Friday, March 27, 2015

Why I don't plan to take PhD students in the future

In a recent Walkabout I let slip that I don't plan to take PhD students in the future. That triggered some comments on FB wanting more details, so here goes.
Before I go any further, I should make one thing very clear. This decision was not a result of my experiences advising PhD students. Max, Adam, and Trevor were great students, and those experiences are not why I am choosing to stop training PhDs.

I'm also currently advising 2 PhD students who fall into a different category altogether. Both of them are already full time employed and are looking at getting a PhD as part of their ongoing professional development. Both are "self-funded" as well. They're also not who I'm talking about; I wouldn't rule out having such students in the future if they can convince me that my help in getting a PhD will benefit their careers.

The question that led to this decision was a simple one: "Why am I training PhD students?". From a decision analysis perspective, what are my objectives? One reason is that I kept hearing that people like me are in high demand as researchers and analysts. So, producing clones of myself* seemed like a good service to the field of conservation and wildlife ecology, and it appeared to be a good career path for the students as well. I was also keen to "pass forward" the great experiences I had as a graduate student in Bernie Roitberg's and Hugh Possingham's lab groups. In both places I benefited from being a part of a bigger group of people interested in hard problems. I wanted to create that at UNL. I failed to achieve that largely because of how students are funded in my academic unit. More recently, I've decided to make a virtue out of a necessity for reasons I'll outline below.

Structural problems with funding

One thing I failed to appreciate before coming to UNL was how much Canada and Australia do to support work like mine by providing direct funding to graduate students. At present, NSF provides 3 years of funding to 2000 students a year with the graduate fellowships program. In 2014 NSERC offered 710 postgraduate scholarships - Doctoral, each of which would provide up to 3 years of funding. In addition, NSERC offers 2000 postgraduate scholarships - Masters, each providing 12 months of funding. Remember that there are approximately 10 x fewer Canadians than Americans on the planet, so NSERC is funding graduate students at double the rate of NSF, at a minimum. Sure there are other places to get postgraduate scholarships in the USA, but there are in Canada too. I think those funding levels from NSERC are likely down from the past as well, in addition to being distributed differently from when I received similar awards**. Australia has a similar system of funding postgraduate students directly.

That postgraduate funding allowed me to spend a significant amount of time learning how to do modelling and statistics without having to produce a product. Even after the NSERC funds were exhausted I was able to get funding from SFU as a teaching assistant and graduate research assistant, neither of which were tied to a grant with specific milestones. In contrast, the School of Natural Resources (SNR) at UNL largely relies on faculty obtaining grant funds to support graduate students. There are a handful of graduate research assistant-ships available (and one of my MSc students received one) each year, but competition for those is steep. So, to get a student you first have to get a grant and then recruit a student to fill the position. The trouble is, doing what I can get money for means that a student has to come up to speed on complex modelling and analysis very fast. Unfairly fast. And, because there is no "backup system" in SNR for supporting students, if I can get money for 1 year it isn't enough to recruit a student on. 

This uncertainty in funding issue makes recruiting quality students harder, and it doesn't just affect me. Other colleagues with experience at other institutions complain about the same thing in SNR. When the competition can offer a guarantee of 4 or 5 years of funding (mostly from Teaching Assistant-ships, it is true), we have to be that much more attractive as a place to be. I had to tell Adam, "I've got two years of funding and I promise I'll try really hard to get you some more funding".  He came anyway, for which I am extremely grateful! 

The best strategy I've found is to sell my time, and then using the resulting salary savings to support a student like Adam. Unfortunately that caps the amount I can spend on students because I only have so much time to sell. 

So, as a result of being in the USA, and more specifically, in SNR, I'm never going to be able to build the kind of large, dynamic lab group that I benefited from. 

Making a virtue from a necessity

So I have trouble funding graduate students. So what? The bottom line is there are too many PhDs already. Here is one list of studies documenting that fact. Here's another chart (admittedly about humanities, but broader point is true too). An excess of PhDs means that getting a job that actually needs a PhD is increasingly difficult according to NPR*** and the Washington Post. So, I begin to think I shouldn't be contributing to this problem. 

