Given that it is well over a month since my last piece in the series, it is an understatement to say that much has happened in the intervening period. The three main developments that are covered here are: the changing positions of the two main presidential candidates, the earlier use of polls by political parties in the selection of candidates to augment or replace the usual nomination contests, and the announced selections of deputy presidential running mates by the two main presidential candidates just hours before the official deadline for doing so. In addition, a brief comment on TIFA’s more recent Nairobi County survey is offered.
‘Horse race’ update
Although several other firms have recently released presidential contest polls, I will ignore them here due to their lack of credibility (but shall take up this issue in a subsequent article) and concentrate on the three “mainstream pollsters” whose results this series has been tracking: TIFA Research, Infotrak and Radio Africa. The table below shows the reversal of fortunes of Deputy President William Ruto and former Prime Minister Raila Odinga found by all three firms:
|Sample Size / No. of Counties
|Undecided / Won’t Vote/NR
|1,719 / 47
|9,000 / 47
|Radio Africa/The Star
|4,780 / 47
Several comments help to explain these figures.
First, regarding data collection dates, although the TIFA survey was conducted the day after the announcement of former Gichugu MP and cabinet minister Martha Karua as Odinga’s DP running mate—with the announcement of Mathira MP Rigathi Gachagua as Ruto’s running mate having been made the day before—not all respondents were able to name them. Specifically, while 85 per cent of all respondents could name Karua, only 59 per cent could name Gachagua. Several factors may account for this discrepancy, the two main ones being the much more public and “celebratory” event revealing Karua at the Kenyatta International Convention Centre compared with the more restrained and entrance-restricted one for Gachagua at the DP’s Karen residence, and Karua’s far larger public and political profile. This contrast notwithstanding, TIFA also provided correlation data that showed no statistical difference in expressed voting intentions for the two presidential candidates between those who did and those who did not correctly identify their running mates. (These correlations were based on the somewhat greater proportion of such Karua/Gachagua running mate awareness among those who elsewhere indicated that they intend to vote for Odinga or Ruto: 90 per cent and 69 per cent, respectively).
Specifically, the TIFA data reveals that slightly more of those aware of Karua as Odinga’s choice declared their intention to vote for him—41 per cent vs. 39 per cent for all respondents, and only 25 per cent among those who did not name her. Further, among women only, there is a 9 per cent difference in this regard: 37 per cent vs. 28 per cent. Yet similar results for Ruto’s running mate invite caution in concluding that knowing that Karua was Odinga’s choice itself made the difference. That is, considerably more of those respondents who could name Gachagua expressed a Ruto voting intention than those who could not (42 per cent vs. 26 per cent), with a nearly similar gap among women (37 per cent vs. 26 per cent).
Of course, if Gachagua were female, one would be tempted to conclude that gender per se accounts for this difference among respondents of both genders. That not being so, however, it would be necessary to explore other variables to account for this clear association of running mate knowledge with the propensity to vote for either Odinga or Ruto. A possible variable here is simply an interest in the presidential election/politics generally. This is apparent in that, among only those stating they are “undecided” about their presidential vote choice, significantly more could not name either Karua or Gachagua as compared with those who could: 23 per cent vs. 13 percent with regard to the former, and 18 per cent vs. 11 per cent with regard to the latter.
Specifically, the TIFA data reveal that a fifth of those who stated their intention to vote for Odinga as president also declared their intention to vote for Sakaja as governor.
Moreover, since neither Infotrak nor Radio Africa provided any such “running mate awareness” profile among their respondents, it was not possible to pose this question with regard to their data. The question as to whether—and how—any of these firms will attempt to measure this “running mate” effect in their forthcoming polls thus remains.
Next, compared to the results in the most recent previous surveys of these three firms, Odinga’s gains are significant. In TIFA’s late April survey, Ruto enjoyed a 7 per cent advantage (39-32 per cent), in Radio Africa’s, a 5 per cent lead (46-41 per cent), while in Infotrak’s previous poll, he and Odinga were tied (42 per cent each). In other words, Odinga’s most recent ratings give him gains of 11 per cent by Radio Africa and TIFA, and a 4 per cent gain by Infotrak.
