Pew Survey on India (religion, etc.)

The Pew Survey on India is long. Here are my main impressions/surprises:

– The strong emphasis on Hindi in the Gangetic plain is pretty striking (“Central”)

– The stuff about mass conversions to Christianity in South India seems overdone. I understand people can/will lie on Census due to reservations, but this seems less plausible for a pollster. That being said, a disproportionate number of conversions are in the South

– South India is less religious, less nationalistic, etc. This seems to apply across religions (that is, Muslims in the South are less focused on religion just like Hindus in the South are)

– Opposition to “inter-caste” marriage is very strong. And, it is strong among non-Hindu groups too

– All Indians seem rather nationalistic

33 Replies to “Pew Survey on India (religion, etc.)”

  1. Reproducing part of my earlier comment from the general thread where Saurav posted this report from The Print.

    …The report itself –

    1. Went to 29000 respondents, this is less than population of Karol Bagh. Has very weak statistical power with respect to the nationwide claims.

    2. Of the 7 people in the methods and research team, only 1 Indian. Read this correctly!! the people who framed the questionnaire are Americans mostly.

    More comments on the methodology –

    Which region has the most respondents? The “North” at 24% of total respondents – consisting of Punjab, Rajasthan, Haryana, Delhi and JK only.

    The respondent distribution is heavily skewed – Only 7% of the respondents were from the “Central” region – consisting of UP and MP. Only 19% of the respondents were from the South which consists of the 4 states. Basically 26% of respondents from the regions of India that host close to 70% of the population.

    Basically a survey that went to an underwhelming section of “North” states (clear and blatant case of mislabelling) which are today primarily ruled by Congress (Rajasthan, Punjab).

    TLDR – these Pewsters were shopping for a specific result

    Why they wanted this specific result? – thats a discussion, Saurav and me did on the general thread.

    1. For mathematically challenged people who have suddenly found “evidence” of their beliefs, I would like to break it down into a “picture storybook for kids”. Should be simple and easy to understand –

      From the states of Karnataka, Tamilnadu, Andhra, Telangana, Kerala, Uttar Pradesh, Madhya Pradesh and Chattisgarh – the Pewsters interviewed 7500 people. The population of these states combined is close to the population of US + Schengen Area.

      For the most recent bypoll election of Maski assembly constituency in Karnataka having 200K voters, KA pollsters went to 3000 people to survey outcome. They still got it wrong 🙂

      Just terrible and amateur tradecraft – don’t even get me started on the questionnaire!

      1. @Ugra Sorry, the argument you are making is plain wrong. If you have a random sample from a large population and you want to estimate the fraction of the population that has a given property like faith in God or R1a haplogroup, the absolute size of the sample is the only thing that matters, not the relative size with respect to the population. A random sample of 7500 people is more than enough to estimate the population fraction with 2 percent margin of error and 99 percent confidence, irrespective of the population size and diversity. The reason is that the sample mean follows a Gaussian distribution and the standard deviation depends only on the sample size.

        1. @Ace of spades

          It’s not only the size but also the deliberate skew which they have introduced.

          They ensured that the smaller states of Rajasthan, Punjab, Haryana and Delhi have the same number of respondents (7500) as the 7 other largest states of India (UP, MP, undivided AP, TN, KA).

          Come on – statistical design cannot be that random! Definite shopping.

          They asked very simple questions in the questionnaire – read it. The conclusions are not supported by the linkage to the individual questions.

        2. @Ace of Spades

          It was not random sampling. Martha McRoy from the Pew team makes that clear. They adopted a model where the probability of all religious adults (any religion) was maximised. So the model ended up downgrading the chances of a Hindu or a Muslim from the more populous states being picked.

          That ties in with the observation I made on the weird regional distribution of respondents.

          1. I was only objecting to your claim that a sample of 7500 people is not large enough. I agree that there are several things in the survey that look fishy. For example, it says there are many Indian Muslims who think offering Namaz is a necessary condition for being a Muslim but believing in God is not. This makes no sense.

