Duke Stat PhD vs. UNC Biostat PhD

[Geococcyx]

I've already talked to several people on the forums about this, but I thought it couldn't hurt to open the topic up to other folks. It also seems like I'm not the only person who's been comparing these two schools, so I hope it might be a useful thread in the future.

My research interests going in are primarily in applications to neuroscience, genetics, and climatology/environmental issues.  Other, smaller application interests would be in SMARTs/precision medicine clinical trials and sports statistics.  Methodologically, I'm probably most interested in robust model selection and record linkage, but let's not pretend I really know much of anything about what those entail.  I'm interested in both academia and industry.

Duke graduates people very quickly (4 years if you're going to industry or are a superstar, 5 years otherwise), and would obviously provide opportunities to branch out into application areas that aren't covered in biostatistics (e.g. time series with Mike West).  I'm mostly concerned about their retirement situation -- Gelfand already retired, I think Wolpert's moving to an active emeritus position this next year, and I'm not sure how active West and Berger will be in terms of taking on advisees in the future.  I know Duke is hiring several new professors this year, but I don't really know their names, so it's difficult for me to judge much of anything about their research areas or how productive they've been so far.

UNC obviously narrows down the available application topics a little bit, but does allow for SMART work more so than Duke.  At the same time, though, that seems to be Kosorok's area of research, and it sounds like he has a ton of students, which isn't necessarily something I'm a big fan of.  UNC sounds like it graduates people roughly a year later than Duke would (~5 years industry, 5.5-6 years otherwise), but they seem to have more professors than Duke, and also seem like they aren't in the same position as Duke in terms of having several potential retirements.  UNC may also provide a better statistical genetics group than Duke, headlined by Danyu Lin, but I don't think Duke's weak on that area (with Dunson, for instance).

I'd welcome any opinions on which program will have more strong research advisers going forward, as well as any other opinions on which program would be a better choice.  I'll note that I'm aware it's usually considered easier to go from a stat PhD to a biostatistics professorship than vice versa, but UNC does also (maybe?) require more measure theory than Duke, so I'm not clear how much that's the case here.

Edited March 15, 2019 by Geococcyx

[gc2012]

I can't fully answer your question, but I'm a current UNC biostat student, so I can offer a few clarifications about the program. First, (basic) measure theory is covered for one month, there is one test over the material (and measure theory is only half of the tested material), and then it's never brought up again (even on the qualifier). So I wouldn't go to UNC with the idea that you will get a lot of instruction in measure theory. I don't know anything about Duke, but for Duke stat to have less measure theory than UNC biostat, they would basically have to not cover it at all. 

Second, if you go to UNC, don't expect to do meaningful statistical research until after you pass your qualifying exam, which will be after 2 years if you come in without a relevant Master's (and if you pass on the first try). A small handful of students do a little bit of statistical research prior to passing their qual, but by and large, you basically won't do any work towards a thesis until passing. You will probably have to do some statistical programming for an applied project (i.e. GRA or training grant), but it's typically not the same as thesis work.

Third, this is more of a feeling than a hard and fast rule, but it seems to me that students don't really choose their faculty research advisor, the faculty choose their research students. So I wouldn't count on working with any specific research advisor at UNC. Also, for the higher profile faculty, they do tend to have a lot of students, and so many of them co-advise their research students along with a more junior faculty member with fewer time demands. 

[Geococcyx]

I gather that Duke typically allows students to choose their adviser, so perhaps that is another benefit of Duke?  Or does UNC have enough strong professors that their allotment procedure isn't an issue?

I should've noted above, but just to be clear, I have no preference for or against Bayesian statistics.

[little white in Stat]

@gc2012 Can you be more specific about that "students don't really choose their faculty research advisor, the faculty choose their research students"? It that is true, that is a not good news?...

[Geococcyx]

@little white in Stat I wouldn't feel quite so strongly about that, myself -- I doubt the professors would be particularly worse at choosing advisees than students are at choosing advisers.  I just want to help ensure I end with a strong adviser and have a decent chance to get a paper or two in Annals, Biometrika, JASA, or such.  

Even so, both students and professors would have some say in an adviser/advisee relationship in general.  It's possible advisers might tend to choose advisees more so than the other way around, but I kinda doubt you'd lose much autonomy.

