If it looks like a statistician and walks like a statistician, it's a political scientist?

December 10, 2008 4:54 PM

Paging ROU_Xenophobe and other applied statistics researchers: Why are people who do a significant amount of applied statistics and statistical computing enrolled in poli sci/sociology grad programs rather than statistics?

This question is fueled by looking at the work that people associated with quant poli sci profs like Gary King at Harvard and Andrew Gelman at Columbia do. It seems that being primarily based in a social science department might be limiting in terms of stats work, but if one has a substantive topic interest in a social science area, is it generally possible to explore that in a stats department? If you're primarily doing statistics, why not get a stats degree instead?

I'm asking this as a student who wants to focus on social statistics and methodology in grad school and eventually work for some big statistical bureau (think StatsCan) or in health promotion. However, I'm worried than having a MA in a social science won't be as well regarded in industry as would a MSc in statistics with an applied focus &emdash; particularly, I don't want people immediately assuming that my studies weren't mostly of a quantitative nature. (I'm also beginning to get really tired of having to explain to people that there are sociology majors who can, you know, do math.)

Also, is there regional variance in where applied stats people enroll? I'm in Canada, where there isn't as much work in political methodology or social science statistics happening in social sciences departments compared to many American schools. Is the trend of having any sort of quant research housed in non-exclusively-quant departments an American thing?

posted by thisjax to education (8 comments total)

Demographics is the most stat-y social science I know.
I am a quant social scientist and I'd venture a guess that a big gment agency would prefer the methods training of a social scientist over a stats person.
posted by k8t at 5:10 PM on December 10, 2008

Also if you're interested in health why not an MPH?
And at my university we have PhD emphases which are like a minor. One of them is a quant emphasis. That would probably help.
posted by k8t at 5:15 PM on December 10, 2008

The word "applied" stands out in your question -- many applied quantitative researchers in the US, as you have pointed out, have chosen to ally themselves with a social science discipline rather than statistics itself. In my experience, statistics graduate students and professors spend a lot of time discussing method in the abstract or in refining increasingly sophisticated and complicated techniques. On the other hand, quantitative social scientists are often interested in substantive questions before they are interested in methods; the methods are simply a means to allow them to answer the questions of interest.

As for the other part of your question, I do know that American social science programs tend to be more quantitatively oriented than their European counterparts; I do not know if this applies to Canada or not. At my university, the sociology graduate students are some of the best-trained quantitative analysts, but this isn't the case, say, in England.

I wouldn't worry about people thinking that your training isn't up to snuff with statistics graduate students. Anyone who is doing the hiring will be able to look at your CV, see your proficiency, and will also know that social science students are often very well trained in quantitative methods. Many graduates from our department end up working for the government (particularly the Census Bureau) and are hired specifically because of their statistical chops.
posted by proj at 5:23 PM on December 10, 2008

My guess is that a statistics department would tend to be very math and theory oriented and would tend to be less applied and more about developing new statistical methods.

I come from a relatively different background (biophysics PhD) but in my field you can't conclude much from the department someone did their PhD in - you rely on their publications (and their CV, letters, etc) to figure out what their background is and what they've done. For instance you can find people from my graduate program who do computational bioinformatics and people who design new microscopes, so the program itself isn't a good indicator of their skill set.
posted by pombe at 5:28 PM on December 10, 2008

Well, Gelman is actually in the stats department too, and I don't think it's just a courtesy appointment.

More generally, advanced methodologists (of which I am emphatically not one of) stay in political science primarily because their substantive interests are strong enough that they still view the various quant or formal methods they specialize in as tools rather than ends in themselves.

It's also probably the case that even very good methodologists in political science rarely have the chops (and recommendations and connections) to get hired by a statistics department as an assistant professor, and by the time they're tenured and have come up with several new estimators and so on, they have a home in a department somewhere and in the discipline and don't want to uproot their professional lives to move to a stats department, where they'd also be a smaller fish in a larger pond. Gary is probably like this -- why on earth would he leave his little fiefdom at Harvard, and renown and respect in the discipline?

More broadly, I don't think you should find it puzzling to see people doing what you think of as applied statistics work outside of stats departments.

Some of them, me included, are just people trying to answer arguably substantive questions. Come up with a theory, extract implications of that theory, then see whether those implications bear out. There are lots of ways to do this, quantitative and not. But for most of the people doing quantitative work in the discipline, it's just a tool to compare data and theory.

