Robustness tests. Confidence in a structural model, such as the one developed here, would be enhanced if the model’s main results can be shown to be robust with respect to specification and variable definitions. The simplest way to do this is to compare it to an alternative model of the demand for borrowed reserves. The only model that exists of discount window borrowing in the interwar years is the one developed by Xxxxx Xxxxxxxx, reported in his 1991 book and in articles that preceded that book. In that model, which was estimated on time-series data and is based on the well-known model of Xxxxxxxx and Xxxx (1966), the change in borrowing, ∆B, is a function of the spread between the commercial paper rate and the discount rate, Spreadi,t = RCPt – Rd i,t changes in unborrowed reserves as a proxy for banks’ reserve need, ∆RN, lagged borrowings, and changes in bank debits, ∆Debits, to capture the influence of economic activity on the demand for borrowed reserves, and a dummy variable added to capture changes in borrowing during the Great Depression years. To adapt this model for estimation with panel data and facilitate comparisons to the estimates reported in Table 4, borrowing and bank debits were scaled by bank capital, seasonal dummies were added, and the dummy variable for the Great Depression years in the regressions reported by Xxxxxxxx was omitted to allow parameter shifts and fixed effects to pick up any change in structure during those years. So modified, the alternative model is:
Robustness tests. Although the DID analyses allow me to be relatively confident that experience is driving support for electoral reform, there are two remaining issues: 1) constituency-level preferences might outweigh regional-level preferences about party performance and 2) placebo tests are necessary to increase confidence that the DID analyses are valid.