In addition, if all the respondents who declined or failed to name any candidate in these three polls are removed from the calculation, the most recent results are as follows:
|Total / Named Candidates Only
Re-Adjusted Margin of Error
|1,719 / 1,324: +/-2.7%
|9,000 / 7,290: +/-1.1%
|Radio Africa/The Star
|4,780 / 4,135: +/-1/5%
In other words, since “Undecided”, “No Response”, and “Won’t Vote” are not options found on ballot papers, removing such survey responses from the calculation gives a more accurate picture of where the race actually stands: Odinga now enjoys leads that are beyond each firm’s margin of error (the slightly greater ones reflecting the reduced effective sample sizes). This is so even if it is reasonable to believe that some of these “no preferred candidate” respondents actually have one but were shy about revealing it for one reason or another. (An attempt to “dig deeper” into the likely preferences of such respondents —at least those who will vote on 9 August—will be included in a subsequent piece in this series.)
At the same time, aside from those respondents who state that they will not vote—and even among those who declare that they will “definitely” vote in the surveys that included the relevant question—it is impossible to predict the levels and variations in voter turnout, even if, as Charles Hornsby has shown, such turnout rates have followed a fairly consistent pattern over the nearly six decades of Kenya’s independence.
Odinga’s most recent ratings give him gains of 11 per cent by Radio Africa and TIFA, and a 4 per cent gain by Infotrak.
All the above notwithstanding, given Odinga’s still quite modest lead in all three of these recent polls, the period remaining between when these three surveys were conducted and the election itself, together with expected (if unknown) differential turnout rates across the country, could still be the deciding factor (with turnout itself being a function of a complex combination of self/communal motivation and “external” mobilization, i.e., by individual candidates and/or political parties).
Nevertheless, it may be concluded that the DP’s campaign team and supporters would have cause for considerable concern if the next round of survey results reveal a further increase in Odinga’s lead. Moreover, it can also be concluded that unless the proportion of votes for all other presidential candidates (of which there are just two as of now) amounts to more than at least 3 per cent, a runoff contest is unlikely. That said, the fact that George Wajackoyah received 7 per cent in a mid-June Nairobi poll conducted by TIFA does at least raise this possibility.
A final set of additional comments about these three polls may be offered.
First, while both TIFA and Infotrak employed their previous methodology of telephone (i.e., CATI) interviews, Radio Africa for the second time conducted its poll via SMS messages. Based on 4,780 respondents, the margin of error is shown as +/-1.5 per cent. This is correct, but contrasts with the incorrect figure in its previous survey of +/-4.5 per cent for a sample of 3,559, a mistake that I had noted in my previous piece. At the same time, however, this Radio Africa survey once again reports a considerably smaller proportion who declined to name a preferred presidential candidate as compared with the other two surveys—13 per cent, as opposed to 19 per cent by Infotrak and 23 per cent by TIFA. Could this be because of the methodology used? That is, if those who receive the initial SMS can tell that it aims to collect data for a presidential elections survey, do many of those who have not made up their minds decline to participate so that a significant proportion of “undecided” responses are not even captured or reported? The fact that a source at the Star indicated that fewer than 10 per cent of those contacted via SMS for these surveys choose to participate suggests that this might be the case. This would also mean that for an achieved sample of 4,780, nearly 50,000 SMSs were initially sent out inviting participation. Further, unless the Radio Africa data-base has fairly detailed demographic details for all the interviewed respondents, it seems it would be impossible to weight the data so as to accurately depict the “population universe” the respondents purport to represent—whether the entire Kenyan population as per the 2019 Census or the total of registered voters, at least as reflected in the 2017 Register (since the current, updated version was not available at the time of these surveys). That the Star’s report on this survey fails to include any demographic information about the achieved sample (even education levels as required by the Publication of Electoral Polls Act, 2012) makes it impossible to know.
The above discussion raises one other question about the achieved samples. As I pointed out in my first piece for The Elephant, there are two basic requirements for “representative” samples. One is that the pool or data base from which respondents are selected must accurately represent the purported “population universe”. The other is that the process of selecting the relatively small number of respondents for interviews must be absolutely random, that is, without any bias, whether intentional or otherwise. While it is safe to assume the latter condition is strictly adhered to for the surveys reported here, a question may be raised about the former. That is, how precisely do these companies’ data bases reflect the “population universe” of adult Kenyans (or of registered voters, if respondent selection is restricted to them)?
