Why Interaction Plots are Preferred over Bar Graphs for Visualizing Interaction Effects in Stata?

[Monica Muller]

I have regression analysis in my paper with interactions between two dichotomous variables. Something like below. (it's the general model. I didn't include the specifications of margins and marginsplot here for simplicity):

regress dv i.m##i.iv
margins i.m#i.iv
marginsplot

This is what I've been doing for years and it's very common in our field. However, a reviewer suggested that I should use bar graphs instead because both variables are indicator variables. I checked many papers in my field, and they consistently use interaction plots similar to the one I made using Stata and not bar graphs. When I mentioned that to the reviewer with citations of the example, they lashed out that it doesn't matter and all those papers are wrong. So, providing citations is not enough for this person and I am looking for the reason.

Given that Stata does not provide bargraphs for interaction plots for visualizing the effects of interactions between dichotomous variables, I believe there should be a good reason for this way of showing interaction effects using continuous lines. Can anyone tell me why interaction plots produced by `marginsplot` are more appropriate for visualizing interaction effects in regression models, as opposed to using bar graphs?

I appreciate your help

[Clyde Schechter]

You can add a -recast(bar)- option to the -marginsplot- command and you will get the same data as a bar plot. You may want to also specify some of -twoway bar-'s options to tailor the appearance to your taste. It's not a big deal to do. And while the reviewer is being obnoxious, it probably is not in your interests to antagonize him/her by arguing about his/her pet peeve when you're not being asked to do something misleading or dishonest. You'll make an enemy of him/her and it can come back to bite you later. Remember, he/she (probably) knows who you are, but you don't know who he/she is. This is the dark side of peer review.

[George Ford]

If you're going to have some sort of line, then it has to run on one of the continuous variables. You've only got 4 possible outcomes here. It looks like a line because Stata connects them, and it is informative.

A bar graph is perhaps better suited to the data, but doesn't tell you anything more.

marginsplot , recast(bar) by(m)

[Monica Muller]

Thank you Clyde for the wise advice. I do have the recast(bar). So, I'll just delete the lines that connect the dots along the x-axis to avoid this person's unreasonable rage. Thank you again and have a great day

[Monica Muller]

Thank you George. I appreciate your response. I agree that it's more informative but I guess I have to remove the lines to please this reviewer.

[George Ford]

for me, the standard approach is easier to visualize. oh well.