Multiple univariate or multivariate analysis?

Hi everyone!

I have a dataset concerning the effect of food group on different dependent variables.

I want to test the hypothesis that the group does not affect each of the measurements (attached is the excel file).

Is the multiple univariate (one way anova for each measurement) the correct way to go?

Thanks,

Giulia

Q3 = xlsread('Q3.xls');
Diet = categorical(Q3(:, 2));
BW = Q3(:, 3);
HW = Q3(:, 4);
LW = Q3(:, 5);
KW = Q3(:, 6);
SW = Q3(:, 7);
CERUL = Q3(:, 8);

[p,tbl,stats] = anova1(BW, Diet);
figure();
cBW = multcompare(stats', 'Alpha',0.05)

[p,tbl,stats] = anova1(HW, Diet);

[p,tbl,stats] = anova1(LW, Diet);
figure();
cLW = multcompare(stats', 'Alpha',0.05)

[p,tbl,stats] = anova1(KW, Diet);

[p,tbl,stats] = anova1(SW, Diet);

[p,tbl,stats] = anova1(CERUL, Diet);
figure();
cCERUL = multcompare(stats', 'Alpha',0.05)
  
  
  
 

Answer 1

[Jeff Miller]

No, that's probably not the right way to go.

The problem with this approach is that you have a 5% chance of making a type 1 error at each comparison, so the overall type 1 error rate across all measurements is more than 5% (much more, if the measures are not strongly correlated). Here's a comic about that: xkcd

This is actually a pretty complicated statistical question rather than a matlab one, and you should probably consult with an expert.