Tuesday, May 28, 2019

Analysing the Differential Voting Behaviour of TN Voters in 2019 LS & By-election Polls

As soon as the election results for 2019 Lok Sabha (LS) polls and 22 by-elections in Tamilnadu (which were held together) were announced, it was apparent that people voted differently between the Lok Sabha elections and the Assembly elections. This is clear from the fact that while ADMK alliance managed to win 9 / 22 assembly elections that went to polls, they managed to win only 1 / 38 Lok sabha that went to polls.  

Research Objective and Questions 

The objective of this post is to share a more critical detailed data analysis of the 20 constituencies where people voted for both Assembly by-elections as well as Lok sabha polls. Please note Gudiyattam and Ambur Assy by elections have to be excluded, because they come under Vellore LS where the Lok Sabha election has been postponed and voters only voted for Assy elections. 

 Our interest is to determine answers to the following questions :

  1. What is the quantified swing in vote % towards DMK (or ADMK) when it came to Lok Sabha polls ?
  2. Was EPS right in tying up with PMK , DMDK and PT for the Lok sabha polls? 
  3. What is the effect of the swing when BJP candidate contested in LS polls ? How did it impact the the performance of the alliance? 
  4. What is the effect of the swing when DMK candidate directly contested in LS polls?  How did it impact the performance of the alliance? 
  5. What is the effect of the swing when Congress candidate directly contested in LS polls?  How did the performance of the alliance? 
  6. What is the effect of the other parties (MNM, NTK and AMMK) on this differential voting? How did the performance of the alliance? 


Study Design and Model 
 We took the published voting details of the 20 assembly constituencies that went into polls as our sample. We took the by-election figures for DMK+ alliance and ADMK+ alliance in these constituencies and the Assembly segment figures for the Lok Sabha polls. We ignored the voting figures of all other parties because they can always be normalised in the study. 

We normalised all the vote numbers to percentage figures so that comparison is meaningful across constituencies. 

We took the dependant variable(s) as follows for the models we were building :- 

  •  % Swing towards DMK+ in the LS polls 
  •  % Swing towards ADMK+ in the LS polls 

We tried a variety of explanatory variables, but in the final model, the following variables were the only biggest explanatory variables that need to be included. The list is as follows :- 

  • Whether it was Theni (=1) where OPS Son contested and ADMK+ won in the LS polls 
  • Whether BJP candidate contested (= 1) on behalf of ADMK alliance in LS polls 
  • Whether DMK candidate contested (= 1) on behalf alliance in LS polls 
  • Whether Congress candidate contested (= 1) on behalf of DMK alliance in LS polls 
  • The % of votes that went to the Other parties in one of these polls. 
Please note we tried whether ADMK, PMK or DMDK or Communist or IUML candidate candidate contested was significant and they did not significantly affect the base case. So those indicator variables were dropped. 

The final Regression model results were as follows :- 

%Swing towards DMK+ in LS Polls = -8.9 - 6.9 (Theni) + 10.5 (BJP) + 18.9 (DMK) + 10.4 (Congress) + epsilon 

Adjusted R-square = 69%, Model F gives p-value ~ 0%, All coefficients are with p-value <  5% 


%Swing towards ADMK+ in LS Polls = -75 + 7.9 (Theni) - 10 (BJP) - 17.3 (DMK) -10.2 (Congress) -74.4 (Others %) + epsilon

Adjusted R-square = 66%, Model F gives p-value ~ 0%, All coefficients are with p-value <  5% 

Analysis 

The model confirms the following key observations which were apparent earlier. 

EPS Effect 

We first examine whether EPS did the right thing by tying up with PMK, DMDK, PT for the Lok Sabha polls and was it useful to boost their performance in Lok sabha polls compared to their performance in assembly by-elections. The result is somewhat counter intuitive.

Ceteris Paribus on all other parameters, there was a 7.5% swing towards ADMK alliance in Lok sabha elections compared to their performance in Assembly polls, when any non-BJP candidate contested on behalf of them. 

Ceteris Paribus on all other parameters, there was a minus 8.9% swing towards DMK alliance in Lok Sabha elections compared to their performance in Assembly polls, when any non-DMK or non-Congress candidate contested on behalf of them. 

This shows the existence of "EPS effect" where people favoured the alliance of ADMK + PMK + DMDK when compared to the DMK allies of VCK, IUML, and Communists. This confirms that ADMK got the right allies for contesting the Lok Sabha polls. 


BJP Effect 

We next examine how did Tamilnadu people accept the presence of BJP in the ADMK alliance when it came to Lok sabha polls. 

Ceteris Paribus on all other parameters, there was a minus 2.5% swing towards ADMK alliance in Lok sabha elections compared to their performance in Assembly polls, when any BJP candidate contested on behalf of them. 

Ceteris Paribus on all other parameters, there was a + 1.6% swing towards DMK alliance in Lok Sabha elections compared to their performance in Assembly polls, when any BJP candidate contested against them. 

