The Digital Poll: How Online Predictions Compared to Actual 2022 Results

In the months before August 9, 2022, Kenyans were drowning in predictions. Traditional pollsters published monthly surveys. Twitter was running dozens of unofficial polls daily. Google Trends tracked search interest. Betting platforms offered live odds. Each painted a different picture of the race. And when the actual results came in — Ruto 50.49%, Odinga 48.85% — the scorecard was revealing.

Here's how each prediction method performed, what they got right, and why most got it wrong.

Traditional Polls: Mostly Right, Widely Distrusted

Kenya's established polling firms — TIFA Research, Infotrak, Mizani Africa, and Radio Africa's Ipsos partnership — published regular surveys throughout 2022. Their track record:

• TIFA Research (final poll, late July): Ruto 47%, Odinga 42%, undecided 8%. Verdict: Directionally correct — Ruto ahead, but underestimated both candidates' final shares.
• Infotrak (final poll, early August): Ruto 49%, Odinga 43%. Verdict: Close on Ruto, underestimated Odinga by 6 points.
• Mizani Africa (final poll): Ruto 50.2%, Odinga 44.1%. Verdict: Best traditional poll — almost nailed Ruto but missed Odinga by 5 points.

The consistent error: every traditional poll underestimated Odinga's final share. This likely reflects the "shy Odinga voter" phenomenon — respondents in mixed or UDA-leaning areas who privately intended to vote for Odinga but didn't reveal this to pollsters. It may also reflect late-deciding voters who broke toward Odinga in the final days.

On balance, traditional polls correctly predicted the winner and roughly captured the margin. But their 3-6 point errors on Odinga's share were significant in a race this tight.

Twitter Polls: Spectacularly Wrong

Twitter (now X) was the most active political discussion platform during the 2022 campaign. Dozens of unofficial polls ran daily, and they almost universally showed Odinga winning — often by 55-65%.

Why was Twitter so wrong?

• Demographic skew: Twitter's Kenyan user base is disproportionately young, urban, and educated — demographics that leaned Odinga in 2022. Rural voters and older voters (who leaned Ruto) are barely represented on the platform.
• Bot and sock-puppet activity: Both campaigns deployed digital operatives who created fake accounts to manipulate online polls. The Digital Forensics Lab estimated that 15-25% of politically active Kenyan Twitter accounts during the election period showed bot-like behaviour.
• Self-selection bias: People who engage in online polls are not representative of the electorate. They are more politically engaged, more urban, and more likely to support the opposition (a global pattern).
• Echo chambers: Twitter's algorithm serves content that reinforces existing beliefs. Odinga supporters saw overwhelming pro-Odinga sentiment and concluded it reflected reality. It didn't.

Twitter polls predicted Odinga would win by 20+ points. The actual margin was 1.63 points in Ruto's favour. That's a 22+ point error. As a prediction tool, Twitter was worse than a coin flip.

Google Trends: Surprisingly Useful

Google search trends offered a different — and more interesting — signal. In the weeks before the election, search interest data showed:

• "William Ruto" search volume consistently exceeded "Raila Odinga" by 15-30% nationwide
• Ruto's search dominance was most pronounced in the Rift Valley and Mt. Kenya regions — his core vote areas
• Odinga's search interest spiked during rallies and campaign events but didn't sustain between events
• The search term "hustler fund" (Ruto's signature campaign promise) consistently outperformed "azimio" as a search term

Google Trends doesn't directly predict vote share, but the correlation between search interest and actual results was notable. Counties where Ruto's search volume dominated were counties where he won decisively. The platform captured something that Twitter didn't: genuine grassroots interest as opposed to orchestrated online activity.

Betting Markets: The Most Accurate Predictor

Kenya's sports betting platforms — Betway, SportPesa, and others — offered odds on the presidential race. While political betting is a grey area legally, the markets were active:

• Final betting odds (August 8): Ruto at approximately 1.55 (implied probability: ~64%), Odinga at approximately 2.40 (implied probability: ~42%)
• Implied prediction: Ruto likely winner, but not a certainty

Betting markets were the most accurate public prediction mechanism in 2022. They correctly identified Ruto as the favourite, and their implied probability range (~60-65% chance of Ruto victory) was reasonable for a race that ended at 50.49-48.85.

Why do betting markets outperform polls and social media? Because people put money where their mouths are. A bettor risking KES 1,000 on the outcome has a financial incentive to be accurate, unlike a Twitter user posting for likes or a poll respondent who may dissemble.

The Information Hierarchy

Based on 2022 performance, here's the accuracy ranking of Kenyan election prediction methods:

1. Betting markets — correctly predicted winner with reasonable confidence levels
2. Traditional polls (Mizani, Infotrak, TIFA) — correctly predicted winner, directionally right on margin
3. Google Trends — correctly indicated Ruto's broader grassroots interest
4. Facebook engagement — mixed signals, slightly better than Twitter
5. Twitter polls — spectacularly wrong, predicted the opposite outcome

Lessons for 2027

The 2022 experience offers clear lessons for consumers of election information in 2027:

• Ignore Twitter/X polls entirely. They are not just inaccurate — they are systematically misleading.
• Traditional polls are directionally useful but expect 3-5 point errors. Average multiple polls rather than trusting any single one.
• Watch betting markets — they aggregate the wisdom of people with skin in the game.
• The best prediction tool is the actual count. Parallel tallying from polling station data — the kind that Votrack provides — replaces prediction with measurement.

Stop predicting. Start tracking. Votrack replaces polls, predictions, and Twitter speculation with real results from real polling stations. Book your demo today.