One of my participants has rediculously long response times. Response time is a dependent measure in my study. With the participant's data (using 1 way ANOVA) I get a p value of .07. Without it I get a p value of .03. So, it's a matter of significant results or not. I only have 27 participants and can't run anymore. Is there a statistical technique to eliminate/reduce the effects of his rediculously long RT? Will this also affect the effect size?
2007-03-24 19:04:43 · 2 answers · asked by Henry B. 3 in Science & Mathematics ➔ Mathematics
The first thing to consider is if the outlier is a recording error.
If you have good reason to think this is likely you are justified in removing it from the dataset.
If it is just a random large number you must keep it in.
In this case you could possibly use a nonparametric test instead of anova as these are less affected by outliers. However, they are not as powerful as parametric tests so you are less likely to detect a significant result when there is one.
2007-03-25 03:51:01 · answer #1 · answered by statstastic 2 · 0⤊ 0⤋
Throw out that participant's data and explain why?
2007-03-24 19:09:42 · answer #2 · answered by jamisojo 3 · 0⤊ 0⤋