Stats PhD at University of Washington vs. Duke vs. Berkeley

[orchidnora]

Hi, everyone. Does anybody have any information to compare these schools? I am very excited to have been accepted to these amazing schools and never expected I would have to make such a decision. Berkeley has been my top choice for a while for various reasons. But I want to be cautious and evaluate all my options carefully. I will definitely be visiting Duke and Berkeley to help decide, but I'm not sure if I want to attend UW. Any information about these schools (especially UW, since I'm less familiar with their program) would be helpful for me. 

[Geococcyx]

It's not information that you (probably) haven't already seen, but it seems like Duke's been offering a few more admissions these past few years (as a proportion of the applicants).  It seems like this might have been, in part, to combat some falling matriculation rates prior to 2017-18, which confuses me, since Duke seems prestigious enough to not have to worry too much about people accepting their offers.  I have a few guesses as to why it might be (higher-ranked programs are getting larger, industry is more open to hiring undergrads, people in your position are more likely to choose Stanford/Berkeley over Duke these days so that they can be near Silicon Valley), but it's something I am interested in looking into a little more.

[bayessays]

They're all amazing schools, so bring your decision down to research fit and location.  If you want to do Bayesian stuff, obviously go to Duke.  If you're interested in ML, UW has some great ML people (Emily Fox and Witten) as does Berkeley (Jordan). I'd visit and follow your gut. You can't go wrong. 

If you are choosing between programs that are current ranked 12 and above on US News (Stanford, Berkeley, Washington, Harvard, Chicago, Duke, CMU, Michigan, UPenn), I don't think there is a one size fits all answer. 

[Stat Assistant Professor]

UW is particularly strong for social science statistics and analysis of network/spatial data. It also has a few outstanding researchers in the machine learning (both theoretical foundations and methodology/applications).

[orchidnora]

Thanks for the info! @Geococcyx: I actually have not heard that, but that's interesting. Maybe they're simply trying to expand their program. 

@bayessays You're right, all of these schools are great. I'll visit to decide, as you suggest, since I like both Bayesian stuff and ML. 

@Stat PhD Now Postdoc Cool, I didn't know about social science applications and network/spatial data at UW. I have some interest in those topics, but not strong enough to justify going there specifically for it. ML research in general though is very important to me.

[galois]

On 2/19/2019 at 12:27 PM, Stat PhD Now Postdoc said:

UW is particularly strong for social science statistics and analysis of network/spatial data. It also has a few outstanding researchers in the machine learning (both theoretical foundations and methodology/applications).

Would you mind expanding on which profs categorize into theoretical vs applied? Often it's hard for me, with limited knowledge, to gather that information based on like a paragraph-length description of research interests. I think my interests would probably fall more into theoretical foundations. Perhaps Sham Kakade falls into this category?

[Stat Assistant Professor]

3 hours ago, galois said:

Would you mind expanding on which profs categorize into theoretical vs applied? Often it's hard for me, with limited knowledge, to gather that information based on like a paragraph-length description of research interests. I think my interests would probably fall more into theoretical foundations. Perhaps Sham Kakade falls into this category?

The distinction is very vague, to be sure. I would look at which journals they are publishing in. If a lot of their work is in places like Annals of Statistics, Annals of Probability, or Bernoulli, it is probably mainly theoretical. If in JASA-Theory and Methods, Biometrika, or JRSS-B, it is probably a mix of the theory and methodology. If in JASA-Applications and Case Studies, Annals of Applied Statistics, Biometrics, Journal of Computational and Graphical Statistics, or JRSS-C, it is mostly methodological/applied (these may have one theorem but usually not more than two).

Some other journals like Statistica Sinica, Journal of Multivariate Analysis, Journal of Machine Learning Research, Bayesian Analysis contain both heavily applied and heavily theoretical articles (and everything in between), so to gauge this, you will need to read the abstract of the article and scroll through it to see how many theorems there are.

Edited February 21, 2019 by Stat PhD Now Postdoc

[galois]

Gotcha, that sounds like great advice, thank you for breaking down the journals like that.

[Stat Assistant Professor]

19 hours ago, galois said:

Gotcha, that sounds like great advice, thank you for breaking down the journals like that.

Nowadays, stat departments are also becoming more receptive to conference proceedings (probably because there are many on machine learning now and this field moves so quickly). The top conferences for statisticians are NeurIPs, ICML, and AIStats, which seem to lean theoretical. There's methodology in these papers, but you typically can't just state a new method -- you also need to prove some theoretical guarantees. And some papers in these conferences are purely theory (e.g. a new error or tail bound, a new convergence rate, etc.).

Edited February 22, 2019 by Stat PhD Now Postdoc

[blehperson]

I have a related question, which is: how steep is the talent gradient going from the very top schools (Berkeley, Stanford) down to the second tier (such as Duke and Washington)? I feel like having very talented and motivated peers should definitely be a consideration, reputation of the program aside. Based on the student profiles, my first impression is that there is a fairly noticeable gap between Berkeley/Stanford and everyone else. 

