Going from Econ Phd to Statistics PhD

[Poisson]

I completed the first year of a top 20 economics PhD program, and I really am unsure if I want to continue. My program is very large, so there is not very much in the way of faculty support, I have found that I am not that interested in economic research. In undergrad, I studied statistics and economics and I enjoyed my statistics coursework much more than economics, I went into economics because I was interested in public policy-that interest has dissipated.

I will post my profile below and would appreciate recommendations as to coursework for the fall and places to apply.

Type of Undergrad: Decent Private University

Undergrad GPA: 3.95

GRE: 560v 790Q 5.5 AWA

Grad GPA: 3.1 (they rarely give out A's in my program)

Undergrad Courses: Standard Statistics Degree (Calc I-IV, Linear Algebra, Regression, Data Analysis, Advanced Calculus, Analysis, Math Stats I-III (at the level of Hogg and Craig). Standard Undergrad courses. A's in all Math/Stat courses

Grad Courses: Standard Year long sequences in Micro, Macro, and Econometrics.

Research Experience: 1 years as RA on an applied econometrics project in undergrad. 1 year as RA on an empirical project as grad school.

Letters of reference: I plan on getting 2 from undergrad stats professors, 1 from graduate school (I think this is acceptable by most schools).

Research Interests: Applied/Computational Statistics specifically biostatistics and high frequency finance.

I plan on applying to: SUNY Stony Brook, Wisconsin, Illinois, Michigan State, Michigan, Toronto, University of Western Ontario, Purdue, North Carolina State and Washington.

Questions:

1) Should I add/remove any schools from this list?

2) Should I take any courses in the fall to shore up my profile? (Thinking of Time Series Analysis and Stochastic Processes)

3) Will adcoms look past my subpar graduate GPA? I am hoping to improve it in the following semester, since grading is a bit more lenient in field courses than the core.

Thanks in advance

[statsguy]

I completed the first year of a top 20 economics PhD program, and I really am unsure if I want to continue. My program is very large, so there is not very much in the way of faculty support, I have found that I am not that interested in economic research. In undergrad, I studied statistics and economics and I enjoyed my statistics coursework much more than economics, I went into economics because I was interested in public policy-that interest has dissipated.

I will post my profile below and would appreciate recommendations as to coursework for the fall and places to apply.

Type of Undergrad: Decent Private University

Undergrad GPA: 3.95

GRE: 560v 790Q 5.5 AWA

Grad GPA: 3.1 (they rarely give out A's in my program)

Undergrad Courses: Standard Statistics Degree (Calc I-IV, Linear Algebra, Regression, Data Analysis, Advanced Calculus, Analysis, Math Stats I-III (at the level of Hogg and Craig). Standard Undergrad courses. A's in all Math/Stat courses

Grad Courses: Standard Year long sequences in Micro, Macro, and Econometrics.

Research Experience: 1 years as RA on an applied econometrics project in undergrad. 1 year as RA on an empirical project as grad school.

Letters of reference: I plan on getting 2 from undergrad stats professors, 1 from graduate school (I think this is acceptable by most schools).

Research Interests: Applied/Computational Statistics specifically biostatistics and high frequency finance.

I plan on applying to: SUNY Stony Brook, Wisconsin, Illinois, Michigan State, Michigan, Toronto, University of Western Ontario, Purdue, North Carolina State and Washington.

Questions:

1) Should I add/remove any schools from this list?

Kind of a weird mix of schools you have there. I'd add Texas A&M, Minnesota, and Ohio State. Those are really good (especially the first two) programs which have historically had excellent reputations and have good (and diverse) research programs. A few of the programs you list are, generally, not considered great. Also, you should probably focus on one research area. Biostatistics and finance are two huge, incredibly broad fields. In fact, there are programs which offer Biostatistics PhDs. This may help you decide on a school. Admission is MUCH more competitive now. Programs are taking fewer students, applications have skyrocketed due to the economy/increased awareness of the field. So apply to a lot of places!!

2) Should I take any courses in the fall to shore up my profile? (Thinking of Time Series Analysis and Stochastic Processes)

Also take a course in Measure Theory/ Lebesgue Integration as well as those two.

3) Will adcoms look past my subpar graduate GPA? I am hoping to improve it in the following semester, since grading is a bit more lenient in field courses than the core.

Probably. I'd say they'll weigh your Math & Stats courses MUCH heavier than your Econ Course. In fact, when I was applying a few years back, most applications asked for two GPAs: cumulative and Math/Stats.

Thanks in advance

[Poisson]

Thanks for the input

Kind of a weird mix of schools you have there. I'd add Texas A&M' date= Minnesota, and Ohio State. Those are really good (especially the first two) programs which have historically had excellent reputations and have good (and diverse) research programs. A few of the programs you list are, generally, not considered great. Also, you should probably focus on one research area. Biostatistics and finance are two huge, incredibly broad fields. In fact, there are programs which offer Biostatistics PhDs. This may help you decide on a school. Admission is MUCH more competitive now. Programs are taking fewer students, applications have skyrocketed due to the economy/increased awareness of the field. So apply to a lot of places!!

Yeah, I am kind of confused about where to apply because I am used to evaluating schools based on econ programs which does not necessarily correlate to having a good statistics program (for one thing, there are a lot fewer stat programs than econ!) Which programs would you recommend I remove? I posted this over at talkstats and received Iowa State as a recommendation so along with your three that brings the total number of prospective programs to 14.

I was planning on applying to 10-12 programs, do you think this is an appropriate amount, or should I go higher? During my last admission cycle, I applied to 16 econ PhD programs and was accepted to 5. Do you think it would be worthwhile to apply to any Master's programs? One thing very different from econ programs is that many stat programs have terminal master's degrees which seem to have roughly the same coursework as the doctorate and seem like they could be used as a stepping stone to the PhD program.

I am still not settled on research interests, I went into a bit more detail on this on the other forum, I will just repost what I wrote below:

Basically, with respect to research interests, I am interested in going beyond analyzing social data, which is all I was really trained to do as an economist. I would like to learn a greater skill-set that has more general applicability than econometrics. For instance, much of econometric analysis is centered around generalized method of moments, but after speaking to statisticians it seems like generalized estimating equations are preferred.

Moreover, use of statistical software within an economics program is expected to be self-acquired; there is very little instruction and many senior professors outsource programming to graduate students and junior professors. In addition, most economists use MATLAB or STATA, two "languages" that I prefer not to use, as they are inferior in almost every respect to R. Not a single faculty member in my department and very few graduate students use R, so I have spend a disproportionate amount of time honing my skills through self-instruction. I still have a long ways to go but I feel that within a statistics department, where R is the unequivocal standard, I would receive a greater degree of instruction and support.

I am interested in applied work, basically in using a wide array of statistical methods to analyze stochastic processes whether they be in social, medical, natural, or financial data. I feel statistics would provide me with a much better skillset of achieving this than economics.

I also have an interest in complex adaptive systems and cellular automation. I am not sure if this falls under the category of biostatistics, but my interests are still malleable. I don't know if I want biostatistics as my primary emphasis, but I would like to learn more about it.

Additionally, one of my macroeconomics courses devoted a module to time series analysis and we scratched the surface of frequency domain/spectral analysis. I would like to learn more about this methodology, but it is rarely used within economics.

