Residuals from Regress

I'm trying to get the residual error from a linear fit between two variables and the residuals I get from the regress function make no sense and don't match what I get from the data fitting gui (which looks correct). What is the difference in the procedures here and how can I programmatically get the residuals I see through the data fitting menu option?

[the cyclist]

Can post the code that uses the regress function and seems suspicious to you?

Answer 1

[Richard Willey]

% Generate some random data
X = linspace(1,100,100)';
Y = X + randn(100,1);

% Use Curve Fitting Toolbox to generate a fit

% In your workflow, you'd create the fit in cftool and then export the
% model to MATLAB as a fit object

foo = fit(X,Y,'poly1')
% Calculate residuals
resid1 = Y - foo(X)

% Use regress to compute residuals
b = regress(Y, [ones(length(X),1) X])
Yhat = [ones(length(X),1) X]*b;
resid2 = Y - Yhat

% Check that the residuals are the same
resid1./resid2

[Jamie Johnston]

Thanks Richard - I was simply missing the second column for the X variable...looks to be working now.

[Richard Willey]

Easy mistake to make. regress gives you really fine grained control over your design matrix. However, in turn you need to do things like add a ones vector for your constant and the like...