The opinion in opinion polling

As if to underline my warning (in Why you should ignore opinion poll seat projections), the latest ABP News-AC Nielsen opinion poll exemplifies how opaque seat projections are in comparison with more straightforward vote share estimates.

Look at the table below that compares AC Nielsen’s vote share and seat predictions over two successive months in 2014:

Screen Shot 2014-02-23 at 11.35.07 pm

The survey estimates that the vote share of the Indian National Congress (INC) and its allies rose by a percentage point between January 2014 and February 2014, but that its Lok Sabha tally fell nine seats to 92. The vote share of the Bharatiya Janata Party (BJP) and its allies stayed about the same while their count rose ten seats to 236. Other parties’ vote share fell by about a percentage point but their projected seats fell only marginally (by a single seat) to 215.

It’s hard to say why this occurred in the absence of more details from AC Nielsen but what is clear is that the seats projection is being driven not by the vote share expectation but by other, unknown, assumptions about the distribution of those votes. These seat share projections create the illusion of momentum in favour of the BJP alliance when in fact the vote share estimates in effect show no change from the previous month.

Buyer beware.

Political parties’ Facebook “likes”

This measure of political party support has some obvious limitations (viz. self-selection and an upper-income/urban/education bias) but is still, I think, worth noting, especially the Aam Aadmi Party’s strong performance.

Why you should ignore opinion poll seat projections

It is a truth universally acknowledged that Indian public opinion polls are much better at predicting party vote shares than they are at extrapolating to seats won in parliament or state assemblies. That’s because vote shares translate into seats in an unpredictable way under India’s multiparty “first-past-the-post” electoral system.

For example, the Indian National Congress (INC) increased its national vote share only 2% between 2004 and 2009 but its seats tally increased by 60 to 206. Uttar Pradesh is even more stark: the Bahujan Samaj Party won 206 seats in the 2007 assembly elections with more than 30% of the vote, but the Samajwadi Party won 224 seats in 2012 with only 29%. As Sanjay Kumar of the Centre for the Study of Developing Societies told The Caravan: “Same state, same voters, same parties. Party gets 1% less votes and gets 20 extra seats.”

The following chart from Rukmini Shrinivasan’s June 2013 article captures this unpredictability:

20501491.cms

This means that people should take vote share projections in opinion polls much more seriously than seat share projections. And yet our TV studios and newspapers spend hours and column inches on seat predictions with only a cursory mention, if that, of respective vote shares.

So let’s take a look at what the four national opinion polls published so far in 2014 are saying about vote shares:

Screen Shot 2014-02-19 at 10.49.20 pm

The average of these four polls projects that vote share of the INC-led alliance will drop by 7% to 24%, that of the Bharatiya Janata Party (BJP)-led alliance will rise 13% to 34% and of others will fall 6% to 42%.

This is a sweeping vote shift by historical standards:

  1. If it occurs this would be the third biggest negative swing that the INC has suffered, smaller than the -9% and -11% swings in 1977 (post-emergency) and 1989 (Bofors, 1984 wave reversal) respectively but comparable with what occurred in 1996 after the Narasimha Rao government. Perhaps not so surprising in the current political environment.
  2. The real news is the 13% increase in the vote share of the BJP and its close allies, which surpasses the +4% swing in 1989 and the +9% swing the party experienced in 1991 after the Ram Temple movement. This is particularly surprising given the absence of the BJP from large swathes of southern and eastern India.
  3. The fall in the vote share of others is also noteworthy given the rise of the Aam Aadmi Party and the resilience of regional parties in Andhra Pradesh, Bihar, Tamil Nadu, West Bengal and elsewhere.

I will look into these numbers in greater detail in the near future, but here are two caveats:

  • With the exception of the BJP, party campaigns only picked up recently. Not all voters may be paying attention at this point and late vote swings have been known to occur.
  • Candidate selection and quality are important, which is why early polls can be misleading.

