Now that the election is upon us, this, my 5th piece for The Elephant, will compare the final round of pre-election survey results released last week—by law no new opinion poll can be published within 5 days to the election—and consider several other issues related to them, individually and collectively.  It will then consider the findings of several dubious survey entities; some examples of weak and incorrect interpretation of poll results by the media; aspects of the Act regulating the publication of polls during the pre-election period; a consideration of several factors related to recent change of presidential voting intentions; and the assumed “bandwagon” effect as a consequence of polls.

To provide some empirical and analytical context, it begins with my (slightly updated) piece that appeared in the Standard of 3 August.

The undecideds: the overall picture 

By all accounts, the forthcoming presidential election is likely to be a close contest. Two factors underlie this assessment: first, not all of the mainstream survey firms agree in giving Raila Odinga a clear lead over William Ruto (beyond their respective margins of error), even if three do so (Infotrak, TIFA and Ipsos—in contrast to Radio Africa). Secondly, those respondents declining to identify their voting intentions either because they declared they were “undecided” or have chosen to remain silent altogether, remain a significant proportion of potential voters in the findings of nearly all the four surveys: between 4 and 8 per cent. In other words, it can be assumed that it is they who will decide who the next president is, whether they turn out to vote, or stay away, thereby leaving this decision to those who do vote.

Such figures raise two questions: which of these two candidates will obtain the most votes, and will either also obtain the 50 per cent + 1 required for an outright win?

As shown in TIFA’s poll conducted during 21 June-16 July, both candidates increased their overall rating by about 5 per cent (Ruto from 39 to 44 per cent, and Odinga from 42 to 47 per cent).  Two factors account for this: a decline in those claiming to be undecided (from 10 to 4 per cent) as well as those who gave no response (from 4 to 2 per cent). Combined with a drop by half in expressed voting intentions for George Wajackoyah (also from 4 to 2 per cent), this explains where the combined gain of 10 per cent for the two main candidates came from.

The remaining proportion who declined to declare their voting intention will be critical. Is there anything about their identity and attitudes also captured in this survey that might indicate which way they will go?

This analysis—based on previously unreleased TIFA survey data from the above-mentioned survey purchased by the Standard—explores two key questions related to the above: just who are these respondents, and does their identification give any clue as to whether and how they will vote? How many Kenyans have changed their voting intentions since the beginning of this year, and for those who have, why have they done so?

Getting inside the numbers

The first point to make is that the proportion of respondents declining to name a preferred presidential candidate has continued to decrease since the beginning of the year. For example, according to TIFA, it dropped by about half, from 30 per cent in January to just 14 per cent in late June. This stands in contrast to the relevant figures for the same period prior to the last three elections: in 2007, 1 per cent; in 2013, 5 per cent; and in 2017, 9 per cent.  (It should be noted that the surveys which yielded these figures from the three previous elections were all conducted face-to-face at respondents’ households, unlike the three at issue here. It may be the case that in the former setting, where interviewers and respondents are able to establish a more “personal” relationship, it would be more “awkward” for a respondent to avoid answering this question.)

As noted, the TIFA figure is 5.2 per cent (among all respondents who said they are registered voters, but excluding those who said that they will “definitely not vote just 2 per cent). Yet with only two weeks remaining when this survey was conducted, it cannot be assumed that all such respondents have not, in fact, made up their minds; they may just be too shy to reveal their voting intentions, for one reason or another.

How many Kenyans have changed their voting intentions since the beginning of this year, and for those who have, why have they done so?

One way to test this assumption is to pose this question: what proportion of these respondents who claim to be undecided about their presidential vote have, in fact, made up their minds about the other choices they will confront on their ballot papers?

To answer it, TIFA asked all “undecideds” (on their presidential choice) if they have decided who will receive their votes for governor, senator and MP.  Altogether, more than half (58 per cent) reported that they had yet to make up their minds on any of these choices. By contrast, just over one quarter (28 per cent) indicated that they had indeed made up their minds about all three.

It thus appears that these latter respondents—those who had made up their minds on the lower races—very likely decided, for whatever reason, not to reveal their presidential preference although they actually have one. By contrast, it may be assumed that the others are being honest, since if they have yet to decide their vote-choice for any of the other positions, their attention to the campaigns—and perhaps to the election itself—is obviously low (even if they are classified as “potential voters”, that is, those who claimed to be registered but excluding those who said that they “definitely will not vote).

