Thursday, May 18, 2006
Redefining the Gap 11, Results
Note: This is a selection from Redefining the Gap, part of tdaxp's SummerBlog '06
|Nalign – 2||0.01||0.58||0.43||0.4||0.55||0.67|
|Nalign – 3||0.001||0.57||0.41||0.38||0.56||0.66|
|LDC – 2||-0.08||0.31||0.55||0.44||0.41||0.49|
|LDC – 3||-0.08||0.38||0.54||0.4||0.62||0.59|
The tables you see above are the coefficients of correlations for the models described in this series to the measures Barnett describes. This study looked at the population of all states, not a sample of states, so the margin of error is +/-0%. These numbers are completely internally valid -- they describe carefully derived measures. The difference between them is significant. However, the greater question of whether or not the correct measures were used is a different subject.
Chirol from Coming Anarchy suggested that I look at the Four Flows instead of brutality, nastiness, etc. It may be that I misconstrued what Barnett meant in the passages of Pentagon's New Map where he gives the definitions.
Regardless of the meaning of these numbers, a short discussion of the results is included below.
Brutality. This was the biggest surprise. For most measures, including Barnett's Core-Gap and Old Core - New Core - Gap, brutality decreases in the Core. This is because the University of Maryland's ICBP database that I used measures the countries involved in wars. Besides ignoring some sub-state conflicts, the project would this could the Kosovo War as mostly a "Core" war. After all, nearly all the combatants -- America, England, etc, - are Core states.
Still, the Afroislamic Gap is the best predictor of brutality. Afromuslim countries go to war more often than any other states. The worst predictor was the Old Core - New Core - Gap model.
Nastiness. Measured through lack of political freedoms and human rights, Afromuslim states fail again. The worst measure is merely defining Lesser Developed Countries (LDCs) as the Gap.
Poverty. Here, Barnett's economic determinist model shines through. The very best measure is Old Core - New Core - Gap, and the second best is a more general Core - Gap. Interestingly, here the Afroislam model scored the worst -- a reversal of our experience with Brutality -- though here at least, both show a positive correlation between being in a "Gap" and general badness.
Solitude. I modified Barnett's measure, from internet hosts in a country to internet hosts per capita. It would make little sense to call a very populous state the most connected state if only a small fraction of its population had access to the internet. The results here are the similar to the ones for poverty -- Old Core - New Core - Gap the best gauge, Afroislam the worst. Interestingly, here a 1st, 2nd, and 3rd world model of the globe does better than Barnett's simpler Core-Gap model.
Shortness. Want to die early? Move to an African or Islamic country. Only looking at the world from t he point of view of Developed -- Lesser Developed -- Least Developed states comes close to this. The very worst predictor is Barnett's Core - Gap model, though Barnett's Old Core - New Core - Gap model is only slightly better.
All in All. Averaging these scores together, the AfroIslam model remains the best for describing the Hobbesian states we fight against and for. All in all, however, the ups in one Hobbesian measure seam to compensate for the downs in others, making all of these pretty good. Still, this shows a danger of just looking at an agregate measure instead of more specific measures.
A Note on the Result. I'm not a statistician. I have advanced training in predicate calculus and relational algebra, but the pseudo-math of statistics is not my forte. I would much rather have my analysis short to pieces than for it to just sit here. Likewise, I used an extremely simple tool to run these numbers.
Please, correct me. Show me where I am wrong. And then, let's shrink the Gap -- Afroislamic or not.
Redefining the Gap, a tdaxp series:
Redefining the Gap 1. Prologue
Redefining the Gap 2. Summary
Redefining the Gap 3. Introduction to Geopolitics
Redefining the Gap 4. First Geopolitical Theories
Redefining the Gap 5. The North and the South
Redefining the Gap 6. Critical Geopolitics
Redefining the Gap 7. The Pentagon's New Map
Redefining the Gap 8. The Research Design
Redefining the Gap 9. Methods and Operationalizations
Redefining the Gap 10. Limitations and Conclusion
Redefining the Gap 11. Results
Redefining the Gap 12. Bibliography
Redefining the Gap 13. Appendix: Computer Code
Redefining the Gap 14. Appendix: National Codes
First off, GREAT work. Putting numbers behind a global-scale theory is hard work. I am currently a grad student in meteorology, and being an inexact science; we live and die by the pseudo-math of stats. We do also have an appreciation for complex, non-linear systems with possible perturbations (Typhoons/Nor'easters, etc)
I would have to raise an issue I have with the analysis. You use distinct values for variables (0, 1, or 2) that are discrete in time and space, and then apply the method of correlations to analyze them. Correlations are truly meant for variables that are continuous in time and/or space (2D correlations are REALLY tricky). Using binary/ternary values can really mess with the correlation coefficient results. I was shot down on a project where I assigned El-Nino years a value of "1", La-Nina years a "-1" and neutral years a "0", so I have gone down this road.
