INGO memberships: raw or cooked?

I’ve been hanging out at Stanford, which is great fun.  One question that came up recently is “how to best measure INGO memberships?”  I’ve been dealing with INGO data for a long time and I have opinions…

First, some background:  John Boli and George Thomas were the first to recognize that International Non-governmental Organizations (INGOs) are a core infrastructure of world society.  The discourses and activities of INGOs are a key embodiment of an emergent global culture, and INGO play an important role in the spread of that culture.  Their book “Constructing World Culture:  International Non-Governmental Organizations Since 1875” makes this point very vividly.

These days, country memberships in International Non-governmental Organizations (INGOs) have been accepted as the standard way to measure national embeddedness in world society.  Countries tied to lots of INGOs are most exposed to global culture, and are fastest to adopt a whole host of policy innovations — new environmental laws, fashionable human rights commitments, and so on.

But, how should one actually operationalize INGO memberships in quantitative analyses?  Suppose citizens of a country are members of 1,500 different INGOs.  Should one use the raw counts?  The natural log of counts?  Memberships per capita?  Or something else?

I usually use the natural log of the INGO membership count.

As a practical matter, raw counts are hugely skewed (except in some cases — for instance analyses focusing on certain particular regions).  Logged INGO memberships are less skewed, and therefore work much better in regression-type models.  Also, one can make a substantive argument:  going from 100 to 200 INGO memberships has a bigger substantive effect than going from 1,100 to 1,200.  The natural log transformation helps take this into account (despite being a somewhat arbitrary correction).

Sometimes people suggest that INGO memberships be standardized by population.  Shouldn’t you correct for the size of the population?  Big countries can have more memberships… and besides, don’t you need lots of memberships to influence the culture of a large country?

These arguments are plausible, but ultimately I’m not sold.  First, the INGO membership variable from the Yearbook of International Association counts organizations that are tied to a country, not individual memberships.  An organization is counted as tied to a country if one or more citizen is a member.  That may not be ideal, but that’s the measure we’re stuck with.  So, if all 1.3 billion citizens of China joined Greenpeace, it would still count as one INGO tie.  Second, most diffusion studies focus on state policy, rather than individual attitudes or activities.  Many INGOs function as advocacy groups of various sorts — and don’t need to be connected to each and every citizen to influence policy diffusion.  Finally, I’ve looked at the actual result of standardizing INGOs by population.  Often it produces a very odd distribution.  Tiny island nations appear to be at the “center” of world society.  (Again, this could vary for different types of INGOs or if you focused on a particular region.)

In short, I’d recommend using logged INGO memberships as a default approach.  I can imagine situations where raw INGO counts or INGO membership per capita could be justifiable… but be sure to check the actual distribution before plowing ahead.

Those are my 2 cents.  If people have other views on this, I’m interested to hear them.

Of course, I’m only talking about count-based measures of INGOs, which are easiest to get.  Pam Paxton, Melanie Hughes, Jason Beckfield, and others, have been working on network-based measures of INGO ties.  That opens up a whole other range of options…

Event History Analysis Resources

Several people have asked me about books/readings on event history analysis (aka survival analysis).  Here’s my recommendation, particularly for people with some prior familiarity with EHA:

Cleves, Mario, William W. Gould, and Roberto Gutierrez.  2004.  An Introduction to Survival Analysis Using Stata, Revised Edition.  Stata Press.

It has the best explanation of the difference between Cox (semi-parametric) models and parametric models.

It has the best explanation of the uses of constant rate (exponential) models.

It does a very good job explaining some other often-confused issues, such as the difference between “normal” and “accelerated failure time” models, and the different types of frailty models.

And, it explains how to do things in Stata, which I use in my classes.  The only thing is that it doesn’t have substantive examples relevant to the social sciences.

That said, there are other very good books out there.  Box-Steffensmeier and Jones (2004) brings some nice substantive examples in political science, and does a good job of explaining the merits of Cox models.  The Blossfeld et al (2007) Stata book has great discussions of conceptual and research design issues — I always assign Chapter 10, which offers very helpful advice.

I also have lecture slides that some people have found helpful, both in my classes on advanced regression and my class on event history analysis:  http://www.socsci.uci.edu/~schofer/teaching.html

Panel Data Book

People ask me for advice about what to read to learn about various kinds of data analysis.  Here’s my recommendation for panel data :

Wooldridge, Jeffrey M.  2001.  Econometric Analysis of Cross Section and Panel Data.  MIT Press.

It was not an easy read for me.  But, it has really good coverage and is done very well.  Wooldridge also has an excellent basic econometrics book.