But even if there are too many PhDs in general, that doesn't mean that people specifically like me won't be highly successful. However, when I look back at my three successful PhDs I couldn't help but notice a difference between Max on the one hand and Adam or Trevor on the other. Max did not acquire a degree in statistics alongside his Natural Resources PhD, while Adam took an MSc in Statistics and Trevor has a joint PhD in Statistics and Natural Resources. Even though all three are successful, I would say that Max had the hardest road to a permanent position; both Adam and Trevor had permanent or next-to-permanent employment straight out of their degrees. And it was the statistics degrees that made that difference. And there's the rub: I can't advise statistics students, because I myself do not have a degree in statistics.

When I look around at the people publishing papers that I wish I'd written, I can't help but notice that they usually have a PhD in statistics (Andy Royle, Jim Nichols, Ken Burnham, Mevin Hooten ... the list goes on). I've had papers rejected by reviewers saying they wanted to see the result produced 'by a real statistician'. So the main conclusion I take away is that I've ventured much, much too far away from my expertise and credibility as an ecologist. If someone really wants to be successful doing what I do (exclusively data analysis and modelling), they are much better served by pursuing a degree in Statistics.

I should say that plenty of my colleagues excel by doing a lot of data analysis coupled with good field ecology (Jim Peterson, Mike Conroy, Larkin Powell, just to name a few). But they largely have students who do field work, and then do a great job of analyzing the resulting data and building simulation models to make predictions. That is still a very viable path for a student to follow, although picking up an MSc in statistics makes it even more credible. In contrast, my students generally have not collected data. I don't collect data. What I do is pretend to be a statistician, and that's the problem. Truth will out.

What's the fix?

I imagine there are a few ways I could try to continue advising PhD students. For example, I could try replicating the experience I had with Trevor, co-advising statistics PhD students. This works great for my spouse, Brigitte Tenhumberg, who typically has 1 or 2 students co-advised with colleagues in the Math department. For that to work, I have to bring something to the partnership. In Trevor's case, co-advising was key to accessing the NSF IGERT funds. Normally however I don't have funding to bring to the table (see above); in fact, Statistics has much more support for graduate students as Teaching Assistants. I also don't really have data or problems to provide -- pretending to be a statistician, remember?

I could try the Adam model again, selling my time as an SDM coach to build up the funds necessary to support a student. Having worked as a consultant a few times, I can say that the market is pretty thin. There are plenty of people coaching for free (like Max!) within the USGS and USFWS. Also, keeping the money within the University is problematic. In the first place, I have to sell a large portion of my time at my salaried rate to make enough to pay a PhD student. In the second place, the University doesn't like it if you hang on to those salary savings too long. I've had such funds disappear from my accounts altogether after fiscal year end, and had to go find them again!

I could also go back to being an ecologist, getting funding to collect data and do a good job of analyzing it. Trouble is, I've never developed the credibility of a good field ecologist either. Funding agencies don't hand out money to just anyone, especially these days. I am trying, but I expect this path was foreclosed a long time ago.

Or, I could simply stop advising PhD students altogether. This frees me up to focus on my special niche, which is helping graduate students develop the basic data manipulation and analysis skills they need to be successful in the 21st century. I seem to be good at that, possibly because I had to learn it all the hard way without the benefit of a solid background in statistics. My colleagues around me have plenty of students that need this training.

So there it is. Funding PhD students is hard, and training students to be like me is not an optimal career path for the student, so I'm not going to do it anymore.

*  Note that I use 'clone' here to be a bit funny; none of my students is exactly like me. In fact, I'd argue that all of them have wildly exceeded me.
**For example, the PhD scholarships come in 2 different levels of support, whereas back in the day there was only the lower level of support. If they discontinued the higher level of support they could fund nearly 50% more people for the same total cost.
***NPR link has PhD comics, if that helps you decide which one to visit first.


