Everyone is aware of the IREV disaster of the Presidential Election which has made people doubt the veracity of results uploaded on there. However, people still say that Tinubu did not even win if we were to add up all the results on IREV. I learnt someone even tried to do so but it was marred with controversies in of itself. The result he finally posted was almost Meaningless. So I decided to do some the collation myself. But because I'm doing it alone, I cannot download and collate over 160,000 polling unit results. That's almost impossible. What I would do is sampling which is a very scientific way to analyse results.

I intend to make sure the sampling gives every polling units equal chance of being selected. It would also take into cognisance, the various geopolitical zones. So that means I would use simple random sampling and also multistage sampling. Sample size would be calculated using an online sample size calculator. The confidence interval will be 95%, margin of error will be 5%. The Population Size would be the total number of Polling Unit Results that has been uploaded into IREV on the day I intend doing the collation and analysis (hopefully tomorrow). There are currently 166,468 results uploaded, that number might increase tomorrow. The proportion will be the percent of registered voters that participated in the election.

Polling Units will be selected from each of the States. The number for each States will be proportional to the number of polling units each States has so that States like Lagos will have the highest number of polling units in the Sample and FCT will have the least number. Multistage sampling will be done on the LGAs within the States and the Wards within the LGAs and the Polling Units within the Wards.

IREV will be the source data. The sampling list will also be on IREV. In a situation where a polling unit result is not clear or unusable, another simple random sampling will be done within the Ward to select another Polling Unit. This will continue until a usable Polling Unit Result is gotten from that Ward.

The collated result will be posted on here. The link to the Spreadsheet used for the analysis will also be provided here.

I decided to post this here so this my intended protocol will be critiqued. Posted this here to eliminate the possibility of bias.

Rubbish methodology for an election.. the concept of an election forbids sampling... the fact that you sample means you cant be certain that whoever comes first from your sampling is the winner...
It is a moot point though, since somebody already tallied everything and posted it on SM

Use MTN Night data and download everything Sir.

Doing a sampling of already uploaded results will not give you the true picture.

If you want to do a good job of this, go ahead and do it well.

Already doomed to be biased from the start point. Take Lagos for example, the turnout was 19% compared to Jigawa which had 42%.

You have 94% of the results on ireV. It might be time consuming but not impossible to tally.

A task as collation of election result has to be thorough to make any sense. Also, it cannot not be done by an individual as there is doubt due to bias. Get a team of 36(or 18) people and divide the states amongst them.

aieromon:
Already doomed to be biased from the start point. Take Lagos for example, the turnout was 19% compared to Jigawa which had 42%.

You have 94% of the results on ireV. It might be time consuming but not impossible to tally.

I know you to be very versed so I will engage you.

Why do you think its biased? The sampling is based on those who voted and the results are on IREV. Even though Lagos had a lower turn out in terms of percentage however Lagos still have more voters in terms of numbers.

Obinoscopy:
I know you to be very versed so I will engage you.

Why do you think its biased? The sampling is based on those who voted and the results are on IREV. Even though Lagos had a lower turn out in terms of percentage however Lagos still have more voters in terms of numbers.

1. If there was 95% turnout, allocating more Lagos PUs in the sample would have been ideal. However, this is not reliable when we look at the actual turnout:-
Lagos - 13,325 PUs vs 1,347,152 accredited voters ~ 101 votes per PU.
Jigawa - 4,522 PUs vs 961,670 accredited voters ~ 213 votes per PU.
I use accredited voter data because some of the uploaded EC8A result sheets were cancelled at the Ward/LGA collation centres for various reasons. What stops Analyst B from pulling more units from Jigawa into his/her sample?

2. For states where the candidate won all the LGAs, bias will be minimised. It becomes complicated when attempting to sample PUs from states where the candidates shared LGA wins. Take for example Kaduna results:
PDP - 554,360 in 14 LGAs
APC - 399,293 in 2 LGAs
LP - 294,494 in 7 LGAs

aieromon:


1. If there was 95% turnout, allocating more Lagos PUs in the sample would have been ideal. However, this is not reliable when we look at the actual turnout:-
Lagos - 13,325 PUs vs 1,347,152 accredited voters ~ 101 votes per PU.
Jigawa - 4,522 PUs vs 961,670 accredited voters ~ 213 votes per PU.
I use accredited voter data because some of the uploaded EC8A result sheets were cancelled at the Ward/LGA collation centres for various reasons. What stops Analyst B from pulling more units from Jigawa into his/her sample?