[Statboy]

9 minutes ago, blehperson said:

I have a related question, which is: how steep is the talent gradient going from the very top schools (Berkeley, Stanford) down to the second tier (such as Duke and Washington)? I feel like having very talented and motivated peers should definitely be a consideration, reputation of the program aside. Based on the student profiles, my first impression is that there is a fairly noticeable gap between Berkeley/Stanford and everyone else. 

I was not aware that Washington and Duke were considered second tier schools.. Although I am just an applicant myself, I always got the impression that Stanford/Berkeley/Chicago/Washington/Harvard/Duke/UPenn etc. were all grouped into the "top tier" and thus had comparable students within them (even though technically they have different rankings) and students accepted into multiple schools in this "tier" made their decision based on other factors (research fit, geography etc.). 

 

Also, I am skeptical about how to objectively measure "talent". Maybe students at Stanford and Berkeley have more work experience or published papers, but I'm not sure if that directly corresponds with talent.

 

I am curious about this topic as well and hope someone with more experience can comment on this! 

[Stat Assistant Professor]

1 hour ago, Statboy said:

I was not aware that Washington and Duke were considered second tier schools.. Although I am just an applicant myself, I always got the impression that Stanford/Berkeley/Chicago/Washington/Harvard/Duke/UPenn etc. were all grouped into the "top tier" and thus had comparable students within them (even though technically they have different rankings) and students accepted into multiple schools in this "tier" made their decision based on other factors (research fit, geography etc.). 

 

Also, I am skeptical about how to objectively measure "talent". Maybe students at Stanford and Berkeley have more work experience or published papers, but I'm not sure if that directly corresponds with talent.

 

I am curious about this topic as well and hope someone with more experience can comment on this! 

I think it definitely is the case that Stanford/UC Berkeley PhD students are more likely to have a paper or two in a top venue like Annals of Statistics, JASA, or NIPS by the time they graduate than are students from other programs. Thus, they do tend to be better-positioned for the academic job market than most others. This makes a lot of sense to me, at least at Stanford, since they not only recruit very talented students, but they also: a) have a plethora of prestigious, productive faculty to choose from as PhD advisors, and b) they get their students started on research and reading papers right away (I believe that all Stanford first-year PhD students have to take a reading course every quarter).

But a lot of future success is also dependent on the PhD advisor and the person themselves (e.g. how driven and self-motivated they are to explore research areas). For example, Michael Jordan of Berkeley doesn't even have a PhD in Statistics/Math, and there are other prominent researchers I can think of who have PhDs from respectable but not "top" programs (like David Dunson who has a PhD from Emory  Biostatistics or R. Dennis Cook -- of the famous Cook's distance -- who got his PhD from Kansas State). So PhD grads from lower ranked programs should not fret their chances of success, especially if they can secure a good postdoc.

Edited February 27, 2019 by Stat PhD Now Postdoc

[blehperson]

14 minutes ago, Stat PhD Now Postdoc said:

I think it definitely is the case that Stanford/UC Berkeley PhD students are more likely to have a paper or two in a top venue like Annals of Statistics, JASA, or NIPS by the time they graduate than are students from other programs. Thus, they do tend to be better-positioned for the academic job market than most others. This makes a lot of sense to me, at least at Stanford, since they not only recruit very talented students, but they also: a) have a plethora of prestigious, productive faculty to choose from as PhD advisors, and b) they get their students started on research and reading papers right away (I believe that all Stanford first-year PhD students have to take a reading course every quarter).

But a lot of future success is also dependent on the PhD advisor and the person themselves (e.g. how driven and self-motivated they are to explore research areas). For example, Michael Jordan of Berkeley doesn't even have a PhD in Statistics/Math, and there are other prominent researchers I can think of who have PhDs from respectable but not "top" programs (like David Dunson who has a PhD from Emory  Biostatistics or R. Dennis Cook -- of the famous Cook's distance -- who got his PhD from Kansas State). So PhD grads from lower ranked programs should not fret their chances of success, especially if they can secure a good postdoc.

 

From what I can tell, Stanford grads completely dominate the academic job market (just looking at the education of profs at top schools). How quickly do the chances at academia drop as you move down the rankings? Harvard seems to produce a decent number of amazing academics (like Avi Feller, Peng Ding, Shane Jensen), but for Washington I can only find two (Drton and Volfosky). 

[Stat Assistant Professor]

1 hour ago, blehperson said:

 

From what I can tell, Stanford grads completely dominate the academic job market (just looking at the education of profs at top schools). How quickly do the chances at academia drop as you move down the rankings? Harvard seems to produce a decent number of amazing academics (like Avi Feller, Peng Ding, Shane Jensen), but for Washington I can only find two (Drton and Volfosky). 

I mean... if you are aiming for an academic job at a place like Stanford, Columbia, or UChicago specifically, then having a PhD from a similar institution is probably much more essential. If you are ok with other less prestigious R1's, R2's, or SLAC's, I think you will find academic placement to be more strongly correlated with PhD advisor or postdoc advisor than PhD granting institution.

 

My PhD alma mater hired two new faculty last year with PhDs from UCSC and UCincinatti (however, their postdocs were at Duke which probably helped a lot). 

Edited February 27, 2019 by Stat PhD Now Postdoc