This paper examines the problem of estimating the parameters of an underlying linear model using data in which the dependent variable is only observed to fall in a certain interval on a continuous scale, its actual value remaining unobserved. A Least Squares algorithm for attaining the Maximum Likelihood estimator is described, the asymptotic bias of the OLS estimator derived for the normal regressors case and a "moment" estimator presented. A "two-step estimator" based on combining the two approaches is proposed and found to perform well in both an economic illustration and simulation experiments.

Canadian data on strikes between 1946 and 1983 are used to estimate linear regression models for the logarithm of completed duration. A thorough investigation of the influence of the business cycle reveals strong support for the hypothesis that strike durations are countercyclical. The cyclical effect is shown to be robust to both the choice of cyclical variable and the econometric specification, and the magnitude of the effect is quite substantial. Experimentation with different representations of the cycle reveals that it is difficult to improve on a simple formulation involving a single continuous variable.