I have a 5-level categorical predictor of a binary outcome and would like to see if there's a monotone trend between levels of the predictor and likelihood of the outcome. Doesn't have to be linear, just reliably increasing or decreasing. How do I code this in SAS?
I suspect your categorical variable is also ordinal.
Do you need a special coding for this? Wouldn't it be enough just using a GLM coding where the lowest category is the reference? You can see if odds ratios are changing accordingly.
I suspect your categorical variable is also ordinal.
Do you need a special coding for this? Wouldn't it be enough just using a GLM coding where the lowest category is the reference? You can see if odds ratios are changing accordingly.
Agreed. This sounds like an ordinal variable.
Instead of creating dummy variables for each level, you can include the variable in your model and the coefficient would represent the slope of the monotonic trend. Goodness of fit statistics will show you how well that assumption of linearity of the categories fits the data presented. The p-value of the coefficient will help to evaluate this too. Additionally, I would suggest looking at residuals and graphs of the data as these will likely be more telling.
One concern is that when using a variable as ordinal you want to make sure that the distance between your categories is equal or logical in some way. For example, you can do categories that are log-transformed (1-9, 10-99, 100-999, 1000+). Here if evaluated as an ordinal, the coefficient would be the trend line for the categorical variable; and related to originating data by exp( beta ).
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Probability
- Probability Theory
- Reliability