Hi there,
I am currently planning my regression analysis, which includes adding interaction terms for my proposed moderators. As one of them is a continuous variable, I plan to use simple slope analysis to probe the interaction further. Is it recommended to split participants by median or mean split (e.g., 1SD from mean for high scores)? I gather the main difference would be that using the median would yield equal sized groups?
Any stats resources outlining the best way to do this?
Thanks a lot,
Esther
Hello Esther,
You can use any number of split methods: Quartile or hinge splits, median split, mean split, mean +/- SD, and so on. The general considerations would be: (a) is it imperative that groupings are (near) equal size?; (b) how much hand work will be involved (e.g., some software will default to a specific method)?; (c) will you have too many subgroups to interpret meaningfully (or which will reduce the precision of any single slope estimate too far)?
Good luck with your work.
- Badges
- Science topic
- Similar topics
- Philosophy
- Ethics