The DOL survey showed that people had been exposed to the TACT campaign and that they said they had changed their behaviors around semi trucks. In order to determine if any true changes in behaviors were occurring around semi trucks, actual driving behaviors around semi trucks were observed. This resulted in the conclusion that actual violation rates were reduced in the intervention sites but not in the comparisons.
Video Data Collection. In order to acquire a sufficiently large sample of vehicle interactions with semi trucks, it was decided to follow semi trucks and videotape the behavior of drivers passing or otherwise interacting with them. The WSP patrol agreed to assign officers to 8-hour shifts of following semi trucks in unmarked cars equipped with video equipment. A protocol was developed to define how the observations were to be made and the judgments desired from the troopers on the scene. WSP troopers were then assigned at each of the four sites and trained to collect data according to the protocol. Video equipment mounted on the dashboard and looking to the front of the unmarked cars recorded data. Collection took place in five waves at each of the intervention and comparison corridor sites. Although each data collection day consisted of eight hours of patrol time, the actual recorded time varied significantly due to external factors such as the availability of semi trucks, compelling needs for the troopers to take enforcement action and equipment malfunctions.
Officers were asked to move in or out of the lane behind the semi truck when possible in order to allow the video cameras to obtain images about vehicle behaviors before, during, and after passing the trucks. This information included signaling in and out of lanes, time for lane changes, and action after passing (e.g., slowing down, accelerating, and space left). Officers were asked to identify any violations around the semi trucks or violations committed by the trucks themselves using the audio channel on the video recorders. Officers looked for speed, improper lane change, failure to signal, following too closely, reckless driving and negligent driving 2 nd degree violations. Officers were asked to verbalize what action they would take regarding the vehicle’s actions so it would be caught on the audio portion of the tape, although they were asked not to stop vehicles unless absolutely necessary. The three possible actions an officer might take when there was a violation included: 1) no stop, 2) driver would be stopped and warned, and 3) driver would be ticketed. The measure of interest was whether the rate of violations (number of violations per unit of observation time) decreased. Rates had to be used rather than raw numbers of violations since the amount of observation time per wave at each site varied due to the extraneous factors mentioned above.
Video Coding. Video data were sent by the WSP to the WTSC where they were copied for protection and then forwarded to Dunlap and Associates for coding and analysis.
A total of approximately 160 hours of video were collected (8 hours x 4 sites x 5 waves). Video violation sequences were then extracted from the video based on the visual and auditory information provided by the troopers. Overall, 1,843 interactions were coded from the video data. The total number of interactions coded includes all instances when an officer indicated a violation in the immediate vicinity of a semi truck, or when a vehicle and a semi truck’s paths crossed. If vehicle and truck paths crossed without a violation, the event was recorded as a “no violation” interaction. Instances where vehicles simply drove by, did not commit a violation and did not cross paths with the semi truck were not coded.
For each relevant sequence, an array of information was coded based on trooper comments, environmental and weather conditions, vehicle type, and observed vehicle behaviors including signaling in or out of lanes, position relative to a semi truck, and the number of highway lanes. Although 1,843 interactions were coded, an interaction was only classified as a violation based on the comments of the troopers.
Violation Rate Results. Several minor issues arose involving the quality of video data. The audio in all of Wave 4 for Lacey was lost. Therefore, the coder made judgments as to the nature of any indicated violations (officers gave a visible signal to the camera when a violation occurred). Also, one and a half tapes of data were lost in Wave 3 for Lacey due to an equipment malfunction. To address these issues, data were normalized among the four sites by calculating a violation rate per observation hour determined by dividing the number of observed violations at each site by the number of video recording hours required to collect them.
As in the analysis of the DOL survey, site data were combined into intervention and comparison condition and by pre and post periods. Violation rates per hour of observation were used to test for changes before and after the TACT program and between intervention and comparison sites using the Poisson log linear offset procedure described in Agresti (2002) and as implemented in SPSS® Version 13 (2004). This technique weighs and adjusts the observed counts for differences in exposure.4 In the present study, exposure is represented by the amount of time (hours) over which each cell count (number of violations for a wave and site) was observed. The analysis addresses whether or not the enforcement program was associated with change in the rate of violations.
