Just half an hour
after polls closed in Karnataka, we were literally bombarded with a variety of
polls from different media houses. Instead of providing us a broad directional sense
what the likely results would be, they served to cloudy the outcome even
further. They were polls calling the BJP as a winner, others that called the
Congress, while most predicted a hung house.
Exit polls serve two
basic purposes. Firstly it permits media to "call"the elections earlier
than when official results are declared. The second purpose as it usually happens in
India is that they function as willing surrogates to political parties to play
mind games designed to deliberately mislead perceptions of the eventual winner.
Politicians use the intervening time between the close of voting and the
declaration of results to identify likely winners in order to woo potential defectors so
that they could be ready with the numbers to stake claim for government in the
advent of a hung house.
Unlike pre-polls,
exit polls are expected to have a +1% margin of error. Exit polls are
normally a better barometer than opinion polls as opinions are recorded of only
those actually voted. If not quite accurate in terms of seats, they do often suggest
the directional trend. We have seen
French pollsters consistently accurate, operating within this +1% range.
On the basis of exit polls, French politicians even without waiting for final
results make out their concession speeches. But this is contrary to the
experience elsewhere in the world, whether in the US, UK or India where exit polls
have an extremely chequered track record. Even a Poll of Polls often could be a
meaningless exercise as if we went by them we should be having President
Hillary Clinton instead of President Donald Trump! Indian pollsters in particular
have an even more notorious record than their US counterparts as we have only a
handful of genuinely polling specialist organizations, with mostly market
research firms masquerading as polling specialists.
Nate Silver shot
into the limelight in 2012 when he correctly predicted all 50 states during the US
Presidential elections and soon achieved cult status within the polling industry.
He observed that among the winners and losers in any election are pollsters.
Ironically, Nate lost his cult status and became a butt of jokes when in 2016
he gave Donald Trump just a 25% probability to be the President of US!
Similarly
reputations of polling agencies are on stake with their Exit Polls. Since it is
not possible for this blog to do a comprehensive analysis of all the exit polls
published, we filtered on the basis of statistical outliers and honed on two
polls - the AxisMyIndia who predicted an outright Congress win and CVoter that
predicted an outright BJP win. That does not mean that we have discounted the
probability of a Hung House. It only means that due to constraints of time and
resources, we have limited our analysis to out statistical outliers. One of
these two agencies would be a loser and another winner on May 15th when the EVMs are
opened. Or both could be losers in case of a hung house.
SUMMARY PROFILE
CVoter: With
respect to Karnataka, this polling agency has a fairly excellent record. They
predicted the defeat of the BJP in 2013 and hit the bullseye for the BPMP polls
of 2015. However in their 2013 Assembly exit poll, while their vote shares were
mostly on the mark, they over-estimated BJP by a whopping 6% and
under-estimated Yedurruppa's KCP by 2%+, suggesting they may have an inherent
tendency to over-estimate BJP. CVoter tends to treat opinion and exit polls solely as a
horse race. The agency is not very transparent of their internals of their
survey in terms of publishing a report of their methodology, raw and modeled
data etc. However, Yashwant Desmukh, founder was kind enough to provide these
data as used in the blog through a twitter exchange.
AxisMyIndia: This is a relatively new kid of the block
who made their entry with a splash. We saw them first in the Bihar Assembly
polls where their predictions of the Mahagathbandhan (Grand Alliance) victory
was initially treated as a statistical outlier which when actualized shot them
into instant fame. Their success rate is very high with one notable failure -
Tamilnadu. But the latter was actually a too close to call election, which
could have gone either way. Calling right and wrong are part and parcel of the
psephology occupation and what sets apart a good pollster from a bad one is a
success rate of 70%+ which Axis has managed to adhere to till now. Axis too is
a horse race pollster but looks more professional than CVoter - the way their
desegregated caste numbers according to individual parties coalition as seen
later in this post was simply brilliant. Like CVoter, they do not divulge much
information on their methodology.
METHODOLOGY
CVoter: Very sketchy
information was available on their methodology exists in the public domain. But
responding to the blog's query this is what Yashwant Desmukh tweeted
"We covered all
seats; but I don’t think a hundred or two hundred samples are good enough for
any particular seat projection. Technically speaking Minimum sample required to
predict any individual assembly segment would be approx 1280 for that kind of
universe.....Exit polls are always face to face."