Main reason(s) for being undecided

Another clue about the “real” position of these self-professed undecided voters is provided by the distribution of responses to the next question that all of these respondents were asked: “What is the main thing that will help you decide which presidential candidate you will vote for?”  Whereas a substantial proportion (28 per cent) identified “more information about the policies/the manifestos of the candidates” as the main factor that would allow them to decide, more than twice as many (61 per cent) said that they are “not sure” what would help them make this decision. A handful mentioned either “guidance/advice from family, friends or respected local leaders” or “instructions/advice from local candidates/politicians”.  At this late stage of the electoral cycle, how anyone could not be certain what they need to know—or whose opinion they need to hear—in order to make this important decision seems improbable, suggesting that this vague response is nothing more than a ruse.

Political party self-identification

Another “window” into the minds (or at least the political profiles) of the “undecideds” is gained by getting the answer to another question: what proportion of them do/do not self-identify with any political party? The assumption here is that far more of those who do may be expected to have made up their mind about their presidential voting intentions, even though, as noted above, they may be shy about revealing this.

The data here indicate that this is indeed the case, with over 90 per cent of those who self-identify with any political party also naming a presidential candidate, as compared to slightly over half of those who do not. Thus, once again, the latter exhibit a much lower level of electoral interest.

At the same time, these “undecideds” regarding the presidential contest, but who do, nevertheless, self-identify with a party are nearly evenly split: 3 per cent doing so with ODM and 2 per cent with UDA, and 1 per cent each with regard to the two corresponding coalitions.

Indeed, only with the benefit of credible official results will it be known if at least some of those declining to reveal their voting intentions have actually concealed them—as happened with significant proportions of respondents in US surveys that were “wrong” with regard to Donald Trump’s victory in the 2016 election and in the UK’s “Brexit” vote the same year. Assuming that is the case, who between Ruto and Odinga will benefit most when the real votes are counted?

Confidence in election integrity

Another factor evidently affecting the naming of a preferred presidential candidate can be put in the form of another question: does confidence in election integrity influence the willingness or propensity to name a preferred presidential candidate? According to the data, only very marginally so. Two sets of figures may be cited. First, slightly more of those who did not name a candidate believe that false election results (for whatever positions) are either “very likely” or “somewhat likely” after the 9 August votes are counted: 56 per cent vs. 52 per cent.  Conversely, rather more of those who named a candidate are “quite certain” that this will not happen as compared with those without a (stated) preferred presidential candidate: 40 per cent vs. 32 per cent. (More of the latter are also “not sure” if any election “wins” will be stolen or not: 12 per cent vs. 8 per cent.)

Two demographic variables: education and English language competency

The impact of education on the inclination or ability to make a decision about one’s presidential vote is significant. For example, among those who mention an intention to vote for any candidate, only 26 per cent have not gone beyond primary school, whereas more than double this figure (i.e., 56 per cent) who do not mention any candidate are within this lower education bracket. And regarding language ability, only 33 per cent of those did not name a candidate say they can speak English, as compared with 57 per cent of those who did (clearly largely a correlate of one’s education level). Taken together, such contrasts related to these two variables suggest that at least part of the failure to name a candidate appears to be insufficient access to political information as opposed to a propensity to conceal one’s intentions.

Flip-side: who has changed their presidential preference, and why?

To probe deeper into the process of deciding whom to vote for (which can be mainly either an individual or collective/communal one, although these two factors vary across the country both between and within various communities), TIFA sought to discover just how many people have recently changed their voting intentions since the beginning of the year. Such changes are of three types: from being undecided to supporting a particular candidate; from supporting a particular candidate to being undecided; and to changing support from one candidate to another.  Even without data, some of the most salient reasons can be suggested: because the candidate dropped out of the race; because one’s most admired or local political/community leader moved into or out of a relationship with one or the other presidential campaign team or alliance; because of campaign or other messaging; or because of the candidates’ choices of running-mates.

In this endeavour, the first requirement was to determine how many people say that they have done this, and who they are, especially in partisan-political terms.

Altogether, a substantial minority—13 per cent—report having done so. Significantly, a clear majority of these—57 per cent—say that they have changed from one candidate to another, as shown in the following chart:

Moreover, slightly more of those now supporting Odinga than Ruto (a 5 per cent margin) are among these 57 per cent. However, the DP has the “upper hand” in terms of winning over those who had initially been undecided (by a 12 per cent margin). Finally, among those who had previously intended to vote for one presidential candidate but are now preparing to vote for another one or are now undecided (the latter, as noted above, just 5 per cent of the total sample), Raila and Ruto have “lost” a nearly equal number of potential voters (36 per cent and 33 per cent, respectively). Nearly all of the remainder (that is, among those who have “switched” to a new candidate) previously supported someone who never entered or who dropped out of the race (especially Musalia Mudavadi or Kalonzo Musyoka).