The analysis is similar in concept to an analysis of variance with the intervention factors of site type (intervention/comparison) and period (pre-TACT/post-TACT). The effects of most interest were whether there was a significant reduction in violation rate pre to post and whether that reduction was significantly more at the intervention sites, i.e., the interaction effect of site type by period was statistically significant. The results of the analysis demonstrated a significant interaction effect (regression coefficient = 0.615, p = 0.002) in the predicted direction.5 Calculating the odds ratio (OR = 1.85) indicates that the comparison group had 1.85 times as many violations per hour than did the intervention group when controlled for their respective rates in the pre-period. Using the reciprocal of the OR, which is an alternative way of describing the findings, shows that there was an approximately 46 percent reduction in violations for the intervention sites when controlled for the pre period rates.
Given that the effect is significant, a simpler way to examine it is by looking at the calculated rates of violations for the pre and post periods for the intervention and comparison sites. As seen in Figure 18, the rate of violations for the intervention sites decreases from 5.8 violations per hour of observation for the pre period to 3.05 violations per hour of observation for the post period. Dramatic drops in the violation rate are seen between Waves 1 and 2 and again between Waves 3 and 4. The intervention sites then show a small increase in violation rate between Waves 4 and 5. Comparison site violation rates stay virtually the same from pre to post with rates of 4.03 and 3.92 respectively.
In order to bound the magnitude of the violation rate reduction observed after the TACT intervention, a second analysis was conducted excluding the pre period data and looking at intervention and comparison sites across each of the remaining post period waves. The same Poisson log linear offset procedure described earlier was used for the wave 2, 3, 4 and 5 data. As expected, the analysis resulted in a smaller but highly significant treatment effect (regression coefficient = -0.262, p = 0.01). The difference between intervention and comparison sites was significant, but none of the differences among the waves reached significance. The odds ratio (OR = 1.30) indicates that the comparison sites had approximately 1.30 times as many violations in the final four waves (the post period waves) as the intervention sites.
Using the reciprocal of the OR indicates a 23 percent reduction in violations for the intervention sites. This result is conservative since it does not take into account the difference in violations at baseline. However, the results including the baseline presented above based on the treatment by period interaction could possibly be inflated somewhat since there is only one wave for the pre-period and there are four for the post. The best estimate is therefore that the actual violation rate reduction lies somewhere in the interval 23 percent to 46 percent.
Summary of Violations Results. Overall, there was a decrease in the rate of violations per observation hour at the intervention sites but not at the comparisons. Since the observation protocol was followed consistently between intervention and comparison sites and for the pre and post periods, these results suggest there was a significant reduction in the actual number of violations at the intervention sites but not at the comparisons.
4 The Poisson log linear procedure is a complex statistical approach that will not be described here. The approach is based on the assumption that the distribution of the counts and rates among the cells follows a Poisson distribution. Although usually a reasonable approximation, this assumption is often not completely valid in analyzing count data; instead one frequently encounters moderate “over-dispersion” in which the variance exceeds the mean. A sensitivity analysis was therefore performed using various levels of assumed over-dispersion. It was found that even if the actual variances were twice as large as those of the Poisson distribution, the program effects observed here would still be highly significant (P<.01). In simple terms, this means that any threats to the validity of employing this approach are extremely small.
5 The results for period and site are not of major interest in the context of the present study but they must be included in the model in order to calculate the interaction. The coefficient for group represents the difference between the intervention and comparison areas on the pre- and post-period rates combined. The coefficient for period represents the difference between the pre and post period for intervention and comparison areas combined. The component of primary interest is the site by period interaction. This interaction reflects the impact of the enforcement program on subsequent violation rates by examining changes in the pre vs. post rates between the intervention and comparison groups.