Using Residential Patterns and Transit
To Decrease Auto Dependence and Costs

by John Holtzclaw

June 1994

Natural Resources Defense Council
71 Stevenson Pl #1825
San Francisco CA 94105

Funded by
California Home Energy Efficiency Rating Systems
1700 Adams Ave #102
Costa Mesa CA 92626

EXECUTIVE SUMMARY

Introduction

This study is a first attempt to measure reductions in automobile usage and personal transportation costs that result from different characteristics of a neighborhood. Its purpose is to support a proposed enhancement to Energy Efficient Mortgage programs, such as those being encouraged by the California Home Energy Efficiency Rating System, Inc. (CHEERS).

Under energy efficient mortgages, the ability to qualify for a mortgage includes a consideration of utility bill savings as well as the direct costs of a mortgage. Utility savings are subtracted from the usual computation of Principal, Interest, Taxes and Insurance to determine qualification. If the characteristics of a neighborhood allow the reliable calculation of transportation costs savings as well, these too could be subtracted from Principal, Interest, Taxes and Insurance when calculating mortgage qualification.

This study evaluates the effects of neighborhood characteristics on motor vehicle usage per household (autos/HH) and total vehicles miles travelled annually per household (VMT/HH). It first defines four neighborhood descriptors that a priori influence personal transportation costs. These are:

• Residential density, or the number of dwelling units per residential land area.
• Transit accessibility. An index of transit accessibility is defined and measured for the neighborhoods under study.
• Neighborhood shopping. An index is developed that defines the ability to perform neighborhood shopping errands with a short walking trip from a home.
• Pedestrian accessibility. Factors that encourage or discourage walking are combined into an index that is quantified for the neighborhoods under study.

In addition, mean household income and household size (people/HH) are evaluated as explanatory variables for transportation costs.

Twenty-eight communities in California, representing a diverse range of variation in the four neighborhood variables being tested, were selected for this analysis. Census data were used to estimate the mean number of autos per household for each community. Analysis of smog check data according to the zip code of auto ownership were used to compute vehicle miles travelled per automobile. Methods described in the text were used to define and evaluate the other neighborhood variables.

Once the characteristics of these neighborhoods were defined, statistical methods were employed to provide the best explanation of the observed values of autos/HH and VMT/HH.

Overview of Results

Using the variables estimated in this study, it is possible to project automobile ownership and usage, and thus average costs, with good reliability. The best statistical correlations we were able to find are based only on density and the transit accessibility index; adding other variables or replacing these variables with others did not explain usage data any better than these two alone. The results obtained were:

Autos/HH = 2.704*(density) -.25	R 2 = 0.85; and
VMT/HH = 34,270*(density) -.25 *TAI -.076	R 2 = 0.83.

Using these two equations, and estimates for annual fixed costs of car ownership and variable costs of an additional mile of driving, allow the calculation of a matrix of average annual household auto expense as a function of density and the transit accessibility index. These are presented in Table 8 of the text.

Application of the Results to Mortgage Qualification

The results described in the equations and in Table 8 are sufficient to allow at least one method of calculating annual transportation cost savings for an individual house or a particular neighborhood compared to a base case of a typical low density suburb. One recipe for doing so could be as follows:

The predicted average annual household transportation cost savings for a particular dwelling unit could be calculated as follows:

• 1) Calculate the average household density (households per residential acre) for the census tract in which the dwelling unit is located by using the enumerated households for the tract, and the acres of residential land measured by the local planning department or regional planning agency.
• 2) Calculate the transit service by identifying each bus line within 1/4 mile walking distance of the dwelling unit and each passenger rail stop or ferry terminal within 1/2 mile of the dwelling unit. For each line within the prescribed distance, calculate the daily number of buses, rail vehicles or ferries on these lines (in both directions) using transit schedules. "Standardize" these vehicles by multiplying the number of vehicles by (# seats on the average transit vehicle)/(50 seats). Divide this by 24 hours per day to get the transit service.
• 3) Look up the average annual household auto costs in Table 8. Values for units with densities or transit service falling between those shown on Table 8 can be calculated by interpolation. Alternatively, the predicted annual auto costs can be calculated using the above equation.
• 4) Subtract the predicted annual auto costs from those for the typical suburban area used as the loan standard.
• 5) Add up the average annual transit costs for all public transit within the city (average annual transit farebox revenues divided by the city's households). Subtract the corresponding "average annual transit costs" for the "typical suburban area" from this. Subtract this transit cost difference from the annual auto savings to get the annual household transportation cost savings.
• 6) Divide the results by 12 months and add to the standard PITI (principal, interest, taxes and insurance) mortgage qualification formula.