Indeed, only with the benefit of credible official results will it be known if at least some of those declining to reveal their voting intentions have actually concealed them.

In partisan terms, an equal (and relatively small) proportion of declared supporters of ODM and UDA (led by Odinga and Ruto respectively) report having changed their presidential voting intentions since the start of the year (11 per cent). However, far more of those who continue to self-identify with Jubilee have done so (26 per cent), and almost as many among those who support any of the numerous other parties (22 per cent), the take-away being that those attached to a party lacking a presidential candidate have been rather more inclined to “move” in one direction or another in terms of their ultimate presidential choice.

But to what do the respondents attribute their changes in voting-intentions? For most (42 per cent), it has been campaign promises and manifesto content, although “the influence of others” and “the choice of running mate” also register significant numbers (13 per cent and 10 per cent, respectively). Less impactful among this group as a whole was the failure of an admired leader to join (or withdraw from) the presidential race (e.g., Mudavadi, Kalonzo, and a few others, at 8 per cent) as well as an evaluation of opinion polls (4 per cent, with Odinga having a definite advantage here, however small the total figure).

Poleni kwa the ‘Undecideds’: What this contest is really about

Even if this is Ruto’s first attempt at the presidency, the considerable duration of his public career even before his decade-long service as deputy president, means that neither of the two main candidates suffers from any “public profile” deficit.  Yet one factor contributing to the higher than usual proportion who have chosen to remain “undecided” in recent surveys may be a higher level of confusion as to what agendas they both represent as compared with those who sought the country’s highest office previously. Such “confusion” may stem from the fact that Odinga and Ruto have largely exchanged their political “clothing”, with the deputy president largely campaigning against his president—and thus the status quo, even if throughout their first term they appeared to be in comfortable lock-step with each other—while the former prime minister usually defends him, even as he, too, tries to offer credible change for an electorate the vast majority of which is currently suffering a plethora of economic, and other, woes. Indeed, in a TIFA survey of June 2021, 50 per cent more respondents identified Ruto rather than Odinga as “the political leader most active in terms of criticizing the Jubilee government and trying to hold it to account”, and in TIFA’s April 2022 survey, some three-quarters of respondents identified Odinga as “Uhuru’s preferred successor”. Such a situation makes it largely impossible for Odinga to assume the anti-government posture he has in the last five elections, notwithstanding his short-lived absorption into Moi’s KANU government and party in 2001.

Finally, in more systemic terms, is the fact that Kenya lacks political parties that exist from one election to the next to which are attached contrasting policies relating to issues that matter to voters and that are fairly consistent over time. In “advanced” countries, such parties are often organized around policies related to such issues as taxation and its application to inequality, or the level of government penetration into private lives. In Kenya, though, the “choice” of candidates, even when belonging to different parties, tends to fall back onto perceptions of their personalities and identities (including but not limited to their ethnic groups), as well as onto voters’ feelings about the status quo and the conviction (or hope!) that one or the other candidate will do more to change things for the better.

The impact of education on the inclination or ability to make a decision about one’s presidential vote is significant.

At the same time (and as is often the case in “established” democracies), some voters will seek to “punish” incumbents as a way of expressing their unhappiness with the status quo, rather than expressing confidence in any particular set of would-be leaders.

Given the above, with less than an entirely clear picture as to just “what” either William Ruto or Raila Odinga represents in terms of what Kenya would be like after five years of having either of them at State House, it seems of little surprise that the “undecideds”—and those who admit to having changed their voting intentions within just the last seven months—are as numerous as the research has revealed.

But with only a few hours remaining until “judgment day”, the “mystery” of just how they will use their votes will soon be resolved, with this group of voters very likely to make the difference—assuming they actually get to their polling stations on 9 August. For as has been pointed out, try as they might, they will not find “undecided” as a vote option in the list of presidential options, nor in those for any of the other five offices to be filled.

Let us now consider some additional issues.

