He has a math/science background and wants to transition to social science. Should he get a statistics degree and do social science from there, or should he get a graduate degree in social science or policy?

Someone who graduated from college a couple years ago writes:

My educational background is almost entirely science and math.

However, since graduating and thinking about what I do, I’ve realized that I’ve always found demographics, geography, urban planning more interesting – and I’d like to pursue research in social science. I’m currently applying to MS Statistics programs with the rationale that it is both necessary for my development as a quantitative researcher and general enough so that I can shift to a PhD program in either statistics or a social science – if I want.

My questions for you are:

1) Do you think going for a traditional MS in Statistics with the goal of doing research in social science is advisable or would I be better suited in a graduate program in social science (public policy, demography, sociology, etc.)?

2) As someone with no formal background in social science, how do I get started?

My quick advice is that there seem to be a lot more people with public policy degrees who can do some statistics, than statisticians who can do policy. So if you can do the statistics degree, that could be a good idea.

But maybe I’m completely wrong on this one.

Any thoughts? What do you, the readers, think this person should do? Or, more to the point, what information do you think he should gather to better make his decision? And should he consider other options, such as continuing to work without getting more formal education?

41 thoughts on “He has a math/science background and wants to transition to social science. Should he get a statistics degree and do social science from there, or should he get a graduate degree in social science or policy?

  1. My educational background is almost entirely science and math…I’ve always found demographics, geography, urban planning more interesting – and I’d like to pursue research in social science

    Do you really need a degree to do that?

    Also, this says you have a background in “science” but find “social science” more interesting. That sounds like you already don’t consider “social science” to be a real “science”.

    • I don’t get that at all, just that he makes a distinction. It would be odd for someone to not consider it a real science who is both interested in it and has a science / math background.

    • He’s just using plain language. Just about everyone makes the distinction between science (physics, biology, chemistry, etc.) and social science (economics, poli sci, etc.). It’s because there is a silent “natural” before science. Also one can have a background in one thing but find something else more interesting.

      Quick test: If someone told you they were a scientist, would you think they worked in the natural sciences or social sciences?

      • It partly depends on which social science. I think an MS in stats is of course helpful but in many doctoral programs they will let you take a lot of stats courses, so one thing is to go somewhere that is both strong in your social science discipline and has a good stats or applied stats program. I agree with the idea of going for the PhD with funding. With a good math background you will be in good shape.

        People think social science is easy, but it isn’t , there are a lot of things you have to learn and the concepts can be hard to get your mind around. Depending on discipline you will have certain things that you will be assumed to know very well coming in. In other disciplines the difference between undergraduate and doctoral programs is huge for most students. But those programs assume you know how to think in a social science way (and again what that means depends on discipline). So in some cases registering for a non matriculated undergraduate course or two is not a bad thought.

        I don’t actually think that the MOOC route is great for that because what you need is a lot of writing and deep engagement with the material. The suggestion of trying to get a job as a research assistant is a nice one, but it can be hard to do that if you aren’t affiliated with the institution.

        I am going to say that if you a social science new PhD on the academic job market it can be helpful to have the MS stats because people have confidence you can teach the department stats course(s). (Whether that is justified is a different question.)

  2. 1) If he is not rich (and if his higher degrees are not funded), he should think hard before going back to school. The opportunity costs are large and, often, underestimated.

    2) Instead, he should self-train in data science. The resources available are amazing. I would start with the Johns Hopkins Data Science curriculum. (Given that he is thinking of spending 100k+ on graduate school, he should put down $500 and get the certificate for the 9 course sequence.) Or pay $10 and do a bunch of DataCamp classes in a month. Doesn’t really matter the exact choice. The important thing is to discover (cheaply!) how much he likes statistics and/or data science. Within these efforts, he can focus in social science applications. Gather more information on your own preferences before paying money for graduate school.

