John Mahama (NDC) is leading, but likely a second round will be necessary
In the period September-November Afroopinion conducted a survey using randomly generated Whatsapp numbers. The mean idea behind this so-called WANGI methodology is that it is impossible to be influenced by national governments or other actors. Since there is a direct interaction between a surveyor from outside of Ghana with a voter from Ghana, there is no way of them knowing that a Ghanian voter is participating and hence it’s also not possible for them to interfere. The main candidates are below.
Even under these circumstances, respondents are reluctant to give their opinion. Only 48.6% of the respondents that we asked “Can it be dangerous if a stranger asks you such political questions” replied with no. Only 39.2% answered “yes”, while 12.2% even refused to answer.
It should be noted that this distrust is still lower than in countries with regimes that are generally considered authoritarian like Uganda (2021) and Zimbabwe (2023), where we did earlier surveys.
The present President is Nana Akufo-Addo, who won the 2020 elections with 51.3% of the votes.
We asked our respondents who were old enough who they voted for in 2020, and they replied
Since the size of our survey is relatively small (we talked to 500 respondents, where 2000 is normal) the 95% error bound is relative large at 9%. This fact, in combination that voters have passed away can explain the difference of 10%. This seems to suggest that Nana Akufo-Addo honestly won the previous elections although not all of our respondents agree on that.
We asked our respondent how Akufo-Addo performed and the responses were mixed.
However, Akufo-Addo is not running, but Mahamudu Bawumia is representing NPP. John Mahama remains the NDC candidate.
According to our predictions, he is leading with 48.5%. But since the 95% error is 8%, we are not sure if he will win in the first round. The considerable amount of voters for Nana Kwame, 12.1% makes it more difficult for either Mahama or Bawumia to win the election with one round only.
Our survey indicates that Ashanti is a stronghold for Bawumia and Great-Accra is a stronghold for Mahama. However, it should worry Bawumia that he scores less in these areas than Nana Akufo-Addo did.
Odinga leads, but likely a second round is necessary
On 9 August 2022 Kenyan Presidential Elections take place. The two main candidates are opposition leader Raila Odonga, with support from president Kenyatta, and vice-president William Ruto. There are two other candidates: George Wajackoyah and David Waihiga Mwaure.
source Wikipedia
According to the most recent polls by TIFA research and Infotrak, it will be a neck and neck race between Odinga and Ruto. TIFA’s latest forecast is 47% for Odinga and 44% for Ruto. To win the presidency in the first round, at least 50% + 1 vote is needed. Whether or not a run-off is necessary will be partly decided by how many votes are taken by Wajackoyah and Waihiga Mwaure.
According to our survey among 450 random Kenyan WhatsApp users it will indeed be neck and neck.
Neither Odinga (49% +/- 4%) nor Ruto (45.6 +/- 4%) are forecasted to get a majority in the first round. The other candidates take 5.4% together, significantly more than in the TIFA forecast.
5% of our respondents indicated to be undecided. 4.5% indicated not to vote. 3.1% didn’t want to share their preference, of which 0.3% gave security as the reason for this. On top of this there was 1.9% insisting on payment which Afroopinion doesn’t do. (Afroopinion was also not paid to do this research)
Interesting about the Whatsapp methodology is that it allows us to record and make transparent why people choose a certain candidate or why they voted Odinga in 2017 but not now.
It should be noted that the scores of the candidates with other survey companies don’t add up to 100%. There is the habit, with which we disagree and break, to include ‘Undecided’ and ‘Non-response’ as separate categories in the addition to 100%. Undecided and Non-response are not possible outcomes listed on the official results of an election. We think it’s the task of survey companies to estimate as well as possible what percentage of the voters will vote for a certain candidate. If they suffer from non-response, then they have to review their procedures and get those non-response percentages down.
Non-response is also abused by some survey companies in other countries to justify surveys that totally wrongly predicted electoral preferences. Having done our first survey in Uganda, we recall how Afrobarometer, two days ahead of the 14 January 2021 elections, predicted a 3 to 1 win for general Museveni versus opposition leader Bobi Wine. When confronted with the official results, that are widely considered to be rigged, the win was not even 2 to 1, the scientists at Afrobarometer blamed it on the non-response of 21%. In this way surveys become non falsifiable, which essentially means that according to the famous Karl Popper theorem, they are not scientific.