          2. @Ace of Spades

            I understood when you talked about Gaussian distribution. That only works when you have discrete, countable outcomes.
            For example, haplogroup distribution assessment – there are only a discrete number of haplogroups possible. Or faith in Jesus – Yes/no. Or the winning candidate in an election – there can be only so many candidates.

            For this survey – the Americans framed questions with “end-points” that they assumed to be correlated negatively. For example, there is a question whether someone “prays” regularly. This is Bible-belt territory that makes complete sense to an American. Saying no to the question implied lack of faith in the view of the Americans. Berkson’s paradox!!

            For favorite Gods, they lined up photos of 15 Gods/Goddesses!! There were only three South Indian controls there – Murugan, Meenakshi and Ayyappa. Where were Bathukamma, Ayyanar, Amman, Siddappa and a million other divinities?? Lack of a favorite deity from those 15 implied lower faithfulness – again American tone-deafness.

            So the Gaussian framework wrt mean fails when you do not have the full set of answers ready for a pick by the repondent.

          3. If they weighted by regional distribution, then the geographical skew shouldn’t be a problem. These are smart statisticians, they know how to correct for these things.

        3. @Ace of spades, re: acceptability of sample size.

          The confidence interval estimate is assuming that the population is not stratified. The population size on the whole doesn’t matter but diversity does for selecting sample size.
          When you have stratified population, both mean and it’s confidence interval estimates can be wrong when estimated assuming a Gaussian distribution of mean error.

          I encourage people (who have basic scripting skills) to try this out with random sampling by creating pseudo populations themselves with variable number of distinct clusters with different sub-population sizes.
          If not so number oriented, read up on differences between stratified sampling and simple random sampling on Wikipedia.

          1. I think you may have misunderstood my comment. It was a response to Ugra’s argument, so I was specifically looking at the situation where the population is extremely large and one has zero knowledge about the way it is stratified. My point was even in this case if you are allowed to pick 7500 sample points you can get a very high quality estimate (2 percent margin of error with p-value less than .01) just by using simple random sampling.

    2. First off, the accuracy of a sample estimate is irrespective of the population size. Other things being equal, a sample of 1000 people is going to be equally informative about a population of 1 million as it is for a population of 1 billion. Now, obviously, in this case other things are not equal: in contrast with China, India is not just a big country but a very diverse one. That’s why they increased their total sample size to 30k, so they could get decent cell sizes for very small groups like Jains, Sikhs and Buddhists.

      It’s true that their sampling was wildly geographically unrepresentative (deliberately so, probably in order to get adequate representation for Sikhs and Jains) but I assume they weighted by census demographic data, so that would have taken care of that problem.

      Incidentally, they didn’t really underrepresent the South, the south is about 20% of India’s total population. They massively underrepresented the Centre, but the weighting issue should take care of that.

      1. “They massively underrepresented the Centre”

        That was my first thought, when they divided central and North zones (or else North would have been even more hegemonic) , because essentially they are the same. But even with under-representation the results and their inferences are not out of the ball park.

      2. @Hector St Clare

        …First off, the accuracy of a sample estimate is irrespective of the population size….

        This is true if you are measuring the blood group or eye colour or height distribution in a population. Where there is no convexity between the questionnaire and the conclusions.
        Take a look at the questionnaire that Pew used and look at their conclusions – complete convexity. This is Statistics 101.

        The second major fault is that they state normatively that this is a “Pew Study on Religion”. But actually it is a study of how politics and religion impact each other. Now the statistical problem with this is that – when you measure the occurrence of a property in the population, you assume that it is normally distributed. Religion could be normally distributed, but political affiliation is not.

        So what they did was to use branching – they asked someoneś religion and then the questionnaire forked to have a different set of questions for Hindus, Muslims etc. What they did not do is to ask someone’s political affiliation and then branch the questionnaire into religion. Therefore we have lost the ability to measure covariance.

