Why do international relations scholars bother doing "Large-N" work?
We discussed this question on the first day of my "Quantitative Security" graduate course.
[THREAD]
We discussed this question on the first day of my "Quantitative Security" graduate course.
[THREAD]
We looked at some standard pieces on this topic, such as the @Prof_BearB & Anne Sartori 2004 piece from this edited volume
books.google.com/books?id=o0iOO…
books.google.com/books?id=o0iOO…
The piece discusses a host of reasons why scholars do large-n work, as well as raise the pitfalls of that work.
Many of the points they raise are found in other work (that, in turn, cites them), such as...
Many of the points they raise are found in other work (that, in turn, cites them), such as...
...the @EuroJournIR piece by @stephenWalt & JJM on the dangers of "hypothesis testing"...
journals.sagepub.com/doi/abs/10.117…
journals.sagepub.com/doi/abs/10.117…
...and the (currently VERY relevant) @HDiplo dialogue on the use of large-n studies to understand nuclear crises (featuring @NarangVipin, @mcfuhrmann, @kroenig, and others)
issforum.org/forums/2-what-…
issforum.org/forums/2-what-…
But all of these pieces discuss two points in tandem that I think we should really separate: (1) collecting a lot of observations & (2) using statistical methods to analyze those observations.
To me, these are separable issues.
To me, these are separable issues.
Statistical methods are a useful and systematic way of evaluating the patterns (or lack of patterns) found in large-n data.
But statistical methods presume that there is merit in looking for patterns in a large number of cases.
Is that the case?
But statistical methods presume that there is merit in looking for patterns in a large number of cases.
Is that the case?
That's why I really like one of the first points raised by @Prof_BearB & Sartori 
Why is aggregating a bunch of observations useful?
Well, I flipped the question on the students: suppose you just want to study a single case?
Well, I flipped the question on the students: suppose you just want to study a single case?
For example, in my new @CornellPress book, "Arguing About Alliances", I dedicate Chapter 4 to the 1901 Anglo-British negotiations
cornellpress.cornell.edu/book/978150174…
cornellpress.cornell.edu/book/978150174…
Notice that I even labeled it "A Key" instance of non-agreement in military alliance treaty negotiations.
How did I know that?
How did I know that?
First, I had a sense of the "universe of cases" because I had collected such data (which I described in chapter 2)
I stress having a "sense" for the "universe of cases" because,
(a) especially when you go further back in time, you really don't know what you're missing.
(b) historians are often implicit about this wider universe: they had a reason for choosing one case over others
(a) especially when you go further back in time, you really don't know what you're missing.
(b) historians are often implicit about this wider universe: they had a reason for choosing one case over others
Second & related, seeing the "universe of cases" made it easier to recognize this negotiation as having more at stake than other negotiations.
In other words, it was an "important case" (or as many historians have called it, a great "missed opportunity")
jstor.org/stable/2440710…
In other words, it was an "important case" (or as many historians have called it, a great "missed opportunity")
jstor.org/stable/2440710…
Third, I could also see that, relative to many other cases in my "universe of cases", this negotiation had the theoretically correct values for all of the key variables (i.e. given the values of the key explanatory variables, the outcome variable had the expected value).
But was the observed correlation due to the "mechanism" proposed by my theory? 🤔
That's why I then had to dig into the documents!
That's why I then had to dig into the documents!
Of course, having a sense for the "universe of cases" is just one reason to collect large-n data.
A second big reason is more fundamental to international relations as a discipline.
A second big reason is more fundamental to international relations as a discipline.
One point of pride for IR scholars is that we try (hope/pray?) to develop general theories of state behavior
[NOTE: set aside concerns that we're only explaining Russia for a moment 👇]
[NOTE: set aside concerns that we're only explaining Russia for a moment 👇]
While someone might evaluate their theory using carefully chosen cases, a good theory should be able to speak to patterns in the wider set of relevant cases (note: this was a key point discussed in the @HDiplo debate).
Hence, at some point, it's useful to see how well IR theories fit the patterns in the broader set of data.
Of course, this presumes that the data used to construct those patterns is accurate and properly formatted, which is a whole other issue 👇
Of course, this presumes that the data used to construct those patterns is accurate and properly formatted, which is a whole other issue 👇
That's pretty much where we left things. Two big reasons to collect large-n data:
1) See the "universe of cases"
2) See patterns that support/rebut IR theories
[END]
1) See the "universe of cases"
2) See patterns that support/rebut IR theories
[END]