Thus, the primary objective of this study has been achieved: providing a first cut at a formula for quantifying the value of location efficiency that can be used in the context of energy efficient mortgages. Several areas for additional research present themselves as a consequence of the results derived in this study.

Discussion

This study confirms and extends the results of a number of previous studies that have suggested household density as the major explanatory variable for variations in vehicle miles travelled, and annual transportation costs. In all of these studies, a community with double the density will have 25-30% less driving per family when the impacts of all the conditions generally accompanying higher density (including better transit, more local shopping, and a more pedestrian-friendly environment) are included.

This study's results are also significant because previous studies focused only on different neighborhoods in a single metropolitan area, whereas this study's results were derived from four different metropolitan areas and one rural county throughout California.

After density, the only other variable that produced statistically significant explanatory power was the transit accessibility index. While all of the other variables, including household income and household size, were statistically significant predictors of observed driving behavior when considered individually, they failed to be significant when the effects of density and transit were considered first. This result may be due to limitations in the sample size. They may also be due to the correlation between the other neighborhood characteristics and the two with the most explanatory power: density and transit accessibility.

The failure of the income variable to be statistically significant is an important departure from previous studies, and suggests further research. It is well established that higher income families drive more, everything else being equal. But this study finds that when everything else isn't equal -- when the characteristics of the neighborhoods in which people live are taken into account -- income fails to provide statistically significant results. This may be in part because none of the neighborhoods studied was extremely poor: one can hypothesize that the inclusion of extreme poverty neighborhoods in the sample would demonstrate an income effect. But for the range of neighborhoods covered by this study, and to which the results are likely to be applied, income variations do not change the predictions of driving.

This is a valuable result for application to energy efficient mortgages, because it suggests the possibility of evaluating transportation savings without regard to the demographic characteristics of a neighborhood.

Summary

Variations in automobile usage per household and in personal transportation costs between communities can be quantified with a good deal of accuracy using two simple equations. These can provide the basis for calculating a first approximation to average transportation cost savings. It appears to be feasible to use this to develop repeatable estimates of transportation cost savings for use in energy efficient mortgages.

TABLE OF CONTENTS

1. Introduction

The Costs of Auto Dependence
The Causes of Auto Dependence
Purpose of This Study
Constraints on This Study

2. Survey of Previous Studies

Density Based Studies
Community Characteristics Promoting Transit and Walking
Walking Distances To Transit

3. The Communities and Their Characteristics

The Communities
Measuring Density, Income, Transit, Shopping and
Pedestrian Access To The Communities
Analysis of Auto Mileage

4. The Results of the Analysis

Transit, Shopping and Pedestrian Indices
Auto Ownership
Vehicle Miles Traveled

5. Predicting Annual Auto Costs

Appendix: Detailed Description of the Communities

References

1. INTRODUCTION

The Costs of Auto Dependence

American families spend a sizable fraction of their income on local transportation, primarily on autos. John Moffet and Peter Miller estimate $775 billion to $930 billion in direct annual auto costs in the United States (Moffet & Miller, 1993). That averages $8,350 to $10,000, or $700 to $835 per month for each household. This averages nearly $11 per commute trip, based on 34% of mileage for commuting (Metropolitan Transportation Commission) and an average of 26 monthly commute roundtrips per household (1990 census).

Driving costs vary systematically between neighborhoods within urban areas. Denser, central areas with good transit and nearby employers and shopping require less driving than sprawling bedroom suburbs with little transit and no nearby shopping or jobs. John Holtzclaw estimated an average annual auto cost per household in dense northeast San Francisco of $4,200 (0.6 autos per household) compared to $17,800 (2.3 autos per household) in suburban Danville-San Ramon, for an annual savings of $13,600, or $1,130 per month, for San Francisco households (Holtzclaw, 1991). These costs are consistent with the Moffet and Miller national averages when Californians' higher driving rates are considered and central city and suburban costs are averaged.

It makes sense to allow families incurring lower transportation costs to apply those savings to mortgage payments. If a family choosing to buy a house and live in the denser area with good transit could apply a $500 per month auto and transit savings to their mortgage payments, that family should be able to qualify for a $50,000 more expensive house at the same combined monthly housing and transportation payments. Implementing such mortgage criteria requires predicting the variation in average local transportation costs, and how they are affected by such measurable neighborhood variables as density, the availability of nearby transit and neighborhood shopping and the safety and appeal of the neighborhood to pedestrians and bicyclists. The calculation of transportation costs includes private autos and light trucks, and local public transit costs.

