Quick question, when comparing between 2 groups ---> before and after treatment in test & control groups, I understand that I should use unpaired t-test. In the graph I present the data as log2 ratio to visualize the increase/decrease following treatment. But for statistical analysis, should I use mean of log2 ratio or mean of actual ratios since when I tried both give different p-values? Which one is more appropriate? Thank you.

Submit your paper to J Biol Methods today!

# T-test: ratios or log ratios

Started by ctfazra, Apr 30 2012 07:03 AM

4 replies to this topic

### #1

Posted 30 April 2012 - 07:03 AM

### #2

Posted 30 April 2012 - 09:29 AM

Use your raw data or transformed raw data (if no normal distribution) but not the means.

One must presume that long and short arguments contribute to the same end. - Epicurus

...except casandra's that belong to the funniest, most interesting and imaginative (or over-imaginative?) ones, I suppose.

...except casandra's that belong to the funniest, most interesting and imaginative (or over-imaginative?) ones, I suppose.

### #3

Posted 01 May 2012 - 07:53 AM

but what if i want to compare the response (ratio) between these 2 groups? Unpaired t-test will compare means (correct me if I'm wrong) so I wonder if it's appropriate to use transformed data (log ratio instead of actual ratio) to get the mean?

### #4

Posted 01 May 2012 - 08:19 AM

yes it does, but your post sounded as if you just want to use the means as input for the test.

Anyway the before-after treatments are paired data and I'd test them either separately using raw data with a paired t-test or using ratios with the ratio paired t-test as offered by some software (e.g. graphpad, prism). The normal student's t-test I would not use for ratio data as they're not normal distributed. If not possible to use the ratio test, use the non-parametric alternative the Mann-Whitney U Test...

Anyway the before-after treatments are paired data and I'd test them either separately using raw data with a paired t-test or using ratios with the ratio paired t-test as offered by some software (e.g. graphpad, prism). The normal student's t-test I would not use for ratio data as they're not normal distributed. If not possible to use the ratio test, use the non-parametric alternative the Mann-Whitney U Test...

One must presume that long and short arguments contribute to the same end. - Epicurus

...except casandra's that belong to the funniest, most interesting and imaginative (or over-imaginative?) ones, I suppose.

...except casandra's that belong to the funniest, most interesting and imaginative (or over-imaginative?) ones, I suppose.

### #5

Posted 02 May 2012 - 01:18 AM

I understand now. Thanks!

btw, sorry for the misunderstanding. non-native speaker's here

btw, sorry for the misunderstanding. non-native speaker's here