In field experiments with sugar beet I found several significant interactions (A*B, A*B*C...) for some of the standard parameters of yield and quality. For lack of new information, I started to ignore interactions with p-values > 0,03.
In my experience, we can try to find some synergism or antagonism between two factors by use of tables and figures presenting the interaction in case they are significant (p>0.05 or p>0.01 as is common, based on your choice ). But the fact is that interpretation of 3 way and more interactions is usually so hard or even impossible that I neglect these interactions at all. Just a star in the table and no other use. I would be happy if anyone adds more regarding interpretation of 3 way and higher interactions.
Thanks for your answer. I agree with you that it is easier to neglect interactions. But I also found usefull ones, so I couldn't ignore them.
For interpretation of interactions A*B*C you have to look at one factor while the other two factors stay fix, e.g. for A:
b1, c1: a1 - a2
b1, c1: a1 - a3
b1, c1: a2 - a3.
...
b2, c3: a2 - a3.
In this way, you can easily structure your data.
But sometimes, I also found significant but weak interactions with p-values near 0,05 (level of significance = 0,05). Here the reason for the interaction was often not detectable. Is there a better way to avoid these effects, than increasing the level of significance to 1%?
How did you collect your data? Did you use some type of factorial design or use observational data?
If it was observational data, you should center your data first, then run your models. I can guarantee the models will be different, especially the significant interactions.
When you look at higher order interactions, I would ask what do they add to the model? If the optimal solution is the same with and without the higher order interaction, it doesn't add much to the model.