    3) Either before (or instead of) 2), he should try to get a job doing data science. If he has a math/science background from a good school, this is straightforward, either immediately or after 2). That will teach him even more about whether or not this is a good career match for him.

    • I think that the best points made so far are about taking on $100k of debt, “becoming a domain expert in a very specific area is where the real hard work is” from Shravan, and similarly being “more concerned about the lack of background in social science” from cwaigl.

      • Given that these disciplines are inundated with researchers, with over 5 million research articles counting all disciplines, I would think that excelling or distinguishing oneself would be a formidable task. This situation made more difficult in that there has been a compounding of statistical errors and their corresponding insight quality lagging in theory and practice.

  3. If I were him/her, I would first find out what social science area I want to work in. Unless that decision is made, it’s hard to know what to do next. Also, becoming a domain expert in a very specific area is where the real hard work is. In the end you may end up doing a one-sample t-test. But knowing when to do that and when you need to dust off and deploy the nuclear weaponry can take a while to figure out.

  4. I’m going to make the assumption that this person’s test scores, GPA, coursework, and letters are good enough to clear the bar everywhere they would like to go.

    This is a big assumption. As people on this site who have gone through a PhD program are surely aware, and with some variation in discipline, only the GPA and test scores are going to matter for the MS, but the importance of each factor is going to vary widely for PhD programs in a way that will seem almost arbitrary.

    However, if the goal is a PhD, in my (quantitative) field I was advised to get a MS in statistics at the same time, and encouraged to do so. It’s definitely something you want to ask about on the interview circuit- some places will train you poorly in statistics, some places will train you well entirely in your department (but not grant an MS in statistics), some places you will end up with an MS in statistics at the request of the department, and at one place you can end up with a dual PhD with only a little bit more coursework and a slight change to a research program (though I personally don’t recommend two dissertations, even if they end up looking quite similar- I didn’t do this, but I know people who did).

    If the goal is academia, get a PhD (obviously your mileage will vary by discipline). I guess on the very front end you want to filter to programs that have a very strong reputation for statistical training. Talk to your undergraduate advisors to get a feel for this, because while people will freely tell you who they think has strong (or weak) quantitative training, it’s not advertised. In my experience, I got honest opinions.

    • “As people on this site who have gone through a PhD program are surely aware, and with some variation in discipline, only the GPA and test scores are going to matter for the MS, but the importance of each factor is going to vary widely for PhD programs in a way that will seem almost arbitrary.”

      Speaking as someone who served briefly as graduate advisor for a Ph.D. program in math, and for about five years as graduate advisor for a MS in statistics program: In both of these, letters of recommendation were very important.

      Re getting an MS in statistics along with a Ph.D. in another field: I agree that this is a good idea. However, I would suggest investigating more widely (deeply?) than just talking with undergraduate advisors

  5. Piggybacking with a related question…

    somebody with a math science background who wants to get into machine learning and data science, but do it the right way with a solid stats background.

    Is there a non multi-year multi degree journey to learn methods, history, pitfalls to avoid, experiment design, and other useful relevant real world and, with hope, high paying skills?

  6. These two choices are a bit of a false dichotomy, because you can do both at the same time, pursuing applied stats through policy-oriented field. The example I am most familiar with is pursuing a degree in economics, with a focus on econometrics. Depending on your school and fields of specialty, the quant side of things is integral with the content matter.

    A pure statistics degree is likely to lead to theoretical statistics, which is not what it seems the question asker is interested in.

  7. [Geoscientist here with bg in physics, now doing a lot of machine-learning applications to earth science data, and looking a little bit towards social impact questions, without much of a hurry.]

    Maybe I’m misunderstanding the situation a little, but I’d be a lot more concerned about the lack of background in social science than the lack of background in statistics. For someone with a good grounding in math and quantitative science, the data science / statistics part can be acquired in self study — the resources are excellent these days. But the same is much less true for the fundamentals of sociology / anthropology / political science, which include the difficult theoretical underpinnings (the parts some of us like to call “soft”). Not to make a fine point of it, the field is dealing with statements about human beings, and any social science research of consequence is liable to ultimately have an impact on the life circumstances of human beings. Sometimes — in the fields close to mine this is common — the very selection of your subjects of study and your relationship to them is an ethical can of worms. The writer really wants to get the basics firmly into place and become part of a community that has a conversation about them.

