I performed a two-tailed t test with Welch correction on a treated cell versus the untreated control. As shown in the figure attached, there is almost a 5-log reduction, but the t test gave a p value of 0.15. But it just doesn't make sense.
Really appreciate any help/suggestion on making sense of this.
Did you perform your test on raw or log transformed data?
Statistical tests are about confidence, not magnitude, and thus a really, really obvious, really really dramatic drop like this might not be statistically significant when assessed using raw data because the untreated values inherently vary by over an order of magnitude. It doesn't matter that the treated values are multiple orders of magnitude lower.
If you log transform your data, it puts everything on a range that is statistically tractable.
Attached is a very rough and ready excel example of this: values of 1000000-50000000 vs values of 100-500 does not reach significance, while the log transformed values (same exact ttest) do.
Thanks John Hildyard! That is spot on. I was using untreated raw values for the t test. Once I used log-transformed values, it is now statistically significant.
- Badges
- Science topic
- Similar topics
- Correction