For Infotrak and TIFA, this boils down to the mobile phone numbers from which their samples are drawn. How many do they have, and do their distributions for both their national and sub-national results reflect the country’s reality on the ground, at least as captured in the 2019 Census? It has been stated that both of these firms regularly collect mobile numbers in the course of conducting household or “face-to-face” surveys, whether the phone number acquired belongs to survey respondents themselves or to another household member (in cases where the respondent does not personally possess one).
It may be concluded that the DP’s campaign team and supporters would have cause for considerable concern if the next round of survey results reveal a further increase in Odinga’s lead.
In any case, two questions arise here. First, how many Kenyan households lack even one mobile phone owner/user? Second, at least in terms of the content of the questions being asked in CATI surveys, would a survey on public and political issues, such as one dealing with the forthcoming election, produce measurably different results if the data obtained came entirely from such phoneless individuals or households? While it is possible to compare the demographic profiles from respondents with and without such phones in previous surveys, the most precise way to answer this question is to conduct two surveys with the same content and at the same time but with respondents drawn from these two distinct “population universes”, an undertaking that I am not aware of having been done by any of these firms. (During the period of my previous employment, I did undertake such an experiment, but that was over ten years ago when mobile phone ownership was far less extensive than it is now. Consequently, I do not feel it would be useful to present those results here.)
One reality that can be reported, however, is that according to the 2019 Census, some 20 million households have (or had then; the number has surely increased) such phones, most of those without being found in parts of the country outside the networks of all of the mobile network providers. Still, even in parts of the country covered by such networks, a few people, mostly the very poor, are without them. With an adult population (i.e., 18 years and above—those who are routinely included in the sort of surveys discussed in this series) of around 25 million, this equates to some 85 per cent. And here I must correct Prof. Karuti Kanyinga who asserted in a recent Sunday Nation article that CATI surveys cannot be representative, since “only 47 per cent” of Kenyans owned mobile phones at the time of the 2019 Census. He evidently failed to notice that such census figures apply to all Kenyans aged three and above, and thus vastly underestimated phone possession by adults. As for Infotrak, in presenting presidential contest results for all 47 counties, one would want to know how extensive—and how representative—their mobile phone database is for each one, especially those mainly northern counties that still lack extensive network coverage.
Also, in the latest Infotrak survey, the 47 counties were divided into three categories: Odinga “strongholds”, Ruto “strongholds” and “battle-grounds”. Yet the lists are slightly misleading, in that the margin of error for these county results in several cases exceeds the difference between these two candidates. That is, the apparently impressively large sample of 9,000, when divided by 47 counties, yields a sub-sample average of only about 190 respondents per county, which has an associated margin of error of +/-7.1 per cent—equal to about a 14 per cent spread. This means that any county results that fall within this range indicate that, in reality, either candidate could actually be leading. For Odinga, this removes Mombasa and Garissa from his list of 20 “strongholds”; for Ruto, the same applies to Laikipia and Isiolo from his list of 16.
One notable aspect of this electoral cycle is that, based on statements by key political actors themselves, as well as in various media reports, there has been an unprecedented level of polling by both of the main campaign teams. These have been conducted not just for measuring presidential candidate and party/coalition popularity as well as the salience of particular issues at the national, regional and county levels, but also, in particular contexts, for the selection of candidates so as to avoid the “messy” process of nomination contests. Just where and when such polls were used, whether as the sole basis for the awarding of nomination certificates, and whether they were used together with “consensus” and even voting, remains a subject for very rigorous research (hopefully by some energetic doctoral students). What is clear, however, is that party and coalition leaders saw their main benefit as eliminating or at least significantly reducing the cost and acrimony of holding nomination elections, and even more potential acrimony when nominations were awarded in the absence of any competitive process at all (i.e., “dished out”, in Kenyan parlance). As the Sunday Nation reported just as the nomination period began, “The issuing of direct tickets to Orange Democratic Movement (ODM) hopefuls has caused unrest in the party strongholds, marking the start of a traditionally tumultuous season of political party nominations.”
Kanyinga evidently failed to notice that such census figures apply to all Kenyans aged three and above, and thus vastly underestimated phone possession by adults.
