How do I decide where is the threshold to split it into levels? I ran an ANCOVA with a continuous moderator and it is significant, but I am wondering how to transform it to a categorical variable? Is it necessary?
In general, no it is not necessary and is generally inadvisable as it create loss of information and loss of efficiency. Additionally, it can introduce bias that is remedied by entering the continuous version of the variable into the model as a predictor in addition to the categorical form (meaning the categorical form is not beneficial).
I am unclear of your use of "moderator" in this case. Do you just mean an x-variable or do you mean a variable that interacts with another to change the relationship between Y and another x-variable?