Five + two poll comparison: Third horse exit and some integrity questions

To no surprise, there has been a flurry of surveys conducted and released in the last several weeks, with two firms—TIFA and Infotrak—doing this twice during this period, as shown in this table:

*In releasing its previous Radio Africa poll a month ago, the Star announced that it would be releasing additional polls “every week between now and the election”, but only the one listed here has appeared.  The fact that this poll is the only one putting Ruto ahead of Odinga in the last two months also raised questions in some quarters<

** In its report, Infotrak shows the margin of error in this survey to be only +/-1%.

Several takeaways can be drawn from this “gaggle” of surveys.  First, Odinga’s margin over Ruto has increased significantly, even if the latter’s rating has also risen, both benefiting from a substantial increase of those stating they are “undecided” or simply refusing to answer the question, so that Odinga’s gain is not dependent on any Ruto loss.  Second, the drop in about half of expressed support for Wajakoyah (together with the failure of Waihiga to register any significant support) means that the prospect of a second round, run-off contest, has ebbed markedly. Indeed, whereas Odinga would just barely cross the first round win-threshold according to TIFA’s late July survey with 50.2 per cent (if the 7 per cent “undecideds” and “no response” are removed), barely a week later in the most recent of all these surveys, this figure rises to 53 per cent, with Ruto at 45 per cent. (The TIFA media Release for this most recent survey of the lot was the only one which included this calculation.) Third, of the six surveys, only the Radio Africa one puts Ruto ahead, but only by 1 per cent, and thus within the survey’s margin of error; nevertheless, this still constitutes its tagging as an “outlier”.  Fourth, even if CATI surveys may have a slight anti-Ruto bias (though that by Radio Africa appears to contradict this), it is clear that larger samples do not yield significantly different results, as discussed in a previous article in this series, even if larger samples allow for more “precision” at the sub-national level (as pointed out by David Ndii, see below).

‘Rogue’ or ‘fake’ pollsters

One issue that arose during a recent workshop in Naivasha on “The Media and Opinion Polls” organized by the Media Council of Kenya and that also came up at the IEBC pre-conference forum on the same subject held at KICC a few weeks later, was that of the appropriate response to the results released by “unknown”/unproven (or so-called “rogue”) pollsters. Unfortunately, no attempt was made to define what these terms mean. On occasion (and as I have mentioned in a previous Elephant piece), a pollster may appear “out of nowhere”, as was the case with the British firm, RealField, that in January released the results of a survey conducted the previous month, but has not been heard from since (even if some reports indicate it has been conducting internal polls for Azimio). Yet being a “newcomer” and thus lacking any (at least Kenyan) track-record does not necessarily mean it lacks integrity. This latter category could more genuinely include firms that release results wildly at odds with at least several other known ones with a history of fairly reliable results. At the same time, the term would more accurately apply to any firm that releases results from an actual survey that have been “doctored”, let alone those based on no survey at all. Unfortunately, for whatever reasons, the professional body to which most “reputable”/established firms belong, the Market Survey Association of Kenya—MSRA—appears to make at most minimal efforts to internally query significantly contrasting results of its members. It is thus largely left to the media to decide whether the results from any particular firm warrant publication or broadcast.

The results from two firms that raised some eyebrows after their results were released recently are the following:

In this connection, at the second of the above workshops, I expressed my disagreement with several other panellists with regard to whether survey results from such “rogue”—or unknown/”out-of-nowhere”—pollsters should reach the public through the mainstream media, however much they deviate from those of the “mainstream” firms.  Whereas others felt that they should be ignored entirely, I suggested that this was a mistake, based on faith (or hope!) that scrutiny would do more to limit their influence than silence, given the increasing use of social media to spread such “misinformation” far and wide. The key issue for me is just how they should be covered. In general, there are three options. One is neutrality, so that their results and methodological details are simply treated as “news”.  Another is to interrogate them, in part by comparing them with other “mainstream” findings, so as to (indirectly but clearly) raise questions about their credibility (assuming they deviate substantially from the latter—which they invariably do). And a third option is to do this, but with an added “pinch” of interrogation (or even ridicule) about the possible motive(s) behind them, including questioning whether any interviews were actually conducted. Such questioning could be accompanied by any information gleaned from communication with those involved in such polls. A place to start with such interrogation is to (at least) report their level of compliance with the “Publication of Electoral Opinion Polls Act” of 2012. Taking these two “outsider” polls, they violate the Act in several respects.

Such “confusion” may stem from the fact that Odinga and Ruto have largely exchanged their political “clothing”.

To begin with, the media release from the Centre for African Progress included no physical address (although their company profile can be located via the Bloomberg News website).  It also claims that the survey “is funded by an endowment fund that it shares with other similar entities around the world”, but the actual identity of this “fund” is not revealed. (The document also claims to be “the only researcher that correctly predicted Kenya’s polls in 2017”, which is false; Ipsos’ final 2017 survey showed Uhuru-Ruto defeated Odinga-Musyoka 52 per cent to 48 per cent.)

