If I keep my current research trend, would it later make me unemployable in the Statistics Departments?

[JDStat]

Hello,

I would like to call myself as a Statistician; I have been in Statistics programs for 7 years now (4 year Stat undergrad + 2 year Stat Master + 1 year Stat PhD). I always thought that if I become a professor after my PhD, I would be hired in a Statistics Department.

But lately I started to doubt the above statement. My PhD research field is Machine Learning, more specifically I am trying to improve predictive performance of the existing neural network model called "Transformer", which is commonly used in Natural Language Processing (NLP) in Data Science. I don't know how math-intensive my PhD dissertation will be, but I think my research is going in a direction that I won't be publishing a lot of (math-intensive) papers in journals like JASA or other strictly "Statistically-oriented" journals, but instead I will be publishing on some of the Deep-Learning specific journals and attend a few machine-learning and/or computer science conferences to present my work. My research involves a heavy Python programming, and I see my research as more of a Data Science than the traditional fields of Statistics.

If I keep on pursuing the current research road that I am on, would this eventually disqualify me from getting employed at Statistics Departments, since my research does not involve a heavy (traditional and math-intensive) Statistical theory? Will I have a better chance, for example, to get hired at an Engineering or Computer Science Departments than the regular Statistics Department with this kind of research background? Would it be possible for "Data Scientist" like myself to be hired at a Statistics Department? The thought of not being able to get hired at a Statistics Department amazes me since Statistics is the discipline that I studied throughout my entire academic career.

Thank you,

Edited December 12, 2019 by JDStat

[JDStat]

To describe more about my current research to provide more information to my previous post, my research through the course of my PhD is going to be improving the performance of a "transformer", a type of neural network (deep-learning) model that is used to predict concurrent words or phrases from a given (incomplete) sentence. My research work will involve some basic statistics such as analyzing and calculating rank correlations, constructing confidence intervals, and a p-test, but what I spend most of my time on is to code out my newly designed "transformer" on Python. When my papers do get published, they are likely to get published in Machine Learning journals, such as the "Journal of Machine Learning Research". I am not sure if I will be able to get my publication out in more Statistics-oriented journals like JASA or "Annals of Applied Statistics". When I attend conferences, I am likely to attend the conference as a deep-learning researcher. I am not sure if Statistics-oriented conferences like JSM would consider having me as one of their presenters.

If I keep this research path, will I have hard time securing an employment at the Statistics Departments of universities?

Thank you,

[Stat Assistant Professor]

I think a lot of your chances in the academic job market will depend greatly on how "traditional" the department is in its makeup. There are some Statistics departments that are partial to hiring people whose PhDs are in Computer Science and/or whose publications appear mainly in venues like IEEE, ICML, and NeurIPS rather than the traditional (bio)stat journals like JASA, Biometrika, Annals of Statistics, JRSS-B, or Biometrics. You will need to take a look at Assistant Professors' profiles on departmental websites to see where recent hires have been publishing to see how amenable a department is to publications in ML conferences or IEEE proceedings.

The academic job market is pretty tough though, and without at least one paper in a top 5 stat journal (including under invited revision) or a top tier conference (acceptance rate < 10%), it can be hard to even make the shortlist at a lot of schools. For Computer Science departments, journal papers are not nearly as prestigious as conference papers, so if you aim for CS faculty jobs, you typically need 3-5 ICML/NeurIPS papers to be a competitive candidate.

[Stat Assistant Professor]

For example, this guy was hired by University of Michigan Statistics Department last year with a bunch of ICML and NeurIPS publications, rather than pubs in traditional mathematical statistics journals.

https://regier.stat.lsa.umich.edu

He has a *lot* of them though. I cannot confirm if this is indeed the case, but based on personal experience, it seems that if your pubs are mainly in journals, then 1-2 articles in top tier journals gets you on the "in discussion" shortlist for many search committees. This is typically because it takes much longer to publish a journal article... whereas with conferences, the review time and rebuttal times are on a fixed schedule (so you only have [x] weeks to reply to the referees who only had [x] weeks to peer review your paper) so they go through much faster. So it may be the case that more conference papers are required to make an impression. 

Is your ultimate goal to work at a research university? If you aim to go to a primarily undergrad institution, then publications in top venues is not nearly as crucial (but you do need some publications and you need to show potential to churn out some more in the future... but they need not be top-tier).

Edited December 12, 2019 by Stat PhD Now Postdoc

[JDStat]

Hello,

Thank you for your insight! This is really helpful. I am only in 1st year of my PhD program, and I still have to take courses to fulfill my requirements so that slows down my research quite a bit, but I guess in the upcoming years I will have to try to publish in the top journals as many as I can.

 

Thank you again,

[Stat Assistant Professor]

38 minutes ago, JDStat said:

Hello,

Thank you for your insight! This is really helpful. I am only in 1st year of my PhD program, and I still have to take courses to fulfill my requirements so that slows down my research quite a bit, but I guess in the upcoming years I will have to try to publish in the top journals as many as I can.

 

Thank you again,

I'd note also that many PhD students do not publish in the very top journals or very top conferences (myself included -- the publications from my PhD thesis research ultimately appeared in quality journals but not the top 4). Most of these PhD students will probably go into industry, but if you aim to stay in academia and your primary goal is to work at a research university,  then you can do a postdoc to improve your profile. If you don't have any publications in the top journals as you're finishing your PhD, I would recommend doing a postdoc at the most prestigious university you can and working with PI's who have a track record of publishing in the top journals.

In my case, doing a postdoc really helped a lot and made my profile much more competitive. As I was finishing my PhD, my profile wasn't competitive at all for most research universities (I did make the shortlist for an R2, more teaching-intensive university though, but ultimately did not get that position and opted to do a postdoc instead).

Edited December 12, 2019 by Stat PhD Now Postdoc

[JDStat]

Of course! I am considering doing the Post Doc, although the university that I am currently attending (University of Waterloo) is a very strong university in Statistics but the overall ranking (e.g. Times Ranking) is lower than many prestigious universities (~250).

I am just curious, is the "Journal of Statistical Software" considered as one of the top journals in Statistics?

Sorry if my question is silly, I guess I don't know a lot of about academia yet.

[Stat Assistant Professor]

Just now, JDStat said:

Of course! I am considering doing the Post Doc, although the university that I am currently attending (University of Waterloo) is a very strong university in Statistics but the overall ranking (e.g. Times Ranking) is lower than many prestigious universities (~250).

I am just curious, is the "Journal of Statistical Software" considered as one of the top journals in Statistics?

Sorry if my question is silly, I guess I don't know a lot of about academia yet.

Most would consider the top journals in Statistics to be:

• Journal of the American Statistical Association
• Annals of Statistics
• Journal of the Royal Statistical Society
• Biometrika
• Biometrics