KroenigEditor’s Note: Back in February I riffed on a post by Erik Voeten in which Erik discussed two articles in International Organization (IO). One, by our colleague Matt Kroenig, argued that nuclear superiority gives states advantages in crisis bargaining (PDF). Another, by Todd Sechser and Matthew Fuhrmann, rejects this claim (PDF).

After the two posts sparked some interesting discussion–both on- and offline–I approached all three about doing a mini-symposium at the Duck of Minerva. They agreed.  Kroenig kicked us off with objections to Sechser and Fuhrmann, and soon after we ran Sechser’s and Fuhrmann’s critique of Kroenig’s article. Next, Sechser and Fuhrmann respond directly to Kroenig’s earlier post. This is Kroenig’s answer.

This is a guest post by Matt Kroenig. He is an Assistant Professor in the Department of Government at Georgetown University.

As I stated in my previous post, my new article, “Nuclear Superiority and the Balance of Resolve: Explaining Nuclear Crisis Outcomes,” develops the first coherent theory for why nuclear superiority matters even when both states possess second-strike capabilities.  The article then goes on to present a wide array of empirical tests demonstrating that states with a nuclear advantage are more likely to achieve their basic goals in nuclear crises.

Readers may be interested to know that when I began this research almost four years ago, I fully expected to find that nuclear superiority did not matter, but over time I became convinced by the unambiguously strong correlation between the nuclear balance of power and nuclear crisis outcomes in my empirical tests.  Turning up what was initially a surprising result, I spent several months racking my brain, and re-reading sixty-years-worth of literature on nuclear deterrence theory, until I was eventually able to develop a coherent logical explanation that accounted for these findings to my satisfaction.  I proudly present the result of these years of labor in my new article.

In their latest post on this subject, Todd Sechser and Matthew Fuhrmann (S+F from here on out) make a number of comments on my article, but none of their observations challenge the compelling, deductive logic linking superiority to victory, nor do they begin to overturn the robust battery of empirical evidence that I provide in support of this logic.  Nevertheless, I will respond to their various observations. Before I do, however, I will briefly return to the serious problems in their study.

Inexplicable Results and an Absence of Explanation

As I wrote in my previous post, I don’t understand why S+F gloss over sixty years of nuclear deterrence theory for an IO article on nuclear threats and, more importantly, the empirical results they report are so perplexing as to suggest something horribly awry in the empirical analysis.  I previously suggested that the problem may be with the new and untested MCT dataset.  Their latest response only deepens my concerns.

In my post, I questioned why the only table presented in a published study included 12 variables, but only 1 variable that consistently reached statistical significance.  S+F responded by pointing out that “other work using the MCT dataset points to the importance of alliances, conflict history, trade interdependence, geography, and even psychological factors.”  But, if these variables have been shown to be statistically significant determinants of compellent success and many of them are also correlated with nuclear possession and thus possible confounds, why were they not included in the nuclear weapons study?

More importantly, I criticized S+F because the statistically significant results they do report are so fantastical as to defy explanation (not just by me, but by any sentient being), they cloister these embarrassing results in an online appendix, and they do not even attempt to provide an explanation for them.

S+F falsely assert that our findings regarding conventional military power are similar, but this statement couldn’t be further from the truth.  To recap, I find that nuclear superior states are more likely to achieve their goals in crises.  I also find that states with a conventional military advantage are more likely to achieve their goals when examining all states, but this finding disappears in a subsample of nuclear-armed states.  This makes sense as conventional forces may be less important in conflicts among nuclear powers.

In sharp contrast, S+F find that: nuclear states are less likely to issue successful compellent threats, states with a conventional advantage are also less likely to compel others successfully, and compellent threats are more likely to succeed against a nuclear-armed target.

I challenged S+F to explain these inexplicable results and they said that weak states “have many incentives for resisting coercion by powerful challengers, such as reputation-building and domestic political gain.”  But, this still doesn’t explain why weak states would be better on average than powerful states at issuing compellent threats.  After all, powerful states have reputations and domestic politics too.

Moreover, if S+F truly believe these findings and their explanation for them, then why did they move these findings to an online appendix and provide their explanation for them in a blog post weeks after the article came out instead of addressing these matters directly in the published study?

The reported findings are truly unbelievable and F+S still cannot offer a plausible story to account for them.  Having carefully read many iterations of their study and their recent blog posts, I remain deeply skeptical that this research contributes anything to our knowledge of nuclear coercion other than confusion.

Addressing Fussy Criticisms and Misleading Claims

Now, I will turn to S+F’s comments on my research.