Edited March 15, 2019 by Geococcyx

[geekstats]

I’m a current student at Duke. Masters, not PhD, but I have done research here and can give you some insights. From all the professors you mentioned, Berger is no longer taking students ( his last student graduated in 2017) and  Wolpert has very few students (only one that I know of). The only professor from your list(that I know of) who is actively taking students is West. Also, measure theory is a big thing here, particularly for the quals. If you need to know more about the department here, feel free to PM me.

Edited March 15, 2019 by geekstats

[gc2012]

@little white in Stat So the way it works is when you come to UNC biostat, you will get an academic advisor and a research advisor. Sometimes these are the same, but they are often not. The academic advisor will be biostatistics faculty, the research advisor may not be (i.e., they might be an MD with a grant for which you help provide statistical support). The work you do for your research advisor is what funds your stipend and tuition. That work is often statistical programming for some scientific study, not statistical methods development. You will meet with this person once a week.  

The academic advisor is (theoretically) the person with whom you will do your thesis research, but you likely won't have a lot of interaction with that person until (or if) you pass your qual. You don't initially choose this person, you are just assigned to someone. I believe the faculty review the application materials of all the incoming PhD students and divide them up. I suspect they try to match the research interests expressed in your application essay to an appropriate faculty member, but it's not a guarantee, and nobody will directly ask your opinion. 

Once you pass your qualifier, you theoretically could change academic advisors, but in practice I think many/most students just stick with their originally assigned person. If you do want to change, you would have to work it out with the faculty members in question. There isn't really a set way to do that, it's kind of ad hoc. For instance, where some programs like Penn have research rotations (i.e., 1-2 month periods where you work with different faculty members and then choose one at the end), we do not. And when it comes to faculty like Danyu Lin, Michael Kosorok, Joe Ibrahim, etc., basically everyone in the program would like to work with them so you can't just 'choose' them, you can express interest in their work, and they decide whether to take you (which for those 3 is not overly likely due to existing volume). 

 

 

[bayessays]

8 hours ago, Geococcyx said:

I gather that Duke typically allows students to choose their adviser, so perhaps that is another benefit of Duke?  

Any school that doesn't let you choose your dissertation advisor is not a school you want to go to.  Unless you have no other options in the top 50, I would have a hard time recommending going to UNC with this system in place. This matches up with my limited knowledge of their program. As recently as a few years ago, UNC biostatistics didn't even guarantee funding for PhD students after the first year. 

[gc2012]

I guess just to clarify, I'm not saying you will have zero agency regarding your dissertation advisor at UNC, it will just be limited. Possibly more so than other programs, but I can't really speak to that. Moreover, for better or worse, if you come to UNC biostat, if you pass your quals, you will probably be so relieved that you won't really care about who you work with for your dissertation. I'm not saying that's a positive trait, but I think a large number of students in the program would say the same thing. 

[Geococcyx]

I recognize that I'm spoiled to already have this many replies to my thread, but since a lot of the debate has concerned how people choose their PhD advisers at UNC biostat, I thought I'd just post here to reopen the floor in case anyone had comments about the other aspects of either of these programs.  If not, then thank you for all the information, everyone!

[StatsG0d]

On 3/15/2019 at 12:35 AM, gc2012 said:

it seems to me that students don't really choose their faculty research advisor

This is false. I do not know a single person in the program who did not choose their research adviser.

 

On 3/15/2019 at 6:55 PM, bayessays said:

As recently as a few years ago, UNC biostatistics didn't even guarantee funding for PhD students after the first year. 

This is true, but I fail to see how this is relevant now. If the admitted student received guaranteed funding, it has nothing to do with whether UNC guaranteed funding to other students or not. 

 

On 3/15/2019 at 5:30 PM, gc2012 said:

The academic advisor is (theoretically) the person with whom you will do your thesis research, but you likely won't have a lot of interaction with that person until (or if) you pass your qual.

Again, this is false. While some students do end up working with their academic or GRA advisor, I would not say it's the norm. Most of the students I know do not end up working with their GRA or academic advisors for their dissertation advisor. Moreover, your GRA advisor may/probably will give you some collaborative work to do.

 

On 3/15/2019 at 5:30 PM, gc2012 said:

Once you pass your qualifier, you theoretically could change academic advisors, but in practice I think many/most students just stick with their originally assigned person. If you do want to change, you would have to work it out with the faculty members in question. There isn't really a set way to do that, it's kind of ad hoc. For instance, where some programs like Penn have research rotations (i.e., 1-2 month periods where you work with different faculty members and then choose one at the end), we do not. And when it comes to faculty like Danyu Lin, Michael Kosorok, Joe Ibrahim, etc., basically everyone in the program would like to work with them so you can't just 'choose' them, you can express interest in their work, and they decide whether to take you (which for those 3 is not overly likely due to existing volume). 