BJP-ruled states more communally violent

Following Chris’s response to my previous post, correctly questioning whether it makes sense to use vaguely defined “incidents” to measure communal violence, it took me a while to locate better data. I found some up-to-date statistics on casualties of communal violence in a table annexed to a 10 December 2013 Ministry of Home Affairs reply to a question in the Lok Sabha. This measure, I think, better reflects the intensity of communal violence. The data cover the period 2010 to October 2013.

Screen Shot 2014-02-15 at 5.16.08 pm

The findings are unambiguous: Bharatiya Janata Party (BJP)-ruled states have higher levels of communal violence than do Indian National Congress (INC)-ruled states. The BJP-ruled states in this sample have an incidence rate of 4.3 per million, which is 59% higher than the INC-controlled states. BJP states have a casualty rate of 13.1 per million, 56% higher than their Congress counterparts.

Of these states, Karnataka and Gujarat are by far the most violent (remember that Karnataka was run by the BJP during this period). Uttar Pradesh shines in comparison, and that too during a period that includes the Muzaffarnagar riots and excludes the 2002 Gujarat riots.

Someone needs to ask prime ministerial aspirant Narendra Modi how “good governance” in Gujarat has led to a communal incidence rate 68% higher than the national average and an intensity of communal strife that is 83% higher.

Needless to say the original premise of the Mail Today story is bass ackwards.

Update on Feb 17

Thanks to Tripti Lahiri for pointing out that the 2012 Kokrajhar riots in Assam are inexplicably excluded from the Home Ministry’s table, an omission that others had also noted last year.

To bring Assam in we need to broaden the analysis to all states that had at least one communal violence death during the period under study, which in this instance are Congress-ruled Assam and BJP-ruled Jharkhand.

This adjustment lowers the casualty rate of the BJP states to 12.1 per million, which is still 46% higher than the 8.3 per million of their Congress-ruled counterparts. The casualty rate for states run by regional parties (Bihar, Odisha, Tamil Nadu, Uttar Pradesh and West Bengal) is lower still, at 5.9 per million.

Update on May 7

The following table incorporates fresh numbers provided by a Ministry of Home Affairs reply to a Feb 5 Rajya Sabha question with data for all of 2013 (the earlier table covered the period between 2010 and October 2013):

Screen Shot 2014-05-08 at 3.40.15 pm

The conclusion is unchanged: BJP-ruled states are still more communally violent than others. BJP states have a casualty rate of 12.4 per million, 45% higher than Congress-ruled states and 70% greater than the national average. Gujarat has a communal incidence rate 70% higher than the national average and a casualty rate that is 82% higher.

How Mail Today got its analysis of communal violence exactly wrong

If anyone needed proof that a course in statistical inference is necessary for journalists, it was provided by this report in the newspaper Mail Today on incidents of communal violence between 2011 and 2013 (written by Abhishek Bhalla).

There was even a tweet:

The main claim in the article was:

The Congress has always found it convenient to attack the BJP for spreading communal hatred in the country, but the Union Home Ministry’s latest data on communal violence is embarrassing for the Grand Old Party.

A look at the Union Home Ministry’s data reveals that there is not much difference between the two major national parties when it comes to the law and order situation in their states.

Exhibit A is this chart:

Of course the total number of incidents of communal violence in a state tells you nothing very interesting about the level of communal violence. A larger state will have more of everything than a smaller one. What you need is the number of communal violence incidents per unit of population.

Once you factor in population the picture changes quite dramatically. The four Congress Party-ruled states had 636 communal incidents between 2011 and 2013. These states had a population of 299 million according to the 2011 census. The three Bharatiya Janata Party (BJP)-ruled states had 641 incidents in the same period and a population of 194 million. We exclude the 38 incidents that fall under “other states” since we don’t know which those states are.

The conclusion is pretty straightforward. Congress Party-ruled states had 2.13 communal incidents per million population, while BJP-ruled states had 3.3. The rate of communal violence in BJP-ruled states is therefore 55% higher than in Congress-ruled states.

Exactly the opposite of what the article originally claimed.