Quantitative Methods II

Spring, 2007

Nathaniel Beck (nathaniel.beck@nyu.edu)

Office: Rm. 407, x88535

Office Hours: T1-2:30, plus any other time (either knock on my door, find me somewhere or make an appointment by email; please note I have offices on 2nd and 4th floor, so try both)

TA: Alejandro Quiroz Flores (contact information available later)

Courese meetings: In Ph.D. lab T and Th 10-12, section on Friday in lab at a time to be determined (most likely 10-12)

This course is a continuation of Quantitative Methods I and a precursor to Quantitative Methods III (taught in the Fall). Note that many topics that were previously in QII are in QIII (and some topics that used to be in QI are now in QII, given the changes in QI). It mostly deals with a series of advanced models, many of which are estimated via maximum likelihood. At some point maximum likelihood becomes easy, and the relevant issue is then the correct one: how does the model you estimate correspond to the political science you care about.

Text: Cameron and Trivedia, Microeconometrics

You also may find the Greene and Wooldridge graduate texts useful; the corresponse of readings should be obvious

Web site: The dynamic syllabus is on the class web site. This will, by week, have topic, reading, links to overheads, other materials, data sets, exercises and anything else of use.

Grading is based on weekly exercises, a final project (to be described later in the semester) and class participation, etc. (and it never hurts to laugh at my jokes). The rough balance here is 50% exercises, 40% project and 10% intangible, but I reserve the right to alter this.

Exercises: These are available weekly on each Thursday, and cover the material covered that week. Typically the section on Friday will discuss issues specific to the exercises (which are typically data analytic excercies using Stata (and its Mata matrix language). These are due after the weekend, so any issues can be discussed on the ensuing Thursday.

The exercises will involve some pencil and paper analysis, some work in Stata that perhaps involves matrix manipulations or writing your own maximum likelihood code (these are early in the course) and some work in using Stata to estimate models, with the exercises then being cncerned with interpretation. The goal is not to type 8 letters in Stata but to understand the underlying models; thus early on I will ask you to make sure that you can reproduce using relatively primitive commands the various high level stata commands (so you have to show me you could produce with more primitives what "predict" will actually produce for most of the rest of your lives).

I assume that exercises will be done in small groups (no smaller than size 2, no larger than size 3). While group exercises can lead to what might be thought of monitoring issues, the issue that really concerns me is that not everyone in the group learns as much as they should. Thus everyone at least must hand in their own written work, and I would like the written work from group members not to be identical (so that everyone has processed the exercises for themselves). Please do not give us more Stata output than is necessary (though of course give us as much as we need to figure out what you did) --- that is, you can edit your Stata log files (and annotate them). The easier it is for us to follow your exercises, the better feedback you will get.

There will also be exercises where you read substantive articles relevant to you and write up a report.

The final project is due at the end of the semester and will be described in more detail when students start to panic and ask me about the paper. Please note that because of my travel schedule that some classes will be devoted to discussion of the project with the TA.

Other web sites: Lots of folks teach courses similar to this one, and lots of folks have invested a lot in creating overheads and data sets and the like. So take advantage of this. Amongst the web pages that I am aware of, I can enthusiastically recommend those of Jan Box-Steffensmeier (OSU) Charles Franklin (Wisconsin), Simon Jackman (Stanford), Gary King (Harvard) and Chris Zorn (Emory). There are also very useful web site connected to the ICPSR maximum likelihood courses (you can find links from Franklin or Zorn) and the Big-10 ITV program (easiest link is from Box-Steffensmeier). There are trillions of other sites devoted to maximum likelihood, and many of them are useful and reliable. So google away.