interaction dummies

I have two variables, time (T) with two values and region (R) with 10 values, and thus two sets of dummy variables. how can I get interaction dummies

dummyvar(T)*dummyvar(R) ?

[bym]

can you give an example of your data?

[Walter Roberson]

Is this possibly a question about covariance ?

[Rabeya]

say T=[1;1;1;2;1;2;2;1;2;1] and R=[1;3;1;2;2;1;3;3;1;2]. So,

group =

1 1

1 3

1 1

2 2

1 2

2 1

2 3

1 3

2 1

1 2

I need dummy variables depending on the rows of group, e.g.,row1 reads as a person in time 1 region 1, row2 a person in time 1 region 3,..etc.

Answer 1

[Tom Lane]

Here is a specific example of how to do it. If you need to generalize it and can't figure that out, let me know. The idea is to multiply the first column of d1 by each column of d2, then the second column of d1 by each column of d2:

d1 = dummyvar(T); d2 = dummyvar(R);
d1(:,[1 1 1 2 2 2]).*d2(:,[1 2 3 1 2 3])

Another thing to consider is the x2fx function, using something like this:

x2fx([T R],'interaction',1:2)

This requests that both T and R be treated as categorical, and that all columns for an interaction model be computed. You'll see the result has a column of ones, then dummy columns for T and R separately, then columns for the interaction. The idea is that these will be used in regression, so certain columns are omitted so that the full matrix is not overdetermined.

Answer 2

[Walter Roberson]

group = [T, R];
row = mat2cell(group, ones(1, length(T)), 2);

Then row{1} would be your row1, row{2} would be your row2, and so on.

If you need automatically generated variables named row1, row2, and so on... then you are probably not programming optimally.

http://matlab.wikia.com/wiki/FAQ#How_can_I_create_variables_A1.2C_A2.2C....2CA10_in_a_loop.3F

[Rabeya]

probably I could not explain my problem properly. Dummy variables created by dummyvar(T) will yield 2 columns, dummyvar(R) will yield 3 columns. I need each of the 3 columns of region_dummy=dummyvar(R) multiplied by each of the 2 columns of time_dummy=dummyvar(T), these are element by element multiplications. I can use the following for loop,

for i=1:2

group_dummy(:,i)=time_dummy(:,1).*region_dummy(:,i);

end

for i=1:3

group_dummy(:,i+2)=time_dummy(:,2).*region_dummy(:,i);

end

but I want to have something efficient. Further how can I sort the final matrix group_dummy in descending order depending on its first column?

[Walter Roberson]

bsxfun() to do the multiplications.

sortrows(group_dummy, -1)

to sort based upon the first column in descending order

In this context, what do you mean by dummyvar(T) and dummyvar(R) ? T and R are numeric column vectors, right? And they are both 10 elements long in your example, so why would the number of columns in the dummy variables differ? Does it have to do with the number of unique values in each?

[Rabeya]

yes, dummyvar is the Matlab command to generate dummy(categorical) variables. It takes the number of unique values in each.