Abstract
Thurner, Paul W., Christian S. Schmid, Skyler J. Cranmer, and Göran Kauermann. 2019. Network Interdependencies and the Evolution of the International Arms Trade. Journal of Conflict Resolution 63 (7): 1736-64. (Original DOI: 10.1177/0022002718801965)
An internal reanalysis of our replication package revealed that we erroneously defined the valued relational variables ‘Path Dependency’, ‘Geographic Distance’, and ‘Absolute Difference in Polity Score’ as binary variables. This was the consequence of using the function network in the R-package statnet in the code lines 1888-1916. By default, this function translates valued cell entries into dichotomous ones. We corrected this miscoding and reran our files. For ‘Geographic Distance’ we take the original values from Gleditsch’s Distance Between Capital Cities dataset (http://privatewww.essex.ac.uk/ksg/data-5.html). We provide all corrected figures from the original article in order to make the results comparable.
As seen in these figures, only the three erroneously defined variables change substantially in size and statistical significance as compared to the original analysis. All other coefficients maintained similar ranges and exhibit nearly identical patterns of statistical significance. A partial exception is the variable Defense Agreements: the weakening of its impact after the mid-1990s is now more pronounced. Therefore, the trade-off between security and economic considerations weakens even more in this period, as can be seen in the new Figure 6. Otherweise, the conclusions in the paper can be maintained except for those related to the three flawed variables.
The altered conclusions that we have to draw as a result of this reanalysis are: ‘Logged Geographical Distance’ is now relevant in this sectoral trade network, just as it is in the international trade network in general. Geographic distance constitutes a transaction cost with the importance of the frictions varying over time. These frictions increase in the period from 1958 to 1963. They then decrease until the mid-1970, oscillate until 2000, and re-increase remarkably in the following eight years, before they weaken again in the recent years. Decreases of these frictions point to a reinforcement of globalization efforts in the weapons transfer system. The ‘Absolute Difference in Polity Score’ exhibits the expected negative sign: the larger the regime dissimilarity, the smaller the tendency to transact weapons. Again, this negative relationship fluctuates over time: from the immediate Cold War period until 1965 these repulsive forces increased. Then, until 1980, we observe a relaxation of this aversion. The following time period (1980 to 2005) exhibits another tightening of reluctance to trade with politically distant regimes. After 2005, we observe again a reversal of this reluctance. ‘Path Dependency’, our measure capturing the sum of all previous trades in the past five past years, shows a persistent strong positive effect as expected. For this variable, a square root-transformation performed comparatively better than a log-transformation in terms of model fit and degeneracy.
In addition, there are some remarkable changes with regard to the assessment of the superiority of the TERGM over naive logit: TERGM proves now to exhibit an even better prediction performance in terms of precision recall for most of the period. The impact of the endogenous statistics is negiglible only during the beginning of the Cold War and the collapse of the Warsaw Pact, corroborating our expectation that these network effects reflect structural conditions that have to crystallize over time.
Online Supplemental Material has also been updated for the article. We reran every model in the original appendix with the new correct coding. As can be seen in the new file Appendix, all original coefficients (except the three miscoded ones) remain similar-to-identical thus allowing us to maintain our original conclusions. We adapted Figure 10 by comparing both degeneracy tests (once with GWESP and once with the number of triangles) for the year 1970. In the original appendix we compared years 2013 and 1970 which was inappropriate.
Time series of the estimated endogenous TERGM parameters for the time period 1956 to 2013.
Estimated value trade-off between security concerns versus economic incentives (green dots: significant parameters; gray area: confidence interval).
Out-of-sample areas under the curve over time of the preceding five years.