Appendix
A
The Congressional Budget Office (CBO) constructed the highway traffic data set for this study from data gathered through an extensive network of electronic sensors embedded in urban and suburban highway travel lanes throughout the state of California.1 The available data describe traffic flow (vehicle counts), lane occupancy rates, and vehicle speeds in every lane at thousands of locations. In the sample that CBO constructed from those data, sensor readings date back at least to 2003, and for many locations, they begin several years earlier. The data can be assembled for periods of as little as five minutes or for much longer intervals. In this study, CBO analyzed 24-hour vehicle totals for Wednesdays, Saturdays, and Sundays, and the distribution of speeds observed over an entire month, on Saturdays and Sundays only, by hour of the day.
CBO collected data on total vehicle flows at 13 representative locations around California; speed data were gathered from 3 of those locations. The locations were chosen to represent traffic conditions in all of California’s major metropolitan areas. They were not chosen randomly, but the data were not inspected before inclusion in the sample.2
The locations selected represent areas of moderate to relatively high (for California) population density. They include locations that are adjacent to rail transit systems as well as those with no nearby rail option. The data exclude locations with the potential for localized congestion, as from lane reduction bottlenecks or merges ahead. The sample includes interstate, U.S., and state highways. Most locations have four lanes in each direction; two locations have three lanes, and two have five lanes. Except where noted, measurements were made on inbound lanes only, where the direction of travel is toward city centers (see Table A-1).
Vehicle Detector Stations in Sample
|
Route |
|
|
Type of |
Rail |
|
(Direction) |
Location |
Comments |
Data |
Option? |
|
|
|
|
|
|
|
|
|
|
|
|
|
Sacramento |
||||
|
I-80 (E)a |
Mace Boulevard, Davis |
Eastern edge of Davis, 9 miles west of Sacramento |
Trips |
No |
|
CA 99 (N) |
47th Avenue, Sacramento |
Adjacent, parallel to Sacramento RT (light rail) 4 miles from city center |
Trips |
Yes |
|
US 50 (W) |
Folsom Boulevard and La Riviera Drive, Sacramento |
Adjacent, parallel to Sacramento RT (light rail) 6 miles from city center |
Trips |
Yes |
|
|
|
|
|
|
|
San Francisco Bay Area |
||||
|
I-680 (N) |
Montevideo Drive, San Ramon (San Ramon Valley) |
Lateral route (neither inbound nor outbound) |
Trips, Speed |
No |
|
CA 24 (W) |
El Nido Ranch Road, Lafayette |
Adjacent, parallel to BART (heavy rail) 7 miles east of Oakland, 11 miles east of San Francisco |
Trips |
Yes |
|
I-880 (N) |
98th Avenue, San Leandro |
Adjacent, parallel to BART (heavy rail) 7 miles south of Oakland, 11 miles from San Francisco |
Trips |
Yes |
|
CA 101 (N) |
Poplar and Peninsula Avenues, San Mateo |
Parallel to CalTrain (commuter rail) 17 miles south of San Francisco (downtown) |
Trips |
Nob |
|
|
|
|
|
|
|
Los Angeles and Orange County |
||||
|
CA 101 (S) |
Barham Boulevard, Universal City and Hollywood |
Adjacent, parallel to L.A. Metro Red Line (heavy rail) 9 miles northwest of downtown Los Angeles |
Trips |
Yes |
|
I-105 (W) |
South Central Avenue, South-Central Los Angeles |
Adjacent, parallel to L.A. Metro Green Line (light rail) 11 miles south of downtown Los Angeles |
Trips |
Yes |
|
I-405 (S) |
Newland Street, Westminster (Orange County) |
Outbound for Long Beach to Los Angeles but inbound for Orange County (Costa Mesa, Irvine, |
Trips, Speed |
No |
|
|
|
and Newport Beach). High-occupancy-vehicle lane not analyzed. |
|
|
|
|
|
|
|
|
|
San Diego County |
||||
|
I-15 (N) |
Scripps Poway Parkway, Poway (North San Diego County) |
Outbound direction, 10 miles north of San Diegoc |
Trips |
No |
|
I-5 (S) |
Lomas Santa Fe Drive, Solana Beach (North San Diego County) |
Parallel to NCTD Coaster (commuter rail), 17 miles north of San Diego (downtown) |
Trips |
Nob |
|
I-8 (W) |
Lake Murray Boulevard, San Diego |
Adjacent, parallel to San Diego Trolley (light rail), 7 miles east of San Diego (downtown) |
Trips, Speed |
Yes |
|
|
|
|
|
|
Source: Congressional Budget Office.
