Alternative clickbait post title: “To a first approximation, there are no fish & wildlife academic ecologists.”
Second alternative clickbait post title: “All ecologists want to live in the Pacific Northwest. Apparently.”
If you want to know the statistical truth underlying those clickbait jokes, read on. 🙂
Recently I invited readers to share their direct knowledge of how many applicants there were for recently-advertised N. American tenure-track faculty positions in ecology and allied fields. I also compiled data provided by the anonymous commenters on ecoevojobs.net from 2014-15 on. I was mostly just curious, and I suspect many of you are too.* These data are a (crude) window into the collective employment preferences of ecology faculty job seekers.
I also compiled, or was provided, various bits of possibly-relevant information about the hiring institutions:
- geographic location (latitude and longitude)
- size of the surrounding city/town (>100K people or <100K people)
- highest degree offered (bachelor’s, master’s, or doctorate)
- country (USA or Canada)
- R1 university or not (according to the 2015 Carnegie classification; I classed UBC as an R1 and the other Canadian unis in the dataset as non-R1. If you didn’t know, “R1” basically means “highly research-intensive university”)
- whether or not it was a fisheries/wildlife/natural resources job
I also looked at the “job season” in which the job was advertised (pre-2015, 2015-16, 2016-17, 2017-18, 2018-19), but there was no obvious effect on number of applicants so I dropped it from further analyses.
I recognize that a few of you may be annoyed that I didn’t collect more/different information. If so, please click this link. 🙂
I ended up with data on 86 positions (48 from ecoevojobs.net, the rest from my survey), with very little missing data. The vast majority of those positions were in ecology or an allied field, but I kept the few in other fields (e.g., evolutionary biology, systems biology) just to increase the sample size. Read the footnote for comments on data accuracy.** This is of course a small sample. It’s large enough to improve on anecdotes. But most of the conclusions I draw below are tentative, and there are other questions I simply can’t address. These data also aren’t a random sample. In particular, I suspect that positions that attracted many applicants may be somewhat overrepresented in this sample. The more people who applied for a position, the more people who received a rejection letter notifying them of how many applicants there were, and thus the more people who might share that information in my survey or on ecoevojobs.net.
Before you read any further, take a guess: which of the following choices do you think is closest to the median # of applicants per TT ecology faculty job? (No peeking at the answer below!)
[scroll down for the answer…]
Here’s a histogram of the number of applicants per job:
You’ll immediately notice the two most important take-home messages of this post:
- Number of applicants varies hugely. That’s the one conclusion that definitely wouldn’t change if you compiled additional data, and that is robust to any plausible inaccuracies in the dataset. There’s a TT ecology faculty position in this dataset that only got 12 applicants–and another that got 1000! The middle 50% of the data ranges from 61-175 applicants.
- The “typical” number of applicants is 100. That’s the median. That’s a lot! But it’s not as many as you sometimes hear (that’s cold comfort to job seekers, I know…). Anybody who tells you that 200+ is common or typical is misinformed. Only 21% of the jobs in this dataset got 200 or more applicants–not rare, but not common either, and definitely not typical. And for reasons I described above, I suspect that 21% is at least a slight overestimate of the true fraction of TT ecology jobs that attract 200+ applicants.
Ok, regarding that 1000 applicant job: that was actually several jobs. It was a broadly-written cluster hire in biology at the University of Washington. Same story for the second most applied-for job in this dataset: it was a broad cluster hire at Notre Dame. Here’s a graph of all the data by longitude of the hiring institution (for reasons that will become clear…), with the two R1 university cluster hires highlighted:
For purposes of further exploratory analysis, I dropped those two cluster hires from the dataset, then ran the remaining data through a regression tree in R using the rpart package. A regression tree seemed like a natural choice. It fits my mental model of how many faculty job seekers decide which jobs to apply for: they apply to any job that ticks certain “boxes”, with different people having different boxes. This was the first regression tree I’d ever run in my life, so I hadn’t the first clue what I was doing and just accepted rpart’s defaults. But the results seem reasonable in light of both my own priors and the exploratory plots I’ll show you in a sec. So here you go: here’s how would-be academic ecologists, collectively, decide which jobs to apply to!
The first split is between fisheries/wildlife/natural resources jobs, which tend to get relatively few applicants, and other jobs, which get relatively many. That’s visually obvious in the data:
Presumably fisheries/wildlife/natural resources jobs get fewer applicants than others (although still many in an absolute sense) at least in part because people with PhDs in fisheries/wildlife/natural resources often go on to work for government agencies, leaving fewer to pursue academic careers.