As for Intel Research Solutions poll, the most obvious violations are the failure to identify “the sponsor of the poll” and its address, as well as the address of the organization itself although it does have a website. (According to one journalist who attended the public release of the survey at the Panafric Hotel, when asked who paid for it, IRS’s CEO, Karen Mwangi, said that it was funded by “EU donors”, but a source at the EU subsequently assured me that the EU “had nothing to do with it.”)  Further, the release is silent on “the education level of the participants” even though it is still unclear why the Act requires declaration of this specific demographic variable and not others. In addition, and while not a violation per se, a closer look at the distribution of the poll’s massive sample reveals some strange anomalies. For example, a fairly populous county such as Embu with over 309,000 registered voters had a sample of 608, while two other counties with far fewer registered voters—Taita Taveta and Isiolo, with around 156,000 and 75,000, respectively—had far more respondents (679 and 682, respectively). The heavily populated Kiambu with nearly 1.3 million voters also had only 605 respondents. Such mismatches would inevitably lead to incomparable results in terms of the margins of errors for such counties.

Whether anyone would bother to seek “legal redress” for such violations seems very unlikely, but they could (or rather, should) be included in any media coverage—having given the Management of those firms every opportunity to explain these failings, of course.

Achieving 50 per cent + 1: Misinterpreting the undecideds

One common error in media reporting of recent survey results relates to the impact of the “undecideds”.  As shown above, none of the mainstream pollsters (TIFA, Radio Africa, Infotrak and “out of nowhere” —after more than three years of silence—Ipsos), show any candidate with more than the 50 per cent + 1 votes needed for a clear win on 9 August, the most recent total average (of the four firms shown above) being 48 per cent. The most visible error is to (evidently) assume that the remaining percentages for “undecided” together with “no response” won’t “vanish” when ballots are actually cast, since these two options will appear nowhere on the presidential “menu”.

For example, reporting the results of Radio Africa’s final pre-election survey in the Star at the start of last week, Bosco Marita stated that with Ruto at 45 per cent and Raila at 44 per cent, “neither of two presidential frontrunners would win on the first round on August 9”.  Likewise, Infotrak’s CEO, Angela Ambitho, added some confusion when she offered that “If most of the undecided vote for Raila he would win on the first round, but realistically I do not think they will all move to one candidate.”

Yet a closer look at the Radio Africa poll reveals that such a first round win by either of the top two candidates is actually very likely. This involves first removing 270 respondents who claimed to be “undecided” from the calculation. When their numbers are then re-calculated, Ruto has 49.5 per cent, Raila has 48.4 per cent, and Wajackoyah has 2.2 per cent, and with a margin of error of +/-1.9 per cent (slightly higher than the +/-1.8 per cent reported by the Star, since the calculation is based on a smaller number of respondents who named a candidate: 2,730 rather than the total sample of 3,000). Moreover, it seems highly unlikely that among all the “undecideds” who eventually do vote, their votes would be split equally between Ruto and Raila, thus giving one or the other of them the margin of victory. As noted above, the DP, at least according to the polls, has been winning over more those who had initially been within the category.

Some voters will seek to “punish” incumbents as a way of expressing their unhappiness with the status quo.

Indeed, among the final polls released from the four (mainstream) polling agencies, only TIFA offered such a calculation (i.e., with the “undecideds” removed). It showed that while without removing the 7 per cent of those “undecided” and those who did not respond to the “no responses”, Raila, despite enjoying an 8-point lead, fails to get over the 50 per cent + 1 hurdle (48 per cent to 41 per cent); when that 7 per cent is removed, he does easily, with 53 per cent as against Ruto’s 45 per cent.

But as explained above, any effort to explore these “undecided” respondents is to question the “honesty” of at least some of them. Further, and where I do agree with Angela Ambitho, is that among those who do turn out to vote, they are unlikely to all move in the same direction, even if some of them who do may indeed deliberately spoil their presidential ballots or even leave them blank (being insufficiently enamoured of any of the four candidates and their running mates), thereby raising the percentage totals among those who do vote by reducing the “denominator” against which such calculations are made. (And recall here one of the 2013 Supreme Court judgments that “invalid votes” must be excluded from the total number used for such calculations.)