To summarize: my main interest is in time series analysis, but I would like to learn more about biostatistics and computational statistics in general. I am not really interested in a pure biostats program because I don't want to pigeonhole myself into a specific subfield without being exposed to the wider body of research.

Also take a course in Measure Theory/ Lebesgue Integration as well as those two.

I still have a few econ classes to fulfill my requirement for the Master's degree, so I don't think that I will have time to take that class in the fall, so I was planning on taking it in the spring. Is it one of the courses that primarily serves as a signal to adcoms re: mathematically ability or is it useful for first year coursework? Is it more valuable than stochastic processes?

Edited August 15, 2010 by Poisson

[statsguy]

Thanks for the input

Yeah, I am kind of confused about where to apply because I am used to evaluating schools based on econ programs which does not necessarily correlate to having a good statistics program (for one thing, there are a lot fewer stat programs than econ!) Which programs would you recommend I remove? I posted this over at talkstats and received Iowa State as a recommendation so along with your three that brings the total number of prospective programs to 14.

Iowa State is a great school but the location sucks. If you are a US citizen, I wouldn't bother with the two Canadian universities, unless you really want to go to Toronto. Also, UIUC and Michigan State are probably the two weakest programs you've listed. I am not familiar with SUNY Stony Brook. Also, don't just go by name alone; several less "prestigious" universities have some amazing faculty members, just not as many as the "better" schools. I am at least slightly familiar with most of those programs you listed and you can definitely get good applied training at them.

I was planning on applying to 10-12 programs, do you think this is an appropriate amount, or should I go higher? During my last admission cycle, I applied to 16 econ PhD programs and was accepted to 5. Do you think it would be worthwhile to apply to any Master's programs? One thing very different from econ programs is that many stat programs have terminal master's degrees which seem to have roughly the same coursework as the doctorate and seem like they could be used as a stepping stone to the PhD program.

10-12 sounds good. I'd apply only to PhD programs if you want a PhD. Most schools have a terminal masters degree; apply to that ONLY if you want to end up with an MS. You're coursework will be a little different (more flexibility, less theory/more applied courses.) The other way to get an MS is Statistics is by leaving the PhD program after your 2nd or 3rd year, or by failing out. However, IMHO, this way of getting a Master's isn't as practical since you will mostly have taken theory courses which will not help that much in industry.

I am still not settled on research interests, I went into a bit more detail on this on the other forum, I will just repost what I wrote below:

R is indeed the academic standard, although SAS is used too. It's not too hard to learn R on your own.

To summarize: my main interest is in time series analysis, but I would like to learn more about biostatistics and computational statistics in general. I am not really interested in a pure biostats program because I don't want to pigeonhole myself into a specific subfield without being exposed to the wider body of research.

From the list of schools above, look at the faculty pages to see who is currently doing research in time series. Apply to those schools, but make sure there are other faculty you can see yourself working with. Check the dates of their last time-series publication to make sure they are still active. I know for a fact that UChicago and Cornell have several faculty doing research in time series, although it's probably more on the theoritical side.

I still have a few econ classes to fulfill my requirement for the Master's degree, so I don't think that I will have time to take that class in the fall, so I was planning on taking it in the spring. Is it one of the courses that primarily serves as a signal to adcoms re: mathematically ability or is it useful for first year coursework? Is it more valuable than stochastic processes?

The course is usuall a two-semester sequence, so you may not be able to take it in the spring. If you want to take the "real" Stochastic Processes course (the PhD level course), you will need to know Measure Theory/Lebesgue Integration, and some PhD level Probability Theory. So I'd drop the Stochastic Processes class and take a year long sequence in Phd-level Real Analysis (ie Measure Theory.) You will have to take it in grad school, and earning a good grade in the course now will definitely look great on your application.

[Poisson]

Thanks for your input again!

Iowa State is a great school but the location sucks. If you are a US citizen' date= I wouldn't bother with the two Canadian universities, unless you really want to go to Toronto. Also, UIUC and Michigan State are probably the two weakest programs you've listed. I am not familiar with SUNY Stony Brook. Also, don't just go by name alone; several less "prestigious" universities have some amazing faculty members, just not as many as the "better" schools. I am at least slightly familiar with most of those programs you listed and you can definitely get good applied training at them.

I am aware of Iowa State's location; I have visited Ames. One of the factors in choosing my current program was location, and I am coming to realize that this was not incredibly wise because I have been so busy with coursework that I have not had a chance to really experience much of the town. So I am relatively ambivalent regarding location. However, I don't think I could live in College Station, Texas, so I don't think I will apply to Texas A&M.

I was mainly considering the Canadian schools as potentially funded Master's programs which could serve as a stepping stone to PhD programs because they looked be true "doctoral stream" master's programs since they do not allow direct entry into their PhD program without a Master's degrees. Toronto looks like a great program as well. SUNY Stony Brook seemed interesting because they are a sort of interdisciplinary program between Applied Math, Quant Finance, and Statistics and it felt like a good "safety." I have found that a lot of schools without much "lay prestige" have great faculty members. Moreover, a lot of top schools have really small and specialized statistics programs, which make them ill-suited towards me.

So organizing schools by relative fit, I will have:

Reach: Wisconsin, Michigan, Washington, Minnesota, Cornell

Target: North Carolina State, Iowa State, Purdue, Toronto, Ohio State

Safety: SUNY Stony Brook (one more)

Does this sound about right?

10-12 sounds good. I'd apply only to PhD programs if you want a PhD. Most schools have a terminal masters degree; apply to that ONLY if you want to end up with an MS. You're coursework will be a little different (more flexibility, less theory/more applied courses.) The other way to get an MS is Statistics is by leaving the PhD program after your 2nd or 3rd year, or by failing out. However, IMHO, this way of getting a Master's isn't as practical since you will mostly have taken theory courses which will not help that much in industry.

That's exactly how I feel about my Econ Master's that I am expecting to receive. Additionally, I have been through courses with both Casella/Berger and Hogg and Craig and I have no interest in rehashing either. When I start a PhD program, will my first year coursework be mostly measure based probability or are the aforementioned two books going to be references?

R is indeed the academic standard, although SAS is used too. It's not too hard to learn R on your own.

I have some knowledge of the basic functionality of R (it was the basis of Data Analysis class), but I am trying to learn it in a Unix environment at the moment to allow for things like batch submissions. I am currently working on rewriting a few of my computationally inefficient programs by calling C subroutines from R (basically outsourcing the for loops). Is there a low-level language that serves as an academic standard for running computationally intensive programs? Most economists use FORTRAN, which I am not willing to learn because the syntax is a mess.

From the list of schools above, look at the faculty pages to see who is currently doing research in time series. Apply to those schools, but make sure there are other faculty you can see yourself working with. Check the dates of their last time-series publication to make sure they are still active. I know for a fact that UChicago and Cornell have several faculty doing research in time series, although it's probably more on the theoritical side.

So a school like NCSU has several statisticians specializing in time series, including D.A. Dickey (of Dickey-Fuller fame). They seem to be on the older side, but some of them are still actively publishing. Should I aim for programs with active faculty on the younger side? I get a bit worried about non-tenured professors because they tend to move around a lot (at least within economics). I will add Cornell to my list, and possibly UChicago, but I don't have much hope of getting in there. I know John Cochrane in GSB does a lot with Time-Series analysis applications in Macroeconomics/Finance, but I have no idea if he works with the statistics program.