I still don't understand your point. This is simply proportional sampling which is a very common statistical principle.

Let's use the instance you provided ie Lagos and Jigawa. Since Lagos has 13,325 PUs and Jigawa has 4,522 PUs it means my sample should have about 3 times more PUs from Lagos. In actuality, my Sample will have 24 PUs from Lagos and 8 PUs from Jigawa. Now using your estimation, 101 voters/PUs in Lagos will mean my sample will have data for 101 × 24 voters which gives 2,424. Similarly, for Jigawa, my sample should have 213 × 8 voters which gives 1,704. So Lagos data will have 2,424 voters while Jigawa data will have 1,704. Is this not comparable to the overall accredited data you just provided?

2. For states where the candidate won all the LGAs, bias will be minimised. It becomes complicated when attempting to sample PUs from states where the candidates shared LGA wins. Take for example Kaduna results:
PDP - 554,360 in 14 LGAs
APC - 399,293 in 2 LGAs
LP - 294,494 in 7 LGAs

That's why I intend selecting the LGAs based on Simple Random Sampling. That gives every LGAs equal chance of being selected. The beauty of Random Sampling is that it eliminates Selection Bias. The Central Limit Theorem in Statistics is the basis for Statistical Deductions from Sampling. I recommend you read that theorem.

Cool

Obinoscopy:
I still don't understand your point. This is simply proportional sampling which is a very common statistical principle.

Let's use the instance you provided ie Lagos and Jigawa. Since Lagos has 13,325 PUs and Jigawa has 4,522 PUs it means my sample should have about 3 times more PUs from Lagos. In actuality, my Sample will have 24 PUs from Lagos and 8 PUs from Jigawa. Now using your estimation, 101 voters/PUs in Lagos will mean my sample will have data for 101 × 24 voters which gives 2,424. Similarly, for Jigawa, my sample should have 213 × 8 voters which gives 1,704. So Lagos data will have 2,424 voters while Jigawa data will have 1,704. Is this not comparable to the overall accredited data you just provided?

That's why I intend selecting the LGAs based on Simple Random Sampling. That gives every LGAs equal chance of being selected. The beauty of Random Sampling is that it eliminates Selection Bias. The Central Limit Theorem in Statistics is the basis for Statistical Deductions from Sampling. I recommend you read that theorem.

u seem to be deeply rooted in statistical analysis.i go like tap from ur wealth of knowledge and experience if opportuned.Thanks

So I have concluded the analysis of the presidential result as a whole and have even created a thread for the result. Below is the excerpt and link to the result thread:

Obinoscopy:
As a statistician, I've always believed in the powers of deductive inference based on sampling.

INEC has uploaded over 94% of the Presidential Results on the IREV Portal. Now there are people who believe that the result announced by INEC is significantly different from the result uploaded on IREV. I decided to randomly select result sheets from IREV and collate. The sample size was 314 result sheets. That sample size was gotten from an online sample size calculator. Click on this link to see my research protocol: https://www.nairaland.com/7614305/statistical-analysis-inec-result-irev

The result is very interesting.

RESULT
So after collating the result sheets from randomly selected 314 polling units, APC got a total of 15394 votes. LP got 10430 votes. NNPP got 2613 votes. And PDP got 11533 votes. So APC got the majority of votes followed by the PDP, then LP and finally NNPP.

This means INEC declaration is in tune with what's uploaded on IREV. However I noticed that the result for Rivers State is very strange. My collated result shows LP has more than 4 times the votes gotten by either APC or PDP.

So in conclusion, the IREV result is almost exactly what INEC declared. The only discrepancy is that of Rivers State. I might have to inspect that of Rivers State even further in my next analysis. But despite the Rivers State discrepancy, APC won fair and square based on the results on IREV Portal. I say this with 95% confidence.

Whether the results from IREV is authentic or not is a matter for another discuss. My focus was just to use the results I saw on the IREV Portal.

https://www.nairaland.com/7626424/inec-result-analysis-based-samples#122034496