When you described earlier how you standardized your results (For each state, it's value will be calculated by taking the difference between that state's value and the lowest state's value, divided by the difference between the highest state's value and the lowest state's value.), I thought that to be quite an odd way of doing things.
A better route may be to use the more widely used of creating a normalized anomaly for each states value (Take the current value, subtract the sample mean, and then divide by the standard deviation of each). This results in a distribution of values with a mean of 0 and a variance of 1. The zero line could then be used to create the dividing line between "haves" and "have-nots" of each variable. Maybe those states whose brutality [or whichever index you choose] is 1.5 stddev below the mean could be quantified as REALLY brutal.
From a data standpoint, can you obtain these same indices for each country for a few years? This would create a time series for each country.
From a methods standpoint, if you can get that data, each country would then be an ensemble member, and you could use other methods that combine both time and space (EOFs which have been used in population studies since the 1930s).
A multivariate regression might also be a good method, being that you have multiple inputs (brutal, nasty, poor, sol, short) trying to "predict" one thing on the left. This would take more work as there really isn’t a value that could be considered ground truth that your research would be trying to find a way to predict.
One problem introduced into all of this is that most statistics are suited for continuous data, not variables with fixed values
Again, I must compliment you. This is great work that will yield great benefits.
Posted by: Matt R | Thursday, May 18, 2006
So, what does this do for the old-core-newcore-gap map? Is AfroIslam a beter model? Whatdoes that say about central and osuth Americ (definetly not core states). Coming Anarchy has a post on ungoverned spaces ( http://www.cominganarchy.com/archives/2006/05/20/neo-medievalism-ii/ ) which I am thinking of as mini-gaps.
Maybe the proper mapping is a set of mini-gap networks and mini-core-networks that are strongly intra-connected, but only weakly inter-connected.
Posted by: purpleslog | Saturday, May 20, 2006
Thank you VERY MUCH. Your comment was extremely informative. I apologize for the delay in getting this reply to you (check the blogs for reasons for that ;-) )
Instead of the ternary variable in the El Nino years study, did you use something else? (I agree with your general comments on statistics, by the way)
I liked your discussion of normalize anamoly. I got my method from the UNHDI, which uses that method in scaling its numbers.
Same apology for the lateness of the reply.
Among other things, it makes the Bush administration's focus on Iran instead of North Korea make more sense.  Focus on Barnett's model, with its inclusive Gap, and the "tailbone of the Cold War" is obviously the gravest threat. Limit the really bad countries to AfroIslamiyya, however, and the implication is that the a second Islamic bomb is the worse future.
Posted by: Dan tdaxp | Thursday, May 25, 2006
I would view the brutality index with suspicion. It strikes me as something akin to trying to discover the arsonists by their proximity to fires. Yes, you'll get the arsonists, but you'll also get the firefighters caught up in the same net. Is the US scored as more brutal for its incursion into Somalia to avoid mass starvation? A group of allies bands together to defend one another and by your measure, they are as equally brutal as the aggressors in the war. That just doesn't make sense. Somebody needs a better measure of brutality.
Posted by: TM Lutas | Thursday, May 25, 2006
I agree with your criticism. You are right that America scores as very brutal -- indeed, as the most brutal state in the world, with a scaled score of 0.0 (0 being the worst. The only country close is Iraq, with a brutality of .58.
This is a limitation of the data. Tom defined brutality based on where the wars are, but the University of Maryland's ICBP (which I took for his "current conflicts database") lists them based on combatants.
It would be interesting to see a "corrected" brutality score. However, that AfroIslam still has apositive correlation even with this complication means that Gap is very, very brutal.
Posted by: Dan tdaxp | Thursday, May 25, 2006