    • I had a math undergrad and comp sci Ph.D. and spent fifteen years in natural language processing and speech recognition coding machine learning algorithms. I found it very hard to self-study stats. Sure, I could plug machine learning packages together and even build machine learning inference like stochastic gradient from scratch. And I could learn the Bayesian theory—that’s nearly trivial for the basics if you understand calculus and can get through conditioanl densities and expectations. But I didn’t understand applied statistics. I knew the tools and technology, but not the art and craft. That’s why I moved to Columbia to learn the material better.

      The exact same thing applies in social science. You can read Freakonomics and take some Coursera courses on econ and read the intro textbooks and work through examples on Quant Econ, but that won’t really teach you to do economics or econometrics. It’s a good start and I wish these things were around when I was younger, but it’s not going to get you all the way there. It’s like learning the rules of chess from a master (e.g., someone on Coursera), then playing a few games at home by yourself with a friend (working some simple problem sets Coursera-style). Not the same as a mentorship with the master or immersing yourself in the culture.

      I am curious how the peer grading works out on Coursera. That could solve some of this problem of not being abel to really wrestle through ideas on one’s own as effectively as with a group of peers.

    • +1, and I’ll add that you should think hard about the political angles. Most people have strong priors about questions of social science; everybody thinks they’re an expert (a few of them have a point). Going into social science is like going to Baghdad if you don’t understand the politics; you’re likely to inadvertently get involved in some multilayered conflict you don’t understand. The ideal of dispassionately analyzing the data will not go over nearly as well as it does in the natural sciences.

  8. A professor of mine from undergrad used to frame his answer to “what should I major in” like this: major in math. If you later decide that you want to do science, engineering, or social science, you will have enough math background to switch easily. If you major in something else and decide you want to switch later, it can be more difficult.

    In this scenario, deciding between statistics and social science, I would argue for getting the statistics degree. You will have opportunities to test the waters in the social science of your choice while getting the statistics degree. It will help you make the decision while giving you the technical tools to do the work in the future.

    FULL DISCLOSURE: I have a degree in statistics, so I’m probably biased. Heh, heh.

    CAVEAT: I do agree with D Kane. You should first spend some time online looking into the wealth of resources out there. That said, I still think there is much to be gained from formal education. Also, even good online materials can be a bit spotty. Case in point, I’m reviewing materials for an internal curricula that I’m putting together for the company I work at. One video series from a highly popular online teaching site refers to “accepting the null hypothesis” in its hypothesis testing segment. A small written note below the video says “never ‘accept’ the null, reject or do not reject”. Sigh.

  9. As someone with similar interests and dilemna I endorse the degree in Statistics advice….that’s what I did and it has enabled me to move from public health to education to public policy to political science (and back and forth). Statistics give you the tools that you can then apply across the board, as your interests and opportunities evolve. So I’m with Andrew’s general take–if you can do statistics, do it, and it will give you a lot of flexibility down the line.

  10. Finding a job involving social science research would be informative to him. Then he could see whether he is more excited by the social science theories guiding the research or the methodologies used to examine the data.

    If going for the MS stat or biostat, go to a dept with many social science collaborations. Take some social science coursework as well, and RA or do class projects on social science data.

  11. 1) Whatever he does, he shouldn’t apply to a master’s program. Just apply to PhD programs; get funding; and if he doesn’t want to stick around, just leave.

    2) If he decides to ignore the excellent advice in (1), he might only be able shift from master’s to PhD if he gets the PhD in the same field at the same school as his master’s (otherwise he will almost certainly be starting from scratch).

