the community site for and by developmental biologists

2 thoughts on “Leaving the bar in five steps”

  1. Joachim,
    Thanks for the write-up; I think these types of graph work well for both publications and presentations, as they are a) more clear to the reader, and b) more professional in appearance.

    I do have one general question. The mantra I heard in stats class was ‘the data must be shown in a way that mirrors the statistics’. So no showing frequency data while testing raw numbers for example.

    In the example above – I typically use mean and standard deviation for graphs where treatments are compared with t-test or ANOVA, as those statistics are making use of mean and variance (in which case, standard deviation is not ideal but can be converted to variance if the sample size is given). In your example, how would you indicate statistical comparisons in the figure, or how would you address them in the text?

  2. Dear Doug,

    Thanks for the comments. I prefer not to perform statistical tests or list p-values (if you must, list them as exact numbers). The notches of the boxplot, indicating the 95% confidence interval of the median can be used for statistical inference. This is quite well explained by George Cumming:
    Instead of the focus on significant differences, the more interesting thing to know is the effect size (what is the magnitude of the difference?). The effect size (and 95%CI) is usually not calculated when comparing conditions, but that may change….

Leave a Reply

Your email address will not be published. Required fields are marked *