Preparation for (Bio)Statistics PhD

[SchoolboyQ]

Hi all, I know there are some posts on this topic already, but I figured other people might currently be making the same type of decision. Right now, I think that I want to eventually go for a PhD in either stats or biostats (I'm undecided, I like the more applied nature of biostats but I would not want to limit myself to only applied work if I can handle the theory). I didn't think my background/research experience was good enough for good PhD programs, so I'm hoping to use a MS in biostats/statistics as preparation for the PhD. At this point, my choice likely comes down to the MS in biostats from UNC or the MS in stats from Minnesota.

I know that having a strong theoretical foundation is of primary importance for PhD applications so my inclination is to go to Minnesota as statistics programs are (in general) more theoretical than biostatistics. However, I've read on this forum that UNC biostats is known for having a pretty theoretical program (not sure if this applies to the Masters?). Also, I think that I'd have better opportunities to be involved in research in a biostatistics department rather than a statistics department.  

Which program do you think is better for PhD preparation? Do you think doing UNC's biostats program would hurt my chances for a statistics PhD? Possibly important information: I graduated with a BA in math and got an A in real analysis, but I haven't taken statistics since high school. Thanks for any help you can provide!

[Stat Assistant Professor]

Those are both solid programs. As long as both have the typical Casella-Berger mathematical statistics sequence and an applied sequence (usually it's on regression/design/categorical data), I would go with the option that is cheaper.

For PhD programs in Stat/biostat, research experience is not a big factor in admissions. Letters of recommendation and grades are the most important. 

[bayessays]

I would disagree about research, or at least add a caveat.  It is certainly not necessary, and opportunities for people to get involved in real stats research before being in a PhD program are very rare, but if you can do it, it will be huge for your application. This probably applies to less than 1% of applicants though - if anything, I roll my eyes at most people on this forum who think their undergrad research is relevant to what writing a dissertation will be like and I think it can come off as a negative. 

[bayessays]

And to answer your question, no, a UNC biostat MS will look great - take as many theoretical classes as you can and do well in them and you will be in great shape.

[GoPackGo89]

20 minutes ago, bayessays said:

I would disagree about research, or at least add a caveat.  It is certainly not necessary, and opportunities for people to get involved in real stats research before being in a PhD program are very rare, but if you can do it, it will be huge for your application. This probably applies to less than 1% of applicants though - if anything, I roll my eyes at most people on this forum who think their undergrad research is relevant to what writing a dissertation will be like and I think it can come off as a negative. 

What do you mean by coming off as a negative?

 

[bayessays]

8 hours ago, GoPackGo89 said:

What do you mean by coming off as a negative?

 

If you think your undergrad research where you ran some regression models on an applied problem resemble what you will be doing as methodological or theoretical research in a PhD program, it leads me to believe you don't know what you're getting into and I'd be worried you're going to be disillusioned quickly and not complete the PhD. This is a huge cause of dropout among stats PhD students - they thought they were getting into something much more applied. 

[Stat Assistant Professor]

I don't think undergrad research *hurts* per se, but it usually isn't given much attention by adcoms. When reviewing applications, many adcoms are most concerned that you are able to handle/complete the coursework. So recommendations, grades, and breadth of math classes taken tend to receive the most attention. 

There are exceptions, of course, but the truth of the matter is that most undergrad research in statistics -- and even Masters projects and theses -- bears little resemblance to completing a PhD dissertation. By construction, undergrad and Masters projects/theses need to be very limited in scope because they have to to be completed in a very limited amount of time. But with PhD projects, the time frame for finishing them is much more open-ended (in fact, research is never really "finished" because there are always new problems to work on and new extensions to be made!), and it takes considerably more time to get a publishable result. Sometimes the projects you start working on in your PhD end up being a dead-end and you have to give up and completely start over, but undergrad/MS projects don't tend to be that way. The papers you write for your PhD thesis also need to be of a certain quality.

[BL250604]

7 minutes ago, Applied Math to Stat said:

I don't think undergrad research *hurts* per se, but it usually isn't given much attention by adcoms. When reviewing applications, many adcoms are most concerned that you are able to handle/complete the coursework. So recommendations, grades, and breadth of math classes taken tend to receive the most attention. 