  12. I’m the person who sent the email. I’ve actually already made my decision but I’m interested in seeing what people have to say! Thanks everyone!

  13. Do a full-time research assistant gig for a social scientist (probably an economist) doing something adjacent to your area of interest. Almost all fields of applied microeconomics hire full-time RAs these days, and having a strong background in statistics will make you very attractive. Look at both the Federal Reserve banks and universities.

    Doing something like this will (a) give you some experience with what empirical social science actually looks like on a day-to-day level (b) afford you the opportunity to take classes, often paid for /subsidized and (c) give you access to good letters of rec if you decide that pursuing grad school in social science is what you’d like to do. If after a year or two you decide that you don’t want to do social science, you still got paid and built some useful skills

  14. Assuming willingness to invest money and time – MSc in statistics, make sure to keep the math background up to date and running (for some people math foundations seem to wash away as they go through statistical and data analytic practice), along the way, look into different specific fields and within fields, specific applications, to figure out the exact thing that will fuel the actual research motivation. If the latter be found, proceed to PhD; if not, work in the industry. When/if applying to PhD programs after MSc, find out at application stage how the program’s obligatory/recommended courses fare against the MSc curriculum, some programs might be willing to count certain MSc credits towards degree prerequisites, in which case, use the freed-up course credits for extra training possibly at other departments/programs/allied institutions (a lot of universities especially in larger urban areas have consortium agreements and the likes, sometimes not very well publicized – so there are ways of really tailoring a curriculum to one’s needs).

    Assuming no willingness to invest money and time, see D Kane’s advice above.

  15. One thing someone will not be able to self-teach is how sampling and data collection work in practice, which is critical for successful social science work. Someone without formal stats or social science training might could likely pick up that experience working in research settings as a research manager or hands-on operations person (and settings like that often want quantitatively savvy employees).

    Side note – a collaborator who has a graduate degree in applied math has made some valuable contributions to my work by having unique viewpoints on quantitative problems that people with graduate stats degrees have said they wouldn’t have thought of (due to disciplinary style, not lack of math knowledge). John D. Cook is a stats consultant whose PhD is applied math, and I suspect that background gives him leverage on a wider variety of problems than one might have with straight stats.

  16. Given what’s reported in the recently published “Statistical Rituals: The Replication Delusion and How We Got There” (e.g. that of the psych faculty teaching statistics surveyed 20% think p(replication | p Credentials

      • Ack! It ate half the sentence but then picked up first word of next sentence then killed last sentence. Should be 20% think p of replication is 95% given p less than .05 and thereafter something about needing Gerd Gigerenzer Seal of Approval for graduate statistics programs.

  17. You can save a lot of money and time by just starting the PhD, but in this case the student doesn’t seem to know what they want a PhD in, and that is a problem.

    One idea is to just start reading broadly. Fall down rabbit holes and decide which are the most interesting. Note what departments the authors are in.

    Another useful thing to do is research salaries and salary differences across fields. You can find salaries by searching big state school systems like “University of California salaries” or Texas or Michigan. Salary may not be the deciding factor for most people, but it’s probably a factor and most students are completely ignorant about it, because we have taboos against talking about money.

    The cost of getting a master’s is often, money and wasted time. Specifically, many PhD programs are fully funded (no tuition and they pay you) while most master’s are not. During the PhD, you can take Statistics classes as electives, so the master’s will waste some time.

    The benefits are that it can help you get into a better PhD program (higher ranked or better funding), you learn stuff, and it will give you time to narrow your interests.

  18. It’s probably a good idea to keep your math chops in working order, which suggests starting in a stats department or something like it. But I suspect that an MS in stats by itself won’t be enough to enable you to participate in a social science conversation in a full way. MS programs can be very full of coursework and it takes a lot of pedagogical care to fit “what is a meaningful research question?” in alongside all that. I can imagine MS in stats + applied stats job in a social science setting + PhD a few years later being a reasonable approach. Starting with a social science and tacking the MS on one course at a time can also work pretty well at some schools, depending on how demanding the two programs are.

