SPSS Workshop: GLMM and Non-gaussian Distributions

When we collect data from our research trials, we do not always have data that is “well-behaved” or that comes from the traditional normal distribution curve.  It can have a number of distributions and with the latest statistical methodological advances, SPSS can handle some of these as well.  Your first job is to be able to recognize when a normal distribution is NOT appropriate and which distribution is an appropriate starting place.   Non-Gaussian distributions are what these are referred to.  Remember Gaussian is the same as calling it Normal.

Where do we start?  Think about your data – what is it?

• A percentage?
• A count?
• A score?

How do we know that our data is not from a normal distribution?

• Always check your residuals!
• Remember the assumptions of your analyses?
  • Normally distributed residuals is one of them!

Let’s work with the following example.  We have another RCBD trial with 4 blocks and 4 treatments randomly assigned to each block.  There were 2 outcome measures taken:   proportion of the plot that flowered, and the number of plants in each plot at the end of the trial.

Generalized Linear Models in SPSS

Those of you that follow the terminology that is commonly used when talking about ANOVAs – Analysis of Variance – may notice that there is one term that is NOT included in this heading?  Any guesses?  MIXED.  In SAS, we talk about and work with Generalized Linear Mixed Models or GLMM (GLIMMIX), but I just realized that the MIXED part is missing here.  That’s way I could not figure out how to add a RANDOM effect to the models!

We will however, still use the RCBD data that was used in the SAS workshop.  It contains, as noted above proportion data and count data – perfect examples of non-gaussian data.

Please download the Excel spreadsheet here and load it into SPSS.

We will work through each of the outcome variables in the workshop using the following steps:

• Analyze
• Generalized Linear Models
  • Generalized Linear Models
  • We will work through the tabs for each variable together and discuss the outputs

Please note that if you have attended the SAS workshop and this workshop – and are trying to compare the results – they will NOT match as we are creating a different model in each program – since we cannot add a RANDOM effect in SPSS yet.  You can, however, create the same fixed effects model and the results match perfectly.

See you later in May!