The Causes of Auto Dependence

Auto use is encouraged by the auto's versatility, speed (as long as congestion is avoided and parking is ample), privacy, comfort and apparent safety. Auto dependence is fostered by subsidies which decrease the marginal or per mile cost of driving; economic policies and social attitudes which encourage moving from denser, convenient central areas to sprawling suburbs; urban renewal and freeway construction; low density settlement patterns and isolation of residential areas from shopping, services and jobs; poor public transit service; and pedestrian- and bicycle-unfriendly residential and shopping areas.

The marginal cost of driving (additional cost per mile) comprises only 14 percent of total direct automobile costs in the first year of ownership, and averages 38 percent (less than 13 cents per mile) over 12 years of ownership according to the Federal Highway Administration (1991). The marginal costs include the fuel, some parking, tolls, maintenance and tires; ownership costs include annual registration and taxes, depreciation, finance costs, insurance and some parking fees.

The low marginal costs encourage driving by decreasing the apparent cost of driving, and, in fact, increased mileage, while raising the total costs, decreases the cost per mile. Many of the hidden subsidies to driving are dependent on the miles driven and would discourage additional driving if their costs were levied directly on auto use, more than doubling the cost per mile. These subsidies include road repair, policing and motorist protection, parking, accidents, noise, vibration damage, pollution damage, global warming, petroleum subsidies, policing the petroleum supply line, and congestion.

Since before World War II, federal housing policies, including those of the Federal Housing Administration (established by the National Housing Act in 1934), Veterans Home Administration and the Federal National Mortgage Association (FNMA), have driven up single-use suburban sprawl by "redeveloping" urban centers, officially or unofficially redlining urban centers, strongly biasing ownership towards single-family homes, encouraging or mandating low density developments, and prohibiting commercial activities in residential areas (Glazer, 1973; Stone, 1973; and National Commission on Urban Problems, 1968 & 1973). Redlining impedes upkeep and rehabilitation of central areas.

Urban renewal removes more affordable housing, usually from the urban center near jobs, neighborhood shopping and transit. Construction of freeways through urban centers and into rural areas removes block-wide swaths homes and businesses from the urban center and opens up rural areas for development.
Lowering residential densities and separating commercial areas from residential areas increase distances between houses, between friends or fellow commuters, and from home to market or restaurant. Increasing trip distances not only increases the length of the trip driven, but also causes more to be driven rather than walked, biked or taken on public transit.

Many suburban areas have development codes and practices which inhibit walking, bicycling and public transit. These include the absence of sidewalks, bus shelters and inviting street furniture, or where buildings are set back behind parking or landscaping and lack attractive fronts. Winding and dead-end streets, cul de sacs, and such pedestrian barriers as freeways, drainage ditches and parking in front of businesses increase trip distances. High traffic speeds and the absence of sidewalks, stop signs and stoplights decrease pedestrian safety and discourage walking or biking. Since most public transit users get to transit by walking or bicycling, conditions which discourage walking or bicycling cut transit patronage.

Purpose of This Study

The purpose of this study is to elaborate the impacts of a community's 1) household density, 2) accessibility to public transit, 3) accessibility to neighborhood shopping, and 4) pedestrian and bicycle accessibility, on the use of private automobiles by the community's residents. This study tests the hypothesis that residents drive fewer miles when they live in communities with higher densities, more transit service, nearby shopping (restaurants, markets, drugstores, etc.) and an attractive, inviting pedestrian environment.

This study tests the proposition that these community characteristics decrease average trip lengths, facilitate transit use and encourage walking and bicycling, thereby decreasing auto use and residents' transportation costs. The neighborhood shopping measure is also a surrogate for job, entertainment, and recreation trips, and visits to relatives and friends.

This study intends to measure the independent impacts of each of these community characteristics on total household driving. If that can be accomplished, the average household auto costs could be calculated for a housing location using these variables.

The calculation might take the following form:

a) Use census tract population density or net household density to look up the average annual household auto expenses in a Density-Auto Expense Table. Census tract density could easily be mapped on a land use map.


b) Calculate the daily average number of standardized buses within 1/4 mile walking distance and the standardized rail cars or ferries within 1/2 mile of the residence. Look up the resulting average annual additional savings or costs in a Transit Access- Auto Expense Table, and subtract from the auto costs derived above.


