• Federal Highway Administration >
• Publications >
• Research Publications >
• 05048 >
• 9.Cfm >
• Safety Evaluation of Red-Light Cameras

Safety Evaluation of Red-Light Cameras

PDF files can be viewed with the Acrobat® Reader®

IX. Data Collection

Following the approval of the basic research design described earlier, the project team initiated data collection in each of the seven jurisdictions. The data items listed in table 14 were sought in each jurisdiction. Team members visited each site and collected available raw data. They then coded the data into computerized analysis files based on protocols and data formats developed earlier.

Crash Data

Crash data for 3 or more years before RLC installation was sought for each treatment, reference, and control intersection in each city. Preliminary telephone interviews with each of the seven jurisdictions indicated that historic crash data were available in most cases; however, discussions held during the site visits indicated that while there were some computerized historical crash data available in some cities, most did not retain computerized (or raw) data for the needed before-treatment periods. The only exception was Charlotte, NC. There, staff could provide 1997 through March 2002 data in Microsoft^®Access files from their current computer system. Earlier 1994-1996 data were not stored in the current system. Working with Charlotte staff, a computer analyst who had worked with the older files was located, and she was able to retrieve the old files and covert them to Access files for project use.

Neither the three jurisdictions in Maryland nor the three in California could provide adequate historic crash data because they either did not store multiple years of older data or were restricted as to what could be released. The California State Highway Patrol (CHP) is the repository of all police crash reports for the entire State. The project team requested and received computerized data from the CHP on all crashes for the three California jurisdictions for the years 1992-2002. Variables extracted for use in this analysis are described in the "Results" section.

The same data deficiency was found in the three Maryland jurisdictions. As noted earlier, the team had learned that the Maryland CODES team retained history crash data for all Maryland crashes. Project staff requested and received a computerized file of all crashes occurring in Howard and Montgomery Counties and Baltimore City for the period 1994-2001. Later, the additional data for 2002 were requested and received.

Intersection Inventory and Volume Data

As indicated, all available raw intersection geometric, signalization, and traffic flow data listed in table 10 were collected from the individual jurisdictions and coded to analysis files by the project team. During the early part of the data collection phase, it became apparent that the effort required and the cost of the data collection was going to greatly exceed the original estimate. This collection and coding required both an onsite visit to each jurisdiction where the project team met with transportation staff and collected the raw data and subsequent extraction and coding of the data into a computerized analysis file. A small part of the increased effort resulted from the addition of a limited number of required variables at the end of the experimental plan effort; however, most of the increase resulted because in most cities, little or none of the data except for the crash data are computerized. Thus, in addition to finding and copying (and later coding) paper files, the team often needed to manually extract data from Computer Aided Detector Design (CADD) files, intersection drawings, and aerial photographs held in different offices.

As the result of internal research team discussions and conversations with the FHWA task order manager, steps were taken to both minimize data collection costs and to examine alternative data collection levels. It was decided that the decision concerning final data collection levels for treatment, signalized reference, and unsignalized control sites would be based on the development of safety performance functions using the data from two cities where full coding was done. The SPFs would be developed with various combinations of variables to determine the minimum set of data items required.

The project team developed these SPFs based on data from El Cajon, CA, and Howard County, MD, analyzed the data, and developed recommendations for FHWA review. (A copy of the internal memo describing this effort is available from the project team or FHWA.)

The findings indicated that the project would require what amounted to full coding of all traffic signal data, coding for all available traffic volume data (which is limited in some jurisdictions), and almost full coding of the intersection geometry data. More specifically, FHWA agreed with the project team that the final analysis files would include the following information:

• All pertinent crash data linked to the appropriate treatment, signalized reference, and unsignalized reference intersections.
• For RLC (treated) intersections, full coding of all non-crash data including coding of signal data (signal timing and changes over the project period), signal data related to left-turn protection, and data related to actuated versus fixed cycle length. Where available, data on light emitting diode (LED) and backplate presence were coded. All data related to the intersection geometry shown in table 14 were coded, including lane and median measurements and speed limits where available. All available traffic volumes were coded and converted to AADTs for each approach. Information on the location of RLC warning signs was also coded.
• For signalized reference intersections, modified coding of all non-crash data. This included all data noted in the paragraph previous for LED and backplate data, and lane and median widths. In addition, the signal timing data to be coded included only indicator variables for left-turn protection and whether a signal is actuated or always works under a fixed cycle length.
• For unsignalized control intersections, modified coding of all non-crash data. This includes volume data as noted in the second paragraph. The geometry data includes number of approaches, number of approaching lanes on each approach (without turn or through designation), and which approaches are stop-sign controlled.

