In pairwise randomized experiments, what if the outcomes of some units are missing? One solution is to delete missing units (the unitwise deletion estimator, UDE). If attrition is nonignorable, however, the UDE is biased. Instead, scholars might employ the pairwise deletion estimator (PDE), which deletes the pairmates of missing units as well. This study proves that the PDE can be biased but more efficient than the UDE and, surprisingly, the conventional variance estimator of the PDE is unbiased in a super-population. I also propose a new variance estimator for the UDE and argue that it is easier to interpret the PDE as a causal effect than the UDE. To conclude, I recommend the PDE rather than the UDE.

In a democracy, it is essential that agents respond to their principals. Because the principals have some control over the agents, the agents are supposed to be loyal to the principals’ wishes. For example, in the context of an election, voters (principals) can hire and fire their representatives (agents), while lawmakers should represent their constituency and legislate accordingly. On this basis, power is delegated from principals to agents and the chain of delegation makes government work, from voters (principals) to legislators (agents), from legislators (now, principals) to ministers (agents/principals), to bureaucrats (agents/principals), and to street-level public servants (agents).

This article reconsiders how to judge judicial independence by using the Japanese judicature, one of the allegedly-most dependent judiciary branches. In their influential work, Ramseyer and Rasmusen (2003) argue that judges who once belonged to a leftist group take longer to reach a ‘moderately prestigious status’ under the long-term conservative rule of Japan. Their method does not, however, deal appropriately with the possibility of judges not reaching this position because the judge dies, retires early, or is still at the early stage of her career. Ramseyer and Rasmusen also mistakenly assume that all judges will eventually obtain this position. This article develops a survival analysis model of judicial careers and attempts to solve the problems of censoring, left truncation, and split population. We also offer a way to utilize a matching procedure to estimate average treatment effects on censored time-to-event as well as event occurrence. We re-analyze a corrected version of Ramseyer and Rasmusen's data using their and our methods. One of the most important findings is that, contrary to what Ramseyer and Rasmusen argue, leftist judges are not discriminated against in terms of the timing of promotion.

Weak electoral registration requirements are commonly thought to encourage electoral participation, but may also promote electoral fraud. As one possibility, candidates and their supporters can more easily mobilize voters who do not reside within the district to register there fraudulently and vote for that district's candidates. We statistically detect this classic type of electoral fraud for the first time, by taking advantage of a natural experimental setting in Japanese municipal elections. We argue that whether or not a municipal election was held in April 2003 can be regarded as an “as-if” randomly assigned treatment. A differences-in-differences analysis of municipality–month panel data shows that the increase in the new population just prior to April 2003 is significantly larger in treatment municipalities (with an election) than in control ones (without an election). The estimated effects are decisive enough to change the electoral results when the election is competitive. We argue that our approach—“election timing as treatment”—can be applied to investigate not only this type of electoral fraud but also electoral connections in other countries.

Legislative scholars have debated what factors (e.g. divided government) account for the number of important laws a legislative body passes per year. This paper presents a monopoly model for explaining legislative production. It assumes that a legislature adjusts its law production so as to maximize its utility. The model predicts that socio-economic and political changes increase the marginal benefit of law production, whereas low negotiation costs and ample legislative resources decrease the marginal cost of law production. The model is tested in two ways. The first approach compares the legislatures of 42 developed and developing countries. The second analyzes Japanese lawmaking from 1949 to 1990, using an appropriate method for event count time series data. Both empirical investigations support the model's predictions for legislative production.