Media coverage and the act

Another aspect of polling during this electoral cycle has been the impact of the Publication of Electoral Opinion Polls Act (2012).  As noted above, several firms have released survey results during this twelve-month period prior to an election during which the Act applies, apparently without any criminal repercussions. (Conviction under the Act invites punishment of about to Sh1 million and one year in prison.) The most relevant part of the Act for this discussion is the ban it imposes on the publication of any election-related results within the five-day period preceding an election, meaning that the last day for doing so was Wednesday, 3 August, although only TIFA released such results on this final day. While they were covered (with interpretation) in an article in the Standard (which included several of TIFA’s graphics) and were included in midday and evening news bulletins by (at least) KBC and NTV, no stories about them appeared in the Nation or the Star on Thursday. While it is unclear whether this was due to an interpretation (correct or otherwise) that the Act prohibits the reporting of such polls within the proscribed five-day period even if they had been released prior to it, a source at Citizen-TV informed me that, based on “legal advice”, such content could not be included in their Sunday evening news “State of the Race” segment which hitherto had provided details of all polls released in the preceding week.  Presumably, such caution is based on an assumption that at least some viewers (however few in number) would be learning of such results for the first time, even if they had been in the public domain earlier. On other hand, political editors at NTV and the Standard indicated that their “legal advice” was of a contrary nature, evidently based on an interpretation that the law applies only to a poll’s initial release. Whether such differences of interpretation will eventually invite an effort to clarify the Act remains to be seen.

The importance of the election by elective office

Another finding relevant to voters’ choice as they enter their polling stations was obtained in TIFA’s final (CATI) pre-election survey. The question (apparently never previously included in any Kenyan survey) asked respondents to identify, among the six positions to be filled on 9 August, which one they consider the “most” and “least” important. Given the media’s overwhelming attention on the presidential race, it came as a surprise that more respondents did not attribute more importance to it.  As shown, it is in a statistical tie with the position of governor, and nearly twice as many respondents ranked their vote for MCA as “most important” as did those who indicated the same for their MP.

While such ratings may support the argument that devolution has at least partly diluted the “do or die” competition for the presidency, they certainly raise the question as to how much these non-presidential races affect the motivation to vote altogether.

A Pre-post-mortem on the polls: Shoddy sampling, late-hour “bandwagons”, and other factors

One question asked increasingly as election day approaches is: “How will you pollsters defend yourselves if the actual (presidential) results prove you wrong?”

To begin with, it is important to correct Steve Otieno writing in last Wednesday’s Nation, when he began by asserting “The pollsters largely got it wrong in the past two elections when they showed ODM leader Raila Odinga in the lead only for President Kenyatta to be declared winner.”  In the final pre-election survey released by Ipsos (where I was then working), Kenyatta led Odinga 52 per cent to 48 per cent. So even if the official result was 54 per cent to 45 per cent, the Ipsos results were certainly not “all wrong”.  He also states that “a similar event happened in 2013”, when “in none of the polls did Mr Kenyatta’s popularity cross Mr Odinga’s even though both of them never passed the 50 per cent hallmark”.  In fact, the final Ipsos poll had UhuRuto at 44.8 per cent and Odinga/Kalonzo at 44.4 per cent. True, it showed that neither could reach 50 per cent, but it should be recalled that some 800 pages of ODM’s evidence meant to accompany its Supreme Court petition-challenge were excluded due to “late filing”.  Moreover, a national exit poll (implemented by a group of Kenyan and expatriate academics and published the following year) confirmed the findings of Ipsos: that neither had reached 50 per cent. So perhaps Otieno should do more fact-checking before trashing pollsters.

Whatever the case, can pollsters sometimes “get it wrong”?  Of course, such an argument would first have to precisely define what “wrong” means.  Specifically, is it mainly a question of who is announced as the winner, or how great the margin between the polls and the official numbers (assuming they are credible) is regardless of who wins?

In a TV discussion about the accuracy of such polls last week, David Ndii argued that a key weakness in the work of most Kenyan pollsters is sampling, even if it is based on IEBC’s distribution of registered voters, since it does not sufficiently match (the eventual) voter turnout, which is certain to vary across the country. This is a valid argument, although he would have to agree that given the absence of a (main) Kikuyu presidential candidate, and the presence of members of this community as running mates, previous turnout variations may not be a very precise guide for this election. So just which turnout projections should be used?

Whether anyone would bother to seek “legal redress” for such violations seems very unlikely.