        Real time political data shows very clearly that the Gangetic Belt societies (UP, Bihar) predominantly vote on the basis of their caste identity and not on the basis of a overarching Hindu identity. There has never been any common societal identity (religious, cultural or political) in the BIMARU zone. A reason for its backwardness!!

        This study suggests the opposite!! Nothing better could illustrate the weakness of its statistical model (branching decision).

        1. Just because some regions in india have no/less Hindu identity, does not mean the north has no Hindu identity.

  2. The stuff about mass conversions to Christianity in South India seems overdone. I understand people can/will lie on Census due to reservations, but this seems less plausible for a pollster. That being said, a disproportionate number of conversions are in the South ✔️

    – South India is less religious, less nationalistic, etc. This seems to apply across religions (that is, Muslims in the South are less focused on religion just like Hindus in the South are) ✔️

    As i have said on the blog umpteen number of times…

  3. If you lok at the data Hindus actually look quite accepting at 60 something percent wont inter marry whilst Muslims are 80% when asked if they are against inter marrying.

    1. Everyone knows it. Everyone who grew up in India atleast.

      “The social evils which characterize the Hindu Society, have been well known. The publication of Mother India by Miss Mayo gave these evils the widest publicity. But while Mother India served the purpose of exposing the evils and calling their authors at the bar of the world to answer for their sins, it created the unfortunate impression throughout the world that while the Hindus were grovelling in the mud of these social evils and were conservative, the Muslims in India were free from them, and as compared to the Hindus, were a progressive people. That such an impression should prevail, is surprising to those who know the Muslim Society in India at close quarters.

      One may well ask if there is any social evil which is found among the Hindus and is not found among the Muslims.”

      ~ Dr. B. R. Ambedkar

  4. It’s a good report. Doesn’t give much for policy prescriptions but overall it’s better than the crap written in editorials by stupid indian Hindus and muslims in Indian newspapers.
    It also removes the perception about the “Central Belt” of being anti muslims while South was hailed as Hail Mary of Unity. Also something missed by Muslim apologists is that “General population” (Hindus, muslims all included) say Hindus are more discriminated than Muslims.
    Some Hindu nationalists will find solace while some may be triggered because Rama got lovers only.

  5. It is weird to me that the strongest self-reported discrimination against Dalits is in the South and North East ?

    Anyone know why that is ?

    1. A hypothesis is in North India , dalits are more ‘Hindu-ized’ and have party of their own, so avenues of channeling their rage.

      In South there is neither, so more anger. Better economic standards of dalits in South means more clash with OBCs and UCs. Less Hindu-ization, more radical ideolgies like Ambedkarism, Communism and Dravidian-ism

      In North East, blantant racism faced by tea tribes who are dalits and adivasis from central India. By the bhumiputras Ahoms and their cousins.

    2. @Sumit

      Read the questionnaire. The conclusions have excess convexity with respect to the simple questions asked.

    1. 5% of Muslims explicitly believe in “more than one god”, which doesn’t seem like a lot, but if you consider that belief in multiple gods is *the worst sin* in Islam, it really is a lot.

      77% of Muslims believe in “karma” as well, although I’m not sure why they differentiated karma vs. reincarnation, doesn’t one imply the other?

      1. Need not be the same. Some of my Hindu family members believe in Karma being just the do-good-things-good-happens-to-u part and not necessarily the accruing karma to have a better reincarnated life.

        1. I can sort of see that, but that vision seems a lot “weaker”. Reincarnation, whether you believe it or not, at least provides a coherent, systematic mechanism for how merit and demerit are accumulated or worked off over multiple lifetimes. My spiritual/religious views are in flux at the moment, but the older I get, the more merit I see to the idea of reincarnation.

          1. Probably in the South and in more traditional settings, reincarnation is given more value. In the North i feel there is this idea of here and now. They see doing good or charity affecting them directly in this life time.

            My parents for example visited all the 4 dhams believing that they will have a good old life.

Comments are closed.