Thread by @mezbeloo

I really commend @StearsInsights for creating a data-driven model to predict the results of Nigeria's 2023 presidential election. Having said that, I think most people are misinterpreting the polling data and distorting expectations. So here's a 🧵with some critiques 1/X

https://twitter.com/StearsInsights/status/1622860212188364800

A good election model answers two questions: (1) who do people support and (2) who's likely to turn out on the day and actually vote? They've done a reasonably good job on the first but a poorer job on the second 2/X

Let's start with 1). @StearsInsights indicates that 40%(!!) of the people polled are unwilling to share their preferred candidate. They try to model these "silent" voters by matching them to "vocal" voters with similar demographics and then assuming they will vote similarly. 3/X

If the non-response rate seemed to be roughly evenly distributed around the country, then I wouldn't be concerned. Perhaps it just reflects some discomfort with polling methods: most Nigerians aren't used to random numbers calling them and asking who they plan to vote for 4/X

But the non-response rate is skewed; it's high in the North-West and South-West and low in the South-East+North-East. To me, this suggests that there's something more substantive driving people's decision not to answer. I'm not sure what that is 5/X

The @StearsInsights theory is that @officialABAT voters are especially likely to be silent bc of the skeletons in his closet. I'm baffled. Nigerians openly support known coup-plotters, kidnappers, thieves, rapists and ne'er do wells all the time. Since when do we have shame? 6/X

I appreciate that it's somewhat stigmatized to express support for Tinubu on social media. But do people sincerely believe that stigma applies in real life? Are the demographics of Nigerians that actually vote embarrassed to say they support Tinubu? Seems very unlikely to me 7/X

Social desirability bias is context specific. So you could imagine that people are silent not bc they're supporting a scandal-prone candidate but bc they plan to vote for a candidate who is unpopular *among their community* i.e. they are breaking against demographic trends 8/X

Until we know why people are silent, we should be cautious. I think the best overall interpretation is that we have a pretty high degree of confidence about how the South East and South-South will tip. Non-response rates are fairly low and the margin for @PeterObi is huge 9/X

But for every other region, we don't have great predictions. In the North East, @officialABAT and @atiku are polling within 2 points of each other. @StearsInsights isn't explicit on this but in most polls, that's within the margin of error. It could tip either way 10/X

In North-Central, the non-response rate is 38% but, more importantly, elections tend to be competitive. My rough metric for "competitiveness" is states where the winner secures less than 55% of the vote. Benue, Nasarawa, and Taraba were competitive in both '15 and '19 11/X

This is especially striking because there was obviously a major realignment between those elections. PDP were incumbents in 2015 but challengers in 2019. This shapes things like the relative ease of rigging, ability to influence @inecnigeria, ability to deploy violence etc. 12/X

In North-West, the non-response rate is high (45%) and @KwankwasoRM is a major player. The margin of support for him (8%) is larger than the gap between @atiku (18%) and @PeterObi (11%). If he decides to drop out and endorse someone, he could shift the dynamics significantly 13/X

In South West, non-response rates were high (38%) and there also seem to be some signs of disengagement or apathy. 7% of people are saying out right that they won't vote. And the South West's biggest city, Lagos, is consistently in the bottom 5 in terms of turnout rate 14/X

This is pretty much the best we can expect given the data available and the lack of certainty about "silent" voters. But people need to bear these limitations in mind. Let's turn now to question 2: who is likely to turn out and vote? 15/X

@StearsInsights have wisely modeled a number of turnout scenarios. Their high turnout scenario is based on the idea that everyone who (1) intends to vote, (2) is sure they will vote and (3) has their PVC will turnout on Election Day. 16/X

This is obviously unrealistic: the no. of people who say they will vote far exceeds those who actually vote. And in Nigeria there are constantly big gaps between the no. of people who register, the no. who pick up their PVCs, and the no. who actually vote on Election day 17/X

More realistic is their low turnout scenario. Here they limit the electorate to those who (1) intend to vote, (2) are sure they will vote, (3) have their PVC, (4) are certain about their candidate choice, (5) are confident about the process, and (6) feel safe voting 18/X

I'm not aware of any data that establishes that voters' responses to these questions are predictive of actually voting in the Nigerian context. But in the abstract, the questions make sense 19/X

What's more interesting is what the turnout prediction doesn't consider: Did you vote in the 2015 and 2019 elections? Generally, the best predictor of future behavior is past behavior. I'm surprised that they didn't weigh this heavily in the low-turnout scenario 20/X

I would also like to see questions that separate those who have a vague& general intention to vote and those who have specific plans. I would ask q's like"where is your polling unit?" "how do you plan to get there?" "what time do you plan to vote?" to try to gauge this 21/X

Overall, how can we assess whether the "low turnout" scenario is likely to reflect real turnout? We could check whether the model produces a 2023 electorate that's demographically similar to the electorate that actually voted in 2015 or 2019. The article doesn't do this 22/X

But let's imagine that you think the 2023 electorate is going to look really different from 2015 and 2019. This is essentially the #Obidient argument: his supporters think he will fundamentally reshape the electorate and win with a surge of support from SS/SE and the youth 23/X

Is there evidence of massive electoral mobilization among those demographics? I have doubts. There have been a lot of rallies, marches, and public expressions of support for Peter Obi. But attending a march is fairly low cost—you can make a last minute choice to go 24/X

Voting isn't like that. You need to plan ahead and deal with a lot of logistical hurdles to have your vote count. (By the way, I'm not blaming ordinary Nigerians for this failure. INEC has done a very poor job of managing the process and I think it should be easier to vote.) 25/X

Which states are showing signs of significant electoral mobilization? Not the ones you might expect. The states with the highest % increase in registered voters from '19 to '23 are: Kwara, Nasarawa, Kogi, FCT, Osun, Cross River, Bayelsa, Taraba, Delta, and Ogun 26/X

The North-Central dominates. Another angle is that registration rates reflect electoral administration as well as voter mobilization. We don't have good data on people who tried to register but were somehow prevented, either by INEC or by partisan interference. 27/X

What about youth voters? Much has been made of the big increase in registered voters, most of whom are in the 18-35 demographic. But people seem to be missing that the relative power of the youth vote has actually declined. 28/X

In 2019, young people were the majority—they made up 51.1% of registered voters. Now, they only constitute 39.65%, still a plurality but not a majority. The demographic that has actually seen the largest increase in relative vote power is 35-50 yr olds 29/X

#EndSARS was obviously a huge development in terms of young people's engagement in politics. But so far I'm not seeing that engagement translate into electoral mobilization. 30/X

To be fair, I don't have data on turnout rates broken out by age for prior elections. (If anyone else does I'd be curious to see it). My sense is that youth voters tend to turn out at very low rates. So even an improvement in turnout could still be a significant shift 31/X

Overall, my 5 takeaways are: (1) people making confident predictions about what will happen outside of SS/SE should be more cautious, (2) North-Central is the critical swing region, (3) we need much more research into this silent voter phenomenon 32/X

(4) We know surprisingly little about what predicts election day turnout and (5) because turnout is so critical we should prepare to document and challenge efforts to depress the vote (rigging, ballot destruction, violence etc.), even more so than normal 33/X

This is where structure comes in: parties need infrastructure to monitor every polling unit and be prepared to raise legal challenges. If this heads to a runoff (to me, the most optimistic scenario) we're in uncharted territory. I hope you all vote and vote your conscience /end