The course is usuall a two-semester sequence, so you may not be able to take it in the spring. If you want to take the "real" Stochastic Processes course (the PhD level course), you will need to know Measure Theory/Lebesgue Integration, and some PhD level Probability Theory. So I'd drop the Stochastic Processes class and take a year long sequence in Phd-level Real Analysis (ie Measure Theory.) You will have to take it in grad school, and earning a good grade in the course now will definitely look great on your application.

So I looked at my schedule of classes, and there is a class called Analysis II which uses Rudin as a textbook and covers things like Convergence, Metric Spaces, Contractions, and differential calculus. I never took a class like this, but I self studied several of the topics as they were used in my macroeconomics sequence. This doesn't seem especially relevant for statistics, but it is a pre-requisite for the class entitled "Introduction to Measure and Integration, which uses Royden, Real Analysis. Ash, Real Analysis and Probability as the primary textbooks which is offered in the spring.

Topics covered in the spring semester are:

• Lebesgue measure on the line: outer measure, measurable sets, nonmeasurable sets, measurable functions.
• Lebesgue integration on the line.
• Monotone convergence theorem, Fatou's Lemma, dominate convergence theorem.
• Almost everywhere convergence, convergence in measure, Egoroff's theorem.
• Differentiation, absolute continuity, derivatives of integrals.
• General measure and integration theory.
• Signed measures, Hahn decomposition theorem, Jordan decomposition.
• Radon-Nikodym theorem, Lebesgue decomposition.
• Outer measure, extension of measures, Lebesgue-Stieltjes measures.
• Product measures, Fubini and Tonelli theorems.
• L-spaces
  

the fall covers:

• More on convergence.
  • Exponential functions, and more on power series.
  • Approximations of the identity.
  • Fourier series.
  • Approximation by polynomials, the Stone-Weierstrass theorem.
  • Infinite products.
  • Stirling's formula and the Gamma-function
  • Compactness in metric spaces.
    • Characterizations of compactness.
    • Arzela-Ascoli theorem.
    • Applications (such as Peano's existence theorem for differential equations).

    [*]The contraction principle.

    [*]With applications, in particular to differential equations.

    [*]Differential calculus.

    [*]Differentiability in normed spaces, the derivative as a linear map.[*]Chain rule.[*]Maps from R^m into R^n.[*]Taylor's formula.[*]Inverse mapping theorem.[*]Implicit functions

    Do you think I could get away with only taking the second one? I took 2 semesters of analysis in undergrad and covered some of these topics, and the first module does not seem that relevant to statistics.

[statsguy]

Thanks for your input again!

I am aware of Iowa State's location; I have visited Ames. One of the factors in choosing my current program was location, and I am coming to realize that this was not incredibly wise because I have been so busy with coursework that I have not had a chance to really experience much of the town. So I am relatively ambivalent regarding location. However, I don't think I could live in College Station, Texas, so I don't think I will apply to Texas A&M. That's perfectly fine. Location is important, especially if you are going to spend 5 years of your life there.

I was mainly considering the Canadian schools as potentially funded Master's programs which could serve as a stepping stone to PhD programs because they looked be true "doctoral stream" master's programs since they do not allow direct entry into their PhD program without a Master's degrees. Just go straight for the PhD. The system here is completely different for Statistics PhDs (I think it's the same for Econ PhDs too.) Toronto looks like a great program as well. It is, but there are great programs here and personally I wouldn't want to bother with the VISA issues/living in another country adjustment. But if Toronto is somewhere you could really see yourself going, I'd definitely appl. SUNY Stony Brook seemed interesting because they are a sort of interdisciplinary program between Applied Math, Quant Finance, and Statistics and it felt like a good "safety." I have found that a lot of schools without much "lay prestige" have great faculty members. Moreover, a lot of top schools have really small and specialized statistics programs, which make them ill-suited towards me.

So organizing schools by relative fit, I will have:

Reach: Wisconsin, Michigan, Washington, Minnesota, Cornell

Target: North Carolina State, Iowa State, Purdue, Toronto, Ohio State

Safety: SUNY Stony Brook (one more)

Does this sound about right?

I'd go:

Reach: Carnegie Mellon, Cornell, UChicago, Washington

Good Chances: Wisc, Mich, Minn, Purdue, Iowa State

Very Good Chances: NCSU, Ohio State

That's exactly how I feel about my Econ Master's that I am expecting to receive. Additionally, I have been through courses with both Casella/Berger and Hogg and Craig and I have no interest in rehashing either. When I start a PhD program, will my first year coursework be mostly measure based probability or are the aforementioned two books going to be references?

You will most definitely be taking a year long sequence at the Casella Berger level (most likely with the Casell Berger text.) This is a GOOD thing. You need to know that course inside-out, front and back etc.... for the qualifying exams, which are very difficult and mostly based off of that material. Hogg and Craig is a good Master's level text, but you will start higher than that. Even if a program allows you to bypass that course (which it probably won't), do not skip it!! Remember, your first goal is to pass the qualifying exam, not to ace or enjoy your classes.

I have some knowledge of the basic functionality of R (it was the basis of Data Analysis class), but I am trying to learn it in a Unix environment at the moment to allow for things like batch submissions. I am currently working on rewriting a few of my computationally inefficient programs by calling C subroutines from R (basically outsourcing the for loops). Is there a low-level language that serves as an academic standard for running computationally intensive programs? Most economists use FORTRAN, which I am not willing to learn because the syntax is a mess.

FORTRAN is still popular among Statisticians since it is so much faster than R and pretty simple. Also, at this point, you shouldn't worry about learning R in Linux but rather focus on knowing R REALLY well in Windows/Mac (not just having a "basic knowledge".)

So a school like NCSU has several statisticians specializing in time series, including D.A. Dickey (of Dickey-Fuller fame). They seem to be on the older side, but some of them are still actively publishing. Should I aim for programs with active faculty on the younger side? I get a bit worried about non-tenured professors because they tend to move around a lot (at least within economics). I will add Cornell to my list, and possibly UChicago, but I don't have much hope of getting in there. I know John Cochrane in GSB does a lot with Time-Series analysis applications in Macroeconomics/Finance, but I have no idea if he works with the statistics program.

So I looked at my schedule of classes, and there is a class called Analysis II which uses Rudin as a textbook and covers things like Convergence, Metric Spaces, Contractions, and differential calculus. I never took a class like this, but I self studied several of the topics as they were used in my macroeconomics sequence. This doesn't seem especially relevant for statistics, but it is a pre-requisite for the class entitled "Introduction to Measure and Integration, which uses Royden, Real Analysis. Ash, Real Analysis and Probability as the primary textbooks which is offered in the spring.

ABSOLUTELY take that spring semester course. You will need it for PhD level probability theory, which you will have to take as part of your PhD coursework. Most PhD programs offer a year long Measure Theory / Probability Theory course where the first semester is mostly Measure Theory and second semester goes into Probability Theory (ie Kolmogorov Zero-One Laws, Convergence Theorems, Conditional Expectation etc....) Having already taken that first semester will be huge.

Topics covered in the spring semester are:

the fall covers:

Do you think I could get away with only taking the second one? I took 2 semesters of analysis in undergrad and covered some of these topics, and the first module does not seem that relevant to statistics. It is INCREDIBLY relevant to probability theory, as I mentioned above. Much more so than the second.