  1. Thanks for the thought-provoking post! After being released into the wild about a year ago (i.e., I defended and graduated) and interviewing for several faculty positions, I see the structural problem with funding as a huge limitation for students in wildlife biology who want to obtain a formal training in statistics. In short, I don’t know how I could produce students in that setting unless they were self-funded or received an NSF fellowship.

    Many statisticians that focus on wildlife or ecological questions were not statisticians from the beginning (Including me!); they received undergraduate and graduate degrees in biology/ecology/wildlife biology. I think one reason statisticians like Andy Royle and Ken Burnham are successful is because of their ability to connect statistics to another discipline. As the number of applicants to graduate programs in statistics increases, it may become more difficult for students who did not major in mathematics or statistics as an undergraduate to obtain a graduate degree in statistics. Without a pre-existing interest in another field, it is not clear to me where the next generation of ecological statisticians will come from.


    1. And was that a soft or hard release? :)
      Interesting thought about the difficulty of getting into a graduate degree in statistics.
      I think the next generation will come from the same place as the last generation -- from the subset of statisticians that cross over in terms of interest. I've seen enough examples of people developing a professional interest in ecology during graduate school in math and statistics that I'm not worried about supply.

  2. If statisticians and mathematicians will become interested enough in ecology to become the next Andy Royle or Ken Burnham, then why should people like us be remotely concerned about training (graduate level) ecological statisticians? I think what you have observed may be more common for mathematicians than statisticians. I think you are under valuing how much time and interaction with scientists it takes to understand a specific scientific field. As we move away from more standard models (e.g., lm(...)) to custom models for specific applications (e.g., N-mixture), having a background in another field is what a large number of statisticians will be lacking. You might get a few academic statisticians to become interested in ecology, but on the whole someone with no experience in ecology is not going to take the time later in their career (post PhD) to understand ecology. At least that is my current projection. This is a good discussion to have!!!

    1. Post PhD you are correct, there are not many that will cross over, and that's true in Mathematics as well. Which is why the co-advising PhD students is a great path to training people with the understanding needed to be successful in both disciplines.

    2. Oh, and exactly, I'm not concerned about training graduate level ecological statisticians anymore! You can do it once you pick which statistics department you're going to work in :)

  3. Thanks for the post, Drew.

    I recall an anecdote when I was taking Bayesian statistics there at UNL. I was the only non-statistician in the class, but also the only student who knew how to program, simulate data and finish the homework. Anyway, our professor told us that there was a special place in hell for consulting statisticians. His thinking was that ecologists should analyze their own data and set up their own research designs. Academic statisticians should develop theory and methods, not analyze other people's data. His comments have stuck with and I personally view ecological analysis as part of being an ecologist.

    You should also consider, Drew, that not all of your students really care about how marketable they are. Some really value and admire what you have to teach them and are willing to make a go of it regardless of the job prospects. I know my supervisor thinks my skill set is unique and I wouldn't think it was as valuable if everyone had the same training.

    1. Yeah, I remember that story too. I didn't mean to imply that I didn't add ANYTHING to your education! I think Trevor and Adam also got a good dose of programming & simulation they wouldn't have from a pure Statistics degree.
      I also view analysis as part of ecology, but the issue is with how much "analysis" vs. "data collection" is the right mix for any one person, and if you're going to fall on the "pure analysis side", what are the best credentials to get. You and I are obvious examples that it is *possible* to succeed as an analyst without the statistics degree. The question is whether following in my footsteps *exactly* is the best advice I can give a future student. I've concluded that no, it isn't the best advice.
      Let me illustrate with another anecdote. I recall hearing from someone in the USGS who was trying to fill a postdoc position for which someone from my lab had applied. They didn't make it to the interview stage because they didn't have a statistics degree, even though the ultimate supervisor thought they would be a great fit for the job with the right skills. That was an eye-opener to the importance of having the right bit of paper, even in academic/research circles.
      I also know that lots of ecologists value what I have to teach them. I hear that a lot! And, what I've decided to do is focus my attention on those students that want to do ecology, but also do a good job of analysis. They're not going to develop new methods or analyse other people's data, but they do need some serious skills to succeed in the 21st century. Given the structural/funding constraints, I will have a much larger impact by focusing my efforts on those folks, rather than trying to train PhD students like you, Adam, and Trevor.

