A Method for Minimizing the Impact of Distributional Assumptions in Econometric Models for Duration Data

Transitions in living arrangements of Canadian seniors: Findings from the NPHS longitudinal data

“…See, for instance, Sarma and Simpson (2007), Börsch-Supan (1990), Börsch-Supan et al (1992), Pezzin et al (1995) and Hoerger et al (1996). The time-invariant individual effect accounts for unobserved heterogeneity (Heckman and Singer, 1984) in the transition rates not accounted for by the included covariates and assumed to come from a discrete probability distribution with a small number of mass points m .…”

Utilization of public health centres in Portugal: effect of time costs and other determinants. Finite mixture models applied to truncated samples

“…To specify the model to accommodate unobserved heterogeneity, a continuous or finite mixture model specification is required, and instead of specify the distribution of ðy i jx i Þ, the specification of the distribution of ðy i jx i ; u i Þ is required [36] along with the specification of distributional assumptions regarding the random variable u i , which represents the unobserved heterogeneity. If the random variable u i is assumed to be continuous, then a continuous mixture model is specified, conversely, when u i is assumed to be discrete then a semi-parametric finite mixture model (FMM) is specified [35,41]. The finite mixture approach has been used by Deb and Trivedi [35,42] [41] show that estimates of a finite mixture might provide good numerical approximations even when the distribution of u i is continuous; second, provides a natural and intuitively attractive representation of unobserved heterogeneity in a finite, usually small, number of latent classes, each of which may be regarded as a 'type' or 'group [35]'.…”

“…If the random variable u i is assumed to be continuous, then a continuous mixture model is specified, conversely, when u i is assumed to be discrete then a semi-parametric finite mixture model (FMM) is specified [35,41]. The finite mixture approach has been used by Deb and Trivedi [35,42] [41] show that estimates of a finite mixture might provide good numerical approximations even when the distribution of u i is continuous; second, provides a natural and intuitively attractive representation of unobserved heterogeneity in a finite, usually small, number of latent classes, each of which may be regarded as a 'type' or 'group [35]'. Moreover, empirical investigation has shown that FMM provide good results [35,36,42,43].…”

Bias properties of Bayesian statistics in finite mixture of negative binomial regression models in crash data analysis

“…First, it can effectively account for unobserved heterogeneity without imposing strong distributional assumptions on the mixing variable (for example, a continuous Gamma distribution assumed in NB model). Laird (1978) and Heckman and Singer (1984) showed that finite mixture models can provide good numerical approximations even if the underlying mixing distribution is continuous (Cameron and Trivedi, 1997). Second, it allows the vector of regression coefficients to vary from component to component.…”