Don’t be taken in by opinion polls used to distort political reality

The recent opinion poll findings released by Ipsos Kenya raised substantial ire and concern among the public. Barrack Muluka, my former student, friend and colleague, an incisive and erudite analyst and critic of public affairs, in an OpEd piece in the columns of The Standard on Saturday, tore that opinion poll to shreds. Few would have faulted Barrack’s arguments, and what I can do is simply amplify on certain points the public has continued to be puzzled about. Foremost in our minds are the following questions. What are opinion polls? What use do they serve? Are they the same thing as market surveys or are they different?

In many ways, opinion polls and market surveys are not really different: they both have to do with trying to determine, understand predict and know human behaviour objectively and, dare I say, scientifically. They both use statistical tools for predicting behaviour.

The first opinion poll company I ever knew about which also engaged heavily in market survey was Marco Survey in the 1960s. My friends Wilson Ndolo Ayah and Shem Migot Adholla both worked for Marco Survey at one time or the other. If I remember well, in the famous Gem by-election of 1969 between the candidate of the Kenya People’s Union Wasonga Sijeyo and the Kanu candidate Rading’ Omolo, Marco did a survey which showed that KPU would win by a landslide. Kanu was not amused while KPU was elated. But even common sense would have shown that Kanu stood little chance of winning a seat in a constituency predominantly supportive of KPU, and following the mysterious death of CMG Argwings Kodhek, the MP whose demise occasioned the by-election. So why was the survey necessary in the first place?

It could well be that someone, interested in the outcome of the election, paid Marco Survey to undertake the poll. That someone could have been leaders of Kanu or those of KPU: there is absolutely no reason to guess one way or the other. It could also be that Marco itself wanted to know how the voting would turn out, and the likely events that would follow the outcome of that event.

Hence, opinion polls carried out to predict, determine or know likely political behaviour can themselves be very dicey since their outcome will always touch on the raw nerves of competitors in the political landscape. When they emphasise the obvious those who are aggrieved by so doing may feel the survey seems to deny them a fighting chance even if being victors is a remote possibility. In the case of the Gem by-election Kanu fell in this category, and blaming Marco for saying the obvious might have been justified at the level of emotions but not at the level of confronting objective reality.

Opinion and market surveys need, therefore, to employ methods which are objective, impersonal, independent of individual biases and aimed at revealing useful knowledge on which further human action can be based with productive or useful results.

For example, when one introduces a new soap product in the market, it is important to try a few samples before one offloads tones on would be consumers. Once the consumers use the samples which are scientifically distributed in the market, then one can take a survey—also scientifically done—to determine how well the soap is received.

The problem we have here is with the words “scientific” and “objective”. In the world of social science, a scientific and objective opinion poll is one that uses statistics within a methodological framework that can produce sound and objective results. Since it is impossible to interview 42 million Kenyans when one wants to know whether they prefer Omo to Surf as a detergent, one can ask a sample that is representative of the whole population. This sample can be taken at random or from specific populations, categories or classes. When one departs from random sampling in order to use stratified sampling, one has to be very careful regarding making general statements about a particular category, especially when, within that category, there are factors or variables other than those unique to that category which influence a particular behaviour.

It is most likely that Ipsos took a random sample from the general population and then stratified it regionally on the assumption that regions coincide with party identity.

We live at a time when most Kenyan households are experiencing enormous economic problems: problems of sheer survival. We live at a time when the security threat is a menace to every Kenyan, and corruption looms large in every public institution. Listening to FM radio programmes one gets the feeling that the public is not very comfortable with the way the government is handling public affairs. Yet, according to Tom Wolf and his ilk, there was never a time in Kenyan recent history when the “common man” was more elated about government performance than now. There was never a time when the ordinary Kenyan was more optimistic about the future than now. In fact 75 per cent approve of the manner in which the President is handling public affairs.

Darrell Huff, as early as 1954, warned us about the possibility of social scientists telling lies with statistics in his book How to Lie with Statistics. This book has been reprinted many times since then. Written by a journalist, it warns us about blind faith in numbers and, well before Mutahi Ngunyi was born, warns: “there is terror in numbers.” The British legendary Prime Minister, Benjamin Disraeli, also cautioned about two equally dangerous type of lies: damned lies and statistical lies. Statistical lies in opinion surveys or research usually emanate from “samples with built-in bias”. In other words, samples chosen with the sole mission of “arriving at the results we want” and not the reality “out there” which we are trying to research about.

We must, of course, respect the statistician when he or she follows the trodden path of scientific research diligently, painstakingly and objectively. When he or she clinically seeks to go beyond appearances into the world of reality remembering, of course, Karl Marx’s famous dictum in the “Eleventh Thesis on Feuerbach”: if appearances coincided with reality science would be superfluous. My caution here is not to throw statistics to the wind but to be careful not to take statistics at its face value.

Sometimes there are just too many variables that explain human behaviour: isolating just a few and creating correlations that are spurious in explaining how certain phenomena happen can be very dangerous.