Dropout Derby update. As of 3 August 2019 with Democratic candidates 20 and 24 having bailed, the following is the cumulative score tally (bottom row). The number/name key is found in bottom figure.
[Addendum] One of the participants requested that I reprise the scoring algo with examples for a couple of the participants. The indicated ‘CumScore’ line contains the updated cumulative scores of the participants that account for all the dropouts to date – currently #20, then #24 as shown in the ‘Actual’ column. CumScore is the sum of the individual scores for each predicted dropout candidate.
A perfect prediction occurs when the participant’s dropout location matches the actual location in the sequence of dropouts. When there is a difference between a candidate’s predicted location and the actual location, then the prediction error is the (absolute value of the) difference between the two locations. The maximum possible difference value with 26 candidates is 25. So if a participant missed it by, say, 5 = predicted position minus actual position, then the relative error is 5/25 = 0.2, and the score for that prediction is the complement, 1 – |5/25| = 0.8. This yields a perfect score of 1.0 if the two positions match, and 1 – |25/25| = 0 if they maximally mismatch. CumSum is then the cumulative sum of all the participant’s individual candidate scores.
As examples, we’ll calculate the indicated CumSum scores for EstFox (1.40) and WaltB (1.24). For EstFox the first dropout difference is |7-1| = 6. Therefore 1 – 6/25 = 0.76. The second dropout yields a difference of |11-2| = 9, giving the candidate score of 1 – 9/25 = 0.64. The sum of these individual scores is the indicated 1.40 = 0.76 + 0.64. For WaltB we have the first score as 1 – |19-1|/25 = 0.28, and the second as 1 – |1-2|/25 = 0.96, for a total of 1.24 = 0.28 + 0.96. A perfect score at this point in the dropout sequence would be 2.0. This says that at the end of the competition when the last candidate ‘drops out’ to be the party’s nominee, a perfect score would be 26.
Along the same vein, we can also compute the ‘normalized score’ at every dropout in the sequence by dividing the participant’s CumScore by the perfect CumScore to date. The perfect normalized score would then always be 1.0 as the competition proceeds. Using the above examples, the normalized scores would be for EstFox 1.40/2.0 = 0.70, and for WaltB 1.24/2.0 = 0.62. The above table has been updated and corrected to show both the CumScore and NormScore for each participant. (H/T to commenter who noticed error in calculating the CumScore of participants who did not update their dropout lists when the last three Dem candidates entered the race. Such missing entries will receive a zero score for the dropped out candidate.)
rebanesruminations.com 
Leave a comment