[Poisson]

I'd go:

Reach: Carnegie Mellon, Cornell, UChicago, Washington

Good Chances: Wisc, Mich, Minn, Purdue, Iowa State

Very Good Chances: NCSU, Ohio State

That looks like a good list. UChicago's website lists that they "highly recommend" the GRE math subject test and that they rarely admit students without the test so I don't think I will bother with them, since they also have like a $100 application fee.

You will most definitely be taking a year long sequence at the Casella Berger level (most likely with the Casell Berger text.) This is a GOOD thing. You need to know that course inside-out, front and back etc.... for the qualifying exams, which are very difficult and mostly based off of that material. Hogg and Craig is a good Master's level text, but you will start higher than that. Even if a program allows you to bypass that course (which it probably won't), do not skip it!! Remember, your first goal is to pass the qualifying exam, not to ace or enjoy your classes.

I went through about the first half of Casella-Berger in my econometrics class, but the selection of topics was somewhat capricious and aimed towards applications to economics. (Basically solely hypothesis testing/maximum likelihood estimation). I went through almost everything in Hogg/Craig except for the "optional chapters." I did not notice a huge difference between the rigor of the two books except that Hogg-Craig was a bit more readable and less compact than C/B. I am guessing the difference is in the later chapters. I don't so much mind going through the material again as long as it is a more holistic approach devoting partial coverage towards non-parametrics and Bayesian methods. I would much prefer to show/derive the various distributions than perform hypothesis test after hypothesis test. Do they typically allow the use of calculators on PhD level exams/will the questions be designed so that the calculation is relatively straightforward? I occasionally had trouble solving double/triple integrals by hand and under time pressure and I also had to memorize all of the distributions and mgfs.

FORTRAN is still popular among Statisticians since it is so much faster than R and pretty simple. Also, at this point, you shouldn't worry about learning R in Linux but rather focus on knowing R REALLY well in Windows/Mac (not just having a "basic knowledge".)

So what is considered knowing R "really well"? I am sure that I will learn a lot of its Time Series capabilities in the class I am taking in the fall, but at the moment, I am able to read data into R, run regressions, perform diagnostics, and basic simulations. I can also write basic functions and if I need to use a specific function from a package, I can generally figure it out by looking at its instruction manual. Are there any other things I should know before entering a program?

ABSOLUTELY take that spring semester course. You will need it for PhD level probability theory, which you will have to take as part of your PhD coursework. Most PhD programs offer a year long Measure Theory / Probability Theory course where the first semester is mostly Measure Theory and second semester goes into Probability Theory (ie Kolmogorov Zero-One Laws, Convergence Theorems, Conditional Expectation etc....) Having already taken that first semester will be huge.

Yeah, I definitely planning on taking it.

There is a year long measure theoretic probability class is taught in the fall and requires only the Measure Theory/Integration course as a pre-requisite. It uses the textbook:Richard Durrett: Probability: Theory and Examples. 3rd edition.

Is this what I will likely take in a PhD program?

It is INCREDIBLY relevant to probability theory, as I mentioned above. Much more so than the second.

Are you referring to the fall course? The spring course explicitly lists in its description that it is geared towards those pursuing higher education in statistics. The spring course also mentioned one of the topics it would cover as:

Probability: conditional probability and expectation, distribution functions, statistical independence. (Optional)

It seems like there is a disconnect between the fall and spring course, and I think that I could handle the spring course without having taken the fall class since they are not part of the same sequence. The main issue is that if I take the fall class, I will have to drop one of my required econ classes, because there is a scheduling conflict,

[statsguy]

That looks like a good list. UChicago's website lists that they "highly recommend" the GRE math subject test and that they rarely admit students without the test so I don't think I will bother with them, since they also have like a $100 application fee. Yeah I forgot UChicago requires that test (it's been a few years since I applied.)

I went through about the first half of Casella-Berger in my econometrics class, but the selection of topics was somewhat capricious and aimed towards applications to economics. (Basically solely hypothesis testing/maximum likelihood estimation). I went through almost everything in Hogg/Craig except for the "optional chapters." I did not notice a huge difference between the rigor of the two books except that Hogg-Craig was a bit more readable and less compact than C/B. I am guessing the difference is in the later chapters. The first 4/5 chapters of CB are similar to HC. Chapters 6-10 are much more rigorous and in depth, and cover material that I don't believe the HC text covers (ie asymptotics.) I don't so much mind going through the material again as long as it is a more holistic approach devoting partial coverage towards non-parametrics and Bayesian methods. You have to choose a program based on your needs then and research each one carefully (for example, Duke is HEAVILY Bayesian.) I don't know of any programs that allow you to skip a course on Bayesian theory/methods. Non parametrics is offered as an elective at most places, but you depending on the curriculum, you may see some nonparametrics in your required applied courses. I would much prefer to show/derive the various distributions than perform hypothesis test after hypothesis test. Do they typically allow the use of calculators on PhD level exams/will the questions be designed so that the calculation is relatively straightforward? I occasionally had trouble solving double/triple integrals by hand and under time pressure and I also had to memorize all of the distributions and mgfs.

In general, most exams are split into two parts: theory and applied. The applied is usually take home. The curriculum of the exam varies by school. Sometimes, it only covers first year material. Sometimes it covers both first and second year material, and possibly may even cover measure theory. Here is an example of a typical qualifying exam: http://qual.stat.ucla.edu/past/ I doubt calculators are allowed (they weren't for mine). You will probably have to memorize the pdfs, mgfs, mean, variances, parameter spaces, etc... Sometimes just knowing the pdf is enough to derive the rest, but sometimes not.

So what is considered knowing R "really well"? I am sure that I will learn a lot of its Time Series capabilities in the class I am taking in the fall, but at the moment, I am able to read data into R, run regressions, perform diagnostics, and basic simulations. I can also write basic functions and if I need to use a specific function from a package, I can generally figure it out by looking at its instruction manual. Are there any other things I should know before entering a program? I meant just know its capabilities (and limitations), know the syntax very well for common uses, programming, writing functions, etc...

Yeah, I definitely planning on taking it.

Are you referring to the fall course? No the spring one. The spring course explicitly lists in its description that it is geared towards those pursuing higher education in statistics. The spring course also mentioned one of the topics it would cover as:

It seems like there is a disconnect between the fall and spring course, there is typically a "fork" after the first semester. Probabilists go on to take a course in Probability Theory, Analysts go on to take the "fall" course. Some people take both for whatever reason. and I think that I could handle the spring course without having taken the fall class since they are not part of the same sequence. As long as you have had some exposure to Analysis (around the Baby Rudin level), you should be fine. Knowing some Topology also won't hurt. The main issue is that if I take the fall class, I will have to drop one of my required econ classes, because there is a scheduling conflict,

Edited August 20, 2010 by statsguy

[Poisson]

The first 4/5 chapters of CB are similar to HC. Chapters 6-10 are much more rigorous and in depth, and cover material that I don't believe the HC text covers (ie asymptotics.) I don't so much mind going through the material again as long as it is a more holistic approach devoting partial coverage towards non-parametrics and Bayesian methods. You have to choose a program based on your needs then and research each one carefully (for example, Duke is HEAVILY Bayesian.) I don't know of any programs that allow you to skip a course on Bayesian theory/methods. Non parametrics is offered as an elective at most places, but you depending on the curriculum, you may see some nonparametrics in your required applied courses.

