Time and causality: A Monte Carlo assessment of the timing-of-events approach

Abstract

We present new Monte Carlo evidence regarding the feasibility of separating causality from selection within non-experimental duration data, by means of the non-parametric maximum likelihood estimator (NPMLE). Key findings are: (i) the NPMLE is extremely reliable, and it accurately separates the causal effects of treatment and duration dependence from sorting effects, almost regardless of the true unobserved heterogeneity distribution; (ii) the NPMLE is normally distributed, and standard errors can be computed directly from the optimally selected model; and (iii) unjustified restrictions on the heterogeneity distribution, e.g., in terms of a pre-specified number of support points, may cause substantial bias. r

Overeducation at the start of the career: Stepping stone or trap?

“…We maximize this log-likelihood according to the procedure described in Gaure et al (2007). In particular, we increase the number of points of support until the likelihood function does not show any improvement and subsequently select the model that minimizes the Akaike Information Criterion (AIC) to reduce the risk of bias induced by an over-parameterized model.…”

“…Even if the time grouping of the data is not problematic, Monte Carlo analysis of Gaure et al (2007) has shown that that it is important to take, as we do (see Appendix B), this time grouping explicitly into account. However, if one follows this advice the aforementioned Monte Carlo analysis has shown that Abbring and van den Berg (2003)'s method is extremely reliable.…”

“…The aforementioned Monte Carlo analysis of Gaure et al (2007) has shown that this assumption is crucial, at least if only time constant explanatory variables are available. 9 We are particularly concerned that the MPH assumption is not satisfied across levels of educational attainment and that this therefore may be a source of bias.…”

The effect of temporary in-work support on employment retention: Evidence from a field experiment

“…Gaure et al (2007) provide extensive Monte Carlo evidence that the parameters of the underlying continuous time model can be recovered using discrete data, so long as the likelihood function reflects the discrete nature of the available data. This allows exit rates to vary flexibly over the duration of the spell.…”

“…16 We were able to estimate the model with up to three points of support. Following (Gaure et al, 2007), it is common to use the Akaike Information Criterion (AIC) to help inform the choice of M. Of the three specifications, M = 3 minimised the AIC and was therefore adopted as the preferred specification. 17 Calculated as exp(0.092)-1=0.0957.…”