Denko

@tundeskie The thing behind the handle. A dunce through and through

@SirJarus The problem is your belief that your analysis is sacrosanct.

I called @SirJarus the GOAT of election forecasting in Nigeria. I insist he is but anyone who has a contrary opinion can disagree with data. A question that has bugged me for a while: how did @SirJarus's forecast for the 2023 Nigerian presidential election compare with forecasts from polls in the 2024 American presidential election? Data source: Source of data for the American forecasts came from @270toWin. They aggregated multiple pollsters and provided forecasts here: 270towin.com/2024-president… source of data for the Nigerian forecasts came from the post I quoted. My analysis evaluating the forecasts for both elections is reported here: docs.google.com/spreadsheets/d… Important to state that this is not an apple-to-apple comparison since it's two different elections in two different countries. We can however use the forecasts from the USA presidential election to establish a baseline. Aggregate Numerical Evaluation (Regression Analysis) American pollsters: Mean Average Error : 2.9130 @SirJarus : Mean Average Error: 9.8939 American pollsters: Root Mean Squared Error: 3.6899 @SirJarus : Root Mean Squared Error: 12.8962 Data for this evaluation is in the spreadsheet above. These numbers indicate that from state to state the American pollsters were more precise than @SirJarus. This could happen for multiple reasons. 1) The American pollsters have a more robust model 2) Nigerian elections are more susceptible to election day anomalies. @SirJarus's biggest misses were in Rivers and Akwa Ibom state. 3) @SirJarus may not have been optimizing for precision in his forecast and his intent was to get directional numbers to forecasts a winner. The Ultimate Test: Calling the Winner The primary purpose of an election forecast is to answer one question: Who will win? The American Baseline: Entering the 2024 election, the consensus of American pollsters pointed toward a Kamala Harris victory. Their national aggregate predicted Harris at 48.4% and Donald Trump at 47.2%. Ultimately, Trump won both the popular vote and the Electoral College, rendering the "consensus" prediction incorrect. The @SirJarus Performance: Operating in a far more volatile three-way race, @SirJarus predicted BAT would win with 38% of the vote. Tinubu won with 36.61%. Sir Jarus passed the ultimate test which the American pollsters failed. The Complexity Factor: Binary vs. Three-Way Races To understand why @SirJarus feat is impressive, one must look at the difficulty of the task. The U.S. election was essentially a binary choice. Predicting a winner in a two-person race is statistically easier than predicting a three-horse race like Nigeria’s 2023 election (BAT vs. Atiku vs. Obi). Final verdict: The USA pollsters presented better finer-grained estimates but failed to interpret major demographic shifts and missed the mark. @SirJarus's fine-grained (state level) estimates were less precise but he was better at interpreting political momentum and structural shifts and ultimately predicting the winner.