That sounds fine. I just want to get exposure to everything since I really wasn't able to learn much about the current methodological debates within statistics while studying economics. A lot of economists use variants of Bayesian methods; it seems much less controversial within econ than statistics, but we don't do much in non-parametrics (aside from bootstrapping).

In general, most exams are split into two parts: theory and applied. The applied is usually take home. The curriculum of the exam varies by school. Sometimes, it only covers first year material. Sometimes it covers both first and second year material, and possibly may even cover measure theory. Here is an example of a typical qualifying exam:http://qual.stat.ucla.edu/past/ I doubt calculators are allowed (they weren't for mine). You will probably have to memorize the pdfs, mgfs, mean, variances, parameter spaces, etc... Sometimes just knowing the pdf is enough to derive the rest, but sometimes not.

I really like that approach of splitting up the exams. I feel like it is ridiculous within econ that we do not have computational/applied portions of exams despite the fact that a majority of our research is in that area. I think that they are really worried about us cheating.

Those questions look difficult, but straightforward and fair. Do they provide you with solutions to old exams to practice from? Econ's prelim questions are designed to trick us, screw with our intuition, not reflective of what we learned over the year and are inherently confusing. They are used to weed people out of the program. Do they do this in stats, what are pass rates generally like? (most programs don't publish them)

Also, how would you consider your general quality of life as a Stat PhD student? Are your classmates very competitive, do professors treat you with respect, are grad students generally happy, are attrition rates high, etc?

Another thing I have noticed is that statistics programs have many more American students than econ programs, which is a stark contrast to econ, where between 70-80% of most programs and 80-90 percent of people who actually finish the degree consist of foreigners (mostly Chinese). I believe that this discrepancy is caused in part by the fact that an undergraduate econ degree is not adequate preparation for graduate work, whereas foreign programs are much more rigorous.

If you don't mind sharing, what was your background like before you entered your respective program If you did your undergrad (math/stats degree) in the US, Do you feel that your coursework prepared you well for graduate study or was there a pretty big culture shock?

I meant just know its capabilities (and limitations), know the syntax very well for common uses, programming, writing functions, etc...

The book required for my time series class is Time Series Analysis and Its Applications With R Examples so hopefully that should provide me with a lot of practice. I also used Data Analysis and Graphics in R as an undergraduate. Do you think that this will provide me with enough preparation?

The main limitation I have found in R is that it handles loops INCREDIBLY slowly, even for an interpreted language. I am guessing that these are the cases that you mentioned of statisticians using FORTRAN or C.

No the spring one. there is typically a "fork" after the first semester. Probabilists go on to take a course in Probability Theory, Analysts go on to take the "fall" course. Some people take both for whatever reason. . As long as you have had some exposure to Analysis (around the Baby Rudin level), you should be fine. Knowing some Topology also won't hurt.

That is the impression that I got. Stochastic processes is also offered in the spring, so I might just hold off on that and take those two classes cocurrently in the spring, since I will have no obligations to the econ dept at that time. Is this a good idea, Will stochastic processes carry enough weight with adcoms to justify taking it in the fall?

Thanks again for all of your advice!

[statsguy]

That sounds fine. I just want to get exposure to everything since I really wasn't able to learn much about the current methodological debates within statistics while studying economics. A lot of economists use variants of Bayesian methods; it seems much less controversial within econ than statistics, but we don't do much in non-parametrics (aside from bootstrapping).

I really like that approach of splitting up the exams. I feel like it is ridiculous within econ that we do not have computational/applied portions of exams despite the fact that a majority of our research is in that area. I think that they are really worried about us cheating.

Those questions look difficult, but straightforward and fair. Do they provide you with solutions to old exams to practice from? I'm pretty sure most do; however they may be in the departments network since some schools may borrow exams questions from one another. Econ's prelim questions are designed to trick us, screw with our intuition, not reflective of what we learned over the year and are inherently confusing. They are used to weed people out of the program. Do they do this in stats, what are pass rates generally like? (most programs don't publish them) Again, depends on the school, but my general impression is they are high (anywhere from 70-100%, depending on the year.) The exams I've seen (they all seem to be similar to the UCLA ones I posted) have all been very fair, some much harder than others, but then again, the pass mark may be lower for those. My prediction is that those UCLA exams require a 50-70% overall score to pass based on what I remember from the qualifying exam. The questions also be much more doable after you take the required courses, although some may still be tough.

Also, how would you consider your general quality of life as a Stat PhD student? Excellent Are your classmates very competitive not at all, especially since a great chunk of student want to go into industry, where competition is very low if you are a US citizen, do professors treat you with respect absolutely, they are extremely helpful, nice, and really want you to succeed. But this may vary by program. Do your research!!!!! Visit the schools! are grad students generally happy yes, are attrition rates high, occasionally a student might voluntarily leave with an MS due to lack of interest/loss of motivation, but failure rates are (again, from what I've seen, so take this with a grain of salt) pretty low etc?

Another thing I have noticed is that statistics programs have many more American students than econ programs, which is a stark contrast to econ, where between 70-80% of most programs and 80-90 percent of people who actually finish the degree consist of foreigners (mostly Chinese). Most stats departments are probably 50-70% Chinese/other Asian I believe that this discrepancy is caused in part by the fact that an undergraduate econ degree is not adequate preparation for graduate work, whereas foreign programs are much more rigorous. The American students probably struggle a little bit more during the first year, but once everyone is caught up, it's pretty even, although when it comes to research, the foreigners seem much more motivated (they seem to really want those academic positions.)

If you don't mind sharing, what was your background like before you entered your respective program BS in Math/BS in Stats If you did your undergrad (math/stats degree) in the US, Do you feel that your coursework prepared you well for graduate study or was there a pretty big culture shock? Nope, first year was a little rough (but not too bad), but after that, it got a lot easier.

The book required for my time series class is Time Series Analysis and Its Applications With R Examples that's a good one so hopefully that should provide me with a lot of practice. I also used Data Analysis and Graphics in R as an undergraduate. Do you think that this will provide me with enough preparation?

The main limitation I have found in R is that it handles loops INCREDIBLY slowly, even for an interpreted language. I am guessing that these are the cases that you mentioned of statisticians using FORTRAN or C. One thing that would be nice in R would be the ability to parallelize loops among CPU cores (now that multicore CPUS are standard). This has been implemented to some extent with the doSMP package in Windows, doMC in Linux, but isn't all that stable, and still is much slower than C (although pretty comparable to FORTRAN, but still slower.) R is indeed very slow. For similar operations, in my experience, FORTRAN has been about 4-5 times faster than R (which is HUGE), C can be as much as 40x as fast as R, but there is a steep learning curve. I don't use C that much.

That is the impression that I got. Stochastic processes is also offered in the spring, so I might just hold off on that and take those two classes cocurrently in the spring, check to see if Stochastic processes prereqs. The class I took required a full year of Measure Theory and Probability Theory, although there may be an introductory class at the masters level that doesn't require it since I will have no obligations to the econ dept at that time. Is this a good idea, Will stochastic processes carry enough weight with adcoms to justify taking it in the fall? Wont your transcripts list your spring courses, even though there is no grade?

Thanks again for all of your advice!

