Comments on: One outlier and you’re out: Influential data and racial prejudice

By: Rense Nieuwenhuis

Rense Nieuwenhuis — Thu, 18 Jun 2009 09:00:38 +0000

Ah, yes, I see what you mean. I misread your previous comment, and somehow thought you argued for how many observations should have been ‘added’ to the data to overthrow the findings. Now that I reread your comment, you’re absolutely right. Thanks!

By: Neuroskeptic

Neuroskeptic — Thu, 18 Jun 2009 08:58:07 +0000

It wouldn’t always be one. I mean if your data were, Group A: 1,2,3,4,5 and Group B: 10,10,10,10,10, deleting any one data point wouldn’t have any impact on the between-group difference. Whereas if it was 1,2,8,8,8 vs. 10,10,10,10,10, deleting two of the data points (1 and 2) would have a big effect.

By: Rense Nieuwenhuis

Rense Nieuwenhuis — Wed, 17 Jun 2009 11:34:21 +0000

Thank you for your comment, Neuroskeptic. Indeed, articles reporting on how stable their findings are, are much to be preferred. However, the measure you propose, would also be exactly 1: a single observation can topple over any research findings, if it is influential enough.

By: Neuroskeptic

Neuroskeptic — Wed, 17 Jun 2009 08:20:45 +0000

Fascinating –
I’ve often thought that there should be a statistic quoted alongside the p statistic (and effect size) meaning “The minimum number of data points that would have to be deleted to reduce the effect to non significance”.