    I don’t particularly recommend the path I took, which was: go for a social science PhD, realize halfway through that you’re interested in statistics too, quickly discover that you can’t keep up with math stat and a dissertation simultaneously, choose the dissertation, a few years later regret this choice and go back for the stats degree, now having forgotten 75% of calculus. (But hey, I survived! A reminder that suboptimal strategies can still work out fine.)

  19. If the goal is a Ph.D. and a career in research, then I’d recommend just finding a good match to your interests in a Ph.D. program at a place with strong stats and strong social science. You’ll get a modest stipend and you’ll get the more theoretical/mathematical classes that are nearly impossible to teach at an M.S. level due to the heterogeneity of student backgrounds. You’ll also get to dive straight into a research environment with peers and upper-level students you hang out with will be focused on research. There are also fewer students in Ph.D. programs, so the faculty-student ratio is much better for mentoring.

    Having said that, we’ve gotten two awesome Stan developers and statisticians through the quantitative social science M.A. (?) program at Columbia and specifically through Ben Goodrich’s courses (when Ben says someone’s good, they’re really good); so I can highly recommend the results from Ben’s courses (which are available online through YouTube, by the way).

    I was a math undergrad with a strong CS, psych, econ, and philosophy background (lots of grad courses as an undergrad in all the fields—Michigan State’s honors college was awesome for an undergrad with interdisciplinary research interests). I then went and did an interdisciplinary Ph.D. in cognitive science/computer science at Edinburgh, which was mostly linguistics and cognitive psych in terms of coursework and mostly theoretical computer science and linguistics in my publications and dissertation.

    I then went into the professor business for nearly 10 years, then worked in industry research, then in industry as a programmer. Finally, I was getting into stats and found myself unable to truly teach myself from online materials. I could learn the math and all the theory easily enough, but that’s a long way from application. So I took a huge pay cut to come work with Andrew and Matt Hoffman so that I could learn the art and craft of statistics in addition to the theory. I can highly recommend immersion as a way to learn stats, but you’ll need to find a good mentor. Although I took a serious yearlong analysis course and had no trouble with measure theory, my actual calculus skills in $latex \mathbb{R}^n$ are rather lacking, and my advisor as an undergrad convinced me “matrices are for engineers, real mathematicians study Galois theory”. So I’m you’re man for groups of ring homomorphisms, but I didn’t even know what a determinant or Jacobian was when I started working on Stan (it was everyone explaining Jacobians to me that led to Stan’s ability to unconstrain variables—it was like a magic trick; of course, Ben had to work out all the crazy Cholesky factor stuff for covariance matrix and correlation matrix types—that was way beyond my matrix skills when it first went into Stan). My point is that if you like learning and have good colleagues, you can keep picking these skils up as you go. Experience learning about densities and changes-of-variables, particularly multivariate ones, leads to a lot of insight into concepts like determinants and identifiability that will be forever buried if you just surf the intro courses online.

    The other thing I can highly recommend if you’re going back for retraining is the core data structures and algorithms class for undergrad comp sci majors. It’ll likely be the first required class for CS majors. Learning that material is key to being able to program effectively and communicate with computational people. So much of stats is computational these days that it’d be a shame to neglect the basic CS side of the education. As Andrew likes to say, you had to be a bit of a mathematician to do stats in the 20th century, but you have to also be a bit of a computer scientist to do stats in the 21st. You can take all these data campy type things, but it won’t give you the same depth. Like others have stressed in comments, learning the basics solidly makes it so much easier to learn advanced things than if you have shaky foundations.

  20. I would probably go for an economics PhD. Highly competitive but highly versatile degree with lots of employment opportunities. Plus you can do lot’s of social science and stats. Good luck!

Leave a Reply to Nik Tyresen Cancel reply

Your email address will not be published. Required fields are marked *