Dear all,

We received a reviewer comment that read:

The software that was used to perform the LMMs also needs to be provided (again relative to the issue of missing data). That is because most LMM software programs will only perform full information maximum likelihood estimation if missing values are contained in the outcome variables but will use listwise deletion if the predictor variables have missing values. Thus, because the likelihood is conditional, most LMM software will listwise delete observations with missing predictors.

We used SPSS to run the LMMs and specified maximum likelihood estimation. I assumed that given that I specified 'maximum likelihood estimation', that SPSS would not be performing listwise deletion when predictor variables have missing values?

Is anyone able to help with this?

Thank-you

The MIXED and GENLINMIXED procedures in SPSS, like other standard mixed models procedures performing maximum likelihood or restricted maximum likelihood for linear models, and maximum pseudo-likelihood or restricted maximum pseudo-likelihood for generalized linear models, require complete data for all variables specified in an analysis, predictors and dependents. If you have something like repeated measures with different time points for different subjects, mixed models are capable of handling this under missing at random (MAR) assumptions on the missing data mechanism to model the relationships over time, but for the observed time points you need the data for all the variables.