Why pollsters face challenge of getting predictions right

The latest opinion polls by Ipsos and Infotrak have exposed structural weaknesses and failures that could erode public confidence in key findings deceptively presented as statistical probability. While Ipsos Synovate has consistently placed President Uhuru Kenyatta ahead of Nasa’s Raila Odinga, Infotrak Research and Consulting would want the public to believe that the latter has galloped four percentage points in just three weeks and achieved a razor-thin edge margin over his rival. In the Ipsos poll, the President had 47 per cent against Mr Odinga’s 43 per cent. Infotrak had Mr Odinga with 47 per cent against Mr Kenyatta’s 46 per cent of the national vote.

NOT CLEAR

The basis of Mr Odinga’s huge gain is not clear and the findings bring closer home the controversy that has hit pollsters in America and Europe, who have been faulted for turning a scientific survey into a self-serving exercise to yield pre-determined outcomes. This is precisely the issue that confronted them when they released their findings on Sunday. Mr Tom Wolf, the Ipsos lead researcher, and Infotrak chief executive Angela Ambitho faced off on NTV, trying to explain how their findings could be so markedly different. Structural weaknesses of the Infotrak poll emerged from several key areas, which could be attributed to either the framework being faulty or the polling controlled to yield certain results. Infotrak sampled 2,000 respondents in 30 counties and 100 constituencies between July 16 and 22. Ipsos interviewed 2,209 respondents in all the 47 counties from July 3 to 12.

SEVENTEEN COUNTIES

Based on the sampling frame, it is apparent that while the opinion of those who responded to the Ipsos survey may be considered representative of the 19.6 million registered voters, Infotrak’s did not capture the views of Kenyans in 17 counties (36 per cent) and 190 constituencies (65.5 per cent!). It failed the test of a poll that could give a representative view of how Kenyans are likely to vote in the presidential race.

ALL RESPONDENTS

Another point of departure was whether the two surveys were randomised and who satisfied the criteria to be included. Mr Wolf explained that Ipsos accepted opinion from all respondents who said they were registered to vote, would most likely vote and named their polling station. Infotrak, according to Ms Ambitho, focused only on registered voters. The problem is that there is marked difference between eligible voters, over 26 million, and the 19.6 million registered by the IEBC. It wasn’t clear how Infotrak was able to determine that its respondents were only registered voters when Mr Wolf identified this as one of the methodological problems in the survey, given that the biometric voting system eliminated voting cards. Moreover, verifying each respondent against an IEBC register would be a daunting task. Also, while Ipsos has been using samples based on the classifications used by the Kenya National Bureau of Statistics in population and household surveys, the Infotrak survey fell way out of the rim by giving more weight to some of the categories of respondents than others. The release of Infotrak’s findings two hours after Ipsos' raised suspicions about the quality and intentions. Infotrak’s appeared timed to dilute the findings of the Ipsos poll and influence the presidential debate on Monday (which eventually featured Mr Odinga alone).

DAMAGE INTEGRITY

While opinion polls are important tools for gauging the feelings of the public before elections, confusing Kenyans into believing that the trend they have seen for many months has suddenly changed, without any credible explanation, damages the pollster’s integrity. With 12 days to go, the fact is that President Kenyatta and Mr Odinga have already established their core voters. Whoever swings most of the undecided voters to his side will carry the day, which both pollsters agree may produce a straight winner with 50 per cent + 1 of the votes cast.

Peter Warutere is a director of Mashariki Communications.

[email protected]