Thank you, Brandon, this is great!
Laurel
(PS don't forget that Harvard students can also take classes at MIT)
From: gov2001-l-bounces(a)lists.fas.harvard.edu
[mailto:gov2001-l-bounces@lists.fas.harvard.edu] On Behalf Of Brandon Stewart
Sent: Monday, April 04, 2011 12:46 PM
To: Class List for Gov 2001/E-2001
Subject: [gov2001] Thoughts on Other Course
Gov2001,
I got the following question from Julie over the weekend (which I reprint here with her
permission):
"I am wondering if you guys have any advice for another stats course I should take
next. While I feel fairly comfortable with the material covered in GOV2000 and GOV2001, I
kind of just want to go over it all again and get more practice applying this stuff. Any
suggestions?"
I figured others might be interested in the answers as well (hence the email to the
list).
First, let me say that I think this question is posed in exactly the right way. Even
though you are "comfortable" with the material in 2000 and 2001 its useful to
learn it again in another setting. When I've spoken with Gary about his methods
education in grad school, that's what he told me he did. He just went to a bunch of
different departments and took classes teaching basically the same material from different
traditions. So just because you've already covered Generalized Linear Models in this
class doesn't mean you should shy away from taking a GLMs course in the stats
department.
Let me start by giving the short, direct answer: Stat 149/249 is the statistics department
equivalent of this class. It covers generalized linear models from quite a different
perspective (closely following the canonical textbook McCullagh and Nelder if that helps).
If you want to review material in this class and get more practice. This is the way to
do it.
Now for the longer answer. Generally speaking the major choice you have to make is which
department to take the class in. The economics perspective is very different from the
statistics perspective (for example). In fact you see the hallmarks of both traditions in
political science. My background has favored that statistical over the econometric
approach, so my apologies if my answer is a bit biased (note my answer is also poli-sci
biased and Harvard-centric because that is what I know!). Here are some thoughts on
individual classes though:
1. Stat 110: Taught by Joe Blitzstein this is the introduction to probability. There is
no data in this course so you won't get "practice in application" but it
gives you a much firmer understanding of probability and the distributions that we assume
for the data in generalized linear models. I think everyone interested in methods should
take this class.
2. Gov 2002: Taught by Adam Glynn and Arthur Spirling. This class is an overview of
various methods focusing mostly on causal inference and light Bayesian analysis.
Definitely worth taking if you want to learn more about causal inference. As it currently
stands its more of a reading seminar than a practical applications course.
3. Stat 139/239: This class covers linear regression from the statistical perspective.
Useful if you want to practice what you learned in Gov2000 but I think you could skip it
and go straight to...
4. Stat 149/249: This class covers generalized linear models. Its the closest
approximation to Gov2001 in the stats department. You cover a lot of things traditional
to GLMs not covered in this class including the exponential dispersion family, Fischer
scoring and analysis of deviance. It also helps you to pick up on nuances in the material
you might have missed the first time around!
5. Stat 220: Bayesian Data Analysis. When you are ready for serious Bayesian work this is
the course for you. It helps to already know some Bayesian statistics going in.
6. Stat 221: Statistical Computing with Edo Airoldi. If you are going to do methods
seriously: take this class. It helps cement the practical logistical issues of Bayesian
model estimation, EM, coordinate ascent etc.
7. Econ 2140: Econometric Methods with Guido Imbens. This is similar to a Gov2001 type
class but taught from the econometric perspective. Imbens is great but it helps to have a
firm foundation befroe going in.
8. Econ 2142: Time Series with Jim Stock. If you are interested in doing Time Series
consider taking this course. Jim Stock is probably the top time series researcher in the
country so its an awesome opportunity. The Stats department also offers a time series
class but it tends to be geared towards situations with considerably more data than we
typically have (e.g. if a machine produces a thousand widgets per minute...etc)
9. Sociology 275: Social Network Analysis with Peter Marsden. This is a great class for
learning about network analysis.
More than anything else the key is to just start reading into different literatures. Come
to Gov3009 (Research Workshop in Applied Statistics) and read Political Analysis or other
methods journals. At a certain point you obviously have to stop taking classes- but once
you have Gov2001 and a firm background in probability theory- you should be able to read
into any literature with enough time.
If you have thoughts on other classes or questions about a specific class, send it out to
the list. Also feel free to come talk to me during office hours (I've taken quite a
few stats classes here).
Brandon
_______________________________________________
gov2001-l mailing list
gov2001-l(a)lists.fas.harvard.edu
http://lists.fas.harvard.edu/mailman/listinfo/gov2001-l