If the equation relating density and transit to VMT is more complicated than addition, a table of auto costs at each density and transit accessibility could be used.


c) If five restaurants, food stores and drugstores are within 1/4 mile walking distance subtract the shopping accessibility bonus from auto costs derived above.


d) Look up the average annual household auto savings for pedestrian access in the Pedestrian Access-Auto Expense Table and subtract them from the auto costs derived above. These savings or costs are based upon the pedestrian grid completeness, hilliness, sidewalks, building setbacks and traffic controls in the census tract. Census tract pedestrian accessibility could be mapped on a land use map.


e) Add up the average household transit costs for the city and add to the auto costs derived above.


f) Subtract the total annual household transportation cost obtained in a-e from the suburban average to obtain the annual household savings; divide by 12 months; and add to the standard PITI (principal, interest, taxes and insurance) mortgage qualification formula.

The average transportation costs associated with a specific location would be added to the house's mortgage, tax and insurance costs to get the total transportation and mortgage expenses for a house. This provides the basis for allowing a family moving into an area with low auto dependence to qualify for larger housing mortgages based on its transportation cost savings. Further, this study intends to define each of these community characteristics in such a way that it can be used to guide land use and transportation planners in designing urban areas to reduce dependence on cars.

Constraints On The Study

This analysis is complicated by the objective of defining variables which adequately predict the extent of auto dependence, yet are simple to measure and apply to mortgage loan policy and urban development policy, all within urban areas that are complex and uneven.

Consider density. Lower household densities result in fewer households nearby, decreasing the average number of family members and friends within walking, bicycling or transit distance. Lower density neighborhoods are harder to serve by transit and in American cities are usually located farther from job centers. So, residential density is also a surrogate for nearness and for public transit access to concentrations of jobs and shopping. One task is disentangling transit access and local shopping from density. Some high density areas are built right next to low density areas. High density apartments might be surrounded by extensive low density housing, giving a low density average. To accommodate this variability, this study looks at communities of 20,000 people or more in order to reflect the larger community environment. At this scale, density and transit service should reflect accessibility to jobs (comprising one-fifth of total trips and one-third of total auto mileage) and to shopping, services and friends.

However, some architects are designing transit oriented developments (TOD), which are medium density neighborhoods with shopping and pedestrian amenities, outside the developed metropolitan area. How much will the higher density of these areas and perhaps better transit service compensate for the area's isolation from job and perhaps shopping centers? A neighborhood in the San Francisco area and one in the Los Angeles area were selected to test this.
High income may be correlated with more driving, independent of neighborhood characteristics. The influence of household income is corrected for.
This study uses reliable measurements of auto mileage, based on odometer readings taken when automobiles receive biennial California Smog Checks. Unlike travel surveys, these data do not depend on the subjects' willingness or ability to remember and report their driving.

There is one important caveat. The high density evaluated here does not fit the media image of monolithic run-down high-rise projects. The highest density area studied, Nob- Russian-Telegraph Hills-Chinatown-North Beach-Fisherman's Wharf in San Francisco, achieved nearly 120 units per net acre primarily with 3 to 6 story apartment houses. The area is popular and well kept up. Its high housing prices show high demand.

Achieving high density need not mean massive redevelopment. It can be achieved by infilling with 3 to 4 story apartments or condos, over restaurants and markets, constructed on lots left empty, on parking lots, and on other underused land along transit corridors and around transit stations. The pedestrian and commercial activities of these higher density areas enhance their neighborliness, excitement and livability. As these core areas grow and densify, new transit corridors between them could provide additional avenues for infill growth. Neighborhoods farther from these transit corridors remain at low density. This is the traditional urban growth pattern.

2. SURVEY OF PREVIOUS STUDIES

No previous studies have analyzed the separate impacts of this study's four independent variables (density, transit, shopping, pedestrian accessibility) on driving. One cluster of studies has focused on the impact on driving of density, with the other variables co-varying. A second cluster has tried to identify the community characteristics which promote walking and transit use. A third cluster has tried to measure how far people are willing to walk.

Density Based Studies

John Holtzclaw evaluated the reduction in driving in the San Francisco region resulting from higher population and household densities (Holtzclaw, 1991). The analysis indicated that neighborhood businesses and improved transit service co-vary with density. It also estimated fuel, pollutant emissions and auto ownership cost savings.

The study used a novel source of vehicle miles traveled (VMT) data: odometer readings taken during California's mandatory biennial auto emissions (smog check) inspections. This data captured all auto travel, including vacation, so should have been an accurate indication of total vehicular use and how much total VMT could be saved by increasing the density of residential areas. And these measurements eliminate concerns about reporting accuracy, trip length estimation accuracy, data completeness, response rate adequacy, response bias and inhomogeneous analytical areas.