Notes: I= Interstate; CA = California state route; US = U.S. route; E = east; N = north; W = west; S = south; NCTD = North County (San Diego) Transit District; RT = rapid transit; BART = Bay Area Rapid Transit; LA = Los Angeles.
According to statistics published by the American Public Transportation Association (www.apta.com), in the third quarter of 2006, average weekday ridership totals for the transit systems were as follows: BART (heavy rail), 355,400; LA Metro (light rail), 129,000; LA Metro (heavy rail), 125,900; San Diego Trolley (light rail), 107,300; Sacramento Regional Transit (light rail), 54,400; CalTrain (commuter train), 36,200; NCTD (commuter train), 6,300. Those figures are not directly comparable because of differences in track miles, but on the basis of passengers per mile of track, ridership on the two commuter train systems is relatively low: LA Metro (heavy rail), 3,690; BART, 1,330; LA Metro (light rail), 1,110; San Diego Trolley, 1,110; Sacramento Regional Transit, 870; CalTrain, 260; NCTD, 80. Because the freeways adjacent to the two train systems (CalTrain and NCTD) carry about as much traffic as the other freeways in the sample, those figures imply that the two train systems carry a much lower fraction of commuters in those locations compared with the light- and heavy-rail systems.
a. Carries substantial Sierra-bound weekend recreation traffic; reported results exclude that route.
b. CBO analysis treats routes served by commuter train as "no transit" routes: Ridership is lower than on light- and heavy-rail systems because of lower capacity and less frequent service.
c. Inbound I-15 (S) has a notorious traffic bottleneck.
CBO collected traffic volume data for Wednesdays, Saturdays, and Sundays through the end of 2006. The sample includes data for Saturdays and Sundays because their typical traffic volumes differ considerably. In CBO’s highway sample, there is about 10 percent less highway travel on Saturdays and 20 percent less on Sundays than on a typical weekday. Wednesday totals are representative of weekday traffic, with Wednesdays least affected by three-day weekend travel. CBO’s sample excludes other weekdays to avoid needlessly introducing holiday-related variation into the data. In urban areas and outlying suburbs, highway travel demand tends to be relatively high on Fridays and Mondays around three-day holiday weekends and lower on the weekends. Travel also varies seasonally (it is typically highest in summer, lowest in winter). CBO’s analysis accounts for all of those effects on highway travel.
Figure A-1 shows daily vehicle flows since 1999 in the westbound lanes of I-8 in San Diego. The data show traffic rising to a peak of about 100,000 vehicles per weekday in mid-2002 and then gently declining through 2005. That could have been caused by shifts in regional development patterns and in economic conditions at that location. CBO accounts for such influences by including a trend line for each of the 13 locations in the analysis and allowing those lines to curve upward or downward as dictated by the data.3
Daily Traffic Volume, I-8, San Diego, California
Source: Congressional Budget Office based on data from the Freeway Performance Measurement Project, https://pems.eecs.berkeley.edu.
Note: Daily traffic recorded on westbound Interstate 8, Lake Murray Boulevard, San Diego, California.
The analysis also accounts for other factors that could explain differences in volume at different locations and on different days. Many potential factors, such as number of lanes, population density, or proximity to employment centers, are accounted for by including a fixed factor for each location–day. Including those factors in the model allows the effect of gasoline prices on traffic volume to be estimated independently of other factors that also affect the demand for passenger vehicle travel.
The several unusual one-day drops or increases in the figure mostly indicate holidays. The extended drop in late 2005 was caused by a detector outage of several weeks. The outliers are included in the analysis, but are attributed to holidays or the effects of offline detectors (thus, not to gasoline prices) as appropriate.
