Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
74 The current study established associations between the independent variables and the outcome (traffic fatalities) but not causal relationships. A series of mechanisms were sug- gested (see Table 8-1) linking the factors to their expected effects. Many of the findings were consistent with those mechanisms, but a truly causal linkage would require a different approach. Moreover, this was an observational, not experimental, study. Thus, there was no control over the explanatory factors to directly test their effect on the outcome (traffic fatalities). The âexperimentâ was, in effect, the Great Recession, which drastically affected certain of the transient variables (unemployment rates, GDP, household income), but had much less effect on more stable factors not directly connected to the economy (belt-use rates, DUI laws). Most of the parameters used in the models were surrogates for factors that could not be directly measured. The general analytical explanatory framework was that a variety of economic pressures changed the distribution of who drives, how much they drive, and where they drive. However, there was relatively little data that bear directly on each of those points. For example, the unemployment rate, particularly for teens and young adults, was substantially associated with the decline in traffic fatalities. It was plausibly suggested that the reduction in employment caused a decline in discretionary and leisure travel by teens and young adults, but there are no comprehensive data showing that occurred over the period. There is evidence from the NHTS showing a substantial drop in self-reported travel by younger drivers, including teens, in 2009. But in the period covered, only two snapshots were available: one in 2001 at the beginning of the period and the other in 2009, which was within the recession period. In addition, it is likely that the decline in median household income constrained the driving of lower-income groups, but there is no direct evidence. However, see Maheshri and Winston (2015) who used insurance data on a sample of drivers from Ohio, showing an association between aggregate unemployment rates and reduction of VMT of individuals in their sample. There is a need for much more granular exposure data, for example, data series on VMT by driver age or VMT by household income. The data showed significant variability across states, and the independent variables in the models did not capture all the variation between states. There are clearly other variables reflective of state differences that are not in the models. The fact that many of the parameter coefficients were stable between the MNCS and MCS models shows that the operation of those variables was stable across states. However, it is clear that some state-to-state differences were not captured. Another source of variability that complicated results was the sheer variability in size across states. The object of analysis was traffic fatalities, for which the counts ranged per year C H A P T E R 9 Limitations
Limitations 75 from 55 to 4,333. Extending the analysis to less-severe crash types would increase data available and reduce relative variability. However, less severe injuries are not as well defined or measured as fatalities. Finally, there was a substantial amount of variability not accounted for in the models. Clearly, there are other factors that affected the number of traffic fatalities not included in the models. The next chapter addresses some areas of future research that may strengthen the results presented here.