Five communities within the San Francisco region were selected to achieve a wide range of density, including one of the densest communities west of Manhattan.

• Nob Hill to Fishermans' Wharf (northeast San Francisco) - somewhat densely settled, richly served by neighborhood business and transit. [Also in this study.]
• San Francisco (all) - moderate to dense settlement, neighborhood business and transit service. [Also in this study.]
• Rockridge (north Oakland - south Berkeley) - moderate settlement density, neighborhood business and transit service. [Part of this community is in this study.]
• Walnut Creek (suburban) - low to moderate settlement density, neighborhood business and bus service, but with two BART stations. [Part of this community is in this study.]
• Danville - San Ramon (suburban) - low settlement density, neighborhood business and little transit service. [San Ramon is in this study.]

Comparing the extremes, the Nob Hill area was found to have 31 times higher net household density, 26 times higher gross population density and 198 times higher local serving job density than Danville-San Ramon, while only about 1/3 the auto ownership per capita and 1/4 the auto ownership per household. The other communities fell within these extremes, except for Walnut Creek's auto ownership per capita, and varied uniformly in the order presented. Comparing communities showed that the higher the population, household or local serving job density, the lower the VMT, as shown in Table 1. VMT measurements for other large counties and the state were included for comparison.

Table 1 1988 Density and Auto Use
	Net HH Density (hh res acr)	Gross Pop Den (people tot acr)	Gro Loc Ser Job Dens ( rt&srv job tot acre)	Annual VMT cap	Annual VMT hh
Nob-Rus-Chin-N Bea-Tele-Fish Wharf	117	52	83	3,462	7,437
San Francisco	32	23	8.4	5,046	11,848
Rockridge (N. Oak-S. Berkeley	14	10	3.1	7,249	15,707
Walnut Creek	6.8	4.2	1.4	8,434	19,054
Danville-San Ramon	3.8	2.0	0.4	10,248	31,291
Sacramento				8,482
Los Angeles				7,993
Orange				9,687
San Diego				8,486
California				8,635
Net household density = households/residential acre
Gross population density = people/total acre
Gross local serving job density = (retail jobs + service jobs)/total acres
VMT/cap = vehicle miles traveled/capita
				John Holtzclaw

This study showed that as the household, population and commercial densities and transit service decrease the auto ownership rate and vehicle miles traveled (VMT) per capita and per household increase. These VMT variations primarily reflect auto ownership rates since VMT/auto varied less between neighborhoods. Using the Hertz Corporation's estimates of auto ownership and operating costs per mile, the average Nob Hill area family annually spent nearly $14,000 less on autos than the average Danville-San Ramon family. San Francisco families spent $11,000 less, Rockridge families nearly $9,000 and Walnut Creek families nearly $8,000 less than Danville-San Ramon families.

The study found that doubling residential or population density reduced the annual auto mileage per capita or per household by 20 to 30 percent. The study concluded that if the population of an area doubled wholly by infill its VMT would likely increase only 40 to 60 percent, rather than the 100 percent it would increase if the city grew retaining its present density pattern. In contrast, doubling population at low density, halving the average density, would likely increase average auto mileage by 150 to 186 percent. At 30% reduction in VMT as density doubles, VMT varies as the reciprocal of the square root of the household density:

VMT2	=	POP2	(0.7)	D2
				1.4 ln D1
VMT1		POP1

or alternatively,

VMT2/POP2	=	( D1 )	.5
VMT1/POP1		( D2)

VMT		1
POP		D

(equation incomplete)

where: VMT = vehicle miles traveled; POP = population of community; D = density of community; VMT/POP = VMT per capita.

The study also evaluated how effectively public transit reduced driving. San Francisco's higher density and better transit service shortened trip lengths sufficiently to allow one mile on transit to replace eight miles of driving compared to trips in Danville - San Ramon. This savings was attributed to the increased convenience of higher density mixed-use areas. Similarly, the suburban areas of Walnut Creek and Danville-San Ramon, which had similar histories prior to Bart opening 13 years before the study data, were compared. In that time Walnut Creek had developed to over twice the density as Danville- San Ramon, with 3 to 4 times the local serving jobs, resulting in 18 percent less auto travel per capita and 39 percent less auto travel per household. The study found that one mile on transit in Walnut Creek replaced four miles of driving compared to trips in Danville - San Ramon, or 13.6 miles of driving per household.

