2020 Statistics PhD Hopeful Looking for Advice

[playingstats]

Graduated undergrad in May 2018 and took a staff job as an RA at an Ivy campus. I passed on applying last year in part because I didn't know much about stat departments or statistics research, but mostly because my grades are a huge anchor to deal with. Since then I have taken additional courses and made a habit of reading 1-2 articles a week to learn what statistics literature looks like. My information:

Demographic: Domestic White Male, late 20's, US Military Veteran

Undergrad Institution: Small state school that no one's heard of

Major: BS Math 3.33/4.00, BS Applied Science 3.60/4.00

CGPA: 3.29

Minor: Business Administration

Awards:

• Outstanding graduating math student
• Outstanding graduating applied science student
• Service award
• Who's Who award twice
• Retention and athletic scholarships

Math courses and grades:

• Calc 1,2,3 (C, A, A)
• Intro stat (C)
• Diff Eq (A)
• Intro to proofs (B)
• Linear Algebra (B)
• Numerical Analysis (B)
• Mathematical Modeling (B)
• Complex Analysis (A)
• Linear Programming (A)
• Abstract Algebra (B)
• Real Analysis (B)
• Mathematical Probability and Statistics (A)
• Topology (B)

Grades are similar in applied science.

GRE: 162V -- 164Q -- 4 AW (91% -- 86% -- 59%)

Graduate coursework: Data science classes at Harvard Extension School

• Mathematical Statistics (A)
• Intro to Statistical Modelling (A-)
• Deep Learning with Natural Language Processing (A)
• Bayesian Inference (In progress)
• Stochastic Methods for Data Analysis, Inference and Optimization (In progress)

Research Experience: 1 science REU, senior thesis in applied science, assisted with projects in my (applied science) department as an undergrad. No research experience in math or stats.

Letters: Undergrad advisors from applied science and math. Probably one from the PI I work for now or from a professor at the Extension School. Each professor has assured me they'd write great letters for me.

 

I'm looking to apply to PhD programs this year or next year, and selecting schools to work with specific faculty. I thought I knew the subject I'd want to research after doing an undergrad thesis but I didn't feel ready to apply yet. Now I broadly know the area I want to study, I've identified great faculty in that subject, and I am pretty confident in what I want to do as a career after graduation. Now I need to convince the adcoms to look past my undergrad GPA, but I have no idea what I look like relative to other applicants.

Biostatistics PhD

• UCLA PhD

Statistics PhD (Sorted by interest!)

• Texas A&M
• Purdue
• Michigan State University
• University of Missouri
• Ohio State University
• U Florida (safety)
• Florida State (safety)
• Bowling Green State (safety)

How bad do I look relative to other applicants at this schools?

Edited September 3, 2019 by playingstats
Posted early

[omicrontrabb]

Realistically, you're not going to be admitted to a top 50 statistics PhD program with a 3.3 GPA from an unknown school. Your coursework at Harvard Extension is unlikely to help. Most students get As in those sort of programs. An admissions committee will be seriously concerned about your ability to do the advanced math required for a stats PhD since you have Bs in linear algebra, real analysis and numerical analysis.

I attended the visit day for Ohio State's stats PhD program this year. This seemed to be a typical admitted student:

• Attended a well known, but not prestigious university. Think something like UConn.
• Math or stats major with mostly As.
• Most had math or stats research experience.

I assume the admitted students at the other programs you listed would be broadly similar.

What is your goal? What are you hoping to gain from getting a PhD in statistics?

[Stat Assistant Professor]

@omicrontrabb nails it.