AxisMyIndia: The agency claims that the covered 222
Assembly constituencies across 30 districts of Karnataka and surveyed 51510
voters.
We would have loved
to have included CSDS-Lokniti in
this comparative analysis who tonight is expected to publish their post-poll findings.
The differential CSDS-Lokniti would have brought in is that they treat polls as less a horse
race and instead provides very rich information on the socio-political context under which the
poll takes place that gives more meaning to numbers projected. Where CSDS
additionally scores is transparency standards on their methodology and other
internals.
Basically both polls project Congress and JDS as identical in vote shares. However CVoter projects BJP 6% higher, the entire difference cannibalized from OTHERS. Here lies the key difference between the two polls.
Considering the
number of rebels and minor parties, OTHERS at 3% look grossly under-estimated. Even
during the Modi Wave, OTHERS had not been reduced to a measly 3% and this time
under an apparently waveless setting, it looks considerably less so going by
probability. Besides, this time NOTA itself is expected to be in the range of
2%+ with many disillusioned BJP voters flocking to this button.
For JDS vote share
to benefit BJP, it should be less than 15%. But at 17% it is more likely to
benefit the Congress as it decreases the Index of Opposition Unity (IoU). Since
other exit polls have marginally projected JDS vote share higher, it may be
possible that JDS vote share could be slightly under-estimated, the effect of
BSP alliance probably not fully reflected. If so, any increase in JDS vote
share in counting, should favor the Congress.
Yashwant Desmukh while
presenting CVoter findings in News9 claimed that the Vokkaliga consolidation was
confined in the Old Mysuru Region and in outside this regions Vokkaligas
tactically voted BJP. However with JDS having 17% overall vote share, there
appears little evidence to substantiate this claim.
According to Axis,
Bangalore looks too close to call with a slight edge to Congress, while CVoter
projects a facile sweep for BJP.
The CVoter vote
share projections for BJP is much below what they registered during the Modi
Wave but substantially higher than what the party registered during BBMP polls
in 2015. Read our archive: HERE. The difference in
BJP vote share is mainly attributed by cannibalizing the vote share of OTHERS
and Congress. The JDS vote at 16% too also looks over estimated as JDS hardly
any presence in Urban Bangalore seats, being confined to 3-4 Rural Bangalore seats.
Bangalore registered
only 40% voter turnout with the popular perception that a considerable chunk of BJP voters
preferring to stay away from voting or pressing NOTA. If CVoter indeed captures
the actual mood of the people, it would puncture this perception.
There exists high
congruence between the two polls regarding this region, the vote share difference
lies mostly within their margin of error of these polls. The Congress according to both these polls
have a comfortable edge over JDS only because the IoU is low.
It is clear from
these numbers that the BJP did not transfer their votes to JDS and instead may
have eaten into the votes of JDS, damaging the latter's seat prospects. Again
we witness CVoter projecting OTHERS with a dismal vote share which raises
eyebrows whether this is a pattern that is catapulting the state into increased
bi-polarism.
In this region, we
find while Axis forecast a sweep for the Congress, CVoter numbers suggest that
it is too close to call. Like Bangalore, only one of these polls could be
right.
What is clear is
that JDS numbers are fairly clear - lying within 8-9% band.
Axis projects a
whopping 12% lead for Congress. Cvoter narrows it down to virtually
neck-to-neck contest by dwindling Congress & OTHERS votes share by 4%+ and
crediting it to BJP.
While exit polls
overall paints a mixed picture, most tend to call this region for Congress.
This is where the Congress AHINDA base is supposed to be the strongest and we
have to wait till May 15th to find out whether CVoter was on track to conclude that this base suffered a degree of erosion by BJP successfully making
inroads into it.
This region we see a
significant variation in projections. Axis calls the elections for BJP while
CVoter projections suggest it is too close to call.
Where there are
consensus is on the vote share of BJP and JDS. The principal difference is the
vote share of the Congress is entirely accounted by difference on OTHERS. Here
too CVoter is projecting just 4.4% for OTHERS that look a gross an
under-estimate.