Holtzclaw's general density-VMT relationship was confirmed by Metropolitan Transportation Commission consultant Greig Harvey of DHS, Inc. He analyzed MTC's 1981 San Francisco Bay Area travel survey. MTC compiled 6200 weekday and 900 weekend responses with a mailed questionnaire and followup phoning. The households surveyed were selected randomly from a reverse directory, with corrections for household size and income.

Mr. Harvey aggregated the survey responses into MTC's 34 superdistricts (175,000 average population) with the results showing the same pattern of per capita VMT with residential density (defined as persons per residential acre) as derived by Holtzclaw. A curve of 30 percent decrease in per capita VMT as density doubles tracked the data well.

The NRDC report reviewed the following eight studies in some detail.

A survey of New York State residents' odometer readings, with telephone follow-up to increase the response rate, found that car owning Manhattan households drove 8,000 miles a year, or 46 percent, less than a comparable household on 10 acres in exurbia (Zupan & Cumella, 1981). But only 20 percent of Manhattan households owned cars.

A New York City energy study (New York City, 1980) showed that the average American consumed 4.1 times as much gasoline as residents of the City, implying a VMT ratio of 4.1:1. They could not measure the fuel consumption in Manhattan alone so they included Manhattan with the rest of New York City. Of course this averages the low fuel consumption of denser Manhattan with the higher consumption of the other four boroughs.

A travel survey in the Greater Toronto Area (University of Toronto/York University, 1989) showed that residents of the dense central area with its excellent transit and wealth of jobs drove or rode in an automobile about 40 percent as much as the residents of the lower density cities bordering Metropolitan Toronto. The survey found that doubling the density results in a decrease in per capita VMT of about 25 percent.

A travel survey in the Chicago area (Boyce, et al, 1981) analyzed morning commuting by concentric rings centered on the loop. The results show that a doubling of density results in a 30 percent reduction in VMT.

Robert Dunphy and Kimberly Fisher analyzed the trips reported in the 1990 National Personal Transportation Survey (NPTS), assigning each household to a density range based on its ZIP code's density. So households in a 10,000 to 49,000 persons per square mile ZIP code in San Francisco were lumped in with similar households in New York, Chicago, Boston, Los Angeles, and elsewhere. This data shows a strong decrease in VMT (Personal Vehicle Miles in Table 5) with increasing density. It seems to show a break at about 6000- 7000 p/sm (10-11 hh/res ac), with a slower decrease in VMT as density increases at lower densities, and a faster decrease at higher densities (maximum was about 200 hh/res ac). At densities above the break, each doubling of density reduced VMT per capita by 40%. Over the urban range of 1500 p/sm (2 hh/res ac) to the maximum, each doubling of density reduced VMT per capita by 28%.

Peter Newman and Jeffrey Kenworthy (1989) surveyed major cities around the world. They found that the residents of the American cities consumed nearly twice as much gasoline per capita as Australians, nearly four times as much as Europeans and ten times that of the three "westernized" Asian cities. The extremely low density urbanized areas of the US and Australia consumed over 6 times as much energy per capita as the very high density areas of Europe and the Far East. The usual concerns about fuel consumption surveys were minimized by including relatively large geographical areas in the analysis.
A travel survey of United Kingdom cities, reported by Newman and Kenworthy (1989), found that as density increases from 8 to 80 persons/hectare the auto travel per person decreases about 64 percent. That is a 25 percent reduction in per capita VMT as density is doubled. They also found that the residents of the low density areas traveled farther at higher speeds, but had to spend more time at it. This striking finding has major implications about the cost-effectiveness of highway programs intended to increase average speeds since most of the assumed benefits consist of saved time.

Newman and Kenworthy (1989) examined fuel consumption in Manhattan and the New York region. Their analysis indicated that Manhattan's residents drive 1/7 as much per capita as the average American. As density increases from 8 to 43 units per acre, VMT is reduced 54 percent, or a 30 percent reduction in VMT as density is doubled. The previous caveat about fuel consumption studies applies.

The U. S. Department of Transportation (McElhaney, 1989) estimated VMT for American urbanized areas. Even at this large scale of aggregation and with much scatter the data suggested a reduction of 30 percent in VMT as density doubles.

Community Characteristics Promoting Transit and Walking

Several analysts have identified the densities necessary to support transit systems. Certainly, transit can be operated at high frequency in low density areas with adequate subsidies or fares. Costs can be cut on low ridership routes by using smaller vehicles or automating the system (automating can backfire and substantially raise the capital and operating costs of complex systems). However, considering the unwillingness of the American public to subsidize "empty buses" in normal operation, these guidelines are useful. These studies provide an indication of patronage changes with density. Other studies have shown the efficacy of mixing uses and locating shopping near housing concentrations on reducing driving.