Edited August 21, 2010 by statsguy

[Poisson]

not at all, especially since a great chunk of student want to go into industry, where competition is very low if you are a US citizen,

I am primarily interested in non-academic employment as well. Should I mention on my statement of purpose that I am interested in going into industry or do they mostly want to train academics? I am guessing that this probably varies by department.

absolutely, they are extremely helpful, nice, and really want you to succeed. But this may vary by program. Do your research!!!!! Visit the schools!

If I have multiple offers my plan is to visit on a day different than the official visit day because I went on 2 official visits for econ and I felt like they hid some of the less appealing elements of the program and the environment created was a bit artificial. I didn't meet any first year students and only a small selection of grad students. I think I will try to visit the professors which share my research interests, the director of graduate studies, and first and second year students.

Also, I plan on applying to the statistics department of the school I am currently attending, do you think it might be worth it to send an email and try to arrange a meeting with the DGS during this semester to get some advice/learn more about the program, or is it standard practice not to contact professors unless you are accepted into the program?

The American students probably struggle a little bit more during the first year, but once everyone is caught up, it's pretty even, although when it comes to research, the foreigners seem much more motivated (they seem to really want those academic positions.)

That sounds fair. Who usually teaches the classes-younger APs or older, established professors? In econ, 2 profs typically split the semester for theory classes, focusing on their specialty. It exposes you to a wide body of research, but it is difficult to keep up.

One thing that would be nice in R would be the ability to parallelize loops among CPU cores (now that multicore CPUS are standard). This has been implemented to some extent with the doSMP package in Windows, doMC in Linux, but isn't all that stable, and still is much slower than C (although pretty comparable to FORTRAN, but still slower.) R is indeed very slow. For similar operations, in my experience, FORTRAN has been about 4-5 times faster than R (which is HUGE), C can be as much as 40x as fast as R, but there is a steep learning curve. I don't use C that much.

I have experienced some minor speed improvements by converting R code to Python, which, along with the Numpy package, is pretty efficient for an interpreted language and the syntax is really simple. Is Python used at all within statistics? It seems like FORTRAN might be worth learning. Is that something that you pick up on your own or is it incorporated in the coursework?

check to see if Stochastic processes prereqs. The class I took required a full year of Measure Theory and Probability Theory, although there may be an introductory class at the masters level that doesn't require it

It is just an introductory stochastic processes class that follows the first 5 chapters in Sidney Resnick's Adventures in Stochastic Processes.

Wont your transcripts list your spring courses, even though there is no grade?

I think so, but I am not sure how much weight this will carry. I know within econ of some people who have received conditional acceptances contingent on completion of an absolute non-negotiable requirement to comprehend graduate theory (requiring passing something like intermediate micro, multivariable calculus, linear algebra, etc. in the spring semester) It doesn't seem like stochastic processes is one of these types of classes, but measure theory might be.

One other question: Most programs don't explicitly state this, but how many letters of reference should I get from undergrad versus grad?

Right now, I am planning on obtaining one from graduate school-an economist who taught one of my theory courses and does a lot of interdisciplinary work (uses explicitly statistic rather than econometric methodology) and 2 stat profs who taught my theory sequences in undergrad (one has a PhD in math and is a probabilist-not sure if that will make a difference. I received A's in all of their classes and I feel like I had a decent rapport with them, so I think the letters should be decent (one of them wrote me letters for econ) but I haven't had much contact with my undergraduate professors over the past year so it might be a bit awkward asking for 10-12 letters and I am hundreds of miles away from my undergraduate campus.

I could also get a letter of reference from the econometrician I did research for as an undergrad. My role involved somewhat intricate wage estimations, but it was nothing especially groundbreaking. Nonetheless, it seems like it might be useful to have someone that can speak to my ability to use statistical software. I still keep in contact with him and he is very well known within econometrics and used his connections to help me get into a good econ program (I think he also has a background in physics).

The main issue with economists writing my letters of reference is that I need to justify to them that I would be better suited in a Statistics PhD program and I also have no idea how much influence letters from non-statisticians will have on admissions committees. If I go with my current plan, I will have 1 math and 2 econ PhDs writing my letters. Would it be better if I asked at least 1 Stat PhD for a letter?

[statsguy]

I am primarily interested in non-academic employment as well. Should I mention on my statement of purpose that I am interested in going into industry or do they mostly want to train academics? I am guessing that this probably varies by department.

I don't think it will matter really. Most departments of course would want every graduate to go and get a prestigious academic position, but they are also reasonable and know that a good portion of their grads will get industry jobs (some by choice, some because they don't get an academic position.) I'm really not 100% sure if you even need to mention your careers goals in the statement.

If I have multiple offers my plan is to visit on a day different than the official visit day because I went on 2 official visits for econ and I felt like they hid some of the less appealing elements of the program and the environment created was a bit artificial. I didn't meet any first year students and only a small selection of grad students. I think I will try to visit the professors which share my research interests, the director of graduate studies, and first and second year students.

Some departments have an official visit day for admitted students. I went to a few back in the day -- basically we sat in on a few classes, met students, listened to a few presentations etc... for two of the orientations, the entire department + the potential new students went out to dinner and then to a bar afterwards.

Also, I plan on applying to the statistics department of the school I am currently attending, do you think it might be worth it to send an email and try to arrange a meeting with the DGS during this semester to get some advice/learn more about the program, or is it standard practice not to contact professors unless you are accepted into the program?

You should definitely contact the DGS/professors if you are unsure about something, BUT make sure the answer isn't already online somewhere!! This looks really bad and my DGS has, on several occasions, mentioned that it's a HUGE pet peeve.I wouldn't arrange for a visit until you get in, however.

That sounds fair. Who usually teaches the classes-younger APs or older, established professors? Both In econ, 2 profs typically split the semester for theory classes, focusing on their specialty. It exposes you to a wide body of research, but it is difficult to keep up. Two-semester sequence courses are typically taught by the same professor, but there are exceptions.

I have experienced some minor speed improvements by converting R code to Python, which, along with the Numpy package, is pretty efficient for an interpreted language and the syntax is really simple. Is Python used at all within statistics? It seems like FORTRAN might be worth learning. Is that something that you pick up on your own or is it incorporated in the coursework?

Yes, FORTRAN is eventually worth learning if you want to do heavy duty simulation / computation. It's pretty easy too, much easier than C. However, don't worry about programming until you need to, which will like be around your 3rd year in the PhD program. I took two courses in object oriented programming as an undergrad and picked up FORTRAN in two weeks or so. If you understand basic programming (if/then, while/for loops etc...) and basic matrix computations, it will be easy for you. For now, just knowing R solid should be more than enough.

It is just an introductory stochastic processes class that follows the first 5 chapters in Sidney Resnick's Adventures in Stochastic Processes.

I think so, but I am not sure how much weight this will carry. I know within econ of some people who have received conditional acceptances contingent on completion of an absolute non-negotiable requirement to comprehend graduate theory (requiring passing something like intermediate micro, multivariable calculus, linear algebra, etc. in the spring semester) It doesn't seem like stochastic processes is one of these types of classes, but measure theory might be. Most incoming students don't have any exposure to measure theory, which is why it will definitely make you app stand out. I know of no school where measure theory is required, although some require Real Analysis at the Rudin level. In Statistics, I have yet to hear of a "conditional acceptance."

One other question: Most programs don't explicitly state this, but how many letters of reference should I get from undergrad versus grad?