The Azimio coalition, and the ODM party in particular, have trumpeted their use of such research tools—albeit in particular, and limited, contexts—in selecting their flag-bearers. This first came to my attention early this year when this coalition’s campaign chairman, Laikipia Governor Ndiritu Muriithi, who is known for his appreciation of “hard” data, was reported to have declared that the services of a pollster had been engaged “to conduct research that will help in determining popular candidates to fly the coalition’s flag in the August General Election”. This same media report, quoted “a source” in the Odinga camp as saying that the data collected would be used to identify the “most popular candidates for all positions from MCA to governor”, and that this method was also aimed at “curbing violence” that is often generated by contested primaries. ODM Secretary General Edwin Sifuna was even more specific, revealing that the (unnamed) pollster had so far covered 15 counties with this figure to double “in the coming weeks.” As he put it, “We want science to guide us. We need not subject aspirants to an elective process when we know who is ahead of the others”, and he indicated that such direct tickets would be given to aspirants with “more than 20-30 percentage points over” their nearest rivals.
Subsequently, however, we saw controversy erupt in such places as Mombasa, where the opinion poll “losers” complained that they were being “unfairly” deprived of the chance to test their popularity with actual party members through primaries—one of the several methods allowed for such candidate selection by the ODM party’s constitution.
While it cannot be denied that such competitive primaries have often triggered intra-party violence among rival supporters, several questions arise in connection with this “opinion poll” option.
First, as the chairperson of ODM’s National Elections Board, Catherine Muma, explained in a TV interview at the time, these polls were not restricted to party members as documented in the list that was filed with the Office of the Registrar of Political Parties, but to all people in a particular electoral unit. Given that actual ballot choices on 9 August are not limited to candidates of one party, this makes sense. But here, one would want to know if any “filter” questions were used, so that only those registered as voters in that unit were interviewed (since, especially in a town like Mombasa, a substantial minority are registered and vote in other parts of the Coast region, and even beyond). Also, were respondents asked to indicate who they thought would be actually vying for particular seats before being asked to indicate their preference? If not, how could the popularity of any candidate be gauged against that of others? And if they were asked, but mentioned only a few of them, then what? That is, with polls conducted a full four months before the election, it was unrealistic to expect that the eventual ballot “menu” for such seats could be known. More generally, at this preliminary stage of the electoral calendar, which factors determine an aspirant’s popularity that a poll could measure? In particular, do incumbents have an advantage because of their name recognition, or a disadvantage, at least among those unhappy with the status quo and who are looking forward to punishing incumbents? And further, if the samples for such polls were not limited to party members, does this not diminish, if not completely negate, the value of such membership itself?
Based on statements by key political actors themselves, as well as in various media reports, there has been an unprecedented level of polling by both of the main campaign teams.
Even if such polls were used much closer to the election—perhaps based on an understanding between coalition partners that those who performed poorly would drop out (at least before the ballot papers are printed!)—how “scientific” are they in terms of reliably indicating actual outcomes? ODM Secretary General Edwin Sifuna, as reported by NTV on 5 April, made the curious assertion that: “There is not a single time when the science has misled us, and those on this podium can bear me witness. When we did the scientific poll in Msambweni [before the 15 December 2020 by-election] the actual results were exactly the same as what the science [of the poll] was telling us. The science and the results were spot-on.”
Since he did not identify the poll (or polls) he had in mind, I can only refer to the two polls I am aware of, both released in early December, just a few weeks before the by-election when voters should have been aware of all the candidates, and knew at least something about the main ones. Their results, together with the IEBC’s official results, are as follows:
Msambweni By-Election Poll/IEBC Results (Rounded Figures – Main Candidates Only)
|Margin of Error
|Radio Africa/The Star
|3 December, 2020
|3 December, 2020
|IEBC: Official Result
|15 December, 2020
Both surveys had a margin of error of about +/- 5.5 per cent based on their respective sample sizes. That means, for example, that in the Radio Africa poll, Boga’s support could have been as low as about 49 per cent, and in the TIFA poll, his support could have been as high as about 51 per cent, an indication that their results were pretty much in agreement. But the conclusion should be clear: both polls were “wrong”.