Two California agencies have guides for developing pedestrian and transit accessible communities: California Air Resources Board (1993 Draft), and Nancy Hanson of the California Energy Commission (1993, with updates).

From their study of 32 major cities around the world, Peter Newman and Jeffrey Kenworthy (1989) report on a United Kingdom study and conclude that below 20 persons/hectare (8 persons/acre, and 8-10 du/res acre (dwelling units/residential acre) at household sizes and land uses common to San Francisco area cities) there is a marked increase in driving, and below 30 persons/hectare (12 persons/acre, 12-16 du/res acre) the bus service becomes poor. They recommend densities above 30-40 persons/hectare (12-16 persons/acre, 12-20 du/res ac) for public transit oriented urban lifestyles.

Boris Pushkarev and Jeffrey Zupan (1982) recommend the following densities (dwelling units per residential acre):

Bus: minimum service, 1/2 mi between routes, 20 buses/day	4 du/res ac
Bus: intermed serv, 1/2 mi between routes, 40 buses/day	7 du/res ac
Bus: freq serv, 1/2 mi between routes, 120 buses/day	15 du/res ac
Light rail: 5 min peak headways	9 du/res ac, 25 - 100 sq mi corridor
Rapid tr: 5 min peak headways	12 du/res ac, 100 - 150 sq mi corridor
Commuter rail: 20 trains/day	1 - 2 du/ res ac, on existing track

The Institute of Transportation Engineers (1989) recommends the following minimums:

1 bus/hour	4 to 6 du/res.acre	5 to 8 msf of commercial/office
1 bus/30 min.	7 to 8 du/res ac	8 to 20 msf of commercial/office
Lt. rail, feeder buses	9 du/res ac	35 to 50 msf of commercial/office

Marcia Lowe recommends at least 7 du/res ac for local bus service and 9 du/res ac for light rail (1992).

Sacramento Rapid Transit recommend at least 10 du/res ac within 1/4 mile and 5 du/res ac outside that for bus service, and 10 du/res ac for light rail service (1987).

Consultants determined that 43 du/res acre within 1/8 mile and 10 du/res acre in the next 1/8 mile would be necessary for rail transit (Barton-Ashman Associates, 1990).

Snohomish county planners similarly found 7 to 15 du/residential acre can support frequent local bus service. They found that a large, pedestrian accessible, area at these densities might also support light rail (Snohomish County Transportation Authority, 1989).

Seattle planners have concluded that transit ridership increases significantly when the density of jobs exceeds 50 employees per acre in centers with at least 10,000 jobs (Seattle METRO, 1987).

The rate of auto travel to a central business district shopping area well served by rail and bus transit was found to be 75% lower than that to a comparable suburban shopping area (JHK and Associates, 1993). Compared to the suburban mall, auto use at the urban center dropped from 95% to 38% of shoppers, while transit use increased from 4% to 32%, and walking increased from 1% to 29%.

A survey of five cities found that over 70% would switch from auto to walking or bicycling for shopping and personal business if the trips were only 1/2 mile and pedestrian walkways were provided (Ferrol Robinson, et al, 1980). Nancy Hanson calculates that if half of the shopping or personal business trips that are between 1/2 mile and 5 miles could be shortened to 1/2 mile, and half those trips taken by foot, then total vehicle trips would decline by over 5% (1993)

While only 3 to 8% of mid-day lunch or errand trips were found to be by walking in typical single-use office parks, walking increased to 20-30% in pedestrian accessible mixed- use areas (David Unterman, 1984).

A survey of suburban centers found that 27-33% of the employed residents living in the center also worked at the center (Kevin Hooper, 1988).

A study of 400 Portland neighborhoods showed "that households in pedestrian friendly neighborhoods make over three times as many transit trips and nearly four times as many walk and bicycle trips as households located in neighborhoods with poor pedestrian environments" (1000 Friends of Oregon, 1994). Households in the highest pedestrian friendly areas drive half as mulch as those in the least pedestrian friendly areas. "The analysis suggests that vehicle miles traveled per household in pedestrian hostile neighborhoods would be reduced by as much as 10% with a significant improvement in the pedestrian environment." The measures of pedestrian friendliness were density, proximity to employment, grid pattern streets, continuous sidewalks and easy street crossings.

The California Air Resources Board has recommended the following actions to reduce auto use (1993).

