I am sure that nobody has any opinions whatsoever on using the placement of significance symbols to...err....emphasize..... the magnitude of the effect.

Haphazardly placed asterisks on graphs bugs the crap out of me. I always put them the same distance from the error bar.

— Dr. Katiesci (@katiesci) August 24, 2015

We're srsly gonna discuss this? Well, ok.

First, exactly how and where you indicate statistical significance on a graph depends completely on the statistical test you have employed, and what is being compared to what.

Second, I agree that haphazardly placed asterisks or other indicators of statistical significance test outcomes are very annoying. But this is purely on aesthetic grounds. I honestly can't even imagine what you are getting at with "placing symbols to emphasize the magnitude of an effect".

Think of a ~linear trend where two means are moving apart. Then use a progressively increasing distance of asterix from the mean value to visually emphasize the difference.

I think we need an example graph here. By "randomly placed" I meant totally random. Not lined up with the bars, one all the way at the top of the graph near the title and one down by an error bar. It was weird and sloppy and that's the only reason it bugged me.

When you say "linear" do you mean when they're place near the error bar they emphasize the differences? If so, it's proportional to the actual difference.

Never thought of this. Seems like a low-level Jedi mind trick to me. Anyone with enough experience reading papers (Jabba the Postdoc?) should see right through it.

Someone needs to run an ANOVA on the variability of asterisk spacing about the center of the horizontal lines connecting two bars on a bar chart.

Why are we still using asterisks when we have smiley faces?

Asterisks are so 2014. I use #significant instead.

It's a matter of information design, and of doing your best to make sure that nothing distracts from the point. I could go on, but I'll just say that reading Tufte is a good place to start.

By the way can we mention that its 2015 and people still use bar/column graphs instead of box-and-whiskers or any other more faithful-to-showing-the-actual-distribution options?

Shame!