The second split on the tree is amusing to us here at Dynamic Ecology, because Brian predicted it. His “all ecologists want to live in the Pacific NW” hypothesis is totally, hilariously true. 🙂 The second split on the regression tree is between “longitude <-122.2 degrees” and points east. Non-fish&wildlife jobs in Berkeley/San Francisco and points west (all of which are also north of Berkeley/San Francisco) get about twice as many applicants on average as non-fish&wildlife jobs anywhere else.*** I suppose I should’ve guessed this, knowing how popular Portland, OR is as an ESA annual meeting location.
The next split in the regression tree is R1 vs. non-R1, with R1 jobs attracting more applicants than non-R1 jobs. I doubt that comes as much surprise. You can see this split in the data, though you have to squint a bit and ignore the Pacific NW jobs:
Finally, the last two splits in the regression tree are by longitude. If you squint at the scatterplots above, you can see ecology faculty job seekers as a group prefer to live near the coasts. The regression tree picks that out, with the aforementioned Pacific NW jobs attracting the most applicants, jobs in the Midwest (operationally defined as “west of Chapel Hill to Kansas”, longitudinally) attracting the fewest applicants, and jobs elsewhere attracting intermediate numbers of applicants. Actually, ecology faculty job seekers’ collective preference for the coasts might even be a bit stronger than the regression tree implies. It looks to me like some of the effect of longitude might be getting absorbed into the R1 vs. non-R1 distinction, because the dataset lacks data from R1 unis in the Great Plains and Mountain West. But it’s a small dataset, so it’s hard to say for sure.
I was slightly surprised that the regression tree didn’t pick out US jobs as getting more applicants than Canadian jobs, and I suspect that it would with a much larger dataset. And I was slightly surprised that the regression tree doesn’t find any effect of city/town size. But a city/town size binary is an extremely crude way to summarize “urban” vs. “rural”, not to mention other geographic factors that commonly enter into faculty job seekers’ decision-making.
Finally, note that there’s a lot of unexplained variation within each of the tips of the regression tree. Number of ecology faculty job applicants varies hugely even among apparently-similar jobs. That’s for all sorts of reasons, only some of them explicable.
Looking forward to your comments, as always.
*I confess I’m not sure what faculty job seekers, or anyone else, could actually do with this information. I mean, why would the likely number of applicants for a given position enter into your decision whether to apply for the position? Plus, it’s not as if knowing the number of applicants really tells you much about how “competitive” the applicant pool is. For instance, jobs that attract large applicant pools do so in part because many non-competitive applicants throw their hats in the ring hoping to get lucky. So if as an applicant you’re just looking for a reason to be optimistic (or pessimistic) about your chances for a given position, I’m sorry, but I don’t think a rough estimate of how many others applied is such a reason. But perhaps the data in this post could be useful to faculty job seekers in other ways I’m not thinking of? I’d welcome comments on this.
**I’m confident it’s accurate enough. I say that for a few reasons. First, there were a couple of cases in which multiple survey respondents gave me information about the same job. In both cases they gave the same number of applicants. Second, the survey emphasized that respondents should only report reliable knowledge, not n-th hand rumors. Third, in a couple of cases a survey respondent who sat on the relevant search committees reported the same number of applicants as was reported on ecoevojobs.net. Fourth, the aggregate statistical properties of the ecoevojobs.net data are similar to the survey data. Fifth, I removed from the ecoevojobs.net data a couple of jobs for which it appeared that the reported number of applicants might have been estimated as 10x the number of ecoevojobs.net users who applied themselves. (Aside: The “10x rule”, often mentioned on ecoevojobs.net, is too imprecise to be useful. Sorry. A linear regression of # of applicants on # of ecoevojobs.net users who applied, forced through the origin, gives a slope of 11.3 (i.e. close to 10) and an R^2 of 0.54, with no visually-obvious model misspecification. So if you insist in having a rule of thumb, “number of applicants = 10x number of ecoevojobs.net users who applied” is about the best rule you can come up with. But despite that impressive-sounding R^2 of 0.54, the prediction interval is quite wide. It’s common for the actual number of applicants to differ from that predicted by the 10x rule by several fold, and order-of-magnitude errors are far from unheard of.)
***Sorry Cal State East Bay. According to objective statistical analysis, you are not in the Pacific Northwest. #kiddingobviously