Recently, I have used Error bar/95% CI chart to show the mean difference among the populations. But, the reviewer commented that it is totally incorrect. Can anyone suggest whether I am right or wrong? or is there any best way to show mean difference graphically?
Bar chart is best way but significance can b established by using t test
Bar-chars in-deed. A bar for significance testing us common in natural science. Think of something along the lines showed in the graph on this page: (http://www.originlab.com/www/products/GraphGallery.aspx?GID=350)
If you have panel data, you can conceder a normal time-series-plot. Mean-differences of panel data can be done with a time-series-plot of the two populations and stars of ttest significance (*, **, and ***) stacked vertically between the lines at each data-point. It usually works visually since the space between the two lines should be too small to host stars only when the means are not significantly different.
http://www.originlab.com/www/products/GraphGallery.aspx?GID=350
Hi again.
Error bars show the spread of a sample for ONE mean. With the error bar you get a feel for how precise the mean is. Error bars were NOT my tip above. Two means in a bar-chart with error-bars show the two means and the errors - yes - but for the mean difference, one would like to see the result of some kind of test. I googled this again and figure 5 in this article shows what I mean. The error bars do not intuitively add to the significance of the difference, but the "significance bars" or "cross-bars", or whatever they're called, with stars from the t-test do. Those are the bars I was talking about above.
"Why we cannot use error bars"? Because they indicate sample spread around the mean of ONE population mean rather than the tested significance of the difference that mean and ANOTHER mean. It's heard to say If that's what your reviewer feels, but I hope it helps.
- Badges
- Science topic