To the OP (and other prospective applicants): It is very important for prospective Statistics PhD applicants to know that Statistics PhD programs (much like math and computer science PhD programs) are designed to train you more in the spirit of a mathematician/scientist, not as an analyst or programmer who implements ML algorithms and does routine data analysis. If doing mathematical proofs and designing new methodologies and algorithms that require heavy knowledge of the underlying mathematics are not things that sound appealing to you, then I would recommend just getting a Masters. Just as you do not need a PhD in CS to become a software engineer, you don't really need a PhD in Stats to work in data science (with a few rare exceptions like research scientist positions at places like Microsoft Research, Google.ai, etc.  and obviously, for international students, it is easier to obtain run-of-the-mill data science jobs with a STEM PhD, but this shouldn't be an issue for you).

Edited September 3, 2019 by Stat PhD Now Postdoc

[playingstats]

This is all helpful. Do either of you (or anyone else reviewing this later) think I'd have a shot at admission into a Master's program at one of the schools I mentioned? There are a few points that I think are lost in translation at the moment which I want to address, because I believe my understanding of Statistics is better than you may perceive it to be.

For slightly more context, I am working in an engineering science research lab now, and I am very aware of what goes into conducting top-tier research in science. Statistics and mathematics is more guess work for me because it is the area I'm weakest in. But statistics is the field I want to work in.

To both of your points, I think I do understand what a PhD in Statistics signifies, and why there is a distinction between Statistics and Applied Statistics among departments. In my mind a PhD statistician from a top university is someone who has expertise in core statistical concepts, such as stochastic processes, sampling theory, estimation theory, and decision theory from both a frequentist and a Bayesian perspective; has graduate level exposure to linear algebra, analysis, and mathematical probability; and has advanced exposure to algorithms and data structures for scientific computing. (I think the quality of training in algorithms and scientific computing is the hardest to estimate among these categories, but it is clear that it is advanced.) Then of course there is the actual subfield that a candidate studied in their dissertation or presently pursues. Imaginably, a PhD will combine their statistical expertise in their area with practical knowledge of how applied research is conducted and what policies and regulations govern the same.

My goals for obtaining a PhD in Statistics are two-fold: one is to learn the specific area I'm interested in for a long-term career, and the second is to develop expertise in the topics mentioned before. The area I want to study is relatively old, but has rapidly expanded due to advances in computing technology. I suspect that it will become much more significant as we continue developing new sensors for measuring and logging data that the field addresses, especially as the market for those technologies grows in private industry. In that time, I think there will be an abundant need for people with expertise in that area in academic, government, and industrial settings, and I would be interested in making a career collaborating with people and organizations in each of those respective settings. My greatest preference would probably be to continue to a Postdoc or Assistant Professor position, but I would not turn away from the right industry or government position.

The place I'm at now is a dead end for the career I want to have. It is a top-tier research environment in an area that I have no interest in pursuing, and there are few perceivable opportunities to transition into a better place from here. I don't want to become a business analyst or a data scientist or a machine learning guru—I want to become a statistician. Realizing how bad the circumstances I'm in now are, I am frustrated that I allowed myself to get here. By next year I want to correct course and start in a program that has higher odds of leading into the area I want to research in.

[bayessays]

Sounds like you have a better understanding of what a PhD would be like than half the people I know who started one. I agree with the above posters that you're not going to get into any of those schools (besides perhaps Bowling Green, and maybe an unfunded offer from FSU).  Realistically, I don't see you getting into a program ranked above 60 right now. If you want to apply to PhD programs, I would start around there. Of you don't want to be a professor at a PhD-granting school, it matters much less where your degree is from, especially for government/most of industry. If you would like to aim higher in a couple years, I think you could likely get into a decent MS program, including at some of those schools you mentioned (although I really don't know much about MS admissions). In order to improve your profile, make sure to go somewhere where you can take mathematically rigorous classes, get As in them, get to know some accomplished professors to get letters from, and perhaps do some research, and you can improve those results. 

[omicrontrabb]

I agree with @bayessays  You mention that you would like to ultimately become a professor. 

