How to extract coefficients values from a regression model results? How to clear?

Hello,

Could anybody help me please to extract the coefficients values (or other relevant data) from the model results such as: mdl.Coefficients

ans =

          Estimate        SE         tStat       pValue  
          _________    _________    _______    __________

    b1     -0.17489     0.025051    -6.9813    2.0592e-07
    b2     -0.47334     0.024024    -19.702    3.7536e-17
    b3    -0.028464    0.0058363     -4.877    4.6488e-05
    b4    -0.023999    0.0072508    -3.3099     0.0027401

I tried mdl.Coefficients (1,2) for instance, but this gives me a table like below instead of just the number (0.025051 in this case).

ans =

             SE   
          ________

    b1    0.025051

I tried get command as well but it didn't work. Also I am wondering how I can clear a part of the regression results. I have a lot of observations that take the whole memory and I need to clear them from the mdl results. (like mdl.VariableInfo, mdl.ObservationInfo, etc.)

Thanks

Answer 1

[Saptarshi]

Hi VD.

I had the very same question! It's a fundamental question, and points to a fundamental complaint that I have about MathWorks, and software companies in general. While Matlab's model-fitting commands are very powerful, what good is a program (or an algorithm) if you can't access the results?

Well, even though your question is "beneath" the geniuses at MathWorks to answer, I may have found a solution. The solution basically revolves around accessing the fitlm results as a sub-table, and then calling the first element (the number) within it.

To extract b1, or instance, try to get it by direct reference:

mdl.Coefficients(1,1).(1) % extracts b1 coefficient. (Or I have the array indices switched...)

mdl.Coefficients(1,2).(1) % This should extract the SE of b1, if that is what you are looking for

mdl.Coefficients(2,1).(1) % should extract b2

Try it out and please let us know if it works.

[Very Determined]

Thanks for the response. I am done with the code as it dates back to years ago. I will try it once I got chance to rework on the same problem and I will let you know then.