in the midst of a pandemic is very crucial, given that the poor were the hardest hit. Second, in contrast to conventional commentary, the empirical results show that the All India Majlis-E-Ittehadul Muslimeen—‘AIMIM factor’ and the Lok Janshakti party—‘LJP factor’ hurt the NDA, while benefitting the Maha Gath Bandan (MGB), with their presence in these elections. The methodological novelty in this article has combined elections data with wealth data to study the effect of poverty on election outcomes.
Introduction
Bihar assembly elections were an important litmus test for the policies of the National Democratic Alliance (NDA), 1 particularly in the aftermath of the socio-economic upheaval caused by COVID-19. Unfortunately, the poor have had to face a disproportionately higher brunt of the crisis (NCAER, 2020, April 12; 2020, July). The results in favour of the NDA came as a surprise (it was counter to what majority of the exit polls had predicted) primarily due to the following factors, first and foremost, the anti-incumbency; second, the humanitarian and economic crisis precipitated by COVID-19; and, third, to a limited extent, a spirited campaign, centred around jobs and unemployment, led by the leader of Rahstriya Janta Dal (RJD), the leading party of the Maha Gath Bandan (MGB). 2 Political commentaries and expert views, while attempting to explain NDA’s surprising victory over MGB, have focused primarily on two classes of voters: (a) women voters, who, recalling the misrule of the RJD, favoured the NDA, and (b) the Muslim vote, which was diverted away from the MGB, by parties such as All India Majlis-E-Ittehadul Muslimeen (AIMIM).
In this article, we perform an exploratory data analysis to isolate factors that have determined outcome in these elections. To our knowledge, this is the first article to explore relationship between poverty and election outcomes. Methodologically, the innovation in the study is to combine data sets across different sources for detailed analysis. In particular, we look at the following factors as determinants of election outcomes:
Sex ratio of the electorate (female per 1,000 male electors): This is an important determinant, given earlier studies that have established the significant role of female voters as agents of change in Bihar elections (Ravi, 2015). Proportion of the poorest households within a state in the district in which the constituency is located: These elections were held in the midst of a pandemic, which has seen greatest suffering among the poorest people in the population, and this article wants to isolate the relationship between poverty and election outcomes. Proportion of general caste households in the district in which the constituency is located: Given the significant role of caste in Indian elections, this is an obvious line of enquiry for Bihar elections. Proportion of Muslim population in the district in which the constituency is located: This is an important line of enquiry, given the belief that Muslims are a distinct electoral demography (Siraj, 1986). AIMIM factor: We analyse the presence of AIMIM as a contestant within a constituency, to identify the effect it might have on performance of other political parties in these elections. LJP factor: We analyse presence of LJP as a contestant within a constituency, to identify the effect it might have on performance of other political parties in these elections.
Data Source and Covariates
A novelty in this article is that we combine data from different sources to analyse election outcomes. In particular, we combine data from the Election Commission of India (ECI) and the National Family Health Survey (NFHS). Our main source of data on 243 assembly constituencies in the Bihar general assembly election in 2020 is from the ECI, which provides data at the assembly constituency level on the candidates that contested, the number of votes secured by each candidate and the political party each candidate is affiliated with. If a candidate is not affiliated to a political party, then the candidate is classified as ‘independent’. In addition to the election results, we use data from the 2015 Bihar general assembly election from the ECI to compute the sex ratio of the electorate (female per 1,000 male electors) at the constituency level. We then classify the constituencies into three groups, based on the sex ratios (i.e., group 1 with electorate sex ratio: <861, group 2: (861–886) and group 3: ≥886).
Next, we use household data from the NFHS, round 4 (NFHS IV), which was conducted in 2015–2016. The advantage of the NFHS IV data set is that the sample size is large enough to construct socio-economic indicators at the district level. First, we construct data on proportion of poorest households in a district using the wealth index. The wealth index was constructed by assigning a score to each household, which was based on ownership of consumer goods such as television, bicycle, car, etc., and characteristics of their house such as source of drinking water, toilet facilities, material used for floors, and walls, etc. The scores are assigned using a principal component analysis. Households are then ranked based on the score and divided into five equal categories (quintiles), and the households in the lowest quintile are classified as the poorest. Then, for each of the 38 districts in Bihar, we construct the proportion of poorest households. Next, we classify districts intro three equal groups, based on the proportion of the poorest households (i.e., group 1 where the proportion of poorest households are <15.5%, group 2 where the proportion of poorest households are between 15.5% and 25.5% and group 3 where the proportion of poorest are ≥25.5%). So, group 3 districts are the poorest in the state and comprise the highest proportion of poorest households in Bihar, while districts in group 1 are the least poor in the state.
