I have a relatively simple 2x2 design. I give a hypothetical example. I have a continuous variable Y (Plant Growth) and I have 2 factors with 2 levels; Temperature (warm/cold) and Condition (Fertilizer, no Fertilizer).

I did an Anova (Plant Growth~Temperature*Condition) and get the following p-values:

  • Temperature: p = 0.01
  • Condition: p = 0.002
  • Temperature*Condition: p = 0.245

As the interaction is not significant, I do not want to do every single comparison in a post hoc test. Instead, I want to see if Condition has a significant effect on plant growth under warm temperatures, and I also want to know if Condition has a significant effect on plant growth under cold conditions.

Now I wondered what the correct way of doing this is? Would I do a simple t-test under Condition cold and another one under Condition warm? If that is possible, would I have to correct for multiple comparisons (I guess so)? It doesn't seem to make sense to do a post hoc test in that case.

  • $\begingroup$ Do the values 0.01, 0.002 and 0.245 refer to p values? $\endgroup$ Apr 5 '20 at 12:51
  • $\begingroup$ Yes. Sorry. I should have mentioned that! $\endgroup$
    – Deschain
    Apr 5 '20 at 13:55

You do not need to do a post-hoc test.

Because the interaction term is not significant, we know that the main effect of Condition is significant at both Temperatures.

  • $\begingroup$ Many thanks for that fast reply and the clarification. That does make a lot of sense. I must have misunderstood this. $\endgroup$
    – Deschain
    Apr 6 '20 at 6:48
  • $\begingroup$ If the answer was useful, please help others by accepting it. Thank you. $\endgroup$ Apr 6 '20 at 7:28
  • $\begingroup$ Sorry that I didn't answer that but I actually do not know how to accept an answer. $\endgroup$
    – Deschain
    Apr 25 '20 at 10:41
  • $\begingroup$ I have, however, a follow up question in terms of how to display the statistical results in a boxplot or barplot. A common thing after a post hoc test would be to add letters over the, in this case, 4 bars. But as I didn't do all the individual comparisons I guess I cannot do that. What is a common way to do that? $\endgroup$
    – Deschain
    Apr 25 '20 at 10:48
  • $\begingroup$ It should be obvious to the reader once you draw the barplot with confidence intervals. The figure legend can state the results of the significance tests. $\endgroup$ Apr 26 '20 at 8:20

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.