7 ms·

Excuse a question from a total statistics illiterate: > Using a transformation of the h-index as our indicator of research output, we find male research output

Excuse a question from a total statistics illiterate: > Using a transformation of the h-index as our indicator of research output, we find male research output to be 0.35 standard deviations (p < 0.001) above female research output. However, the gap falls to 0.13 standard deviations (p < 0.001) when years publishing is controlled for. What does this mean? What does "years publishing is controlled for" means?

jjitz 4y ago
It basically means instead of measuring "papers published" they chose to measure "papers published per year"
- davisoneee 4y ago
  No, they mean that they control for 'career length' ... e.g. an academic with 5 years of history is likely to have a better score than one with 2 years of history. So they effectively compare women-with-1-year vs men-with-1-year, etc. etc, rather than 'women with h-index 5' vs 'men with h-index 5' edit: should be "Women with h-index of X after Y years" rather than just "Women with h-index of X" (e.g. they control for Y years' publishing between men and women, assuming that time-in-academia is correlated with h-index. Quick glance at the paper suggests that time-in-academia has R2 of 0.62 with h-index)
  - dandare 4y ago
    Thanks. And what does the standard deviation 0.13 means? I just don't know how to interpret the result.
aidenn0 4y ago
Assume for a moment that research output is affected by time in the field; either older people publish less due to resting on their laurels, or older people publish more due to greater experience allowing them to publish with less effort; I don't know without reading the paper which is true. Now consider what happens if men are (on average) more experienced because up until recently more men entered the field than women, but then the trend reversed. If you just compared female to male research output you could come to the wrong conclusion; the difference in experience could overwhelm the difference between sexes. So there are various statistical tools you can use to measure the difference due to experience, and subtract it out from the difference between sexes to get a more accurate measure of the difference due to sex. Doing this is called "controlling for a variable"

jjitz 4y ago

It basically means instead of measuring "papers published" they chose to measure "papers published per year"

davisoneee 4y ago
No, they mean that they control for 'career length' ... e.g. an academic with 5 years of history is likely to have a better score than one with 2 years of history. So they effectively compare women-with-1-year vs men-with-1-year, etc. etc, rather than 'women with h-index 5' vs 'men with h-index 5' edit: should be "Women with h-index of X after Y years" rather than just "Women with h-index of X" (e.g. they control for Y years' publishing between men and women, assuming that time-in-academia is correlated with h-index. Quick glance at the paper suggests that time-in-academia has R2 of 0.62 with h-index)
- dandare 4y ago
  Thanks. And what does the standard deviation 0.13 means? I just don't know how to interpret the result.

aidenn0 4y ago

Assume for a moment that research output is affected by time in the field; either older people publish less due to resting on their laurels, or older people publish more due to greater experience allowing them to publish with less effort; I don't know without reading the paper which is true. Now consider what happens if men are (on average) more experienced because up until recently more men entered the field than women, but then the trend reversed. If you just compared female to male research output you could come to the wrong conclusion; the difference in experience could overwhelm the difference between sexes. So there are various statistical tools you can use to measure the difference due to experience, and subtract it out from the difference between sexes to get a more accurate measure of the difference due to sex. Doing this is called "controlling for a variable"