A statistical technique for determining ‘hazard functions’, or the probability that an individual will experience an event (for example first employment) within a particular time-period, given that the individual was subject to the risk that the event might occur (in this case, given continuing initial unemployment).
Since the hazard or risk of re-employment is likely to change over time, it is necessary to decide upon the functional form by which the hazard may depend upon time and the different explanatory variables, although there may in fact be little information upon which to base this decision. Cox's ‘proportional hazard model’ is a particular approach to a class of such models which is easily estimated. In the above example, for any two individuals the ratio of their ‘hazard’ in terms of finding employment is considered to be constant at all points in time, hence ‘proportional’. The technique allows for individual observations to be ‘censored’ (the event of interest has not occurred) or ‘uncensored’ (the event of interest has occurred) in producing estimates (see P. D. Allison, Event History Analysis, 1984). See also event-history analysis.