At the same time, it is strange that Ndii appeared so uncomfortable with these polls showing Odinga leading, since he also claimed that “our internal polls show the opposite”. Unless he, too, believes in this “bandwagon” effect, it seems such “incorrect” polls would help Kenya Kwanza win by giving Azimio a false sense of confidence which would likely lead to reduced voter turnout among Azimio supporters, either because of “laziness” in mobilizing pro-Azimio voters on 9 August, or insufficient motivation among such voters due to overconfidence (both of which determined the result in the Msambweni by-election, as recalled below).

(One frustrating aspect of the entire discussion was that at no point was any attention given to the challenges the media sometimes faces in interpreting survey results, and even deciding whether to publish/broadcast them, as indicated above. Nor was the question raised at to the contribution, or otherwise, of these polls to Kenya’s democracy, whether they be internal/confidential ones conducted and analysed by campaign teams or those made public so as to allow voters to make more realistic choices, or be influenced by them.)

Another potential factor—that everyone says they hope will not occur—is “flawed” results, although in order for it to be taken into account in explaining a significant variance from the polls, it would be necessary to verify that such “rigging” did indeed take place and at what magnitude—not an easy task.

Still another would be any significant movement in voting intentions between the final polls and election day. While it is generally agreed that nearly all voters have usually made up their minds by this time, last-minute “waves” can occur, nonetheless.  For example, Peter Kagwanja, speaking on another TV panel show a few days ago, said that Odinga’s “numbers” in Mt. Kenya will increase dramatically in the final days, since “what matters most to our people, especially in this situation where we don’t have our own presidential candidate, is to be with the winner so that we will be as close as possible to Government”. (This also constitutes an interesting hypothesis that once elected, whoever controls the Executive branch will pay more attention to the needs of those areas/communities that voted for the winning pair than to those of other parts of the country.)  In this regard, he noted that nearly all recent (credible) polls showed Odinga-Karua opening up a greater gap between them and Ruto-Gachagua, so that this “bandwagon” effect is bound to materialize, thus making even such final polls look inaccurate.

One recent survey finding by TIFA (in its late July household survey, and published exclusively in the Standard) is that a significant proportion of respondents (13 per cent, as noted, all claiming to be registered voters) reported that they had changed their voting intentions since the start of the year. The following chart shows how these respondents changed.

Taking just those who moved from being “undecided” to supporting any candidate together with those who changed from one candidate to another (but excluding those who changed from supporting any candidate to being “undecided” now) —who comprise two-thirds of all those who reported some voting intention changes—just a handful (4 per cent) said they did so due to opinion polls, as compared with nearly half who cited policy-promises/manifesto content and another quarter who mentioned either “influence of other people” or “choice of running mate”:

True, among this “handful”, near twice as many say they now intend to vote for Odinga than for Ruto, but this is hardly a sufficient number to talk about any significant “bandwagon” effect, at least from polls.

Yet a closer look at the Radio Africa poll reveals that such a first round win by either of the top two candidates is actually very likely.

Still, the belief in this effect is clearly very widespread. Ever since the Deputy President was overtaken in most of the polls by the former Prime Minister, he and his associates have bashed them, often by alleging they were part of the “System’s” efforts to ensure an Odinga win. For his part, Senator Kipchumba Murkomen asserted in a recent TV discussion that “Odinga was leading in an Ipsos poll just a few months before the 2017 election, yet he lost”, although a review of the record reveals that at no time before that election did any Ipsos survey report any such thing.  Whether Murkomen’s statement was deliberate or a consequence of a faulty memory is unclear, but the point is that polls—especially those from the more “established” firms that show one’s preferred candidate trailing, invariably cause discomfort. On the opposite side, Karanja Kibicho, CS for Internal Security, was shown in a 2 June news broadcast saying that Azimio had the support of 60 per cent of Kenyans, according to an unnamed survey, although Ruto was then shown refuting this, claiming that he still has personal access to such National Intelligence data, which showed that he was leading “by 8 per cent”. He also dismissed other unnamed polls not showing him in the lead as falsified.

(It is also said that Kenya Kwanza/UDA had been conducting its own surveys, under the “expert guidance” of the former Energy CS, Davis Chirchir. However, some time ago, a UDA candidate approached TIFA to ask for a quote for a survey in his constituency, although he subsequently reported that he was told “not engage any local firms”, but was not given any alternative.)

At the same time, I received an e-mail message from a UDA stalwart a few weeks ago asking for my opinion about survey results that had just been released by one of these “non-mainstream” survey firms showing the Deputy President with a comfortable lead, as she wanted to know whether I consider their work reliable. When I told her I did not, and gave my reasons, she replied,

“I suspected as much; it seemed that they are just a random pollster trying to make a name by saying Ruto will win. A bit opportunistic. And none of the main media houses carried it, so I guess it flunked. These are the ones who need to be called out.”

