Re: st: significance stars on stata plots

[Nick Cox <njcoxstata@gmail.com>]

1 piece of information per square cm translates for US citizens as
about 6 or 7 per square inch.
Not quite so catchy, unfortunately.

I go along with Ronan, as usual, and you won't find support for stars
or kisses in my graphics programs, except by accident.

However, I note that with two groups we should really be showing on
graphs the _difference between the means_ and its confidence interval.
That doesn't seem to be a widely used plot but there are some ideas in

Cumming, G. 2012.  Understanding the new statistics: Effect sizes,
confidence intervals, and
meta-analysis.  New York: Routledge.

Disclaimer: I agree with everything you might possibly say about
Cumming's use of the term "new statistics".

Nick
njcoxstata@gmail.com

On 3 July 2013 18:39, Ronan Conroy <rconroy@rcsi.ie> wrote:
> On 2013 Iúil 1, at 19:49, Slade, Alexander N wrote:
>
>> Hello,
>>
>> Is there a simple way to add significance stars to the various plots (e.g., bar graphs, line graphs, etc.)? For instance, if I wanted to have a bar chart comparing the mean years of education between men and women, is there a way to put a star (or two) on top of the female bar to denote a significant difference?
>
> You could just as readily put it over the other bar. And that is a problem with adding these things to bar charts. With multiple bars, you need to draw attention to the bars being compared, and even with two bars, where do you put the star?
>
> Putting the information in a note is easier.
>
> With only two bars I would argue that showing two mean values using bars is a waste of graph space - why not show the data, the means and their confidence intervals using Nick Cox's -stripplot- from ssc?.
>
> I have a personal rule that a graph should show at least one piece of information per square centimetre. Given that the smallest graph will still be about 12 square centimetres, this means that no graph should have less than 12 pieces of information. Graphs that break that rule are generally better as short sentences in the text. (Mean age was 32 years for men and 27 years for women, P =0.021.)
>
> And written that way, you can see that you would really include some measure of dispersion in the text, hence my recommendation to do likewise in the graph.
>
> But you may be stuck with a researcher who wants two bars and a kiss…

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/