There are exceptions, of course, but the truth of the matter is that most undergrad research in statistics -- and even Masters projects and theses -- bears little resemblance to completing a PhD dissertation. By construction, undergrad and Masters projects/theses need to be very limited in scope because they have to to be completed in a very limited amount of time. But with PhD projects, the time frame for finishing them is much more open-ended (in fact, research is never really "finished" because there are always new problems to work on and new extensions to be made!), and it takes considerably more time to get a publishable result. Sometimes the projects you start working on in your PhD end up being a dead-end and you have to give up and completely start over, but undergrad/MS projects don't tend to be that way. The papers you write for your PhD thesis also need to be of a certain quality.

Just a question, since we're on this topic. Do you think it would be a benefit for me that my undergraduate research will be a total of 2.5 years, resulting in publications, and presentations. I intentionally made the choice to focus on my spatial statistics topic (rather than undertake multiple different projects), and tried to go as in depth as possible over the 2.5 years working on the project, to show some maturity in my research methods/thinking (if that makes sense).

[Stat Assistant Professor]

6 minutes ago, BL250604 said:

Just a question, since we're on this topic. Do you think it would be a benefit for me that my undergraduate research will be a total of 2.5 years, resulting in publications, and presentations. I intentionally made the choice to focus on my spatial statistics topic (rather than undertake multiple different projects), and tried to go as in depth as possible over the 2.5 years working on the project, to show some maturity in my research methods/thinking (if that makes sense).

Your case sounds like one of the exceptions. If your work is resulting in publications and you're intimately involved with the whole process, then it definitely helps your application. What you have is quite rare though. A lot of PhD students in statistics enter with Bachelor's degrees in mathematics where they did not do any research of note. 

[BL250604]

7 minutes ago, Applied Math to Stat said:

Your case sounds like one of the exceptions. If your work is resulting in publications and you're intimately involved with the whole process, then it definitely helps your application. What you have is quite rare though. A lot of PhD students in statistics enter with Bachelor's degrees in mathematics where they did not do any research of note. 

Thank you! And thanks for all of your insight on the forums, your tips, and the survival guide were great. All the best with finishing up your work, and landing at the post-doc you want!

 

[SchoolboyQ]

Thanks for the help guys. I just checked and both of these programs do teach the Casella-Berger sequence and an applied sequence. It seems like the biggest difference (other than the obvious bio/health focus of UNC's program) is that UNC's program has more required classes with fewer electives and UMN has more flexibility. Both require two upper level statistics electives but Minnesota also requires two electives outside of the department (they suggest math if interested in the PhD).

UNC is cheaper but I'm drawn to the colder weather/bigger city of Minneapolis - it's going to be a tough decision!

[Stat Assistant Professor]

I would agree with the UMN program description that if you want to get a PhD in Statistics, your best bet is to take graduate level math classes... possibly Real Analysis II, a measure theory class, an advanced linear algebra class, complex analysis, or a partial differential equations class.

 As a Masters student, I wouldn't bother taking an upper division PhD-level theoretical stats class (like Advanced Statistical Inference/decision theory, theory for Generalized Linear Models, or large sample theory), since you would just need to take these courses again in the PhD program. And some schools seem to want to teach these to you *their* way  -- a lot of it is the same at different schools, but the qualifying exams are likely to cover slightly different material. Showing mathematical maturity and getting great letters of recommendation which speak to your potential to excel as a researcher are the most critical components of your PhD application.

Edited March 21, 2018 by Applied Math to Stat

[SchoolboyQ]

On 3/20/2018 at 8:54 PM, Applied Math to Stat said:

I would agree with the UMN program description that if you want to get a PhD in Statistics, your best bet is to take graduate level math classes... possibly Real Analysis II, a measure theory class, an advanced linear algebra class, complex analysis, or a partial differential equations class.