At least in this case then, and whatever politicians say in public, there is clearly a sober need to know which poll results can be trusted, at least in terms of not being driven by any partisan or personal agenda.

However influential (or otherwise) such a “bandwagon” effect, some have argued that these polls have substantially underestimated Ruto’s eventual numbers, given the current president’s overt opposition to his ambitions. That is, they say that a significant number of survey respondents will be too “shy” to reveal their intentions to vote for the de facto “opposition”, by saying either that they are “undecided”, or even that they will vote for Odinga. This argument is somewhat similar to that offered by many analysts in the US as to why Donald Trump’s numbers were higher than what he received from pollsters in both of the last two elections, but especially in 2016. But in this case, it seems the main weakness was that too few of his voters were interviewed in the first place, since Trump himself had frequently trashed them, as well as the motives of those conducting them. In addition, post-2016 research revealed that a significant proportion of women who were interviewed live (in phone calls) were reluctant to reveal their pro-Trump intentions due to his history of sexual molestation and rhetoric, whereas in automated (non-live) survey calls, more women with the same demographic profile named Trump as the recipient of their forthcoming votes. Since all polls in Kenya are conducted “live” by phone or in person (aside from a few SMS polls that have their own problems), whether this kind of reluctance to reveal one’s choice (generally known as “the Bradley effect”) will be evident when the official results can be compared with those of the final polls remains to be seen, but a surprise Ruto win—perhaps at least in part based on a further move of a majority of the “undecideds” in his direction (and shown above)—would certainly invite scrutiny of this nature.

The Proof of the pudding: Voter turnout, last minute surprises, and election integrity

However, the most widely recognized and probable cause for such contrasts between final polls and the official results is voter turnout. As suggested above, making any such predictions based on past elections is challenging. However, it seems that turnout in Kenya is based on a combination of individual motivation and community (and/or political party/campaign team) mobilization. The latter factor raises the question as to which side will be more successful in doing so. Here, as discussed in my third piece for The Elephant, two surveys conducted in Msambweni before the December 2020 parliamentary by-election there (by Radio Africa and TIFA) both found the ODM candidate, Omar Boga, leading the independent (but UDA-backed) candidate, Feisal Bader, by around 20 per cent, yet he lost by nearly that much. With only 40 per cent turnout, it was clear that the Bader campaign was far more effective in getting his supporters to the polls. With turnout in Kenya’s general elections usually above 70 per cent (although the exact figure has varied considerably), such a dramatic “turnout advantage” is far less likely, but given the high numbers of registered voters in certain areas—starting, perhaps, around Mt. Kenya with some 5 million out of a national total of 22.1 million—this factor could well compensate for the current Odinga-Ruto margin, especially if combined with some level of reticence on the part of the latter’s supporters to reveal their voting intentions, as noted above.

Any effort to explore these “undecided” respondents is to question the “honesty” of at least some of them.

One additional factor that could “pollute” analyses that compare these final pre-election polls with the official results is the anticipated judgment on Monday—just a day before the election – that will determine whether the case filed by “a group of activists” at the Constitutional and Human Rights Division of the High Court a week ago seeking to prevent Rigathi Gachagua, Johnson Sakaja and twelve others from being sworn in should they win their respective contests (excluding out-going Governor Okoth Obado) on the grounds that they are facing criminal investigation for various forms of corruption, with reference to the integrity provisions of Chapter Six of the Constitution. Assuming the case is allowed to proceed, it is not possible to predict—or perhaps even gauge after the affect—its impact on the presidential outcome, but it seems inevitable that it would have some.

Finally, with specific regard to individual motivation, it might be asked: will Ruto gain more from the distress so many Kenyans have felt about current economic conditions for most of which the Kenya Kwanza “brigade” blame the Uhuru-Raila “handshake”, combined with a determination to “punish” the president for opposing his deputy, without whose assistance as a running-mate he probably could never have ascended to power? Or will Raila gain more from a combination of a “bandwagon” effect, combined with an admiration of his past struggles and suffering for the sake of better “governance”, with such credentials evidently buttressed by the choice of Karua as his would-be deputy?

The actual results of the election cannot themselves answer such questions fully, but together with sufficient post-election research of both a qualitative and quantitative nature—all elements of psephology (the study of elections) —they should provide an understanding of their salience and interaction than we have now, but again, depending upon whether the official results are without any (verifiable) credibility deficit, a scenario that both sides in this contest have vociferously championed.