Tales Left Tails Tell: A Natural Experiment Involving the Truncation of Nonignorable Missing Data
Richard L. Hunt, D. Lerner · Academy of Management Proceedings · 2012
Failures are much more common than successes in business. And yet, reliable data on failures is seldom available for analysis, meaning that empirical studies and the supporting theories are often based exclusively on observations involving survivors. Much of management research implicitly assumes that the missing observations substantively resemble the available observations. But, what if the missing data bears little resemblance to the available data? Under these circumstances, the conventional arsenal of parametric and semiparametric corrections may prove to be insubstantial remedies if fundamental distributional assumptions are violated by unobserved non-linear relationships. In order to demonstrate the empirical and theoretical hazards of truncation, we examine all 612 firms, 56,240 permitted projects and 12,778 licensed individuals, spanning the entire twenty-five year history of the Colorado asbestos abatement industry. Using this natural experiment involving the creation of a new industry as a consequence of legislative action, we present empirical evidence to show that the truncation of events preceding an observation window can result in invalid empirical findings, which in turn can lead to radically different conclusions about the efficacy of the underlying theory. Truncation, which is best understood as situations in which observations on both the dependent variable and regressors are missing, is neither new, nor obscure. It is, however, largely ignored outside the research methods community because the consequences have seemed remote. By tracing specific truncation effects through a specific theory -- the theory of entrepreneurial spinoffs – this study constitutes an important advance in describing and quantifying truncation effects using non-simulated data.