Second, the NFHS IV household data also identifies the religion and the caste (which are classified into other backward class, scheduled caste and scheduled tribe) of the head of the household. If the household head reports her or his religion as Muslim, then the household is classified as a Muslim household. We classify the household as belonging to ‘general category’ if the head of the household does not belong to either of the caste categories (other backward class, scheduled caste, scheduled tribe). Then, for each district, we construct the proportion of Muslim households and classify them into three equal groups (i.e., group 1 districts are those with proportion of Muslim households <8.6%; group 2 are districts with proportion of Muslim households between 8.6% and 15.33%; and group 3 are districts with proportion of Muslim households ≥15.33%). Similarly, we classify each district into three equal groups by the proportion of general category households (i.e., group 1 has districts where the proportion of general category households are <14.5%; group 2 has districts where proportion of general category households are between 14.5% and 19.6%; and group 3 has districts with proportion of general category households ≥19.6%).
Third, to gauge variations in voting behaviour across rural and urban areas, we classify all districts in Bihar into three equal groups. We create groups of districts based on proportion of households residing in rural areas (i.e., group 1 has districts where the proportion of rural households is <86.6%; group 2 has districts where the proportion of rural households is between 86.6% and 92.3%; and group 3 has districts where proportion of rural households is ≥92.3%). Therefore, districts in group 1 have least rural households, while districts in group 3 have most rural households.
We, then, classify each constituency as belonging to one of the groups in terms of proportion of poorest households, proportion of Muslim and general caste households, and proportion of households residing in rural areas, depending on the districts in which the constituency is located.
In addition, we use the data from the ECI, to construct a dummy variable, which takes a value of 1 if AIMIM contested the election in that constituency, and 0 otherwise, similarly we construct a dummy variable, which takes a value of 1 if LJP contested the election in that constituency, and 0 otherwise.
Results
The first set of results are the summary statistics, reported in Table 1. There were 243 assembly constituencies, where elections were held in Bihar. The contest was primarily between the NDA 3 and the MGB. 4 In terms of the total seats, the NDA won in 125 out of the total 243 assembly constituencies, while the MGB won 110 assembly constituencies. However, in terms of the total votes, the NDA received 15,701,226 (37.3%) out of the total 42,137,620 votes that were cast, while the MGB received 15,688,458 (37.2%) of the total votes. The difference in the votes was a mere 12,768. However, there was a high degree of variance in the performance of individual political parties, as the analysis reveals. For example, the BJP of the NDA had the best strike rate (defined as the number of seats won as a proportion of number of seats contested) of 67.3%, while the Indian National Congress (INC) of the MGB had the worst strike rate of 27.1%. In terms of the percentage of votes received by a political party across the constituencies that they contested in, the BJP received 42.6% of the votes, while the RJD received 38.9% of the votes.
Summary of the Results
For our next set of results, we construct three variables at the level of the alliance (NDA and MGB) and the political parties (Janata Dal United—JDU, BJP, RJD and INC): (a) total seats contested, (b) total seats won and (c) strike rate, which is the ratio of seats won to seats contested and relate their performance to the factors considered (see Table 2). For example, in 81 assembly constituencies that NDA contested in, the sex ratio of the electorate was <861 female per 1,000 male electors, and NDA won in 36 of those constituencies with the strike rate of 44.4%. Similarly, the NDA contested in 81 constituencies that had a sex ratio of the electorate ≥886 and won in 52 of them with a strike rate of 64.2%. If we look at the performance of INC in the constituencies where they contested and where the AIMIM also contested, there were seven such constituencies. Of these 7, INC won 4 of them with the strike rate of 57.3%, while in other 63 constituencies in which the INC contested where AIMIM did not contest, it won 15 seats with the strike rate of 23.8%.
Summary Statistics at the Constituency Level Based on the District Level Data from NFHS IV, Across Alliances and Political Parties
To sharpen our analysis, we perform a logistic regression, where primary outcome of interest is winning a constituency in which the political party (and alliance) contested the election. In particular, we ran the following regression:
where subscript i denotes the constituency, Covariatesd are the factors at the district level d where the constituency is located; these covariates have been explained in detail in the data section. AIMIM factor is a dummy variable, which takes a value of 1 if AIMIM contested the election in that particular constituency, and 0 otherwise; LJP factor is a dummy variable, which takes a value of 1 if LJP contested the election in that particular constituency, and 0 otherwise; and error terms are the error terms that are clustered at the district level to account for similarities in the constituencies within a district. The summary data of the covariates are presented in Table 3. The results of the logistic regression are presented in Tables 4 and 5; Figures 1 and Figure 2.