 As a Masters student, I wouldn't bother taking an upper division PhD-level theoretical stats class (like Advanced Statistical Inference/decision theory, theory for Generalized Linear Models, or large sample theory), since you would just need to take these courses again in the PhD program. And some schools seem to want to teach these to you *their* way  -- a lot of it is the same at different schools, but the qualifying exams are likely to cover slightly different material. Showing mathematical maturity and getting great letters of recommendation which speak to your potential to excel as a researcher are the most critical components of your PhD application.

Would a MS in Math be better preparation in your opinion? I got into a funded MS math program at a well-regarded but unranked institution (they don't grant PhDs) but I wasn't seriously considering it because I got into UNC and Minnesota. You're really stressing graduate level math/mathematical maturity for the stats PhD so maybe I should reconsider?

[Stat Assistant Professor]

13 minutes ago, SchoolboyQ said:

Would a MS in Math be better preparation in your opinion? I got into a funded MS math program at a well-regarded but unranked institution (they don't grant PhDs) but I wasn't seriously considering it because I got into UNC and Minnesota. You're really stressing graduate level math/mathematical maturity for the stats PhD so maybe I should reconsider?

Well.. PhD admissions committees in Statistics would certainly view more mathematics courses (not necessarily more statistics) as a positive, but they also admit a lot of students with degrees in Statistics.  How much funding are we talking about? If you are able to take the Casella-Berger sequence and maybe one additional stat course in the Mathematics MS program to familiarize yourself with statistical concepts, then this is not a bad option. On the other hand, if the Statistics MS programs give you enough flexibility to take math courses, then that is also a good option.

[bayessays]

You said before that you didn't think your research background was good enough - getting a math degree isn't going to help you there. I'd slightly disagree with AppliedMathtoStat maybe, depending on your exact situation. If you've taken real analysis and did very well (and perhaps a few other courses stats programs sometimes look for like numerical analysis), then the marginal benefit of getting a MS in math is going to be small.  Getting a top MS in stats where you take theoretical classes shows you are interested in statistics, know what you are getting into and can complete the coursework. If you want to be a statistician, do you really want to spend two years getting a math master's? Are you sure your profile isn't good enough where you could apply to PhD programs next year without getting a master's at all and saving money?

[SchoolboyQ]

37 minutes ago, Applied Math to Stat said:

Well.. PhD admissions committees in Statistics would certainly view more mathematics courses (not necessarily more statistics) as a positive, but they also admit a lot of students with degrees in Statistics.  How much funding are we talking about? If you are able to take the Casella-Berger sequence and maybe one additional stat course in the Mathematics MS program to familiarize yourself with statistical concepts, then this is not a bad option. On the other hand, if the Statistics MS programs give you enough flexibility to take math courses, then that is also a good option.

Funding is full tuition remission + 1600/month. They don't use Casella-Berger (they use Wackerly-Mendenhall-Scheaffer instead?) but I would have the flexibility to take at least three (probably more) stat courses. It's a joint math-stats department so I think it's really flexible.

8 minutes ago, bayessays said:

You said before that you didn't think your research background was good enough - getting a math degree isn't going to help you there. I'd slightly disagree with AppliedMathtoStat maybe, depending on your exact situation. If you've taken real analysis and did very well (and perhaps a few other courses stats programs sometimes look for like numerical analysis), then the marginal benefit of getting a MS in math is going to be small.  Getting a top MS in stats where you take theoretical classes shows you are interested in statistics, know what you are getting into and can complete the coursework. If you want to be a statistician, do you really want to spend two years getting a math master's? Are you sure your profile isn't good enough where you could apply to PhD programs next year without getting a master's at all and saving money?

Well, I applied to a couple statistics PhD programs this year between 10-30 and got into 0 of them. I think it's probably because I've been out of school for a few years now and I lacked the direction and some of the math/stats background that other candidates might have had. I agree that I would prefer to get a MS in stats/biostats over a MS in math but if math makes me way more competitive then it might be worth it.

Thanks for your help guys - leaning towards UMN at the moment.