The analysis expresses traffic volume as a percentage of the average baseline volume (from a period before the study began) at each location. That accounts for the likelihood that a change in weekly average gasoline prices will have a similar percentage effect on daily traffic volumes that week at locations with different amounts of traffic, as opposed to affecting similar numbers of trips at locations carrying different numbers of vehicles. Gasoline prices also are expressed in relative terms, as a percentage change from a baseline price, to allow a given percentage price change to have a consistent effect on traffic volumes at different times.
CBO’s analysis assumes that gasoline prices are independent of local demand for highway travel. Changes in the price of gasoline are determined largely by changes in global supplies of and demand for oil and in the cost of refining that oil into gasoline. In recent years, the price of gasoline in U.S. markets has been substantially influenced by growth in the demand for oil in countries with rapidly developing economies, such as China.4 Regional constraints on supplies can cause gasoline prices to be higher in some areas than in others, but within California, supply constraints do not differ substantially.
CBO analyzed four years of monthly average gasoline prices against the same months’ characteristic vehicle speeds at different locations and times of day. Figure A-2 provides an example, showing vehicle speed data for April 2006 for a northbound section of Interstate 680 near Montevideo Drive in San Ramon, California—one of the three sampled locations. For each location, CBO collected vehicle speed data from January 2003 through December 2006. Figure A-2 shows the median, 5th percentile, and 95th percentile speeds observed at each hour of the day on Saturdays and Sundays in April 2006.
Weekend Speeds on I-680, San Ramon, California, April 2006
Source: Congressional Budget Office based on data from the Freeway Performance Measurement Project, https://pems.eecs.berkeley.edu.
Note: Data come from sensors located on northbound Interstate 680 at Montevideo Drive, San Ramon (in East Bay, San Francisco).
The figure indicates, for example, that between 2 p.m. and 3 p.m., the 5th percentile weekend speed was 68.1 miles per hour (mph). The median speed was 70.7 mph. The 95th percentile speed was 71.4 mph. Tests of hypotheses that drivers react differently to gasoline prices according to their value of time actually are tests for whether prices affect speeds more at the slow end of the distribution than at the median or at the faster end. Support for that interpretation is given with the results.5
Sometimes there is mild traffic congestion on weekends. Figure A-1 shows that speeds peak in the predawn hours and gradually decline through 2 p.m. Another peak occurs at 6 p.m. Other months and other locations exhibit slightly different patterns. The 5th percentile speeds in Figure A-2 provide stronger evidence for congestion: They are as much as 10 mph slower than the median speeds. That pattern could be due to a one-time slowdown (as from an accident, road work, poor weather, special event, or simply a random surge in traffic), or it could originate in recurring congestion that affects only a fraction of vehicles.6
CBO’s analysis accounts for the effects of congestion so that they do not influence the estimated effect of gasoline prices on speeds. The analysis accounts for slowdowns caused by relatively severe congestion and for differences in traffic volume from one month to the next.7 It also accounts for congestion that is related to the time of day by incorporating fixed factors for location and hour into the statistical model. (Those factors also account for differences by location in speed limits and in the physical characteristics of the roadway, such as grade, curvature, distance from ramps, number and width of lanes, and type and condition of pavement.8) Time-of-day variation in traffic speed cannot be attributed to fluctuating gasoline prices, because prices generally do not change much over a single day. However, accounting for time-of-day effects reduces the amount of unexplained variation in the data, thus improving the precision of the analysis.
Seasonal differences in weather, amount of daylight, and weekend recreational travel can affect driving speeds, and inasmuch as they are related to the demand for gasoline, those differences can also affect the price of gasoline. CBO’s analysis accounts for seasonal effects.
Figure A-3 shows the overall structure of the data. It reports median weekend speeds from 2003 to 2006 (for the sake of clarity it shows only one location and two periods: 6 a.m. to 7 a.m. and 1 p.m. to 2 p.m.) and the statewide monthly average gasoline price for all grades and formulations.9 At the I-405 location, median speeds are often 4 mph to 5 mph slower in the afternoon than they are in the early morning. The difference could result from higher traffic volume in the middle of the day. It also could be that traffic enforcement is more rigorous at that time of day.