Summary Statistics at the Constituency Level Based on the District-Level Data from NFHS IV
Odds Ratios for NDA
Odds Ratio for MGB
These results are presented in the form of odds ratios (OR)—an odds ratio measures the association between an outcome and an exposure, where an OR >1 is a positive association between the outcome and the exposure, while an OR <1 is a negative association. Here, the outcome of interest is the likelihood of a party/alliance winning a constituency, and the exposures are various factors that have been considered in the analysis. We find that NDA had 4.25 times higher odds of winning an election in constituencies that were located in the poorest district (where proportion of poorest households was ≥25.6%) as compared to wealthier districts (where proportion of poorest was <15.6%), and these odds remained similar even after adjusting for other factors associated with election outcomes. So, while the unadjusted OR was 4.25 (95% Confidence Intervals [CI]: 2.17–8.30), the adjusted OR (aOR) was 4.39 (95% CI, 1.79–10.81). This indicates that even when we control for all other observable factors, the odds of NDA winning in the poorest districts remain more than four times as compared to their winning in wealthier districts. However, the results show the reverse for the MGB: the unadjusted OR and aOR were significantly <1 for the constituencies in the poorest district: 0.21 (95% CI, 0.10–0.41) and 0.26 (95% CI, 0.11–0.63).
The results, similarly, show that the AIMIM factor reduced NDA’s odds of winning an election, and this is reflected in the unadjusted OR and aOR of 0.37 (95% CI, 0.14–1.01) and 0.09 (95% CI, 0.02–0.33), respectively. While, for the MGB, the AIMIM factor improved the odds for winning an election, and this is reflected in the OR and aOR of 0.99 (95% CI, 0.39–2.48) and 2.97 (95% CI, 0.91–9.73), respectively. The results of LJP factor are similar—it reduced NDA’s odds of winning as reflected in OR and aOR of 0.37 (95% CI, 0.22–0.63) and 0.29 (95% CI, 0.15–0.54), and it increased MGB’s odds of winning as reflected in OR and aOR of 2.80 (95% CI, 1.65–4.76) and 3.49 (95% CI, 1.87–6.49).
Next, we study the effect of women voters on the likelihood of winning of political parties and alliances. We find that constituencies with higher sex ratio of the electorate (i.e., more eligible women voters) have voted in favour of the NDA. However, this effect becomes lower and statistically insignificant once we adjust for the proportion of the poorest households in the district in which the constituency is located. The results show an OR of 2.24 (95% CI, 1.19–4.21), whereas, when adjusted for other covariates, the aOR is 1.80 (95% CI, 0.78–4.13). For the MGB, however, the results were reversed. The MGB was less likely to win in constituencies with higher sex ratio of the electorate or where there were more eligible women voters.
Discussion
In this article we explore the association between socio-economic factors and the likelihood of win at the level of the constituency. Our key finding is that NDA was more likely to win in constituencies that were the poorest. This is an important finding, given that these elections were held in the midst of a pandemic, where the poorest population was affected significantly (NCAER COVID-19 surveys, 2020). Bihar is the poorest state of India, so the significance of these results is particularly striking. To gauge the impact of poverty on election outcomes, we carefully study the variations in poverty across districts of Bihar, using wealth index scorecard for households. Since Bihar has a large concentration of poor households within India, we created wealth quintiles for households within the state and picked up the poorest (lowest) quintile among these to study the impact on election outcomes. These are, therefore, the poorest households within the country, and their voting preferences become particularly critical in the midst of a debilitating pandemic, which has been particularly harsh on the poor. The results of our analysis show that the poorest districts were more than four times as likely to vote for the NDA as compared to richer districts within the state. Controlling for all other observable factors in our analysis did not affect the economic and statistical significance of this robust result.
The hardships faced by the poor in the beginning of the pandemic have been well documented by the popular press as well as through well-designed household surveys (NCAER COVID-19 surveys). The declaration of a national lockdown, as a strategy to contain the spread of the COVID-19 infection in India, led to significant reverse migration of informal workers back to their home states. The general trend was migration of workers from western industrialised states, including Gujarat, Maharashtra, Karnataka and Tamil Nadu, towards the eastern poorer states of Bihar, Uttar Pradesh and Jharkhand. While it is difficult to estimate the exact number of Bihari migrants across India, there are some estimates, which peg this number to 50% out of approximately 20 million households in the state. 5 The early nature of government intervention, in the midst of the national lockdown, was primarily focused on humanitarian support to poor affected households. The relief measures included cash transfers to women, elderly, construction workers and farmers. The interventions also included widespread emergency distribution of food grains through the Public Distribution System (PDS). Existing literature (Mukherji, 2020) on the assessment of these various relief measures suggest that despite the drawbacks, Bihar government managed to reach over 2 million people within few weeks of the relief announcements, through its Corona Sahayata Scheme among other means. The analysis of the election results seems to suggest an overall success of these policy interventions in reaching the poorest within the state. The different welfare and humanitarian schemes, which were rolled out as immediate relief measures after the sudden imposition of national lockdown, seem to have been effective in reaching the poorest section of the population in Bihar, as endorsed by the assembly election results.
