Jyoti Rahman
About a quarter century ago, when I first started following Bangla politics, I used to wonder how various Prothom Alo columnists, or TV talk show guests, spoke so authoritatively and confidently about what the people wanted.
How did they know that Sheikh Hasina was wildly popular (or hated) because the public was happy that the country was finally returning to the spirit of 1971 (or that they were livid about the way she was selling the country to India)? How did they know that Khaleda Zia was doomed (or poised to emerge as a political genius) because she had joined forces with the hated killers of 1971 (or the nationalist forces were united)? How can all these diametrically opposed views be simultaneously correct? Where were the polls that supported these views?
The naive and young me soon learned that there were no polls. These pundits basically sprouted their views, asserting that their prejudices reflected the public opinion!
Fast forward to our time, and we have social media superstars doing exactly the same thing — everyone has their views, and everyone knows that their views reflect the public pulse, never mind that there is no evidence!
In fact, the problem now is far worse. Back before the 2001 election, we had a decade of electoral democracy — two free and fair national elections, plus many local government ones and bye elections, to gauge at least the relative strengths of major parties around the country. In contrast, we haven’t had any meaningful election since 2013. Back then, parties used to test their strength with large rallies, meetings, and hartals. Of course, Hasina ended not just elections but also politics as we knew it.
We are beginning afresh, but we still have the same information problem — how do we know what the people want? The biggest test of public will is, of course, election! But right now, we have a debate about when the election should be held. How do we know that people are happy with the interim administration and don’t want election until 2029?
And even when we have an elected government, how will we know whether their policies are popular or not? Clearly we can’t have an election every time there is a contentious issue!
This is why polls are an important part of functioning democracy. They tell us not just which party or candidate is likely to win at a particular point in time, but also what people think and how they feel about issues over time.
Polls, in turn, allow political parties to recalibrate their policies, positions, and platforms. It is common in the west for major political parties to change even their leaders ahead of elections because of bad polls — Joe Biden and LBJ declined to run for re-election, for example!
Of course, for polls to be credible and trusted, they need to be run by professionals. And thanks to the Hasina despotism, trust and professionalism are both in short supply in Bangladesh. Any poll is likely to be greeted by politicians, or random folks in social media, with cries of ‘whose agenda is being served’!
Brave indeed would be the professionals who would want to venture into this field.
It is against that backdrop that Innovision — a management consulting firm in the development sector — conducted a detailed poll and released the results yesterday. The detailed methodology and results are here.
The headline results have been widely reported in the social media. I will provide my analysis later. For now, let me note my involvement, and dispell some wild speculation (willfully malicious or just out ignorance) doing the round in the social media.
First, the poll is an Innovision product. Their executives are professionals who can be approached directly. There Managing Director, Md Rubaiyath Sarwar, is pictured below and is a very amicable fellow. He would be happy to engage on the substance.Second, as a data nerd and a political junkie, I have seen the poll develop from the concept stage to execution. It has been a thrilling ride. But I have no financial or pecuniary involvement with Innovision. I am not making money from this. I doubt Innovision is either, but you can ask them.
Third, Rubaiyath and I are both in the Bangladesh Research Analysis and Information Network (BRAIN), which has supported Innovision with the dissemination of the poll results. However, BRAIN has not conducted the poll. Indeed, it is simply beyond the organizational ability of BRAIN to do anything remotely like this. Hosting an iftar or seminar stretches us.
Fourth, neither is the poll carrying water for any political party, nor is BRAIN a front for one! There are people affiliated with both BNP and NCP in BRAIN, but most members are not even in politics. I have seen wild speculations in social media that BRAIN is a BNP front who ran this poll to make NCP look bad. I have seen wilder ones that the whole thing is Jamaat’s effort to make BNP look bad!
Such speculations might have been funny, but is in fact ironic. And that leads to my fifth point. Our political discourse needs to rise above such petulance and recognize professionalism. Reform is not just something wise men write in the constitution. It also involves a collective effort to see facts for what they are.
There are two further, separate but linked, issues that needs to be addressed. I am sure Innovision folks would do it, but one can never have too much education on these things.
First is about the sample size. A common refrain I have seen is that “in a nation of 180 million, 10,000 is too small a sample size and therefore the poll is meaningless”. People who say this with confidence perhaps think they sound smart, but are really displaying their ignorance about basic statistics.
Let me illustrate.
Suppose the true support for Amar Bangladesh Party among the 180 million people (or 120 million voters) is 5 percent. Obviously, if we were to ask the whole population, we would get this 5 percent. Equally obviously, this would be a huge undertaking that is unlikely to happen. The question is, roughly how large does the sample need to be before we can be confident that the estimate we are getting is the true representation?
One way we can answer this question is to observe the measure of interest (support for ABP in this hypothetical) as we increase the sample size. If our sampling method is reasonably good, as the sample size increases, the measure of interest should stabilize around a figure. Once that figure is reached, increasing the sample size usually does not add any further information.
Anyone with first year statistics would get this from the Law of Large Numbers. Let me just turn over to google:
The law of large numbers is a fundamental concept in probability theory. It states that:
The average of the results obtained from a large number of independent random samples converges to the true value, if it exists.
As the number of identically distributed, randomly generated variables increases, their sample mean approaches their theoretical mean.
The average is closer to the expected or theoretical value as the number of trials or observations increases.
As it happens, the numbers that were reported yesterday had remained roughly similar after the first 1200 or so responses came in. If nothing else, this fact alone makes me reasonably confident that the Innovision team did a reasonably good job! Incidentally, well reputed western polls tend to have sample sizes of 900-1200.
A flipside of this, however, is that the national averages reported in this poll cannot readily be translated into granular, district or constituency level details. Let me illustrate again.
‘Support for AB Party’ is not the same thing as support for ‘AB Party candidate in Dhaka-17 constituency’. The Law of Large Number implies that in a well designed poll with a sample of 1200 from people from across the country would do a good job of estimating the true support of AB party nationally. If now choose the few dozen people from that sample of 1200 who live in Dhaka-17, we would not get a good estimate of the support for the AB party candidate in Dhaka-17! To find out that candidate’s support, we would need to survey 1200 or so people in Dhaka-17.
This means that we should not take this survey to draw any firm conclusions about individual candidates’ chances in any particular seat or any party’s chances in any particular district!
We can be very confident that two-fifths of voters have not made up their mind yet. We can be very confident that of the remaining three-fifths, one-third is not willing to disclose their choice. We can be very confident that of the people who reveal their choice, BNP has over 40% support. We can be reasonably confident that this 40% is widely distributed across the country. We can speculate, with reasonable confidence, that if BNP were to receive 40% votes across the country in an election, then our first past the post electoral system would confer it a landslide.
But we have no way of knowing from this poll which seats BNP will win!
Of course, this is just one poll, and the results are an estimate of the snapshot of things as of mid-February 2025. Much will change between now and the election. No individual poll is a prediction of election outcome, particularly when the election has not even been called yet!
Nonetheless, the poll gives valuable information about the state of politics. Our political discourse will be much improved if we analyzed the information instead of bloviating about small sample size or speculating about the pollsters’ agenda. And our politics will improve if the parties and politicians took polls like these seriously.
Let a thousand polls bloom in our spring of democracy!
Originally Published: Mukti