first argument is the variable to cut, second argument is the number of bins you wish to group the data into. "labels" is an optional argument, as well as "ordered_result" (which I suggest to set to TRUE).
There is nothing that makes a variable inherently either ordinal or interval, although you can certainly group the original interval level scores into fewer categories in just about any way that you want.
Could you explain more about what your goals are with regard to the SEM?
As a follow-up to David's question, do you believe there may be nonlinearity in the X->Y relations? That's the only reason I can think of offhand to do this. Depending on your sample size, there are other methods for modeling nonlinear relations that don't "throw away" information by categorizing.
I take that back -- really odd distributions, especially multimodal, can be another reason to do this, and one I've used. For example, "on how many days of the last 30 did you smoke cigarettes" typically gets mostly zeros, a smaller number of 1s, 2s, and 3s, and then very sparse responses until a second mode at 30. So sometimes I'll cut it as 0, 1-2, 3-29, 30, for example, depending on the actual distribution and sample size, reflecting qualitative differences between non-smokers, occasional smokers, and daily smokers.
Thank you Michael et al., I was a bit worried about how to combine interval data to ordinal data for using in SEM. Noted , I can use the excel formula.