  4. Thanks for this post, Drew. I had only a foggy idea of the challenges of funding PhD students in the U.S.A. and absolutely no idea that lacking a formal statistics qualification could be such a hindrance.

  5. I was wondering if your decision to stop taking on PhD students impacts your decision to serve on PhD committees. I feel very lucky to have you on my committee because it allow me to pick your brain on a semi-regular basis. It seems like serving on committees might be a great way to foster a community of students interested in tackling ecological statistics issues, similar to the lab experience you mentioned you had in Australia. It would also eliminate the pressure of trying to find funding or produce “clones.” Another added bonus would be that it would allow you to have control over the number of students asking you for help outside of EcoStats.
    I’d also like to add my perspective as someone who came into academia excited over ecology, and then got interested in the contribution statistics and mathematics can make to decision making. I’ve found it very challenging to dive from an ecology background into statistics. If I knew back in freshman year of college the skills I would want down the line, I probably would have taken a different path, but my passion for natural resource conservation issues was (and remains) the strongest driver of what I study. People with your skill set are invaluable to students like me. We have tried to improve our stats skills through stats/math coursework, but at the end of the day we are not about to “start over” by committing to a stats degree and will need folks like you and Trevor to be our translators.

  6. Noelle, short answer is no, I'll still be on PhD committees. That's pretty much an extension of helping people develop the skills they need to succeed by teaching my courses. That's why I need to be careful about how many committees I agree to be on; all that 'semi-regular brain picking' adds up fast!
    The idea of pulling people together more regularly is an intriguing one. I do already control the number of people I help outside ecostats by restricting help to those whose committees I'm on. So by trying to meet at a regular time I might concentrate the 'semi-regular brain picking' into one time slot. On the other hand, people's problems tend not to occur at regular times, and often people have long stretches with no problems at all (really!).
    The other issue is finding a time that many different people could come to, and avoiding the appearance of interfering with other people's students.
    Your second point is spot on; ecologists need to learn more math/stat/computing than they typically have, but not necessarily enough to get a statistics degree. And, people like Max, Adam, Trevor and I, people that understand both statistics and ecology will be necessary. The point I was trying to make was that I may not be able to meet both of those training needs myself, given the circumstances I find myself in. But that's OK.

  7. This is from Chuck Frost, who had some IT troubles but could post to FB.
    Wow, my federal google account blocks me from commenting in your blog...haha. Anyway, a couple brief points: I was once offered a job with a state agency that was eventually blocked by their Human Resources because "what would the other biometricians think if we hired a biologist to lead them?" I made it through several interviews and left them thinking I was the best statistician for the job...then they saw my degree. I also got a job that had a ridiculous botany requirement (that I certainly wouldn't need to succeed in it) by claiming Ecological Detection and EcoStats were actually plant courses. Thankfully no course descriptions existed at the time. And today my position description says biologist, but last I checked, I'm a computer programmer, statistician, and general naysayer. You don't have to tell people not to be like you...your students (me included, even though you don't claim your co-advisorship...sheesh) learn to just want to. That's one of the few things I took from grad school, but I can attribute at least 90% of my success to it. So thanks.

    1. Chuck, thanks for the comment, especially the anecdote about the state agency. It sort of drives home the point that I was trying to make: the piece of paper matters more than the skills, at least in some cases.
      I didn't forget that I was your co-advisor! I thought about including you in there but didn't for a couple reasons. First, I don't know your post-PhD story as well as the others. Second, I became your co-advisor sort of late in the game, and so didn't have the influence on your PhD that I did for the others. Your PhD fell into the category of "collected data, did awesome job of analyzing data, built cool simulation models". AFTER your PhD you parleyed those skills into a pure data analysis/programmer career, which is awesome.
      To be honest, I'd have interacted with you about the same amount if I'd just been a member of your supervisory committee. That's the main reason why I need to be careful about agreeing to be on supervisory committees!