Professional Documents
Culture Documents
Fallibility of Opinion Polls in India
Fallibility of Opinion Polls in India
Fallibility of Opinion Polls in India
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
https://about.jstor.org/terms
Economic and Political Weekly is collaborating with JSTOR to digitize, preserve and extend
access to Economic and Political Weekly
Sabha thus may become a choice terrain Lok Sabha, will stick out as sore thumbs legislature with their abi
for intrigues among these abominable ef- from the midst of this cesspool. But they the masses in the streets, i
fluvia of self-serving, criminal and cor- can make a difference if they are sincere sist the domination and
rupt politicians which will flow from the in their commitment to the secular and of society by religio-polit
ongoing polls. democratic values embedded in our the Sangh parivar, as well as the corrup
The handful of honest and courageous Constitution. They will have to combine tion of our political system by the cor
individuals who may get elected to the their debating skills on the floors of the rate boss-politician-bureaucrat nexus.
did not respond. This helps in avoiding successes as well as failures. Seat pre- led by Congress cam
non-response error. dictions based on election surveys done Thus poll predictions based on election
by various media houses during the surveys during the 2004 elections (as
A History general elections held in 1998 and 1999 seen in Table 3) went completely haywire
The popular media surveys started in the were fairly accurate, and that gave a big in most cases.
1980s when Prannoy Roy (beginning in boost to opinion polling. Tables 1 and 2 With such a failure of pre-polls and
1979-80), conducted opinion polls during show seat predictions by the various exit polls, questions were raised about
elections to find out the mood of Indian surveys and the level of accuracy vis-à-vis the failure to predict nda's defeat during
voters. The proliferation of electronic the actual results. the 2004 Lok Sabha elections. Were the
media in the 1990s made the electionTable
Table1:GeneralElections1998-AlmostAccurate
1: General Elections 1998 - Almost P°"s biased, incorrectly done or w
Accurate
Outlook/CMS
319-329 135-145 34-39
media further increased the number of Outlook/CMs 319-329—135-143 34-39 wjj[ cross tjje 200 everybody fa
„„il- 1 . J . .. IndiaToday/Insight
India Today/Insight332-336 332-336
132-146 132-146 70-80 „ ,. . ■ c c ,
70-80
opinion polls being conducted in the -— to predict the upsurge in favour of the
,,7, ,, , ,. . HT-AC Nielsen
HT-AC Nielsen
300 146 300
95 146 95 „ . ,
country. What
Pioneer-added
RDI 313-318 to
140-150 this
84-86 growing Pioneer
demand wasActual
the result advent
296 134 113 of various "Act
television channels. With a large number The seat pr
of news channels competing against conducted
Table 4: General Elections 2009 - Predictions b
each other, the race for conducting During
Failed to See the Congress Upsurge the L
Seat Forecast BJP Allies
election surveys and airing them as in Congress 2004, Allies Others
all t
STAR News-A C Nielsen 197 199 136
quickly as possible after election dates houses pred
CNN-IBN
CNN-IBN 165-185 185-205 165-195
are announced has become the NDTV norm cratic 177 216
Allian
150
of the day. Janata Party (bjp) would be able to retain 180 191 172
Headlines Today
Headlines Today
While the election polls are of different power at the centre. The only difference
NewsX 199 191 152
kinds, it is the pre-poll and the exit poll between different polls
Times was
Now that while
183 198 162
which catches the attention of most some suggested that theActualnda would come
result 159 262 79
people. The reason is simple - people are back with an increased tally, others pre- The seat predictions were made on o
. _ 1 1 . 1 . 11. . j.j 1 c ^ c ^ 1 conducted during the various phases of the elections.
