- Permanent researchers publish more when they are in smaller labs.
- Having many Ph.D. students fails to improve productivity.
- Funding has little effect on research productivity.
Reference:Â Carayol, N. and Matt, M., Individual and collective determinants of academic scientists’ productivity, Information Economics and Policy 18 (1), 2006.
Further reading (on this blog): To be smarter, ignore external rewards, Is collaboration correlated with productivity?, Big schools are no longer giving researchers an edge?
@Mikael
Fred Brooks book on the “design of design” hints that “teamwork” is bad for design. There needs to be one individual who understands the whole thing, whatever it is.
Moreover, because of communication overhead, throwing more people at a research project is not necessarily going to help it along.
My *personal* experience with large, generously funded projects is that they are wasteful. Usually, good stuff comes out of it, but almost by accident and often because people decided to do “their own thing” on the side.
I’m not entirely surprised by the results: once you start accruing a whole bunch of PhD students, your energy’ll go into management and advising, instead of research.
Does the report say anything interesting about where the sweet-spot for size of research group is at? Does your own experience?
Interesting conclusions. Thank you.
However, from a PhD candidate’s point of view, it’s still better to be at the receiving end of a big group.
@Waleed – is it though? Why?
It’s not at all clear to me that being part of a big group is necessarily more beneficial for a PhD candidate. How much is the size of the group support, and how much does it end up being about everyone desperately vying for any attention at all from the professor?
What is the argument behind it being better?
Being in a big group means exposure to different but truly related topics in the field, meaningful discussions over lunch, colleagues to consult next door, networking and co-authoring with more people …etc.
A friend of mine in a large IR group in DCU is really happy with what he’s getting from his peers.
I agree it’s hard to get supervisor attention, though.
Are these results different in industry? For example:
* SDEs are more productive in small teams
* Adding many junior SDEs fails to improve productivity
* Budget has little impact on team productivity
I think those are almost certainly all true as well?
@Linden
Yes. I expect these observations to hold true for any design-oriented work.
Nevertheless, conventional wisdom is that (1) you should try to join or create large teams (2) recruit as many junior people as you can (3) attempt to increase your funding as much as possible.
@Lemire
Mythical Man Month tried to debunk that conventional wisdom 35 years ago, at least for software projects, but, right, it continues.
@Greg – In my experience, I have witnessed evidence that first and third results are perfectly valid in industry. But I don’t have evidence to support or reject the second (i.e. adding junior developers).
waleed
I don’t think the data supports this conclusion, though it might be true anyway.
Of course, there is no absolute truth is such matters. Scientific productivity is an extremely complex object to measure.
the average researcher had less than a single Ph.D. student (…) In other words, none of the researchers had “many Ph.D. students.â€
I would expect the distribution to be very uneven (skewed). In fact, I expect it follows a power law where 80% of the Ph.D. students are with 20% of the faculty members.
The question is whether the individuals who supervise many Ph.D. students fare better than those who do not. Notice that the authors correct for co-authorships. Hence, if you have many co-authors, you are penalized accordingly.
I’m also a little confused about the European system. Is it the case that researchers that advise Ph.D. students are in university-professor-type positions and are required to teach classes, whereas full-time researchers don’t teach but also don’t advice Ph.D. students?
Full-time researchers in France do supervise Ph.D. students.
@Anonymous
The correction for co-authorship part is important.
If you don’t introduce this correction, then collaboration always boost automagically the productivity, as a paper with 5 authors counts 5 times.
So, yes, if you have 6 students and write a dual author paper with each one of them, this is 6 papers. But you could have just easily written 3 papers on your own.
“Having many Ph.D. students fails to improve productivity.”
I don’t think the data supports this conclusion, though it might be true anyway. I noticed that the average researcher had less than a single Ph.D. student. In other words, none of the researchers had “many Ph.D. students.”
I’m also a little confused about the European system. Is it the case that researchers that advise Ph.D. students are in university-professor-type positions and are required to teach classes, whereas full-time researchers don’t teach but also don’t advice Ph.D. students?
@Lemire
The correction for co-authorship part is important. I was just assuming that researchers got full credit towards their production for their students’ papers, but instead the credit is shared.
Also, besides the average being less than one, none of the researchers had more than three Ph.D. students. I would have liked to see some groups in the 5-10 range. The mean, SD, min and max are in the paper.
What I have found is that in mid-tier universities having more students helps in the following way. If you have one student and the student is bad, then, your time is wasted on that student and not much gets done beyond what you are doing yourself. On the other hand, if you have more students, you can see more of your ideas implemented because although some students do not contribute others do. So, from an absolute productivity point of view with respect to the lead researchers perspective, having more students help. Of course, the average may be less because inevitably in a large group there are some students who are slackers and the lead researcher does not have the time to “drive” him or her.
A lot of the variables may suffer from endogeneity (or reverse causality), so I think the paper should use instrumental variables to check for robustness.