I am comparing temperature curves in two groups, n=8 in each group. I have measured fever at 29 time points after LPS stimuli, and want to test if one group respond different than the other. One group had a higher mean temperature at all time points during the study, but the difference was only significant at a few time points (multiple t-test, after confirming normality). Is there a appropriate statistical test for comparing the two temperature lines as a whole, and thereby get an overall p-value for a potential difference?
(I'm using GraphPad).
Yes, it is called a hierarchical (linear) model.
See: http://www.stat.columbia.edu/~gelman/arm/contents.pdf
Your test did not confirm normality. You only failed to reject the null hypothesis.
I would think that you could also do a stratified randomization test. At each time point the values are randomized. You then calculate two regression curves, find the difference between the regression coefficients, and repeat. Sort the differences, and see where the observed values are located relative to the random outcomes. I would suggest 10,000 randomizations. Eight values in each of two treatments is insufficient data for this type of test. However, with 29 evaluation points there are 373230 possible randomizations.
You may find this book useful.
http://www.stat.columbia.edu/~gelman/arm/contents.p
Good luck!
If I understand your problem correctly, you are stating that there are two groups with 8 subjects each whose fever temperatures are obtained at 29 different points. Did you verified that the initial temperatures (say mean temperatures) were comparable between the groups, if not, the initial temperature itself can cause the difference from the begining between the two groups. You can do analysis of variance with repeated measures (8). The result of ANOVA can give you over all differences being significant or not between the groups.
- Badges
- Science topic
- Similar topics
- Geoscience
- Atmospheric Sciences
- Climate Science
- Climate Change