Beauchamp:anovan

Matlab ANOVAN
For performing sophisticated statistical analyses, Matlab's anovan is a useful function. Other options are dedicated stats packages like R, SPSS, etc.

These notes are adapted from /data1/UT/STP/STPNotes.rtf

The values can be read into Matlab from a text file or from Excel with the command correct = xlsread('tomatlab.xls')

We feed ANOVAN a 1-D vector which contains all of the data. To make sense of the vector, Matlab also needs other vectors which list the factor-level for each datum. There is a separate 1-D file for each factor.

For instance, if the data is input by subject with three data points (corresponding to stimulus types) for each subject, our data vector would be s1d1 s1d2 s1d3 s2d1 s2d2 s2d3 etc.

Then, our subject factor vector would be 1 1 1 2 2 2 etc. and out stimulus factor would be  1 2 3 1 2 3 etc.

Note that these can also be text labels (e.g. "Face","House"). We use numbers for ease of automation. These factor vectors can be automatically generated in Matlab with stimnum = 5 subjnum = 37

gsubj = reshape(meshgrid(1:1:subjnum,ones(1,stimnum)),stimnum*subjnum,1) gstim = repmat(meshgrid(1:1:stimnum,1),1,subjnum)'

In a more complex example, we have two different groups with different diagnosis. numgrp1 = 20 numgrp2 = 17 total1 = numgrp1 * stimnum; total2 = numgrp2 * stimnum; ggroup = [ 1*ones(total1,1)' 2*ones(total2,1)' ]'

Next, we can try to load in data to run the ANOVA on: cd /Volumes/data1/NIH/autism/point_light/AKGroup/TimeSeries load NCSTSv6.2D This contains the average timeseries from the STS in 17 NC subjects (group 2). load ASDSTSv6.2D This contains the average timeseries from the STS in 20 ASD subjects (group 1). We average the appropriate time points to get a single number for each stimulus:

clear alldata ctr = 1 for i=1:numgrp1 alldata(ctr:ctr+4) = [mean(ASDSTSv6(3:9,i)) mean(ASDSTSv6(16:22,i)) mean(ASDSTSv6(29:35,i)) mean(ASDSTSv6(42:48,i)) mean(ASDSTSv6(55:61,i))]' ctr = ctr + 5; end for i=1:numgrp2 alldata(ctr:ctr+4) = [mean(NCSTSv6(3:9,i)) mean(NCSTSv6(16:22,i)) mean(NCSTSv6(29:35,i)) mean(NCSTSv6(42:48,i)) mean(NCSTSv6(55:61,i))]' ctr = ctr + 5; end

resdata = alldata' ggroup = [ 1*ones(total1,1)' 2*ones(total2,1)' ]' gstim = repmat(meshgrid(1:1:stimnum,1),1,subjnum)' gsubj = reshape(meshgrid(1:1:subjnum,ones(1,stimnum)),stimnum*subjnum,1)

[p,table,stats] = anovan(resdata,{gsubj ggroup gstim},'model','interaction','random',1,'varnames',{'Subject','Diagnosis' 'StimType'});

[p,table,stats] = anovan(resdata,{ggroup gstim},'model','interaction','random',1,'varnames',{'Diagnosis' 'StimType'});

[p,table,stats] = anovan(resdata,{ggroup gstim},'model','interaction','varnames',{'Diagnosis' 'StimType'});

Include another stimulus type: stimnum = 6 subjnum = 37 numgrp1 = 20 numgrp2 = 17 total1 = numgrp1 * stimnum; total2 = numgrp2 * stimnum;

clear alldata ctr = 1 for i=1:numgrp1 alldata(ctr:ctr+5) = [mean(ASDSTSv6(3:9,i)) mean(ASDSTSv6(16:22,i)) mean(ASDSTSv6(29:35,i)) mean(ASDSTSv6(42:48,i)) mean(ASDSTSv6(55:61,i)) mean(ASDSTSv6(68:74,i))]' ctr = ctr + 6; end for i=1:numgrp2 alldata(ctr:ctr+5) = [mean(NCSTSv6(3:9,i)) mean(NCSTSv6(16:22,i)) mean(NCSTSv6(29:35,i)) mean(NCSTSv6(42:48,i)) mean(NCSTSv6(55:61,i)) mean(NCSTSv6(68:74,i))]' ctr = ctr + 6; end

resdata = alldata'; ggroup = [ 1*ones(total1,1)' 2*ones(total2,1)' ]' gstim = repmat(meshgrid(1:1:stimnum,1),1,subjnum)' gsubj = reshape(meshgrid(1:1:subjnum,ones(1,stimnum)),stimnum*subjnum,1) [p,table,stats] = anovan(resdata,{ggroup gstim},'model','interaction','varnames',{'Diagnosis' 'StimType'});

We can then use ttest2 to get probabilities for any given combination: If stimuli 1 are movies and 2 are pictures, then we can perform a t-test of movies vs. pictures: [h,p,ci,stats] = ttest2(alldata(gstim==1),alldata(gstim==2)) h =

1

p =

1.4378e-13

ci =

0.1230   0.1922

stats =

tstat: 9.0872 df: 72 sd: 0.0746

or, more complex: [h,p,ci,stats] = ttest2(alldata(gstim==1 & ggroup ==2),alldata(gstim==2 & ggroup == 2)) [h,p,ci,stats] = ttest2(alldata(gstim==4 & ggroup ==2),alldata(gstim==5 & ggroup == 2)) [h,p,ci,stats] = ttest2(alldata(gstim==4 & ggroup ==1),alldata(gstim==5 & ggroup == 1)) [h,p,ci,stats] = ttest2(alldata(gstim==4 & ggroup ==1),alldata(gstim==5 & ggroup == 1))

[h,p,ci,stats] = ttest2(alldata(gstim==1 & ggroup ==2),alldata(gstim==3 & ggroup == 2))