using ANOVAN to analyse categorical data

Hello I have two climate data files, each of them is one column (temperature and moisture) and four plant types existence and absence results (0 for absence, 1 for existence), I made each variable in separate txt file in one column . I want to test the significance effect of the interaction of two predictors climate factors (temp and moisture) on the plan occurrence. for example: (temp*moisture)on plant1, and so on. I have used this code p=anovan (plant,{temp moisture}, 'model, 'interaction'); but it didn't work, I think the mistake is of my plants data which is zeros and ones. any suggestions plz. Regards

Answer 1

[Tom Lane]

You should be able to adapt this for your logistic model:

load hald
[b,dev,st] = glmfit(ingredients,heat);
dataset(b, st.se, st.t, st.p, 'varnames',{'coef' 'se' 't' 'p'})

Answer 2

[Tom Lane]

You could use a logistic regression to predict the probability of "existence" using predictors temp, moisture, temp.*moisture. You can fit this regression using glmfit, or in the most recent release using GeneralizedLinearModel.fit. The latter might be the easier of the two because it has a number of built-in features for plotting, model building, and diagnostics.

Other choices for binary data include knn classification, classification trees, and discriminant analysis.

Now, this would be for a single binary response or dependent variable. If you have four of them, you could do it four times.

[Yasmine]

hi Tom thank you so much

can you write me the code of using GeneralizedLinearMode in Matlab?

[Daniel Shub]

In general we do not write code for people, but help them along.

Answer 3

[Tom Lane]

I don't mind supplying code comparing glmfit and the new feature, but of course you will have to adapt it for your own use. Here's an example where I fit a single binary response using a model with interactions:

a = rand(100,1); b = rand(100,1); y = binornd(1,max(a,b));
glmfit([a b a.*b],y,'binomial')
GeneralizedLinearModel.fit([a b],y,'interactions','distribution','binomial')

[Yasmine]

And you know what I've tried many other options for the

GeneralizedLinearModel.fit, but the program considers it as undefined variable.

[Tom Lane]

You should use glmfit then. GeneralizedLinearModel.fit is new in the R2012a version of MATLAB.

[Yasmine]

Thanks Tom that is so helpful, BUT..I've got a coeffecients but not the P values, I need to test the significant of these veg distribution ..ANY SUGGESTIONS

[Yasmine]

YES YES tom I've done it, I just want to ensure that the (st)values are the significancy values, right?

Thanks so much Tom