[Stat Assistant Professor]

7 hours ago, bayessays said:

You said before that you didn't think your research background was good enough - getting a math degree isn't going to help you there. I'd slightly disagree with AppliedMathtoStat maybe, depending on your exact situation. If you've taken real analysis and did very well (and perhaps a few other courses stats programs sometimes look for like numerical analysis), then the marginal benefit of getting a MS in math is going to be small.  Getting a top MS in stats where you take theoretical classes shows you are interested in statistics, know what you are getting into and can complete the coursework. If you want to be a statistician, do you really want to spend two years getting a math master's? Are you sure your profile isn't good enough where you could apply to PhD programs next year without getting a master's at all and saving money?

For admission to Statistics PhD programs, taking many more statistics classes and having research experience are usually not necessary (the latter is especially rare, and it is also not uncommon for students to be admitted to PhD programs who have never taken a statistics class before but who only have an extensive background in pure math). In fact, I have a suspicion that taking *too* many advanced PhD-level stat classes as a Masters student can hurt your PhD application. I think a lot of programs want to teach topics like advanced statistical inference, measure-theoretic probability, and theory of linear models/generalized linear models  to you "their" way in order for you to pass their qualifying exams.  But I agree with you that getting a Masters in mathematics is not going to automatically guarantee admission to a Statistics PhD program, so if statistics is the professional track the OP wants to go down, it may be more worthwhile to get a MS in Stat, supplemented with advanced math courses. 

7 hours ago, SchoolboyQ said:

Funding is full tuition remission + 1600/month. They don't use Casella-Berger (they use Wackerly-Mendenhall-Scheaffer instead?) but I would have the flexibility to take at least three (probably more) stat courses. It's a joint math-stats department so I think it's really flexible.

Well, I applied to a couple statistics PhD programs this year between 10-30 and got into 0 of them. I think it's probably because I've been out of school for a few years now and I lacked the direction and some of the math/stats background that other candidates might have had. I agree that I would prefer to get a MS in stats/biostats over a MS in math but if math makes me way more competitive then it might be worth it.

Thanks for your help guys - leaning towards UMN at the moment.

If that is the case, I would try to take a holistic view at your application. If other factors were strong (math grades, GPA, GRE score), then it may have been recommendation letters that did not "stand out." You can only speculate at this point, but if you eventually do decide to apply for PhD, you should do your best to ensure that everything within your control is strong.

[bayessays]

Everyone I know, including myself, who got a top master's degree in stats improved their results when they reapplied to PhD programs. I think there's a big difference between saying that the future PhD program will make you retake their class (which is almost certainly true because it will cover different material) and saying that they won't value that background (which I think is completely false). I don't think there's much evidence that having a math background beyond real analysis/numerical analysis helps in admissions, except maybe at a few top programs -OP isn't going to be applying to Stanford or Penn if he didn't get into a top 30 this time around. If you have good undergrad grades in analysis, I strongly think that graduate statistics courses would be looked upon more favorably than taking a bunch of abstract algebra.

[Stat Assistant Professor]

41 minutes ago, bayessays said:

Everyone I know, including myself, who got a top master's degree in stats improved their results when they reapplied to PhD programs. I think there's a big difference between saying that the future PhD program will make you retake their class (which is almost certainly true because it will cover different material) and saying that they won't value that background (which I think is completely false). I don't think there's much evidence that having a math background beyond real analysis/numerical analysis helps in admissions, except maybe at a few top programs -OP isn't going to be applying to Stanford or Penn if he didn't get into a top 30 this time around. If you have good undergrad grades in analysis, I strongly think that graduate statistics courses would be looked upon more favorably than taking a bunch of abstract algebra.

Maybe it depends on the program, but there are many PhD students in Statistics who never took any stats before matriculation (maybe they took an undergrad probability class but that's it).  If someone were to tell me, "I want to get a PhD in Statistics but I haven't taken much/any stats before. Can I still get into a PhD program?" I would probably tell them that this is not a huge issue as long as they have a very strong math background and great letters of recommendation. As per the OP, the UMN Statistics Masters description even recommends that those considering a PhD in Statistics take math electives to strengthen their application. 