If we go by other
exit polls, Mumbai-Karnataka, a region that is traditionally a bastion for the
BJP, looks mostly a sweep for the BJP but could be closer than 2013.
Here both polls
converge to suggest a BJP sweep in this region. However CVoter numbers suggest
a more comprehensive sweep of the region for BJP. This is on boosting BJP numbers
largely from OTHERS and to a smaller extent, the JDS. Here we see OTHERS as
projected by CVoter to be pathetically reduced to 2.3%!
Here again both
polls converge to suggest a BJP sweep in this region. Axis numbers suggest a
more comprehensive sweep than CVoter
The gender data of
Axis is not available but during their presentation of their findings in
IndiaToday Exit Poll program, they did mention Congress had substantially more
share of women voters, leading by a margin by 5%+.
On the other hand,
CVoter projects women marginally preferring BJP than the Congress.
If CVoter has indeed
accurately captured the mood of women voters, then it is akin to women giving a
thumbs up to rape, sexual crimes as seen catapulting in BJP ruled states.
Caste Data
In the caste desegregated
data, we see evidence the methodological superiority of Axis over Cvoter.
LIBRA (Lingayats plus Brahmins)
BJP is trying to
create a social coalition of Lingayats, Brahmins, non Kuruba OBCs and poach a
section of Dalits from Congress, Vokkaligas from JDS.
AHINDA
Congress is trying
to create a rainbow coalition of Dalits, STs, Muslims, Kurubas and Christians
while poaching a section of Lingayat voters from BJP and Vokkaligas from JDS.
VOKKALIGA
JDS doesn’t have plus
votes apart from Vokkaligas.
Instead of a
religious polarization as seen in the cow belt states, it is clear from the
caste data the Battle for Karnataka is one between AHINDA and LIBRA with
Vokkaligas as spoilers. Since the AHINDA
social coalition represents the poor & the marginalized and LIBRA as mostly
representing the upper castes, the Battle for Karnataka is one mainly between
the poor and the rich, with Vokkaligas sitting on the fence.
Within this context,
the JDS becomes very important, not merely as they could probably be decisive
in state government formation but depending on their alignment, can tilt the
state decisively for the Lok Sabha 2019 for either the Congress or BJP
AxisMyIndia
projecting Congress having an edge is based on the paradigm that their social
coalition AHINDA represents 49% of the population while BJP's LIBRA coalition
trails behind at 41% and voter turnout confirms that the Congress succeeded in catalyzing
high consolidation of AHINDAs.
CVoter projecting a
BJP win rest on the paradigm though BJP's LIBRA(41%) trails behind Congress AHINDA
social coalition, the degree of consolidation within LIBRA is much higher which
offsets the overall net vote share gap.
However, it should
be noted the assumptions of allocating different weightage to different castes
and sub-castes is most likelynot to be fully accurate. The AHINDA social coalition is much
larger than 49% as assumed by Axis as per the new caste census undertaken by
the Siddaramiah but not officially released yet. This statistical error if any, can cause huge
changes to vote share projections. Secondly, polls by and large tend to
underestimate Congress and JDS while over-estimating the BJP. Being poorer and
vulnerable sections, the AHINDA sections of voters may not be as articulate or truthful of their
voting pattern fearing reprisals from higher castes. In fact, hours after
voting close, instances of reprisals on Dalits by higher castes have been
reported.
Overall Sense
Karnataka is
vertically split with Coastal, Central & probably Mumbai Karnataka regions
being swept by BJP and with the Congress sweeping Bangalore, Old Mysuru &
Hyderabad Karnataka.
The winner would be the
one, who relatively is able to more comprehensively sweep their respective
regions than the other. Whoever wins a simple majority will still need JDS, if
not for giving the next state government more stability in numbers, but also for
enhancing their winnability for Lok Sabha 2019.
It is difficult to
see either AxisMyIndia or CVoter hitting the bullseye. Only one of them can
emerge directional right. Perhaps
tonight's Lokniti-CSDS Post Poll could give us which way Karnataka would swing
tomorrow.
- Monday, May 14, 2018
- 1 Comments