Monday, June 7, 2010

Make those numbers talk!

This just in (the Poetry Bus is below) :

A manuscript I'm going over cites an "extensive literature" on a particular topic, and illustrates this by citing a paper that reviewed 900 abstracts. The author of my manuscript didn't go on to explain, however, that of 900 abstracts drawn from six different sources, many were counted up to five times because they appeared in multiple databases, and then a majority were eliminated as being abstracts of papers presented (or not) at meetings* for a total of just 46 different articles. A 20-fold exaggeration.

But then of the 46 articles, 10 were reviews of the others, and 21 were set aside for not asking questions we're interested in. For a grand total of 15 unique & original articles out of the initial 900. Quite the exaggeration.

So. Is 15 articles an "extensive" literature? No (though it's certainly a start).

But I would never have known that the 900 figure was a mirage if I hadn't gone to the trouble to look it up. People rarely go to the trouble to look this stuff up. It's taking most of our time already to read the paper in front of our eyes. Chasing down the references inside it is way too much.

Anyway. I just wanted to show you how numbers can be just as deceiving as revealing sometimes. Even among people trained to know better.
End of rant. I just needed to say.
*The same data can be presented at meetings over and over again, in addition to becoming a real article someday. In fact, when you have something interesting to say, it's pretty routine to go around to different meetings presenting your data to different audiences. There's nothing bad about this - it's good to reach as wide an audience as possible, and discuss with many different colleagues. Even at my modest level, I might present the same work, with modifications as data accumulates or gets published, at a local meeting, a national meeting, and an international meeting (for three abstracts, which is why abstracts don't count, only articles do).


Dominic Rivron said...

To a fan of Ben Goldacre's Bad Science column in the Guardian this is a familiar story!

NanU said...

And it's sadly ironic that this study was designed and written up by our staff statistician! Our very number-guy.