topic generalized least squares? in Discussions

generalized least squares?

fever — Fri, 09 Jun 2023 00:53:44 GMT

I'm looking to run a generalized least squares (GLS) model to replace my OLS model. I believe the residuals of the OLS model have spatial autocorrelation (I'm not positive, but the map definitely looks so). I read that GLS models correct for autocorrelation (any kind?) of residuals. I don't see GLS as an option under the Generalized Regression platform. I don't have Arc or QGIS, R, or SAS. Suggestions?

Re: generalized least squares?

statman — Mon, 15 Aug 2022 20:10:03 GMT

You mean GLM? Start here:

https://www.jmp.com/support/help/en/16.2/?os=mac&source=application#page/jmp/generalized-linear-models.shtml#218524

Re: generalized least squares?

fever — Mon, 15 Aug 2022 20:13:05 GMT

Thanks, but no, the generalized least squares (GLS) is different from generalize linear models (GLM) as I understand it: https://stats.stackexchange.com/questions/272336/what-is-the-difference-between-generalized-linear-models-and-generalized-least-s

Re: generalized least squares?

peng_liu — Tue, 16 Aug 2022 14:28:41 GMT

You may want to check out Mixed Models and Random Effect Models

Note: If you click the corClasses link in the top answer on that StackExchange page, it leads you to a R package "nlme", for "Linear and Nonlinear Mixed Effects Models".

Now look at the description of "Classical Linear Mixed Model" on Mixed Models and Random Effect Models . Here is a screenshot of the formula. For such data, observations are not independent. The model allows modeling the correlation among observations.