So, was Sifuna referring to some other poll, perhaps one never made public? (Surely not the poll results released by the Bader campaign on social media on 18 November showing him with 74 per cent as against just 23 per cent for Boga!) In that case, and if the official results mirrored it “exactly” (according to Sifuna—that is, an “upset” ODM loss), then we must assume either that the ODM campaign team, believing the contest was a lost cause, simply gave up (which surely would have caused Bader’s margin of victory compared to the poll to increase), or continued its campaign efforts to try to overcome Bader’s lead, but without any effect whatsoever.
There are several possible explanations, none mutually exclusive, as to why the two published polls failed to match “the reality”, perhaps most important among them: that, as in any election, differential voter turnout across any electoral unit cannot be predicted precisely. Moreover, by-elections invariably attract fewer voters than do general elections. In this Msambweni case, it was just under 40 per cent, and quite varied across the wards. As such—and especially in such a by-election context—even “scientific sampling” can often yield misleading results. In addition, however—and evidently more important—is that according to other reports at the time, Boga’s team was not nearly as active on the ground as Bader’s, in large part (I was told during a visit to the area this January) because the Boga campaign team “relaxed” after seeing these two polls, whereas Bader’s put even more energy into their efforts.
This is not to say that such polls are useless when parties/coalitions seek to identify their most-viable-candidate options. But it is clear that their use should be accompanied by considerable caution, since their “science”, in terms of predicting popularity—and thus an election’s outcome—may be limited at best, especially when voters are not yet aware of the full and final ballot menu.
Choosing running mates
Let me turn finally to what has been perhaps the most remarkable use of surveys so far in this pre-election period: that conducted by UDA in identifying a running mate for their presidential candidate, William Ruto.
The DP himself made clear just prior to his announced choice at his Karen residence 15 on May that a “general public” poll of some 10,000 respondents had been conducted (although it remained unclear by whom) in the greater Mt. Kenya region, an exercise that was supplemented by an “electoral college” poll among area MPs. In both cases, we were told, Tharaka Nithi Senator, Prof. Kithure Kindiki, was the overwhelming choice (“90 per cent”). Yet almost in the next breath, the DP revealed that he had settled on Rigathi Gachagua (as noted above). In fact, prior to this announcement, it was reported that the DP had made this decision some time earlier—the “rag sheet” Weekly Citizen had headlined this choice in its 1-7 November publication! It remains unclear to me why such contrary results were made public, or even why the polls had been undertaken in the first place. The fact that Senator Kindiki chose not to attend this event, and only issued a statement later, served to underscore his own dissatisfaction with the process, or at least with the decision. Whether he would have attended had these poll results not existed or been released must be left to speculation; he subsequently re-joined Kenya Kwanza’s campaign activities.
Aside from whatever internal polling was conducted by the main presidential campaign teams, publicly released surveys exploring this issue were at least equally off the mark. In both Radio Africa’s early April survey and TIFA’s a few weeks later, the “best” running mates for Odinga in terms of bringing in the most additional votes as identified by respondents were Peter Kenneth, Kalonzo Musyoka and Karua, in that order. Likewise, both Radio Africa and TIFA identified Musalia Mudavadi as Ruto’s “best” choice, with the ultimate “winner”, Rigathi Gachagua coming in second in the Radio Africa poll—followed by Kiharu MP Ndindi Nyoro—and third in TIFA’s, behind Kirinyaga governor Anne Waiguru.
Closer to the actual day of the announcements, Infotrak results were nearer to the eventual reality with regard to Odinga’s “best” choice, Karua placing first, followed by Musyoka and Kenneth. For Ruto, however, the results were less “accurate” than in either of the two other polls on the issue, with Gachagua nowhere among the top three, those positions being taken by Mudavadi, Kindiki and Waiguru, respectively.
In none of these three cases would it be “fair” to criticize the pollsters for “getting it wrong”, since it would be expected that respondents’ choices are far more a reflection of their general familiarity with these political figures, rather than of any intimate knowledge of the factors being considered by the presidential candidates themselves and their closest associates. For the latter, while adding “the most votes” is clearly a critical one, other factors include, especially, the “depth” of one’s (and one’s friends) pockets for campaign resources, the personal “chemistry” between the would-be president and deputy as well as the “fit” of their policy agendas and priorities, and the perceived capacity of the deputy to run the executive in the president’s absence (or even permanent exit from office for whatever reason). In other words, the “ideal” running mate must have enough political stature and following to help the two of them win the election and then govern, but not so much as to constitute a rival or threat to the dominant position that the constitution bestows on the president.
