Slashdot Mirror
Scientific Papers With Shorter Titles Get More Citations
sciencehabit writes: Articles with shorter titles tend to get cited more often than those with longer headers, concludes a study published today, which examined 140,000 papers published between 2007 and 2013. It appears in the journal Royal Society Open Science. Citations are a key currency in the academic world. The number of times other researchers cite a scientist’s work is often an important metric in hiring and workplace evaluations. Citations also play a role in determining a journal’s place in the scholarly pecking order, with journals that publish more highly cited papers earning a higher “impact factor” (although many critics challenge that measure).
Oh, maybe that's because algorithmically generated papers tend to generate long titles.
Check out the generated phrases here.
"First they came for the slanderers and i said nothing."
Yes, exactly. the R squared (variance explained) is tiny. So, yeah, the effect is there, but it's unimportant (and, as you point out, skewed by outliers). There's no assessment of normality of the data (it pretty clearly isn't), which also affects the validity of the results. And, finally, when you have a very large sample size, getting a "significant" result is very easy (20,000 data points is a very large sample size, for statistical purposes). Honestly, with 20,000 data points, I could "prove" pretty much any theory I chose about that data.
Many confounding explanations for the small correlation are ignored that might also have eliminated the observed correlation.
FWIW, I have a PhD, I do this stuff for a living. I got a "significant" result for one of my theories that had an R-squared of 7%. While I of course reported the significance, I also pointed out that it was of no real consequence, and probably due to sample size rather than a real relationship. Especially with the problems of Popper-style hypothesis testing, one should be very careful about what one reports as "real" connections.