NUCLEAR DIFFUSION SURVIVAL MODEL
A final check on the results from the nuclear weapons chapter is conducted using an event history, or survival, model. Survival models are especially useful for a task like measuring the acquisition of a weapons system because they are designed to predict failure rates, or the length of time until a given event occurs.1 Survival models are also helpful for dealing with rare events such as the acquisition of nuclear weapons (less than fifteen states have acquired nuclear weapons over a sixty-one-year period), where the passage of time matters in predicting outcomes (Box-Steffensmeier and Jones 2004; Singh and Way 2004, 871). Creating a survival model involves outlining a “hazard” rate based on the underlying distribution of risk, or the probability that as time passes an event will occur—in this case, the initiation of a nuclear weapons program or acquisition of a nuclear weapon. This makes survival models well suited for measuring something like nuclear proliferation.
Unfortunately, as explained above, the multistage nature of acquiring nuclear weapons is not something that survival models can easily measure, making it necessary to run independent survival models for each stage of the process. Given that a cohesive selection survival model is not available, the survival models are used as a robustness check rather than as the main analytic tool. Building on prior survival analysis on the spread of nuclear weapons by Singh and Way, the appropriate description of the underlying hazard function for the spread of nuclear weapons is a parametric Weibull distribution. One difference in the dependent variables for survival models is that they only register movement when something changes. In other setups, the United States is coded as a nuclear power beginning in 1945, meaning that the United States would register a 1 for a nuclear weapons variable in 1945 and then again in 1946, and so on. In the survival setup, it is only during the year of acquisition that the variable changes. So the United States is scored a 1 in 1945 but not in 1946.
Table A.2 imports the three-tiered Singh-Way coding scheme for levels of nuclear weapons proliferation. The first level is the initiation of a nuclear weapons program, the second level is the active pursuit of a nuclear weapons capability and the third level is the acquisition of nuclear weapons. The results, with one survival model for each stage, replicate the Singh-Way analysis, but insert the financial intensity variables that adoption-capacity theory predicts
1 They were originally developed for predicting patient behavior in the medical field.