Right now, I am planning on obtaining one from graduate school-an economist who taught one of my theory courses and does a lot of interdisciplinary work (uses explicitly statistic rather than econometric methodology) and 2 stat profs who taught my theory sequences in undergrad (one has a PhD in math and is a probabilist-not sure if that will make a difference. I received A's in all of their classes and I feel like I had a decent rapport with them, so I think the letters should be decent (one of them wrote me letters for econ) but I haven't had much contact with my undergraduate professors over the past year so it might be a bit awkward asking for 10-12 letters and I am hundreds of miles away from my undergraduate campus.

I could also get a letter of reference from the econometrician I did research for as an undergrad. My role involved somewhat intricate wage estimations, but it was nothing especially groundbreaking. Nonetheless, it seems like it might be useful to have someone that can speak to my ability to use statistical software. I still keep in contact with him and he is very well known within econometrics and used his connections to help me get into a good econ program (I think he also has a background in physics). I would actually go with the two stats professors and the econometrician for letters. You want people to speak highly of your math/stats ability. Also, the research is a good thing to mention in your statement, even if it wasn't groundbreaking. You could mentioned how it helped influence your decision to pursue a Stats PhD / exposed you to topics you didn't cover in class, etc...

The main issue with economists writing my letters of reference is that I need to justify to them that I would be better suited in a Statistics PhD program and I also have no idea how much influence letters from non-statisticians will have on admissions committees. One non stats letter is fine, especially if it's from an econometrician who can speak highly of you and the work you did for him. If I go with my current plan, I will have 1 math and 2 econ PhDs writing my letters. Would it be better if I asked at least 1 Stat PhD for a letter? Yes.

Edited August 27, 2010 by statsguy

[Poisson]

I don't think it will matter really. Most departments of course would want every graduate to go and get a prestigious academic position' date=' but they are also reasonable and know that a good portion of their grads will get industry jobs (some by choice, some because they don't get an academic position.) I'm really not 100% sure if you even need to mention your careers goals in the statement.[/color']

Yeah, for econ it is standard practice only to mention that you want to go the academic route. Statistics as a discipline seems better geared towards industry employment especially many programs require some sort of internship or statistical consulting course. Right now I have my statement of purpose consisting of about 1/4 my undergraduate background, 1/2 why I feel an econ PhD is a bad fit for me and a stat PhD would be a good fit and 1/4 my potential areas of focus within statistics/future goals. I know that a lot of schools don't really seem to put much weight into them, but it seems like in a case like mine where I am switching disciplines, it might play more of a factor.

Some departments have an official visit day for admitted students. I went to a few back in the day -- basically we sat in on a few classes, met students, listened to a few presentations etc... for two of the orientations, the entire department + the potential new students went out to dinner and then to a bar afterwards.

I went to 2 for econ one had about 20-30 other prospective students, one had like 6. They were similar to what you described except I did not sit in on any classes or meet many graduate students. Conveniently, there were no first year students at either one and basically both schools seemed to harp on the fact that they were fast developing and looking to move up in the rankings and hire more faculty members rather than their actual student placement or quality of life.

One other question. I know stat programs typically have much smaller entering classes, but when Duke says that their PhD program "admits between 6-10 students per year" should that be interpreted as they aim for an entering class of 6-10 students or they actually only admit 6-10 students? Because if it is the former, then I feel like I really need to apply to a lot of places. Even Harvard or MIT economics admit like 30-40 students a year in hopes of an entering class of like 25. My program probably admits about 70-80 out of 500-600 applicants for an entering class of 30-40. Is the admit rate in stat programs really that low, or are these figures that websites post not factoring in yield? If Duke only makes 6-10 offers, are they aiming for an entering class of like 2 or 3 people? I can't imagine them having anywhere near 100% yield.

You should definitely contact the DGS/professors if you are unsure about something, BUT make sure the answer isn't already online somewhere!! This looks really bad and my DGS has, on several occasions, mentioned that it's a HUGE pet peeve.I wouldn't arrange for a visit until you get in, however.

I had a few questions about an internal transfer, so I met with the grad secretary and set up a meeting with the DGS. They made it sound like transferring into the Master's program would be relatively simple because I am already admitted into the external graduate school, so admission is solely at the department's discretion. Direct entry into their PhD program does not seem possible right now, but being able to finish the MS in 2-3 semesters and then applying for entry to the PhD program, either at the institution I am at or somewhere else seems like a tenable option.

Two-semester sequence courses are typically taught by the same professor, but there are exceptions.

That is good to hear. It would be nice to have at least some degree of consistency and a single person writing your prelim questions.

Yes, FORTRAN is eventually worth learning if you want to do heavy duty simulation / computation. It's pretty easy too, much easier than C. However, don't worry about programming until you need to, which will like be around your 3rd year in the PhD program. I took two courses in object oriented programming as an undergrad and picked up FORTRAN in two weeks or so. If you understand basic programming (if/then, while/for loops etc...) and basic matrix computations, it will be easy for you. For now, just knowing R solid should be more than enough.

That is kind of what I figured. I have just found C to be a major pain since it has no built-in functions (not even something to find the maximum in an array). I will just hold off on all of this stuff for awhile and concentrate on R since it seems like it will be required for my fall classwork.

Most incoming students don't have any exposure to measure theory, which is why it will definitely make you app stand out. I know of no school where measure theory is required, although some require Real Analysis at the Rudin level. In Statistics, I have yet to hear of a "conditional acceptance."

The program at my school seems to emphasize that measure theory is a strict requirement for entry into the PhD program since the PhD level Mathematical Statistics is measure-theory based. I also think that the conditional acceptance is very rare. but I do know that if you haven't finished your undergrad degree, your admission offer can be rescinded if you decide to drop out or flunk a class in the second semester or something else along those lines.

If I understand correctly, the typical PhD Stat program follow a structure similar to:

Year 1: Master's Level Probability\Math-Stat (2 semesters) Generalized Linear Models\Anova\Regression (2 semesters), Computational Statistics (1 semester), Measure Theory (1 semester), elective.

End of year 1, take Master's prelim (not always required).

Year 2: Measure-based probability\math stat (2 semesters), Higher Level Estimation\Computation Class (2 semesters), electives

End of year 2: PhD prelim

Years 3,4,5: Field Classes, Seminars, independent research

I am under the impression that some people can enter directly into the second year if they already hold a Master's degree of have substantial undergraduate coursework, but most Americans probably start at year 1. But it seems like a lot of the elective classes don't necessarily require measure theory/an extensive mathematical statistics background, so you can start taking some of them right away. Do most people tend to finish in 4-5 years? I imagine that an advisor would be less reluctant to hold a student back to polish his dissertation if he/she was not aiming for the academic job market.

I would actually go with the two stats professors and the econometrician for letters. You want people to speak highly of your math/stats ability. Also, the research is a good thing to mention in your statement, even if it wasn't groundbreaking. You could mentioned how it helped influence your decision to pursue a Stats PhD / exposed you to topics you didn't cover in class, etc...

The graduate secretary at the Stat department made it sound like I should have at least one person in my current department write me a letter of reference to make sure that I am not leaving on bad terms or anything like that. Otherwise, I think that admissions committees could really misinterpret why I am leaving my current program. I explained my situation to one professor who does mostly econometrically and statistically oriented research and whose class I took was primarily statistics based and he offered to write me an excellent letter of recommendation. I also spoke with the econometrician and one of my stat professors who both agreed to serve as references.

