How to correct p-value in statistical analysis?

31 visualizaciones (últimos 30 días)
Kyle
Kyle el 14 de Ag. de 2014
Comentada: Ben11 el 15 de Ag. de 2014
Assume that I want to compare brain states A (e.g., wake) and B (e.g. sleep). To do that, I first allocated 10,000 brain sites (locations on the scalp) of my interest. Then at each state, at each brain site, I measured 5 data. The 10,000 brain sites are the same locations for both experiments A and B. So in a summary:
A has
A1 = [a1_1 a1_2 a1_3 a1_4 a1_5]; % 5 data recorded at brain site #1
A2 = [a2_1 a2_2 a2_3 a2_4 a2_5]; % 5 data recorded at brain site #2
A3 = [a3_1 a3_2 a3_3 a3_4 a3_5];
...
A10000 = [a10000_1 a10000_2 a10000_3 a10000_4 a10000_5];
B has
B1 = [b1_1 b1_2 b1_3 b1_4 b1_5]; % 5 data recorded at brain site #1
B2 = [b2_1 b2_2 b2_3 b2_4 b2_5]; % 5 data recorded at brain site #2
B3 = [b3_1 b3_2 b3_3 b3_4 b3_5];
...
B10000 = [b10000_1 b10000_2 b10000_3 b10000_4 b10000_5];
in which ai_j and bi_j are numbers ranging from 0 to 1.
Then I compared statistical difference (p-value) between A and B over all brain sites:
p1 = A1 vs B1;
p2 = A2 vs B2;
p3 = A3 vs B3;
...
p1000 = A10000 vs B10000;
I'm wondering
  • if I should correct these p values via multiple comparison test? If so, how should I do it in Matlab? Any method better than bonferroni correction? because bonferroni correction is too conservative.
  • how I can tell if brain state A is significantly different with B?
  2 comentarios
Ben11
Ben11 el 14 de Ag. de 2014
what test do you use to compute the statistical difference?
Kyle
Kyle el 14 de Ag. de 2014
I used Wilcoxon rank sum test to calculate the p-value:
p = ranksum(x,y)
since this method doesn't need a specific data distribution (I just have 5 data points for each x and y).

Iniciar sesión para comentar.

Respuesta aceptada

Ben11
Ben11 el 14 de Ag. de 2014
With the Statistics Toolbox you want to use mutcompare for your statistical tests. There are quite a few options for corrections. Other than Bonferroni, you might want to use the "dunn-sidak" test, which is less conservative than Bonferroni.
Are you comparing an experimental group with a control group for example or 2 experimental groups?
  2 comentarios
Kyle
Kyle el 14 de Ag. de 2014
Hi Ben11,
Thank you for the comments. Could you please specify how to get the corrected p value from mutcompare() function? Seems like it doesn't compute p-value? I'm comparing two experimental groups. The experiment background has been added in my post.
Cheers Kyle
Ben11
Ben11 el 15 de Ag. de 2014
Hi Kyle,
I've been digging a bit and it seems that we can't get the corrected p-value using multcompare in Matlab, at least in a somewhat easy/straightforward way.
Can you use another software like Prism by Graphpad? It's quite user friendly and versatile. There's a 30 day trial you can get here .
Sorry if I could not be of more help.

Iniciar sesión para comentar.

Más respuestas (0)

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

Translated by