It remains unclear to me why such contrary results were made public, or even why the polls had been undertaken in the first place.
At the same time, in this 2022 context, several of these factors—whatever the weight given to each—are unlikely to apply equally to considerations by the Ruto and Odinga sides. This stems from the widespread assumption that whereas if he wins, Ruto would certainly attempt to extend his incumbency in 2027, there is rather less certainty that Odinga would do so due to his age, however healthy and vigorous the 77-year-old appears to be at present; indeed, some unconfirmed reports have indicated he has agreed to a single term. As such, there is likely to be more jockeying for influence—and ultimately, executive power in 2027—on the Azimio side of the partisan equation following the inauguration ceremony, eventually if not immediately.
TIFA’s Nairobi County Survey (20 June)
|Candidates and Percentages
|J. Sakaja – 40%
|P. Igathe – 32%
|Others – 9%
|Undecided/NR – 28%
|R. Odinga – 50%
|W. Ruto – 25%
|Others – 8%
|Undecided/NR – 18%
The results of TIFA’s Nairobi County Survey carried out on 20 June have invited numerous queries as to how Johnstone Sakaja, the UDA/Kenya Kwanza candidate for governor, could clearly be more popular than Polycarp Igathe of Jubilee/Azimio, whereas the former’s presidential candidate is only half as popular as the latter’s (25 per cent vs. 50 per cent). While the data does not suggest an explanation, several “common sense” factors do. First, Sakaja has been in Nairobi public life for nearly a decade, first as a nominated MP, and then as the current senator. He has also been highly visible, frequently appearing on TV talk shows and taking well-publicized positions on various county and national issues. Prior to the 2013 election, he served as chairman of presidential candidate Uhuru Kenyatta’s The National Alliance Party (TNA), at the time “the youngest person in the world” to hold such a position (according to Wikipedia). By contrast, relatively few (in Nairobi, at least) had heard of Igathe before he was selected by then Nairobi gubernatorial candidate Mike Mbuvi Sonko to be his deputy governor running mate for the 2017 election on behalf of the winning Jubilee party. He then largely vanished from public life after only six months, when he resigned from this position, having fallen out with Governor Sonko, who would later be removed from office through impeachment. More broadly, given the general absence of clear policy or ideological differences in Kenya’s political parties, especially perhaps in terms of county-level issues, it is clear that “personality”, including one’s personal “track record” is more salient for many voters. Specifically, the TIFA data reveal that a fifth of those who stated their intention to vote for Odinga as president also declared their intention to vote for Sakaja as governor. As such, while party/coalition identity certainly has some salience, it is far from overwhelming, at least for this forthcoming Nairobi gubernatorial contest. Regardless, how much rhetorical attention/support Sakaja will give to Ruto during his Nairobi campaigns between now and the election will still be interesting to watch. However, should the current investigation into Sakaja’s Uganda Team University degree result in his disqualification, this issue may become moot though at the time of TIFA’s survey, an impressive 20 per cent of those who declared their intention to vote for him also indicated they do not believe his Team degree is valid!
Concluding point: heading to the home stretch
Whatever the frequency of both public and internal polls so far, it may be expected that it will increase in the remaining five weeks or so prior to 9 August. In addition to the regular, roughly monthly, surveys by Radio Africa, TIFA and Infotrak, it is rumoured that Ipsos will emerge from its nearly four years of “survey silence” with at least one poll before the election. The main media houses have also shown an increased interest in these surveys, in terms of sponsorship as well as coverage (in both news bulletins and talk shows).
It is rumoured that Ipsos will emerge from its nearly four years of “survey silence” with at least one poll before the election.
Just how much such poll results—and their accompanying/reactive commentary—may influence actual voting patterns remains a subject for a future piece in this series. So, too, does a consideration of how such devices might help to answer another major question about Kenyans’ voting behaviour: how can the split about presidential choices, especially within particular ethnic groups, be understood? In this context, it will be argued that simply posing this question should help to dilute the greatly overstated mantra that “all Kenyan politics is ethnic”!