Linear Regression: order of estimated coefficients does not match input order

For example,

mdl = fitlm(data,'EXPOS ~ X1 + X2 + HEIGHT + AGE')

The output

mdl.Coefficients

produces the estimated coefficients in the following order: HEIGHT, AGE, X1 and X2. Why does it do that? It really messes up some calculations! Please help!

Answer 1

[John D'Errico]

If you have created the dataset with those variables in the order (height, age, x1, x2) then that is the order they will appear in the final model, even though you specified the model as

'EXPOS ~ X1 + X2 + HEIGHT + AGE'

[Isabel Chen]

Oh I see! Thanks. So obvious now that you say it. The reason why this is an issue is because I'm trying to use the function

linhyptest()

which requires an input matrix specifying the linear combinations of the coefficients that I'm testing for (eg, testing the null hypothesis that beta_2=beta_4=0). Clearly if I choose my matrix based on how I specified the input, which is different from how Matlab actually outputs the coefficients, I'm going to be in trouble. Do you have any ideas on how to work around this, apart from checking the data manually?

[John D'Errico]

You can extract the variable names from the data set, then use regexp (or some other tool, like strtok, or even my own function allwords on the File Exchange) to take apart the model string you will be using. Then sort the two sets of variable names to be consistent.

For example:

allwords('EXPOS ~ X1 + X2 + HEIGHT + AGE',' +~')
ans = 
    'EXPOS'    'X1'    'X2'    'HEIGHT'    'AGE'