Troubleshooting
Problem
I am trying to set up an Analysis of Variance model using Analyze>General Linear Model, but as soon as a covariate is added to the model, the Post Hoc button becomes unavailable. How can I control for covariates and still obtain post hoc tests?
Resolving The Problem
When performing post hoc comparisons of means when covariates are present, many of the options for adjusting probabilities (Duncan, SNK, and so on) that can be requested are either known to be invalid, or at best of uncertain validity. For this reason, the button becomes disabled.
However, you can use Estimated Marginal Means to obtain all pairwise comparisons of the adjusted cell means for any factor in your model. The output will be very similar to what is produced using Post Hoc: you will obtain a table listing, for each pair, the Mean Difference, an estimate of the standard deviation for that difference, its significance, and a 95% confidence interval.
Click on the Options... button. In the box labeled Estimated Marginal Means, you will be able to select the factor(s) for which you want pairwise comparisons. Be sure that "Compare Main Effects" has a check mark in the box next to it.
If you wish, you may also request Bonferroni or Sidak adjusted probabilities. The default, Fisher's LSD, simply gives unadjusted probabilities. These adjustments are known to remain valid when covariates are present. Moreover, they are also viable for other models, such as repeated measures.
Since Bonferroni simply multiplies the probability by the number of comparisons being made (with a maximum significance of 1), if there are more than a few comparisons, Bonferroni will be needlessly conservative.
If you are using command syntax, you would add subcommands similar to the following examples:
/EMMEANS=TABLES(factor) WITH(covariate=MEAN) COMPARE
/EMMEANS=TABLES(factor) WITH(covariate=MEAN) COMPARE ADJ(LSD)
/EMMEANS=TABLES(factor) WITH(covariate=MEAN) COMPARE ADJ(BONF)
/EMMEANS=TABLES(factor) WITH(covariate=MEAN) COMPARE ADJ(SIDAK)
Related Information
Historical Number
20370
Was this topic helpful?
Document Information
Modified date:
16 April 2020
UID
swg21476752