This paper analyses and compares alternative approaches  for analysing event history data when the dynamic features of the data generating process are the central focus, as is the case in the model of completed fertility considered here. Event history data in their raw form are generated by a non-Poisson point process operating at irregular intervals. Two leading alternatives are the autocorrelation conditional duration (ACD) model and the modified dynamic event history (MDEH) model. ACD models analyse inter-event intervals of time using methods of survival analysis while allowing for serial interdependence between observations. Our proposed MDEH variant reformats the irregularly spaced event history data into a panel data format and applies the dynamic discrete hazards panel data methodology.  However, the dynamics are modelled partly thorough lagged dependent variables and partly also through qualitative features of past birth events. The paper models fertility data from the British Household Panel Study that began in 1991 and ended in 2008.  The paper provides a detailed empirical analysis of the determinants of birth and birth-spacing and the role of drivers high-lighted in human capital theory literature. Several specifications of a dynamic panel logit model are estimated and compared with a flexible variant of the logarithmic autoregressive conditional duration (Log-ACD) model.  The framework allows us to evaluate separate contributions of state dependence, serial correlation and birth history on birth outcomes

Alternative models for the analysis of dynamic event histories (with Alfonso Miranda)

Fri 8 Apr 2016 3:30pm5:00pm


Room 103, Colin Clark Building (#39)