It is noted that the sources of different intersection inventory variables differed from jurisdiction to jurisdiction. In most cases, the basic data source was a paper file; however, in some cases, one or more of the jurisdictions did not have paper files, and alternative sources were found and used. For example, in Baltimore, MD, electronic CADD drawings were not available for any intersections, and hard-copy drawings were only available for a limited number of intersections (i.e., less than 20 percent). There, aerial photographs were located in the city's geographic information system (GIS) office, and project staff coded intersection geometrics from those photographs to the extent possible. Appendix A provides information on the basic source of data for each of the key variables in each jurisdiction.

As indicated, annual traffic volume data for each approach at all intersections is important in the development of the SPFs and the subsequent analyses. As was the case with other inventory data, the full array of needed volume data did not exist in data files in the jurisdiction. Unlike State systems where traffic volumes (i.e., AADT) for each section of roadway are generally updated each year through a series of counts and estimating procedures, only limited volume data were found in city and county files. While there were a limited number of midblock traffic counts at set locations in some jurisdictions, most traffic data were collected in the form of intersection turning-movement counts, and done as-needed. Thus, the available counts were often either multiple-hour turning movements (e.g., 4-hour or 11-hour counts), or such counts had been converted by the jurisdiction to average weekday daily traffic (AWDT) counts.

Thus, it became clear early in the data collection effort that estimates of annual approach volumes would need to be developed by the project team from the available count information. The types of available data, and thus the gaps in annual approach counts, varied from jurisdiction to jurisdiction. In all cases, the requirement was that there be at least one count for all approaches in either the before- or after-period, with the goal being to have at least one full count in each period.

Following is a list of the methods used to develop AADTs and fill in gaps:

• In limited cases, missing counts for city or county intersections were obtained from the State Department of Transportation (if the road was a State-system highway), or from local metropolitan planning organizations (MPOs).
• When hourly turning movement counts or AWDTs were available, they were converted into AADTs using factors either supplied by the jurisdiction (often on the count record) or factors developed by the project team based on information provided by the jurisdiction or the State Department of Transportation for that jurisdiction.
• When only one traffic count was available for the before- or after-period, that AADT was assigned to all years in that before-or after-period. (We chose not to use estimated growth factors between years because none were provided by the jurisdictions, and the project team felt that the presence of multiple intersections with counts in different years would allow the development of sufficient SPFs for the analysis.)
• When counts in differing years in either the before- or after-period existed for a given approach, the approach counts were averaged and that average was used for all years in either the before- or after-period.
• When only one count was available for an intersection, the AADT developed from it was assigned to all years in the before-and after-periods. This occurred in less than 10 percent of the total intersections in the study.
• When an approach count existed for one approach but not the opposing one (usually for the minor roads), the opposing approach was made equal to the existing count.
• When a count of entering vehicles for the full intersection was found, but not individual approach counts, data from other years when approach-specific data were available at the same intersection were used to calculate a percentage at each approach, and the entering-vehicle counts were distributed using those percentages. (The project team analyzed one limited set of intersections where both entering vehicles and approach-specific counts were present for different years. The individual approach counts were predicted based on proportions from another year, and then compared to actual counts. There was fairly good agreement, with approach count errors usually between 5 percent and 20 percent. The larger percentage errors were usually associated with low counts, and thus did not represent large differences in frequencies. Given this level of agreement, the fact that most large errors seem to be for small counts, and the fact that the only alternative was to delete these intersections without suitable replacements, the project team felt this method was suitable.)

The AADT data ultimately developed and used in these analyses cannot be considered as accurate as what might be found in some State files. All available sources for additional data were explored; however, the project team considered the AADT data to be sufficient for use.

Previous | Table of Contents | Next