multcompare and anovan result in zero and nan

I want to perform a multcompare test on my data set and find which parameter or the cobination of parameters can change the mean value of my response value. Here is the code I use:

%%
X = readtable('HHH.xlsx','sheet',3);
y=[X.UL]';
g1=X.Type;
g2=X.ThicknessSP;
g3=X.ThicknessDP;
g4=X.Weight;
g5=X.Adhesion;

[~,~,stats] = anovan(y,{g1 g2 g3 g4 g5},'model','interaction',...
    'varnames',{'g1','g2','g3','g4','g5'});

But what I get is all Nan and zeros.

Can you please help me?

I have attached my data.

Thanks

[sarah Abdellahi]

Any Ideas?

Answer 1

[Jeff Miller]

You can't use anovan with numerical predictors like thickness, weight, and adhesion. Have a look at regression models. You will probably need a lot more data, though, to separate out the effects of these different predictors.

[sarah Abdellahi]

Thanks for your response.

but dont we have a numerical example here? under Multiple Comparisons for Three-Way ANOVA?

https://www.mathworks.com/help/stats/multcompare.html

[Jeff Miller]

Not that I can see. In that example, the g1/g2/g3 vectors all have two distinct categories each, and there are lots of scores in each category. Your thickness, weight, and adhesion vectors each have many different numerical values rather than a few distinct categories.

[sarah Abdellahi]

Thanks Thanks Thanks!

[sarah Abdellahi]

Hi Jeff,

I changed my data to groups. My data looks like this:

But I still have problem with anovan shown below. Still Nan!

Here is the code I used to converd data to groups

Data= readtable('HHH.xlsx','sheet',1);

Th_Weight=4;Th_adhesion=0.8; Th_SP=2.8; Th_DP=5.5; Th_UL=65; % threshholding values


Data_digitized=table(Data.Type, double(Data.ThicknessSP>Th_SP), double(Data.ThicknessDP>Th_DP), double(Data.Weight>Th_Weight), double(Data.Adhesion>Th_adhesion), double(Data.UL));
Data_digitized.Properties.VariableNames =Data.Properties.VariableNames; %generating new table

y=Data_digitized.UL;
gg1=Data_digitized.Type;
gg2=Data_digitized.ThicknessSP';
gg3=Data_digitized.ThicknessDP';
gg4=Data_digitized.Weight';
gg5=Data_digitized.Adhesion';

[~,~,stats] = anovan(y,{gg1 gg2 gg3 gg4 gg5},'model','interaction',...
    'varnames',{'gg1','gg2','gg3','gg4','gg5'});

Thanks

[Jeff Miller]

I suspect you don't have enough data to estimate all the two-way interactions (i.e., empty cells in some of the 2x2 designs). Does it work with 'model','linear'? This might be all that can be computed with your data set. Or maybe you can get some of the 2-way interactions using a 'terms' matrix. But evidently you cannot get all of the 2-way interactions, which is what you are asking for with 'model','interaction'.