 

 

It is true that it is ad-hoc and very informal. I simply sent out emails to some professors I thought about working with. All the professors whom I emailed responded positively. Ibrahim has a rule that you must be co-advised by a more junior faculty member. Not sure about Lin, but I never felt like he was that popular of an advisor. Kosorok does have many students, but my impression is that he is always willing to take on more students.

[biostat_student]

I wanted to second the above message. As as current PhD student in the department, I can tell you that the claim that "students don't really choose their advisor" is completely untrue. Everyone I have met has picked his/her advisor, and the vast majority of students I've interacted with thoroughly enjoy their research and has a positive relationship with their advisor. Additionally, almost all admitted domestic students are guaranteed funding, and if you have already received confirmation of funding yourself, the claims about non-guaranteed funding are irrelevant.

Apart from that discussion, a positive about UNC is that many more students are funded through GRA's than teaching assistantships. This allows for much more exposure to research and programming skills in early years rather than teaching obligations. Additionally, though research assistantships across PhD programs technically expect work for up to 20/hrs a week, this is rarely true at UNC except for specific cases. Not guaranteed, but in your first and second year, you can generally expect to have to work no more than a couple hours a week. Most students who do more than that do it not because of requirement, but through their own desire to expose themselves to research practices early. Thus, UNC offers the opportunity to get involved in projects at an early stage, while giving students the opportunity to focus more on coursework early on if they so desire.

I don't know about your interest in data science specifically, but UNC Biostats has also revamped its data science curriculum over the last year, in order to provide strong data science training as part of the PhD. Helpful especially from an industry recruitment perspective. All first-year students now take a revamped introductory data science course (course website) which is updated yearly to include modern methods and keep up with industry practices. Additionally, you can follow up that course by taking Statistical Computing (also a revamped course so message me for the updated syllabus) and PhD electives that build on this knowledge, such as BIOS740 (Precision Medicine). Again, don't know about your interest in machine learning, but UNC Bios houses a lot of UNC's overall expertise in machine learning, perhaps as much as the computer science department itself. Along with the revamped data science focus, this department is one of the strongest theoretical departments within biostatistics, which bodes well for academic prospects. You'll work hard, but the program is well-balanced between theory/application and built to set up both academic and industrial prospects.

Edited March 18, 2019 by biostat_student

[gc2012]

It's a big program, we seemingly know different people. Most of the people I know in the program have stayed on with the faculty academic advisor to whom they were assigned when they came in. I also would agree that most of them are satisfied with their advisors. That being said, it's clearly a debatable point and a bit of a semantic argument, but I still argue that the student/advisor pairing is far more related to faculty preferences than students. And I think it's misleading to suggest that any student can simply choose to work with any advisor. 

Second, I think it's valuable to be clearer about what 'guaranteed' funding means. I also have, in writing, 'guaranteed' funding. However, that does not mean the department has a specific grant designated to fund me throughout the rest of my time in the program (they might have this for some students). Rather, I've had year to year GRA appointments, and at the end of each year, I have to talk to the student service managers, they send out my CV to the faculty and other affiliated researchers, and I hope that something comes through. This has worked out thus far, but it's stressful and it's not clear what would happen if nothing materialized. Other students, I think, have more stable sources of funding, but at least in my case, this is 'guaranteed funding' at UNC. 

Also on that note, it varies wildly by GRA as to whether you will be asked to work 20 hours per week. If your GRA advisor is kind and can afford to pay you for nothing, you might only have to work a few hours per week. However, they are fully within their rights to demand 20 hours of work per week (I have had these types of GRAs), and I know people who have had GRAs where the researcher basically bullied them into working more than 20 hours. I don't think that's common, but it's common enough that student service managers have to intervene (and brought it up to me, unsolicited). Sometimes a 20 hour GRA is workable, but other times (like your qual year), it isn't. 

Last, again a matter of opinion, I wouldn't say the data science curriculum is a particularly noteworthy feature of the program. Basically they now require first year students to take an introductory course in R and (a little python) instead of SAS (which was previously required, and still heavily used in core courses). The precision medicine and statistical computing courses are newer, so I can't speak to them. I would note that they are electives and therefore might not be offered every year (although I think they plan to do so for computing).