Median Weekend Speeds on I-405, Orange County, California, and Gasoline Prices
(Miles per hour) (Dollars per gallon)
Sources: Congressional Budget Office based on data from the Freeway Performance Measurement Project, https://pems.eecs.berkeley.edu.
Note: Speeds were recorded at 6 a.m. and 1 p.m. each Saturday and Sunday from 2003 to 2006 by sensors located on southbound Interstate 405 at Newland Street, Westminster, California. Prices are nominal average California retail gasoline prices for all grades and formulations.
Median speeds at the I-405 location appear to have fallen slightly from 2004 to 2006, as gasoline prices were rising. CBO’s analysis estimates the influence of higher gas prices on vehicle speeds, controlling for other possible factors that also could have caused a change in median (and other percentile) vehicle speeds. Those factors are described in Appendix B.
The figure shows that median speeds at the I-405 location dropped sharply in mid-2003 and again in mid-2006. Such a pattern (also found at different times at the other two locations) can be caused by chronic congestion (lasting a month or more), as from road construction. That data pattern also could have been caused by an offline vehicle detector station (in which case imputed data are substituted). In all such cases, CBO flagged the data, thus neutralizing their effect on the analysis. For technical reasons, CBO did not exclude those observations altogether, because doing so would have removed all of the contemporaneous data from the other locations.
The data are provided by the Freeway Performance Measurement Project, a joint effort of the University of California at Berkeley, the California Department of Transportation, California Partners for Advanced Transit and Highways, and Berkeley Transportation Systems. See Freeway Performance Measurement System, https://pems.eecs.berkeley.edu.
In some cases, data from a location were of insufficient quality, so a nearby location was chosen instead. Data quality suffers when electronic detector stations go offline. In those cases, the Freeway Performance Measurement Project imputes values from nearby detectors. For lengthy outages, however, the imputed data are not usable for CBO’s purposes because their day-to-day variance is too low (often zero).
Technically, the analysis fits a second-degree trend line to the data, so that the trend can be a straight line or have the shape of an upward or downward "U." Restricting the trend to be a straight line (constraining it from curving) resulted in very similar conclusions.
See Congressional Budget Office, China’s Growing Demand for Oil and Its Impact on U.S. Petroleum Markets (April 2006).
The results are also consistent with the possibility that drivers of less-fuel-efficient vehicles tend to drive more slowly and are more responsive to increases in gasoline prices. However, such a possibility also would imply that those drivers have lower values of time than do owners of more-fuel-efficient vehicles.
Most months have eight weekend days, so a one-time congestion event lasting 45 minutes would affect observed speeds only up to the 10th percentile (45 minutes out of 8 hours of traffic observed), and only for that time of day.
For the purposes of its analysis, CBO defined congestion as 5th percentile speeds below 55 miles per hour, median speeds of 60 mph or slower, and 95th percentile speeds slower than 65 mph. Between 1 percent and 3 percent of observed speeds are below those thresholds. Of the three locations surveyed, I-405 in Orange County experienced the greatest frequency of temporary slowdowns: Its slowest 5th percentile speed (for all times of day) was below 40 mph in 32 of 48 months, and it was faster than 55 mph only twice. On I-680 in San Ramon there were 7 months below 40 mph, 27 months above 55 mph, and 12 months above 60 mph. In contrast, I-8 in San Diego had only 1 month below 40 mph and only 5 months below 50 mph; it had 41 months above 55 mph and 31 months with 5th percentile speeds above 60 mph.
Fixed effects also would control for differences in the stringency with which speed limits are enforced, if the agency charged with that enforcement, the California Highway Patrol, consistently allocates its enforcement resources on the basis of historic differences in accident rates or other criteria.
The prices are averages of nominal posted prices from a survey of gasoline stations around California. CBO’s analysis adjusts prices for inflation. Prices vary slightly by metropolitan area in this study, but price movements are highly correlated.
|
|
|