0
$\begingroup$

I am currently working on my thesis and have found an non-significant interaction effect. I am wondering if it is "allowed" or "recomended" to visualize (graph) and explain/interpret the insignificant interaction effect with a simple slope analysis.

(The only effect that is significant is the main effect.)

Can someone plz help me?

Extra info:

I predicted that moderator C had a significant interaction on the relation between variable A and B.

But my multiple regression analysis showed that my interaction term is non-significant. How should i report this? Just in text or can I also visualize the insignificant interaction effect? (is this recommended or?)

$\endgroup$
  • $\begingroup$ Could you tell us why you would belabor the explanation of a phenomenon you found to be "insignificant"? The usual meaning of "insignificant" is that you do not find enough evidence for the phenomenon (here, an interaction) to take it in consideration. Do you perhaps meaning something else by this word? $\endgroup$ – whuber Jul 3 '17 at 19:32
  • $\begingroup$ No i mean just exactly that. I am testing a hypothesis. I predict that moderator C has a significant interaction on the relation between variable A and B. But my multiple regression analysis show that my interaction term is insignificant. How should i report this? just in text or can i also visualize the insignificant interaction effect? $\endgroup$ – Peter Jul 3 '17 at 19:41
  • $\begingroup$ I think you mean "non-significant," which means the p-value has failed to reach some threshold. "Insignificant" means not interesting and is unrelated to the statistical significance of your estimate. Insignificant things can often be ignored, but nonsignificant things may be worthy of discussion. $\endgroup$ – Noah Jul 3 '17 at 20:49
  • $\begingroup$ Thx i didn't knew that insignificant met something else. I suppose i should have said statistically insignificant. But yeah i found a nonsignificant effect. So you say i can (i am allowed to) intrepret and visualize a non significant interaction effect with a simple slope analysis? $\endgroup$ – Peter Jul 3 '17 at 22:30
  • $\begingroup$ You are "allowed," but it is critical that you don't mislead your readers. If your lines have obviously different slopes visually but the interaction is not statistically significant, then you can easily mislead readers into focusing on the depiction rather than the statistics. What benefit do you see for depicting the simple slopes? $\endgroup$ – Noah Jul 4 '17 at 4:10
1
$\begingroup$

This can be thought of more simply. Since you apriori thought that the slope of one variable might vary by the levels of another variable, stick with this more complex model throughout and plot the predicted values as a function of the two predictors, and include pointwise (or even better: simultaneous) confidence bands. You would have B on the y-axis, A on the x-axis, and C depicting different B vs. A relationships.

This is the 'estimation' point of view that gets away from many of the problems with hypothesis testing and significance.

$\endgroup$
0
$\begingroup$

Noah has answered my question.

You are "allowed," but it is critical that you don't mislead your readers. If your lines have obviously different slopes visually but the interaction is not statistically significant, then you can easily mislead readers into focusing on the depiction rather than the statistics. What benefit do you see for depicting the simple slopes?

$\endgroup$

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.