Other people are doing more straightforwardly statistical work. Some of these folks largely import and explain established techniques from elsewhere. Other people put together new statistical estimators or techniques to deal with questions of particular importance to the discipline, but narrow enough that statisticians probably aren't interested in. Estimating ideal points of legislators from votes, like Keith Poole did, is a good example; these estimators include a lot of political science "content" in them. Likewise, ISTR but could be wrong that a lot of item-response models came out of education research.
posted by ROU_Xenophobe at 8:45 PM on December 10, 2008

And no, quant research isn't just American, at least not in political science. There's a big quant program in Essex every year, sort of the Euro equivalent of the ICPSR school.

You're in Ontario. Are you in UT? UT doesn't have or do much of anything in the way of quant work in political science, but that's just UT being weird. Other programs in Canada, like Western, wouldn't be out of place in the US.

As far as credentials for employers to see go, you could always try to go to Essex or ICPSR and take some relevant classes.
posted by ROU_Xenophobe at 9:01 PM on December 10, 2008

....if one has a substantive topic interest in a social science area, is it generally possible to explore that in a stats department?

You certainly could, but you would be expected to fully understand and possibly extend the "why" of the statistical analysis, whereas in the social science simply understanding how the model works is sufficient, and the focus is more on the topic. As ROU_Xenophobe notes statistics in the social sciences is often seen as simply a tool, and there is definitely not a limit - most departments are keen to promote stats.

The stats folks are still working on specific real world issues, for example, look at this recent edition of Journal of the American Statistical Association. It has got Irish electoral analysis, risk studies, and a whole bunch of bio-medical work. If look over one of the articles you will find the emphasis is on the model used, the range of topics covered is not indicative of how well rounded stats folks are, but that to them the data is largely interchangeable. You referenced political science research examples, and in that field a great deal of the statistical analysis models are borrowed from econometrics, which is where economics intersects with statistics, so there really isn't a hard boundary between these fields.

Current interest in statistics is dominated by the bio-medical end of things- I'm basing this on the focus of the top stats journals (Bioinformatics, Biostatistics, etc.) although there is lots of action over in education - the Item Response Theory folks (The Rasch Model or Death!) are really stirring the pot.

Why are people who do a significant amount of applied statistics and statistical computing enrolled in poli sci/sociology grad programs rather than statistics? Mainly because calculus isn't a prerequisite for the political science programs. If you didn't focus on math in your undergraduate studies you have a hard time getting into a stats program, as they expect you to have a relatively high level of mathematical abilities coming in.

Bear in mind that for a great deal of social science research, the statistical analysis stage takes up only a small portion of their work. As previously noted, the CV is key, and I want to emphasis the importance of internships - if you want to work for a particular agency find a way to get an internship there or someplace just like it.

American academic departments reward publishing, and quant work is often quicker to complete than qualitative, so the incentive is to churn quant out. This isn't the only factor, but explains some of the discrepancy between US and Canadian schools.
posted by zenon at 11:08 AM on December 11, 2008

Late to the party as usual, but chipping in anyway. Others have noted the more theoretical and technical focus of a lot of pure statistics programs. I can back this up: the statisticians I have known who come from a pure statistical educational background have more often than not been great at knowing the ins-and-outs of various methods, but not necessarily so great at asking/answering research questions. Economists and political scientists, on the other hand, work in fields in which there are generally massive amounts of data readily available, and so these fields lend themselves to extending statistical methodology in the course of doing the research work. I will admit to having major stats-crushes on both Gelman and King.

The wording in your question suggests that you're planning to get a master's degree but not necessarily a PhD. There's a lot of variation in what kinds of stats training different programs include at various levels of the program. So if your real goal is to do statistical analysis for the government, look carefully at how many and what type of stats courses each program you consider will get you. For StatsCanada or health promotion, you might extend your search to include MS programs in Epidemiology, or you might prefer to specialize in complex survey design & data analysis, which is a special kind of beast. You could also look at MS programs in Biostatistics, although some of these tend to focus heavily on preparing people to work on clinical trials.

The rub is that a lot of terminal masters' programs in the social sciences are NOT going to be heavy enough in methods training...so if you're thinking along the lines of a social sciences masters degree, you might be better off getting accepted into a PhD program and then quietly giving yourself the option to hit the eject button once you complete your master's degree...or just keep going.

It's fun and rewarding work - good luck!
posted by shelbaroo at 5:44 PM on December 11, 2008