## Technote (FAQ)

## Question

I am having problems running GENLINMIXED (Analyze->Mixed Models->Generalized Linear) with an ordinal target (dependent) variable and the NLOGLOG link function. The data distribution and

linking function are supposedly incompatible,

My target variable, GRADE, is ordinal with values 0, 1,and 2, so I chose DISTRIBUTION=MULTINOMIAL. The distribution of GRADE is positively skewed, so I so I chose LINK=NLOGLOG.

The following syntax runs without error statements.

GENLINMIXED

/FIELDS TARGET = grade

/TARGET_OPTIONS DISTRIBUTION=MULTINOMIAL LINK=LOGIT

/FIXED EFFECTS = training homework tutor.

However, the following command aborts with the error message:

" glmm: An invalid combination of distribution and link setting is specified.

Execution of this command stops."

GENLINMIXED

/FIELDS TARGET = grade

/TARGET_OPTIONS DISTRIBUTION=MULTINOMIAL LINK=NLOGLOG

/FIXED EFFECTS = training homework tutor.

Perhaps SPSS assumes that the distribution is NOMINAL MULTINOMIAL (for which the nloglog link is not valid)? If so, how would I specify that the target distribution is ORDINAL MULTINOMIAL (for which nloglog IS valid)?

## Answer

GENLINMIXED is treating the target as nominal and this likely reflects the measurement scale of the target as defined in the data file dictionary.

In the Variable View of the Data Editor, check the Measure column for the target variable (GRADE, in this example). The Measure value is likely "Nominal". Click the right side of that table cell and choose "Ordinal" from the pop-up menu. Then rerun the command with NLOGLOG as the link.

Newer commands in SPSS Statistics, including GENLINMIXED, are generally quite strict in enforcing analysis choices that correspond to the measurement scale. Note that GENLINMIXED will also treat your predictors as categorical (Nominal, Ordinal) or continuous (Scale) on the basis of their Measure level in the data dictionary.

You can also set the measurement level for a variable with the VARIABLE LEVEL command, as in

VARIABLE LEVEL grade (ordinal).