Friday, March 07, 2008

Statisticians are verifiably insane

R is a great free statistical software package. I'm trying to use it frequently enough to get comfortable with the basic commands and data objects, so that I don't have to dig through the help manual every time I want to do some small thing. I'm trying to break my Excel habit - if you want real statistics, you don't use Excel, right?

I'm writing my paper, and need to make a simple bar graph. The basic kind, with error bars, which you see in every single issue of every science journal published today. In about 1 minute I can make one, using Excel:

I'm thinking, 'wow, that was easy in Excel, but I should really learn to do it in R.' Of course in R, 35 minutes later, I still haven't figured out how to get the chart to look the way I want. So I go to the trusty R Help mailing list, confident that someone else has shared my frustration. And there I learn that statisticians hate these charts. Do we need any more evidence that statisticians are insane?

(I understand their point, but in most cases - come on! Bar charts are great!)


S Denslow said...

I have always had a problem with the editorial admonition to summarize data and avoid lists of data. True, lists take up precious room, but for small datasets, the list seems like the only honest way to present the data. Dynamite plots are indeed really good for bigger sets. Even better are box and whisker charts.

Unknown said...

I'm partly joking, because I'm a biochemist surrounded by computational biologists, and I like to give them a hard time, since they are amused by my troubles with R.

In this case, I'm plotting multiple experimental replicates of yeast strains from different genetic backgrounds, so I believe a box and whisker chart would work fine here. But you rarely see them in a typical molecular biology paper - it's all dynamite plots.

Unknown said...

Useful to know that barplot() returns the bar centres. I took a long time to discover this.