eager to know which party or alliance is dieted a loss of some seats for th
likely to win the elections, and how many alliance. Among the polling age
seats they will win. Exit polls became pundits there was complete unanim
very popular in 1996, when Doordar- in their predictions that nda
shan, the government-owned television win the elections. But the result
channel, commissioned an all-India exit shocker - the nda lost the election
poll. The fieldwork and data collection the United Progressive Alliance (u
for this poll was done by the team at
CSDS, and its findingsTable
were3: General reported
Elections 2004 - UPA Completely
Table3:General Elections2004-UPAC
j ,. j • r- , Underestimated
Underestimated , , , , , ,
and discussed
SeatForecast
Seat Forecast in
BJP Allies aCongressfive-hour
Others progra
aired live on Doordarshan. Since then, Ate voter
Allies
concerns and preference, are the
NDTV-lndian ExpressExpress
230-250 190-205 100-120
there has been no election in India NDiv-indian 230-250 190-205100-120 best means of deriving an accurate
Aaj Tak-ORG Marg 248 190 105
where exit poll results have not been Aaj Tak-ORG Marg 248 190 105 of the voting intentions and polit
Zee-Taleem 249 176 117
televised the day polling gets over. Zee-Taleem 249 176_ 117_ attitudes of the Indian electorate (B
Star-C-Voter 263-275 174-186 86-98
Lahiri and Roy 1995), opinion polls that
A Mixed Record Sahara-DRS 263-278 171-181 92-102
are used to predict poll outcomes are, in
Outlook-MDRA
Otrf/oo/r-MDRA 280-29 159-169 89-99
The history of opinion polls and seat Wf/00*-IVIUKA —49—159-169 B9'99 Indian conditions, fallible. Ho
Actual
Actualresult result 189
189 222
222 132
predictions during the last four general ^,ctualtresu!t . ———222, . ^2 exit polls and post-poll surveys3
" ° The seat predictions were made on opinion polls/exit polls r r r j
elections has been a mixed bag of conducted during the various phases of
timing; they gauge the voting intention with 70% of its population living in sometimes result in the failu
of voters after they have already voted. rural areas. That means any election and any data analysis
Psephologist Yogendra Yadav says that survey in order to be representative comes fallible and question
in addition to timing, other crucial factors, should try to conduct about 70% of its
including the sample size, sample design interviews in the villages and rural Factors behind Inacc
and the representativeness of the sam- areas and only about 30% interviews in Predictions
pie, ensure accuracy (Nath et al 1999). the towns and cities. The reasons for election pred
The choice of the survey method, sam- But this is not followed by most of wrong can be analysed at fo
pling method and the level of training of the market research agencies that are First, there is th
enumerators who collect data determine engaged in election polling in India. The diversity and volatility o
the accuracy of the survey conducted. reasons for this anomaly can be attributed Do voters form their vo
Election surveys, whether pre-poll or to inconvenience and high costs of con- after the elections are d
post-poll, are based on a random sample ducting surveys in rural areas. Most of there a significant number
drawn from the voter list of the Election the polls suffer from an urban bias, which are floating voters and m
Commission of India and are generally ultimately results in their sample being sions at the last moment? D
accurate as they yield a representative unrepresentative. A sample with more voters reveal their voting
sample ruling out coverage errors and interviews in towns and cities is most the surveyors or do they c
minimising sampling errors. On the other likely to have more educated, rich and arising out of extraneous
hand, exit polls, done on the day of elec- middle-class respondents. Thus the whole Second, can the surv
tions, are based on quota sampling and sample becomes biased and skewed, complexities of elections arisi
are fraught with the risk of some sections/ making it completely unrepresentative multipolarity of contests, p
subsamples of the population being left and rendering the survey inaccurate. and transfer of votes to e
out completely. Thus, the sample frame The third important component of an geographical concentratio
of exit polls is in most cases unrepre- election survey, crucial for conducting it some parties in some re
sentative and suffers from both coverage accurately, is the training of enumera- Similarly, Indian election
and sampling errors. tors who carry out field investigations, tionalism in parties, rebel candidates
The adequate number of respondents Unlike in the United States and other and local-level settings that are difficult
that should be interviewed in an elec- European countries where an election to ascertain in a survey,
tion survey is statistically determined, survey is conducted telephonically, in Third, is the sample selection repre
and the size of the sample does not to a India an election survey is conducted by sentative of the demographic of the vot
great extent determine its quality and enumerators who contact the respond- ers and does the size of the sample de
credibility. The election surveys (as against ents and conduct interviews in person, termine the accuracy of the survey?
pre-poll opinion surveys) in India do not So for every opinion poll or election sur- Fourth, how accurate are the various
go wrong as most of the polls have big vey, training of enumerators a few days statistical models for seat predictions
samples. However, even if the sample before field investigations is imperative developed by pollsters?
size is large it has been observed that the so as to ensure standardisation.