It’s unfortunate that in dictatorships like Uganda and Zimbabwe there are very little serious surveys, and that’s why we created Afroopinion. When you can’t trust the officially announced results, you need a reliable reference. You need a survey that the government cannot influence. In her book “Another Fine Mess”, Helen Epstein describes how the Uganda government controls surveys. The surveys by Afroopinion are done through Whatsapp. Whatsapp conversations are encrypted end-to-end and the Afroopinion surveyors are outside of Africa, making it very hard to impossible for African governments to influence the results. Even in Kenya some respondents have fear for the consequence of sharing their opinion, although by far not es much as in Uganda & Zimbabwe.
Kenya is definitely ahead of Uganda and Zimbabwe when it comes to democracy, but that doesn’t mean Kenyans trust their democratic process. We actually asked Kenyan voters if they think that the IEBC will organize free & fair elections. 40% of our respondents have no trust in the IEBC. Among Odonga voters even a majority of 60% have that opinion, whereas only 18% of Ruto voters distrust the IEBC.
Demographical corrections
It’s common in field surveys that the locations are chosen such that a representative sample of the population is constructed. In the same way equal amounts of women and men are allowed to the poll. In the WANGI methodology we invite someone to participate in a survey without knowing in advance the residence or gender of this person. The sample may therefore not be representative of the full population and the raw results shown in the tables below have to be corrected for that.
Odinga
Ruto
Other
Raw sample
50.2%
43.6%
6.4%
Male
52.1%
41.8%
6.1%
Female
43.6%
50.9%
5.5%
Kenya, gender weighted
47.9%
46.4%
5.7%
In the same way Whatsapp penetration and willingness to participate is not equal per region on Kenya. We get twice as many response from Nairobi than to be expected on the amount of registered voters. In Northern, we get twice as little. Since the Odinga & Ruto supporters are strongly concentrated in certain zones weighing of the zone results is important.
Odinga
Ruto
Other
Raw sample
50.2%
43.6
6.4%
Nairobi
68%
27%
5%
Central Rift
16%
73%
11%
Kenya, region weighted
49.0%
45.6%
5.4%
Our sample is not big enough to apply gender and zone correction at the same time. As regional differences are much bigger than gender differences, we have chosen to apply regional weights. Since all the results give Odinga as being in the lead and the difference between the two corrections is only 1% the effect of this choice is not drastic.
Looking back at 2017
We also asked our respondents how they voted in 2017. We found the following:
In the official results Kenyatta got 54%. There were multiple counties such as Nyandarua, Nyeri, Kirinyaga and Murang’a where he scored approximately 99%. In some of these counties our respondents tell us to have voted Odinga, sufficiently many to conclude that fraud did take place in these counties.
There are also some suspicious Odinga wins with 99% in Siaya and Home Bay. We need to increase our sample to proof that fraud took place there as well.
Methodology
Interviews
23-31 July 2022
Geography
Since Whatsapp penetration is not equal across Kenya, ‘zone’ results are weighted by the number of registered voters.
Data collection
Randomly constructed Whatsapp numbers with Kenyan mobile operators
Sample size
450 WhatsApp users
Statistical error
4%
Client
None
We asked the following questions:
May we know your age?
Gender (male/female)?
Which county of Kenya do you vote in?
Who did you vote for in the election of 8 August 2017?
In the next elections, will you vote for Raila Odinga, William Ruto, David Waihiga Mwaure or George Wajackoyah?
Does IEBC organize free and fair elections? (yes/no)
The following 9 geographical zones have been used. There are the same as the ones used by TIFA.
Last October we started to survey the Ugandan voters through Whatsapp. Since nomination day we have been going back to these voters and ask how they feel now.
We find that the support for Museveni has reduced from 34% in October to 28% in November (after nomination day but before the arrest of Kyagulanyi). The p-value of this research is 0.003, meaning it is scientifically significant. There were 300 respondents.
Where do these voters go to? We see that 2% of them are going to Kyagulanyi, he grows from 61% to 63%. But we also see the other opposition candidates, Muntu, Amuriat & Kabuleta going up.
We have also registered the home district of the voters, allowing us to aggregate according to the various regions. We find that Museveni still has the majority in the West, but hardly any support in Central and the East.
Results per Region (copyright: FreedomUganda.com)
Lastly, we looked at the age of the voters. Mainly younger voters (below 35) choose Kyagulanyi.
These surveys that FreedomUganda.com is performing monthly are separate from our human rights activities, but sometimes they coincide. One of our respondents told us he was himself victim of police brutality.
The survey was discussed in our “Democracy in Uganda” program with Nic Cheeseman.
