When running post-hoc pairwise comparisons for a generalized linear model (GENLIN with log link and Poisson family error distribution and robust standard errors) I am confused about whether the estimated means presented are adjusted for the other variables in the model. I typically understand marginal means to be accounting for the other variables in the model by holding covariates at the mean and factors at 0 (or whatever is specified). However, the SPSS documentation states the following, saying comparisons are "made on unadjusted values." 

"Post hoc multiple comparison tests. Once you have determined that differences exist among the means, post hoc range tests and pairwise multiple comparisons can determine which means differ. Comparisons are made on unadjusted values. The post hoc tests are performed for each dependent variable separately" - page 7 https://www.sussex.ac.uk/its/pdfs/SPSS_Advanced_Statistics_22.pdf 

So, if I present the means from the post-hoc comparisons, will they be accounting for the other categorical and/or continuous variables in the model (i.e., holding other groups at 0, leading to a very specific sub-group analysis of people with all 0s), or are the estimated means in this case unadjusted for those variables? 

Thank you!

------------------------------
Emily Cherenack
------------------------------

#SPSSStatistics

Hi, Emily.

This reply is from one of our statisticians, who I took the liberty of contacting in reference to your question:

"The referenced pdf discussion is of the POSTHOC subcommand in GLM, while the analysis of interest is EMMEANS in GENLIN. The former is done on unadjusted means, as indicated, while the latter is done on adjusted means (estimated marginal means based on the model). Perhaps of relevance here is that when we use "GLM" we're referring to the general linear model, not generalized linear models."


------------------------------
Rick Marcantonio
Quality Assurance
IBM
------------------------------

Original Message:
Sent: Tue May 18, 2021 03:26 PM
From: Emily Cherenack
Subject: GLM Marginal Means and Post-Hoc Pairwise Comparison

When running post-hoc pairwise comparisons for a generalized linear model (GENLIN with log link and Poisson family error distribution and robust standard errors) I am confused about whether the estimated means presented are adjusted for the other variables in the model. I typically understand marginal means to be accounting for the other variables in the model by holding covariates at the mean and factors at 0 (or whatever is specified). However, the SPSS documentation states the following, saying comparisons are "made on unadjusted values." 

"Post hoc multiple comparison tests. Once you have determined that differences exist among the means, post hoc range tests and pairwise multiple comparisons can determine which means differ. Comparisons are made on unadjusted values. The post hoc tests are performed for each dependent variable separately" - page 7 https://www.sussex.ac.uk/its/pdfs/SPSS_Advanced_Statistics_22.pdf 

So, if I present the means from the post-hoc comparisons, will they be accounting for the other categorical and/or continuous variables in the model (i.e., holding other groups at 0, leading to a very specific sub-group analysis of people with all 0s), or are the estimated means in this case unadjusted for those variables? 

Thank you!

------------------------------
Emily Cherenack
------------------------------

#SPSSStatistics