One non stats letter is fine, especially if it's from an econometrician who can speak highly of you and the work you did for him.

At this point I have one econometrician (who also publishes in statistics journals), one economist who is a highly active researcher in some interdisciplinary areas like biostatistics and time series analysis ( and who is ranked in the top 200 in IDEAS for publication counts), and my undergraduate advisor who is a probabilist and taught 2 of my stat courses. I think this is a good list, but I could also ask the professor who taught my Master's sequence in Mathematical Statistics. I haven't talked to him in a while and I think he is currently on leave, so I will probably only go that route if it is absolutely necessary. I am not sure what he would write beyond the fact that I got A's in his courses since I did not really talk to him very much outside of class.

The econ professor also advised me to apply to public policy PhD programs, since I like the political analysis oriented research of Andrew Gelman et al. I know that Carnegie Mellon has a joint PhD program in statistics and public policy, but I have not heard of any other places like this. Do you know of any other programs like this, or should I just shoot for programs that are strong in Bayesian methods, since it seems like that methodology has the most direct applicability to public policy/demographic related statistical questions? Columbia seems out of the question since they are one of those programs that "highly recommends" the math subject exam.

Edited August 29, 2010 by Poisson

[statsguy]

Yeah, for econ it is standard practice only to mention that you want to go the academic route. Statistics as a discipline seems better geared towards industry employment especially many programs require some sort of internship or statistical consulting course. Right now I have my statement of purpose consisting of about 1/4 my undergraduate background, 1/2 why I feel an econ PhD is a bad fit for me and a stat PhD would be a good fit and 1/4 my potential areas of focus within statistics/future goals. I know that a lot of schools don't really seem to put much weight into them, but it seems like in a case like mine where I am switching disciplines, it might play more of a factor.

I went to 2 for econ one had about 20-30 other prospective students, one had like 6. They were similar to what you described except I did not sit in on any classes or meet many graduate students. Conveniently, there were no first year students at either one and basically both schools seemed to harp on the fact that they were fast developing and looking to move up in the rankings and hire more faculty members rather than their actual student placement or quality of life.

One other question. I know stat programs typically have much smaller entering classes, but when Duke says that their PhD program "admits between 6-10 students per year" should that be interpreted as they aim for an entering class of 6-10 students yes or they actually only admit 6-10 students no? Because if it is the former, then I feel like I really need to apply to a lot of places. Even Harvard or MIT economics admit like 30-40 students a year in hopes of an entering class of like 25. My program probably admits about 70-80 out of 500-600 applicants for an entering class of 30-40. Is the admit rate in stat programs really that low, or are these figures that websites post not factoring in yield? If Duke only makes 6-10 offers, are they aiming for an entering class of like 2 or 3 people? I can't imagine them having anywhere near 100% yield.

http://gradschool.duke.edu/about/statistics/admitstat.htm

I had a few questions about an internal transfer, so I met with the grad secretary and set up a meeting with the DGS. They made it sound like transferring into the Master's program would be relatively simple because I am already admitted into the external graduate school, so admission is solely at the department's discretion. Direct entry into their PhD program does not seem possible right now, but being able to finish the MS in 2-3 semesters and then applying for entry to the PhD program, either at the institution I am at or somewhere else seems like a tenable option.

That is good to hear. It would be nice to have at least some degree of consistency and a single person writing your prelim questions.

Questions are written by a bunch of faculty members, but there is a committee consisting of the faculty who taught the 1st/2nd year courses who put together and finalize the exam. They will reject questions which involve material they didn't cover.

That is kind of what I figured. I have just found C to be a major pain since it has no built-in functions (not even something to find the maximum in an array). I will just hold off on all of this stuff for awhile and concentrate on R since it seems like it will be required for my fall classwork.

Yeah but C has tons of libraries to do just about anything, including Statistics.

The program at my school seems to emphasize that measure theory is a strict requirement for entry into the PhD program since the PhD level Mathematical Statistics is measure-theory based. I also think that the conditional acceptance is very rare. but I do know that if you haven't finished your undergrad degree, your admission offer can be rescinded if you decide to drop out or flunk a class in the second semester or something else along those lines.

If I understand correctly, the typical PhD Stat program follow a structure similar to:

Year 1: Master's Level Probability\Math-Stat (2 semesters) Generalized Linear Models\Anova\Regression (2 semesters), Computational Statistics (1 semester), Measure Theory (1 semester), elective.

End of year 1, take Master's prelim (not always required).

Year 2: Measure-based probability\math stat (2 semesters), Higher Level Estimation\Computation Class (2 semesters), electives

End of year 2: PhD prelim

Years 3,4,5: Field Classes, Seminars, independent research

I am under the impression that some people can enter directly into the second year if they already hold a Master's degree of have substantial undergraduate coursework, but most Americans probably start at year 1. But it seems like a lot of the elective classes don't necessarily require measure theory/an extensive mathematical statistics background, so you can start taking some of them right away. Do most people tend to finish in 4-5 years? I imagine that an advisor would be less reluctant to hold a student back to polish his dissertation if he/she was not aiming for the academic job market.

Look around at various schools. Not all programs are the same. Sometimes PhD prelims are at the beginning of year 2, end of year 1, or even the summer before year 3.

The graduate secretary at the Stat department made it sound like I should have at least one person in my current department write me a letter of reference to make sure that I am not leaving on bad terms or anything like that. Otherwise, I think that admissions committees could really misinterpret why I am leaving my current program. I explained my situation to one professor who does mostly econometrically and statistically oriented research and whose class I took was primarily statistics based and he offered to write me an excellent letter of recommendation. I also spoke with the econometrician and one of my stat professors who both agreed to serve as references.

At this point I have one econometrician (who also publishes in statistics journals), one economist who is a highly active researcher in some interdisciplinary areas like biostatistics and time series analysis ( and who is ranked in the top 200 in IDEAS for publication counts), and my undergraduate advisor who is a probabilist and taught 2 of my stat courses. I think this is a good list, but I could also ask the professor who taught my Master's sequence in Mathematical Statistics. I haven't talked to him in a while and I think he is currently on leave, so I will probably only go that route if it is absolutely necessary. I am not sure what he would write beyond the fact that I got A's in his courses since I did not really talk to him very much outside of class.

If you think that those recommendations are strong, then go for it. I wouldn't worry about the Stats departments thinking you left on bad terms; people change grad programs/fields all the time.

The econ professor also advised me to apply to public policy PhD programs, since I like the political analysis oriented research of Andrew Gelman et al. I know that Carnegie Mellon has a joint PhD program in statistics and public policy, but I have not heard of any other places like this. Do you know of any other programs like this, or should I just shoot for programs that are strong in Bayesian methods, since it seems like that methodology has the most direct applicability to public policy/demographic related statistical questions? Columbia seems out of the question since they are one of those programs that "highly recommends" the math subject exam.

I applied to two schools back in the day that strongly recommended the subject test. I didn't take it and got into both. Don't let that stop you from applying. You don't want to choose a program based on it's applicable field; go into the one whose Statistics research interests you. What happens if you have a change in heart again and want to go into, say for example, Biostatistics? If you a Stat PhD student, you will not be doing a dissertation on Statistics and its application to public policy, but rather developing new Statistical methods/tools/theorems in a sub area of Statistics (examples: Linear Models, Machine Learning, Time Series Methods, Non-parametric/Semi-parametric Regression, etc...)