Re:... posted byRilwan (TX)[21 April 2014] IP: 198.*.*.* United States | Comment Reference Number: 34517
This is an excellent article. You have correctly questioned the validity of the opinion polls and their scientific validations behind it.
There are various ground factors to be established for any statistical analysis to be valid. Otherwise the statistical measurement could be intentionally or unintentionally biased and skewed.
a) Population sampling size – The theory of large sample size suggests that as much small is the sampling size the margin of error is very high and the margin of confidence will be low (or the error ratio to be low). It is unclear how JuVi kind of magazines just take a sample size of 100 and assumes that is good enough to represent 1.3 million strong population in a constituency.
b) Control grouping – This tend to group generally homogeneous population. For e.g. Muslims, Hindus. Among those, could Shias and Sunnis. Among those other subsequent demographic categories etc. This grouping determines the group characteristics and sizing. Sometimes this is derived through targeted questions and how the questions themselves have been designed.
This leads to the question – what were the questions asked and how these questions were designed. Where the questions misleading or broadly generalist. One group may assume a question differently than the other group and potentially provide answers unintentionally wrong – may be quite rightly to say, mislead to answer wrongly.
c) Population dispersion – How do they fix the sample size among different control groups? Is that quiet representative? How these folks acquire data from rural India? Did they visit the “internally displaced” refugee camps in Chhattisgarh, Orissa etc.? Did they question adhivasis?
d) None of them publish margin of error. Typical margin of error is 3% - which would eventually give a swing in the observation. It was laughable with Thanthi TV presented like 72% back Modi without providing this error margin. There is no 100% error free (1.0 Confidence) statistical result. Which would typically require the analysis to be subjected to error analysis and most often statisticians assume the result is “biased”.
e) Relevance of data – Simply taking opinion polls from big city doesn’t represent entire India; however, it does matter that who votes in large number than who may not. In online opinion polls people like me vote without any relevance to the actual population who tend to vote. Such would typically fall under the margin of error. Most statistically analysis the accuracy is reliable with 3-5% of margin of error – however, this margin could vary – could be set even as high as 20% in some cases and even 40% in some cases – depending on the usefulness of the analysis.
f) Data quality – to have opinion poll like that of the size of Indian population, it would take months of preparation, months of execution and months of data analysis and validation. It would require the data to be validated for key criteria. While the government runs the election for couple of months, how does these folks where able to release opinion poll every month or every week? Definitely you don’t need to ask every citizen; even not quarter of the citizens, but significantly representative size is required for 1.3 billion large population.
Without producing any such – when the media presents the public with opinion poll, it does poses question on the integrity.
It is also that it makes us to think how much India has to travel to become a true democracy – when criminals go to parliaments and people like Irom Sharmila is refused rights to vote. How many of us know who Irom Sharmila is or do we really care?
Would we really care about the world when DCW expand their poisonous capabilities?