[bayessays]

1 hour ago, Applied Math to Stat said:

Maybe it depends on the program, but there are many PhD students in Statistics who never took any stats before matriculation (maybe they took an undergrad probability class but that's it).  If someone were to tell me, "I want to get a PhD in Statistics but I haven't taken much/any stats before. Can I still get into a PhD program?" I would probably tell them that this is not a huge issue as long as they have a very strong math background and great letters of recommendation. As per the OP, the UMN Statistics Masters description even recommends that those considering a PhD in Statistics take math electives to strengthen their application. 

Right, I think we are just extrapolating the data here in a different way. Yes, I agree you can get into a stats PhD with a limited stats background. Yes, taking additional math classes related to analysis will help.  OP already has this met. The question is, once you have this background, what is most helpful additionally? The UMN thing you mentioned says most people take an intro analysis sequence but doesn't say anything about higher level math. Outside of a the top 5-10 schools, I just don't see any evidence that taking additional math beyond the couple classes will help.  Whereas I have seen lots of evidence that graduate level statistics classes show you can succeed in a grad stats program and know what you're getting into.

I think that the benefit of getting an MS in math is small compared to the amount of time put in and if you get the stats MS, you'll be ahead of your peers in the PhD applications and you'll have an advantage over other students in the coursework when you start your PhD.

Edited March 27, 2018 by bayessays

[statfan]

3 hours ago, Applied Math to Stat said:

I have a suspicion that taking *too* many advanced PhD-level stat classes as a Masters student can hurt your PhD application.

I disagree with this. Doing well in grad level statistical inference and/or probability serves you two purposes. (1) It shows your aptitude for grad level research. (2) Grad classes tend to be much smaller and it is easier to establish close relationship with your prof. This can result in a great letter from a well-known faculty member cuz grad course instructors tend to be more senior. Taking grad level analysis courses would definitely help too and if you do well in them, you will be well ahead of most applicants. However, this does not necessarily requires a math master. Taking abstract algebra/geometry certainly wouldn't hurt but under time constraint, you would want to focus on the most important things.

[Stat Assistant Professor]

Fair enough. I can't say I know for sure whether a Masters in Statistics or a Masters in Mathematics looks better to a PhD admissions committee in Stats, but anecdotally, my current program has admitted a number of students into the Statistics PhD program who obtained Masters degrees in (pure) mathematics (including some alumni who are now faculty/postdocs at places like Duke University, University of South Carolina, etc.). My MS is also in Applied Math.

Based on personal experience, a Masters in Mathematics would certainly not be a handicap in admissions to statistics PhD admissions. I'm not sure if a Statistics Masters is "better" for admission to PhD programs though.

[StatsG0d]

On 3/20/2018 at 8:54 PM, Applied Math to Stat said:

I would agree with the UMN program description that if you want to get a PhD in Statistics, your best bet is to take graduate level math classes... possibly Real Analysis II, a measure theory class, an advanced linear algebra class, complex analysis, or a partial differential equations class.

 As a Masters student, I wouldn't bother taking an upper division PhD-level theoretical stats class (like Advanced Statistical Inference/decision theory, theory for Generalized Linear Models, or large sample theory), since you would just need to take these courses again in the PhD program. And some schools seem to want to teach these to you *their* way  -- a lot of it is the same at different schools, but the qualifying exams are likely to cover slightly different material.

This is 100% true. Taking PhD-level courses can sometimes hurt an application.

[statfan]

35 minutes ago, footballman2399 said:

This is 100% true. Taking PhD-level courses can sometimes hurt an application.

Many students who went to top schools in the US in my school had taken multiple graduate level stat courses such as probability theory/statistical inference. I agree that real analysis/linear algebra are very important but other math courses are much less related to statistics. It may be a good idea to go to stat masters as a stepping-stone to phd and make up the real analysis background. 

Edited March 28, 2018 by statfan

[bayessays]

1 hour ago, footballman2399 said:

This is 100% true. Taking PhD-level courses can sometimes hurt an application.

Unless you do poorly I don't possibly see how this could be the case. In mathematics graduate school, taking grad classes is essentially a prerequisite for attending a good program. But somehow all stats programs have gotten together and decided that taking advanced courses in the discipline you're going to spend your life studying is somehow a negative thing?