PSA on indicating statistical significance

Sep 09 2015 Published by under Uncategorized

Do not, under any circumstances, for any reason, EVER use the same symbol to indicate different comparisons between points on sub-panels of the same figure that depict similar data sets.

[ e.g., if you have a dose-response function for one genotype depicted in Panel A and for the same measure in a second genotype in Panel B, don't use # to indicate a difference from the vehicle condition on one panel and a difference from the highest active dose on the other panel. ]

14 responses so far

  • This is just a symptom of poor visual design. If you have more than two bars in a bar graph, then you should indicate the paired statistical comparisons by unambiguous indicators, such as brackets that clearly indicate the two bars being compared.

  • drugmonkey says:

    That doesn't always work out well when it is important to compare all the groups. I'm okay with a few different symbols to indicate different contrasts between points. But for goodness sake, use a given symbol to specify the same comparison if it is in the same Figure! (and preferably be consistent across an entire paper if you can possibly manage it).

  • physioprof says:

    If you're comparing every group to every other, and there are more than a few groups, then this should be done by assigning a letter to each group and then specifying by those letters which group is different from each of the others.

  • drugmonkey says:

    That is an acceptable option, sure.

  • drugmonkey says:

    Are you not going to admit that there is a difference between something that actively confuses the reader versus the minutia of acceptable design choices?

  • physioprof says:

    Jeezus fucke, you fucken paranoid loon. I am obviously against confusing the reader with shitty design choices. So yes, I "admit" it. Hahahahah. Fucken loon...

  • chemicalbilology says:

    I love you guys.

  • Draino says:

    If you are comparing every group to every other group, better just make a table for all the p-values in your paper. Even better, you can display it in the supplemental section so decent don't have to witness your nerdy enslavement to statistics.

  • physioprof says:

    DoucheMonkey is such a dumasse lame troll he makes this shittio easy.

  • drugmonkey says:

    The truly talented always 'make it look easy', my friend.

  • jmz4gtu says:

    I use the old-timey Batman comic word bubbles (that say things like "BAM!") to indicate statistical significance. Is this wrong?

  • drugmonkey says:

    Yes.

  • jmz4 says:

    Lightning bolts, at least, are surely acceptable? Although, then your really significant results look anti-Semetic.

  • yikes says:

    and please use a correction for multiple hypothesis testing when needed.

Leave a Reply