The second focus of our analysis is women voters in Bihar. This is primarily driven by the fact that women votes have been an important determinant of NDA win and, in particular, have benefitted the JD(U) significantly in recent past assembly elections in Bihar. This trend has been particularly noticed in assembly elections since 2005. However, contrary to our expectations, the analysis for 2020 elections reveals that women votes did not make a significant difference to the odds of NDA winning, once we control for poverty levels. In the analysis, when we include poverty distribution within the district, then women votes (proportion of female electors within constituency) become insignificant at the conventional 5% level. Hence, the results reinforce poverty as the key driver of election outcomes in Bihar in 2020. These results also hold some deeper insights when we disaggregate the NDA into BJP and JD(U) and then analyse the impact of women votes. As Figure 2 shows clearly, while women votes did not have any significant impact on winning odds of the BJP, they did have a significant impact in raising the odds of JD(U) winning within a constituency. This is an important result as it reinforces women as a consistent vote base for Nitish Kumar’s party in assembly elections in the state since 2005.
Another striking finding in this study, which goes against the popular narrative, is of Muslim population’s voting preferences. The results strongly show that NDA was more likely to win in constituencies with higher proportion of Muslim population, for example, the aOR was 3.40 (95% CI, 1.24–9.34) for constituencies that were located in districts with a Muslim population ≥15.33% as compared to the reference group, which were constituencies located in districts where the Muslim population was <8.6%. Furthermore, the AIMIM and LJP factors significantly reduced the odds of NDA win in the constituency in which AIMIM and LJP contested the election, where the aOR was 0.09 (95% CI, 0.02–0.33) and 0.29 (95% CI, 0.15–0.54), respectively, as compared to those constituencies in which they did not contest an election. It is important to mention that there were 86 assembly constituencies in the 12 districts that had a Muslim population greater than 15.33%. However, AIMIM, contested in only 16 of those assembly constituencies, which perhaps were the constituencies that had the highest concentration of the Muslim population. Our results suggest that it was in these constituencies that NDA had significantly lower odds of winning—the aOR was 0.09 (95% CI, 0.02–0.33). Therefore, even though the NDA performed well overall in districts with a higher proportion of Muslim population, it performed particularly poorly in the constituencies, which had the highest concentration of Muslim population.
Post elections, there was much noise on the impact of AIMIM on performance of MGB with the popular belief that the AIMIM played the saboteur to the INC in particular. Our analysis refutes this consistently because the ‘AIMIM factor’ improved the odds of MGB winning significantly, as reflected in the aOR of 2.97. Within the MGB, the AIMIM factor seems to have benefitted the INC in particular, as shown clearly in Figure 2. The odds of INC winning are significantly higher in constituencies where the AIMIM contested than elsewhere. There is no such clear relationship for the other MGB partner—the RJD.
Limitations
First and foremost, while this article establishes a strong relationship between socio-economic factors and election outcomes, it does not indicate neat causality due to numerous other observable and unobservable factors, which might determine election outcomes. Political preference of people and their voting behaviour is a complex phenomenon, whereas research and analysis are limited by availability of data. Second, we do not have data on socio-economic factors at the level of the constituency. Socio-economic data are only available at the level of a district within a state. In Bihar, there are 243 assembly constituencies across 38 districts. On average, therefore, there are six to seven assembly constituencies in each district. There is, however, a widespread distribution like elsewhere in the country. For instance, Sheohar district has 1 assembly constituency, whereas Patna district has 14 assembly constituencies within it. Third, while our results are suggestive of factors that are strongly associated with a win in a constituency, there may be other factors that are excluded from the analysis, which might affect the results.
Conclusion
Our article suggests that in Bihar assembly elections of 2020, poverty was the key driving factor, over and above female voters. The results show that the poor were significantly more likely to support the NDA in these elections. The odds of NDA winning in the poorest districts of Bihar were more than four times when compared to richer districts in the state. The relevance of this result for an election held in the midst of a pandemic is very crucial. It seems to suggest that even though the poor were the hardest hit by the COVID-19 pandemic and the national lockdown, the various relief measures announced by the central and state governments for the benefit of the poor might have been an important factor in NDA’s victory. Second, the results also show that women voters in Bihar did not affect the fortunes of NDA overall, but within that coalition, they were a significant driver of JD(U) victories. Third, in contrast to conventional commentary, the analysis of the elections data reveals that AIMIM factor and the LJP factor hurt the NDA much more and, in contrast, benefitted the MGB from their presence in the Bihar assembly elections of 2020.
Footnotes
Declaration of Conflicting Interests
The authors declared no potential conflicts of interest with respect to the research, authorship and/or publication of this article.
Funding
The authors received no financial support for the research, authorship and/or publication of this article.