So we can see, not only did all the good work of EPS effect (base case) got vanished, but the alliance with BJP led to a 10% vote share loss to ADMK+ and a 10.5% vote share gain to DMK+ (the balance 0.5% coming from other parties) when it came to Lok sabha polls. 

This is as clear as it gets that Tamilnadu people did not like ADMK and BJP alliance and hence punished ADMK in the Lok sabha polls in the order of loss in 10% vote share. We can all argue how their perception is wrong about BJP etc., but perception is reality and that is what hit ADMK the most in the Lok Sabha elections! 

Stalin Effect 

We next examine the DMK effect where whether there was any positive impact when DMK contested the election directly, as opposed to giving the seat to the allies. There was a lot of criticism that Stalin was extremely generous and gave far too many seats to the allies. 

Ceteris Paribus on all other parameters, there was a minus 9.8% swing towards ADMK alliance in Lok sabha elections compared to their performance in Assembly polls, when any DMK candidate contested against them. 

Ceteris Paribus on all other parameters, there was a + 10swing towards DMK alliance in Lok Sabha elections compared to their performance in Assembly polls, when any DMK candidate contested on behalf of them. 

So we can see, not only did all the good work of EPS effect (base case) got vanished when DMK contested directly by putting their candidate, but DMK would have gains 10% vote share when they directly contest the seat. 

This is clear that Stalin made an error in giving far too many seats to its allies whereas it was the perception of DMK among the voters that actually tremendously helped the UPA alliance. 

Rahul Effect

We next examine whether there was any impact when the congress candidate contested in the seat. 

Ceteris Paribus on all other parameters, there was a minus 2.7% swing towards ADMK alliance in Lok sabha elections compared to their performance in Assembly polls, when any Congress candidate contested against them. 

Ceteris Paribus on all other parameters, there was a + 1.5swing towards DMK alliance in Lok Sabha elections compared to their performance in Assembly polls, when any Congress candidate contested on behalf of them. 

So we can see, not only did all the good work of EPS effect (base case) got vanished when Congress contested directly in that seat, but UPA would have gains of 1.5% vote share when they directly contest the seat. But the corresponding ADMK loss was 2.7% indicating some of the ADMK vote loss went to "Other parties" when congress contested there. 

This is clear that there is a positive Rahul effect in Tamilnadu, but this is not as strong as the DMK effect. In fact when congress candidate was there, some of the people switched to other parties instead of voting for DMK alliance. 

Theni Effect 

 Theni went different to other seats because there ADMK won the Lok sabha seat. This may be due to OPS son contesting there as ADMK candidate, or because the opposition candidate EVKS Elangovan was a complete outsider imported from Erode district. But whatever was the reason there was a Theni effect which took Theni in the opposite direction. 

Ceteris Paribus on all other parameters, there was a + 15.4% swing towards ADMK alliance in Lok sabha elections compared to their performance in Assembly polls, in Theni. 

Ceteris Paribus on all other parameters, there was a minus 15.8swing towards DMK alliance in Lok Sabha elections compared to their performance in Assembly polls, in Theni

So we can see that the EPS effect was accentuated in Theni with a OPS effect presumably and the swing was 15.4% in Lok sabha elections (compared to the base case of  7.5%. In fact DMK suffered 15.8% vote share loss here, indicating that 0.4% of the DMK votes went to other parties (presumably AMMK) instead of going to EVKS Elangovan. 

Other parties Effect 

We next examine whether the presence of the other parties (MNM, AMMK, NTK etc.) had any impact on the differential voting between assembly by-elections and Lok sabha polls. 

There was no impact of the presence of other parties on DMK alliance performance. 

But this impacted ADMK alliance performance by a degree of  minus 0.744 % for every 1% swing that went towards the other parties. 

Since these numbers are materially insignificant, we can say the presence of other parties did not matter when it came to differential voting. 

Conclusions 

The summary of this differential voting behaviour reveals some interesting insights. 

  • EPS did the right thing by roping in PMK, DMDK, PT etc for the Lok Sabha polls as it gave them a positive advantage over their performance in Assembly by-polls. 
  • Presence of BJP in the alliance hit ADMK alliance very hard when the candidate was a BJP candidate. People punished ADMK alliance in LS polls more when BJP contested. 
  • The performance of DMK alliance is largely due to the perception people have on DMK. In fact Stalin was wrong in giving far too many seats to his allies. 
  • There was a positive perception against Congress as well. It may not be as strong as DMK, but it exists. 
  • Theni seat witnessed an accentuation of EPS effect. We can explain it as EPS + OPS effect or a EPS + EVKS effect. Whatever it is, Theni was different from others.  
  • The presence of other parties like MNM, AMMK, NTK in the lok sabha seats was immaterial in explaining the differential voting. Their voters are their voters.