The next specification includes controls for TIME_AT_SCENE and TTME_TO_HOSP. The TIMEJTO_HOSP variable can be thought of as a control for the distance from the patient to the hospital, although we show later that the hospital allocation (and thus expected travel time) arc conditioned on the patient’s severity. The TIME_AT_SCENE is more difficult to interpret.
It might represent the extra time required to administer treatments which are only available on some ambulances, in which case longer TIME_AT_SCENE should be associated with longer TIME_TO_SCENE, since we expect a longer wait for the scarce resource of a better ambulance. It might also represent some features of the patient’s location, such as the presence of elevators or stairs in a high-rise building. High-rises might be located closer to hospitals. However, when the largest counties are excluded, there are probably fewer high-rises in the dataset. further
The last specification considers only counties who changed their 911 system during the year. Since a fixed effect is included for each county, the coefficients on the 911 dummies can be interpreted as differences in the mean response time as a result of the change. Of course, all time-invariant variables are dropped from this regression, and in addition several other control variables were dropped due to the small number of observations. Since an alternative explanation for any findings in the first three specifications is that unobserved differences in counties drive the results, our findings for within-county changes are particularly interesting despite the limited size of the dataset which considers such changes.
Consider now the results of our analysis. The first result is that TIME_TO_SCENE is lower in counties with no 911 or basic 911 than counties with E911. In the base specification, counties with no 911 are about 10% slower than counties with E911, while counties with Basic are approximately 8% slower than counties with E911. The magnitudes vary somewhat in different specifications, and the result for no 911 is not always significantly different from zero. Nonetheless, the signs of the coefficients are robust to a variety of specifications. When interpreting these results, it is of course important to observe the caveat that results may be driven by unobserved differences between counties, such as the distribution of residences relative to hospitals. However, as shown earlier in Figure 3, many adjacent counties in similar geographical areas have different 911 systems, and further, when the four largest counties are excluded, the counties are fairly comparable in terms of demographics. Of course, controls are included for several important demographic variables as well as the number of hospitals per mile in the county (which decreases response time, as expected).