MATLAB Weighted Multiple Regression

Question

I have a set of data that includes 821 observations, each with 20 measurements. I would like to regress this set data against a set of single dependent variables using a multiple linear regression in MATLAB. However, I would also like to weight each observation differently in the regression based on my own calculations. For example, I would like to give the first observation a weight of 1 and the second observation a weight of 1.6, which would ideally pull the regression towards the more heavily weighted second observation.

Is such a computation possible in MATLAB? If so, what function(s) would be best to carry out this type of computation?

Thanks for the help!

Answer 1

You do not actually need the Statistics Toolbox to do this. The built-in function lscov will do everything you want.

[b,bse] = lscov(X,y,w)

will provide weighted OLS estimates and their standard errors. If you would like a constant in the regression then include a column of ones in X.

Answer 2

With Statistics Toolbox, you can use fitlm to create a linear regression model, applying the Weights option to supply your weights.

NB in older versions of MATLAB, you'll need to use LinearModel.fit rather than fitlm, but they do the same thing.