I conducted my analysis in SPSS as follows:
I fitted a generalized linear mixed model based on multinomial distribution with a logit link function.
I did that using repeated measures longitudinal data with an ordinal target variable: duration of hallucinations with levels of 0= "N/A" (no hallucinations), 1="Seconds", 2="Minutes", 3="Hours" and 4="Continuous". I
The model shows a significant effect of
Event.Name, which is actually
Time measured in weeks.
FINAL.DIAG is my 5 different patient groups.
My problem is that I am struggling to interpret the coefficients.
Some sources online seem to say that I can only state something like the following:
For a one unit increase in time, the odds of observing any of the lower categories of duration (N/A, Seconds, Minutes, or Hours) versus the highest category of duration (Continuous) are 0.958 times less or decrease by 4.2% (1-0.958), given that the other variables in the model are held constant. Due to the proportional odds assumption of ordinal variables, the same decrease in odds of 0.958 times, is found between any category below a certain threshold and any category above the same threshold. For example, the odds of observing the lowest category of duration, “N/A”, are also decreased by 4.2% compared to observing any of the other categories (Seconds, Minutes, Hours or Continuous).
But other sources say that I can interpret my findings like this:
An increase in time (expressed in weeks) was associated with a decrease in the odds of a longer VH duration, with an odds ratio of 0.958 (95% CI, 0.958 to 0.978), t = -4.115, p < .001.
My questions are:
- Can I say that the odds of a subject being in a higher category of duration decrease with increasing time?
- Or should I say that the odds of observing any of the lower categories of duration (N/A, Seconds, Minutes, or Hours) versus the highest category of duration (Continuous) decrease by 4.2% as time increases but the same can be said for the odds of observing N/A vs any of the other levels in my ordinal variable?
- And finally, what can I even conclude from that? The proportional odds assumption confuses me and I would really just like to use the second version of the interpretation if that is correct as well.