1. Perhaps the strongest critique raised by S+F is that they can find a model specification that results in a statistically insignificant relationship between nuclear superiority and crisis outcomes if they code a state’s nuclear arsenal size as being the entire combined arsenal size of all the nuclear arsenals of the states with which that state shares a formal alliance.  But this is an unreasonable coding decision.  Were the United States, Britain, and France all players on one big unified team in the Suez Crisis of 1956 because they were in an alliance?  Obviously, they were not.  Britain, France, and Israel aimed to seize the Suez Canal from Egypt without notifying the United States.  When the operation was underway, President Eisenhower pressured Britain and France to withdraw their forces, as did Soviet Premier Khrushchev who also made explicit nuclear threats against London, Paris, and Tel Aviv.  It would be nonsensical to argue that Britain failed to achieve its goals in this crisis despite having nuclear superiority over the Soviet Union due to America’s nuclear weapons given that Washington and London’s goals were at direct odds in the crisis.  In addition, if allies could simply rely on each other’s nuclear weapons, then why were Britain and France so determined to build independent arsenals in this period?  Aggregating nuclear weapons within an alliance and assigning the same nuclear arsenal size to every state in the alliance, therefore, does not make any sense. After all, IR scholars do not code conventional capabilities within an alliance in this way.

The concern about states with a nuclear patron potentially driving the results is a valid one, but a better method for dealing with this concern is to simply run the analysis after removing the states that have a nuclear-armed patron.  This is one of many robustness tests that I ran and as I report in the article, the core findings hold.

2. Next, S+F claim in both posts that nuclear superiority does not improve crisis outcomes because nuclear superior states only achieve their goals in roughly 50% of crises.  This number is true, as I clearly state in my article (by my count, the number is 53%), but the winning percentage for nuclear inferior states is much lower (only 15%).  This is possible because in many crises, neither state achieves its goals.  The fact that nuclear superior states achieve their goals in nuclear crises roughly half the time, therefore, provides strong support for my argument and it was misleading for S+F to suggest otherwise.

3. Third, S+F accuse me of inflating my number of observations in order to get statistically significant results, but this comment reflects an apparent misunderstanding of standard practices in the scholarly crisis outcome literature and misrepresents the strength of my findings.  It is an accepted standard (e.g., Beardsley and Asal 2009) to focus on the crisis participant, not the crisis, as the unit of analysis.  This is because in a crisis, it is possible that both states achieve their goals, that one state wins and one state loses, or that neither state achieves its goals.  This variation would be completely lost if one focused on the crisis as the unit of analysis.  I simply follow standard practice by focusing on crisis participants.  Moreover, and contrary to S+F’s claims, my results do not depend on inflating observations to get statistically significant results.  The results are equally compelling when simply eyeballing the list of historic nuclear crises.  As I report clearly in the article, if one wanted to focus on twenty nuclear crises as the unit of analysis, one would find that 55% of crises had at least one nuclear superior state that achieved its basic goals, but only 20% of crises had at least one nuclear inferior state that achieved its goals.  Again, this provides strong support for my argument.

4. Fourth, S+F report a problem with the way one of my variables is coded, but, as they also report, this issue in no way affects my results.

5. Finally, S+F suggest that our datasets are geared toward different questions and my data are somehow inappropriate because I do not focus explicitly on “compellent” threats.  By drawing such an overly fine distinction between compellence and deterrence, however, S+F may be too clever by half.  As Thomas Schelling himself argued, it is often difficult to distinguish between these two theoretical concepts in the real world.  In MCT, Sechser records compellent threats as “interstate demands designed to change the status quo, which are backed by the threat of military force.”  But, was the Cuban Missile Crisis a case of deterrence or compellence? Does one define the status quo as the moment immediately before Moscow placed missiles in Cuba, or the moment immediately after?  If the difference between deterrence and compellence is not entirely clear in the most carefully-studied nuclear crisis in history, then it is probably even less clear in other cases.  Rather than dropping a bunch of cases of nuclear coercion because a researcher decided to code them as “deterrence” instead of “compellence,” an alternate and arguably better method would be to simply examine all serious confrontations between nuclear-armed states.  This is the approach I adopt in my article.

Our Best Answer to an Important Question

In sum, S+F make a number of fussy criticisms of my study and in doing so seem to want to set an unrealistically high bar for research standards in this area – a bar that their own research does not come close to even reaching, to say nothing about surpassing.  If we had thousands of nuclear crises to study (and thank goodness we do not), or if we could design an experiment in which we randomly assign nuclear superiority to a treatment group, we might be able to draw stronger causal inferences, but this is simply not possible.

Nevertheless, some topics are important enough that it is worth having a best answer, even if we can’t have a perfect answer.  I believe that my study provides the best available answer to the question of what determines the outcomes of crises between nuclear-armed states.