Your most realistic path to that is probably something like this:

• Getting a masters degree. You could probably get into somewhere around the level of FSU. According to the FSU website, their average admitted GPA was around 3.3. You're probably not going to get funding.
• Do well in your masters. Get As in mathematically rigorous classes and get some research experience.
• Some programs allow you to transfer from the MS to PhD or apply to other programs around that level.

If you look at the placements from FSU, almost all of their graduates go into industry. That might be selection bias since they can earn far more in industry. The ones who go into academia generally go to schools ranked around FSU or lower.

I don't think you have a realistic chance at places like Texas A&M or Purdue even with a masters degree.

I'm not very familiar with masters admissions, so other people might have better advice than me.

[Stat Assistant Professor]

@playingstats Thanks for the clarification. As others have mentioned, your biggest hurdle for PhD admissions will be convincing adcoms that you can handle the rigorous mathematical coursework. Apart from possibly some applied regression, categorical analysis, and statistical computing classes, the majority of classes you take in your first two years of a Statistics PhD program will tend to be very mathematical and proof-intensive (particularly classes like advanced statistical inference, measure-theoretic probability theory, theory of linear models, and large sample theory ). Later when you do your dissertation research, you can choose to focus on something more methodological/applied, but to get to that point, you have to pass written qualifying exams on proof-intensive courses. And even if your research is more applied, you still need to have advanced knowledge of the math behind it (and some basic knowledge of theory is also helpful). Just as an example, if your research is on Bayesian nonparametrics (say), you could write applied papers that use a Dirichlet process (DP) prior without any theorems/proofs, but you still need to understand the mechanisms that make the DP prior work.

Since you are currently employed by a university, I assume that they give you some sort of benefits to take courses there at a reduced tuition? In order to strengthen PhD applications down the road, you could take a few advanced undergraduate math classes and get A's in them. For example,  you could retake Real Analysis and take an Advanced Linear Algebra (with proofs) class. If you can get A's in these courses, that would greatly assuage Statistics adcom's concerns that you cannot handle the math, particularly the mathematical proofs component. Masters degrees at reputable universities are also something to look into, and strong performance there would certainly help your profile a lot. Since you are a veteran, you may be able to get some educational benefits and not go into a ton of debt to obtain a Masters degree. If you are unable to retake Analysis or take advanced LA now, you should definitely retake Real Analysis in a Masters program (it can be real analysis at the undergrad level, it does not have to be the PhD-level measure theory class) and maybe a few other upper division math classes. As the others above me mentioned, you need to get A's in these courses to have a chance at some Statistics PhD program.

In your case, a Masters in *Mathematics* may actually boost your profile more than a Masters in Statistics for admissions to Statistics PhD programs, because this would prove that you can get A's in math classes and partly mitigate the B's from your undergrad. And Mathematics MS programs are more likely to be funded than Stat ones, since they often need TA's for  large college algebra/trig, pre-calc, and Calculus courses. So this is something to also consider.

Edited September 4, 2019 by Stat PhD Now Postdoc

[jmillar]

On 9/4/2019 at 10:02 AM, Stat PhD Now Postdoc said:

@playingstatsAnd Mathematics MS programs are more likely to be funded than Stat ones, since they often need TA's for  large college algebra/trig, pre-calc, and Calculus courses. So this is something to also consider.

Absolutely. I came from a combined math/stats department, and the difference in TA appointments was great. For incoming masters students, there would be around 10-15 appoints for math students, and 1-3 for statistics.

[Stat Assistant Professor]

8 minutes ago, jmillar said:

Absolutely. I came from a combined math/stats department, and the difference in TA appointments was great. For incoming masters students, there would be around 10-15 appoints for math students, and 1-3 for statistics.

Most Mathematics MS programs allow you to do a concentration too. I would suggest the OP do a mathematics Masters degree with a concentration in statistics. This way, he can take the Casella and Berger mathematical statistics sequence, while also taking the usual real analysis, abstract algebra, and topology classes. Getting A's in all those would help mitigate the B's he previously received and show that he can do proof-based work.