Three way interaction without significant two way interaction?

[Johan Anderson]

Hello,

Does anyone know if it is legitimate to proceed with threeway interactions when the two-way interactions are non-significant? There is a significant threeway intercation however.

All the best,
Johan

[Jesse Wursten]

Theoretically it is definitely not an issue. In practice it depends on the context. Are there reasons to expect the 3-way to be significant but not the 2-ways?

[Rich Goldstein]

I ignore the issue of whether "statistical significance" is the right criterion here and just say that deleting lower levels will make the final result, at best, very hard to interpret; thus, if you include a 3 way interaction you should also include all nested 2 way interactions and each individual variable making up that interaction; using factor variable notation helps but it automatically does this for you; see

help fvvarlist