TBLP and Correlation Statistics Explained

The image on the left demonstrates the rollup statistics for eTBLP. It shows the total number of figures scored, how many of those figure scores eTBLP replaced with the average from all of the judges, and how many of those figures eTBLP adjusted toward the average. For the adjusted figures, the rollup shows the average weight of the adjustment. The clipped count shows how many times the scores for a figure were so far in agreement that the standard deviation had to be clipped to a lower bound of 0.03 times the average.

The figure statistics come from the TBLP Phase I calculation.

The judges statistics after the figure statistics show the same values for the TBLP Phase II calculation.

The image on the left demonstrates the rank correlation metrics given by the contest list and individual contest pages. These compare the rankings for both individual flights and for overall contest categories.

The first column lists possible pairings of the ranking methods, for example, Mean and TBLP. The next column shows the number of times the paired ranking methods came up with pilot ranks that were statistically correlated. That means they are closer than random chance would have them.

The columns after show what we really care about. The column with a one over it shows the number of times the paired methods agreed on the first place winner. The column with a two over it shows agreement on the second place pilot. The column with a three over it shows third place agreement.

The image on the left demonstrates the correlation shown for judges on each of the flight pages. This shows a 'T' if the individual judge's ranking of the pilots correlated with the ranking computed by each of the three ranking methods. We applied the Spearman Rank Coefficient to determine correlation.