Estimation of infrastructure distress initiation and progression models from condition surveys

Infrastructure distress models predict the initiation and progression of distress on a facility over time as a function of age, design characteristics, environmental factors etc. Examples of facility distress include cracking, rutting etc. Facility condition survey datasets typically include a large number of structural zeros indicating absence of distress at the time of observation. Most distress progression models in the literature are simple regression models that are estimated using the sample of observations for which distress has been initiated. These models are statistically erroneous because they suffer from selectivity bias due to the non-random nature of the estimation sample used. In this paper, we apply an econometric method to estimate joint discrete-continuous models of infrastructure distress initiation and progression while correcting for selectivity bias. An empirical case study demonstrates this method for the case of highway pavement cracking models. It is shown that selectivity bias can be a very serious problem in such models.