choice of sampling methods has been However except for a few academic Sociocultural Diversity and Volatility:
unscientific and unable to statistically institutions, most of the market research Election studies conducted in the past
cover the universe of the study. As a organisations do not spend time on and have indicated that Indian voters are
result, the surveys are not représenta- invest financial resources in fieldwork highly heterogeneous, with different socio
tive and the seat predictions have gone training and practices. Instead they pick cultural practices and demographic
wrong on several occasions. one-time trained enumerators from backgrounds, and their voting patterns
Random sampling methods which are their pool and ask them to do the field- and preferences are varied. But at times,
popularly used for pre-poll and post-poll work for subsequent rounds of election the multiple identities of Indian voters -
election surveys have a greater chance surveys. But there are certain aspects of region, caste community, language and
of getting a representative sample as training - rapport building with sampled religion - overlap, making it difficult to
compared to a sample drawn purposively. respondents, reading out questions from ascertain their political affiliation and
Purposive or quota sampling of voters is interview schedules and using survey electoral choices,
popularly used by market research agen- instruments, following accepted fieldwork For example, Muslims in a state like
cies for exit polls where a quota is fixed procedures and practices, standardised Uttar Pradesh (up) do not vote for the
for sampling the respondents based on methods of asking questions and record- bjp as reported by election surveys con
gender, education, caste, communities, ing responses, and do's and dont's of ducted in the state in the last decade,
different age groups, occupational back- surveys - that need to be done afresh for However, there are variations in voting
ground and class. each round of an election survey. The patterns of the Muslims residing in dif
What accounts for such unrepresenta- absence of rigorous training leads to ferent regions of the state. Muslims in up
tive sample selections? India is a country inaccuracies in data collection that do not form a homogeneous group and
Economic & Political weekly GSd may 3, 2014 vol xlix no 18 15
there are differences among them based two main political parties, provided the data one requires for a
on region, language and religious sect, surveys get their vote shares correctly, as the sample size
which are reflected in their voting However, in states where there are tive, seat predictions b
preferences. Thus, a large sample survey multipolar contests, making predictions small sample size can c
would capture that certain sections of become difficult as a slight error in esti- curate results. On the con
Muslims in the state have voted for the mating the vote share for any one party based on a large but u
bjp. Similarly, the voting patterns of can completely upset the accuracy of size will yield a wrong
voters in India vary from state to state, the seat prediction. Yogendra Yadav has argued that there is
For example, certain issues in elections Parties that contest elections in alii- no guarantee that a bigger sample size
may have regional and sectional appeal ance also pose a problem in seat predic- will get you the right result. Bigger
and can form the basis of voting deci- tion, as it is very difficult to compute surveys only multiply errors 10 times
sions. For others, the issues might not how the votes are shared between one (Nagaraj 2008)
have any appeal and may not affect another in an alliance. The geographi- The method of sampling used in an
their voting decisions. cal concentration of votes for some election survey and its accuracy also
Pre-poll surveys might make seat pro- parties in some regions of the states also plays an important role in makin
jections that are off the mark. Election makes it difficult to make a correct seat reasonably accurate election forec
studies, done before or during the elec- prediction in spite of getting the vote Thus a smaller representative sample c
tion campaign, have revealed that estimates correctly. The result of the help make an accurate predicatio
around one-fourth of the voters do not Karnataka Assembly election held in compared to a bigger unrepresentative
decide beforehand about who they are 2008 revealed that bjp had a 1% vote sample. The method of sampling used in
going to vote for and they have been share less than Congress but it went on surveys done by market research org
termed as floating voters by pollsters, to capture the majority of seats. The sations is usually not the most scientif
The floating voters make up their mind Congress got more votes than the bjp in method of survey research,
during election campaigns, and if there the state, but its vote was evenly distri- This becomes clear when measurin
is a bandwagon effect in favour of any buted throughout the state which led to voting preferences and intentions
particular political party, the initial pre- huge losses in winning seats. castes and communities. For example
dictions can seriously go wrong. up, a majority of the dalits, especially the
Though pollsters while making election Method of Sampling and Size of Jatavas, hav
forecasts based on pre-poll surveys put in Sample: The most significant feature in while the m
a rider that the predictions can change in the preparation of an opinion poll sur- voters have b
subsequent weeks, its readers remember vey is the sample size and method of Congress. Sim
only the predictions and accordingly sampling. The sample size for any na- to the Yad
judge them after the actual outcome, tional-and state-level election study de- porters of th
During the up assembly election of 2007, pends upon the level of analysis one in- sample in u
a post-poll survey revealed that a large tends to do. Thus if one wants to analyse these cast
number of voters from the upper castes the voting behaviour and attitudes of mating the p
who actually voted for the Bahujan voters only at the state level, a survey of these commu
Samaj Party (bsp), a dalit party, did not of 1,500 respondents would be good So if a sa
reveal their voting behaviour due to fear enough. But if one also wants to do a the opinion
of caste-based violence. As a result most region-wise analysis in the state, then community,
of the election surveys conducted during the sample size should be bigger as there be highly v
that election failed to predict a clear-cut should be sufficient number of cases for Yadav (2008
win for the bsp led by Mayawati. disaggregate analysis. through which the sample is selected
Similarly, exit polls, in which the vot- Thus sample sizes for any election sur- is crucial for the survey. Most Indian
ers are interviewed outside the polling vey depend upon the level of disaggregate polls go wrong because their sampling
booth after they have voted, do not al
ways reveal voters' actual decisions.
Web Exclusives
This happens because there can be a
fear that a correct revelation of the party The Web Exclusives section on the journal's website (http://www.epw.in)
they voted for could be used by other po features articles written for the web edition. These articles are usually on
litical parties in subsequently identify current affairs and will be short pieces offering a first comment.
ing and targeting them. The articles will normally not appear in the print edition.
All visitors to the website can read these short articles. Readers of the print edition
Multipolar Contests, Alliances, etc: It are invited to visit the Web Exclusives section which will see new articles
is easier to forecast the election out published every week.
come in a state where there are only
16 may 3, 2014 vol XLix no 18 QBS9 Economic & Political weekly
India's 'Dutch Disease' and would reduce the level of its quantitative
easing fin other words, the open market