Revisiting the notion of induced traffic through a matched-pairs study

In investigating the question of the existence of "induced demand" in connection with highway expansion projects, Hansen et al. (1993) studied eighteen California state highway segments whose capacities had been improved in the early 1970s. For the present study, these segments were paired with control segments that matched the improved segments to unimproved ones with regard to facility type, region, approximate size, and initial volumes and congestion levels. Taking annual data for average daily traffic (ADT) and design-hour-traffic-to-capacity (DTC) ratios during the 21 years 1976–1996, three approaches were used to compare growth rates between the improved and unimproved segments: overall growth comparisons for the matched pairs, repeated measures analysis, and analysis of matched mean profiles. We found the growth rates between the two types of segments to be statistically and practically indistinguishable, suggesting that the capacity expansions, in and of themselves, had a negligible effect on traffic growth over the period studied. Reasons for the differences between these results and those of aggregate cross-sectional models finding a significant induced demand effect are discussed. Our analyses suggest that the aggregate models may overestimate induced traffic due to the attribution of at least a fraction of the observed traffic growth to "induced demand" rather than to some of the confounding factors which were not controlled for in such studies. At the same time, it is noted that the traffic induced by capacity expansion may in certain circumstances be larger than that observed in the present study, with the effect of new highway construction on traffic growth being a prime candidate for scrutiny in this regard. The results of this study nonetheless suggest that, for existing facilities, the size of the induced-traffic effect that can be attributed to capacity enhancements may be sufficiently small that its detection in a case-control study would be difficult, if not impossible, without a substantially larger sample size.