red queen's race

[zwol]

A nice simple outlier rejection method — bin the data according to experimental condition, throw away everything more than two standard deviations from the bin mean — throws out fifteen data points and gives me four bad residuals.

A clever outlier rejection method — pcout from mvoutlier — throws out nineteen data points and gives me six bad residuals. It might work better if it would let me bin the data according to experimental condition, but it won't.

The even cleverer method — not actually throwing data away, just downweighting it (using pcout again), which would let me be rather more aggressive about counting data as dubious — does not work.

Time expended: six hours. Forward progress: zero. Grumpy emails sent to statistical software authors: three.