	VMT or trip  reductions at site	VMT or trip   reductions in region
Bike, pedestrian, traffic flow improvements	1 - 10%	1 - 2%
Mixed uses, higher densities %	20 - 50%	4 - 11%
Improved transit, ridesharing, traffic flow		5 - 10%

Walking Distances To Transit

How far people are willing to walk to work, shop, visit friends or to transit depends upon many factors which make up pedestrian accessibility, including hilliness, the availability and condition of sidewalks, trees and such street furniture as awnings for protection from sun or rain, seating and other amenities, other pedestrians and interesting stores or vistas along the walk, the amount and speed of the street traffic and the ease and safety of street crossings. Studies should find greater willingness to walk as the pedestrian accessibility of an area increases. As communities improve neighborhood shopping and achieve higher densities with more pedestrians, the distance its residents are willing to walk should increase.

Boris Pushkarev and Jeffrey Zupan report that the median (half are longer and half are shorter) walk to the New York subway is .35 mi, and the median walk to New Jersey commuter rail stations is .5 to .6 mile (1980). They use 1/2 mile walking distance as "rail territory".

The National Personal Transportation Study found that 70% of Americans will walk 500 feet for normal daily trips, 40% are willing to walk 1,000 feet (1/5 mile), and 10% will walk a half mile (David Unterman, 1990). This study shows little willingness to walk in the pedestrian-unfriendly environments of most Americans.

The NPTS also found that 10.3% of those living within 1/4 mile of public transit used it to get to work, while only 3.8% of those living within 1/4 and 2 miles used it, and less than 1% of those living farther away used it (U.S.DOT, 1986). Michael Bernick found that 30 to 40% of apartment residents living within 1/2 mile of Walnut Creek and Pleasant Hill BART stations took BART to work and another 25% used other public transit, compared to 13% using transit regionwide (1990).

Pedestrian analyst Michael Replogle found that Montgomery County, Maryland residents will walk 1/4 mile median distance to a bus and 1/2 mile to a rail stop, and recommends assuming those distances for analyses (1984).
A trip survey in the San Francisco area gave an average time for all walking trips of 12.5 minutes, which is 0.625 mile at 3 mph, a common average walking rate (U.S. DOT, 1988).

3. THE COMMUNITIES AND THEIR CHARACTERISTICS

The Communities

This study compares eleven communities in the San Francisco area, eight in the Los Angeles area, five in the San Diego area and three in Sacramento. Five of those in the San Francisco area were studied in the 1991 NRDC analysis (although Rockridge and Walnut Creek were each shaved by one ZIP code, and the three Danville-San Ramon ZIPs pared to one San Ramon ZIP in this analysis). Those five communities were selected because they vary uniformly from high density with good transit, neighborhood shopping and pedestrian accessibility at one extreme, to low density with poor transit, shopping separated from residences and poor pedestrian accessibility at the other. These variables co-varied within these communities. The communities were selected to minimize the impacts of variations in income and ethnicity on driving. Those communities are shown in bold on the tables.

The additional San Francisco area communities were selected with the advice of planners at the Association of Bay Area Governments. These communities were selected to diverge from uniform co-variance on one characteristic so the impact of that characteristic could be measured. For instance, Lafayette is similar to San Ramon except for its higher income, and Central Berkeley matches Rockridge but has lower income. Alameda matches Rockridge but has less transit. Daly City matches Rockridge but has less neighborhood shopping. Los Altos-Los Altos Hills matches Lafayette but is more pedestrian friendly. Morgan Hill was selected to match Alameda, except that it is located far from the urban core: a test of the non-centrally located but transit-oriented "traditional village" concept. It will be seen later that the data do not support many of these comparisons. For instance, while Lafayette does have higher income than San Ramon, it also has somewhat better transit service and neighborhood shopping but less pedestrian accessibility--muddying the direct comparison.

The Los Angeles area communities were selected to co-vary from the high-medium density south Long Beach through medium density south Santa Monica and south central Pasadena and low-medium density Alhambra to Moreno Valley at the lowest density. Downey was selected to match Alhambra but with less transit. Beverly Hills matches Alhambra but has higher income. Riverside was selected to match Alhambra, except that it is located far from the urban core: another test of non-centrally located but transit-oriented "traditional village" concept.

The San Diego area communities were selected to co-vary from medium density Uptown through Escondido and Clairemont to lowest density Bostonia-Crest-Flinn Springs- Blossom Valley. La Costa was selected to match Bostonia, et al., but with higher income.

The Sacramento communities were selected to co-vary from medium density Central City through East Sacramento-North Land Park to low density South Sacramento.

These community descriptions were circulated to the city planning departments for review.

Table 2 Communities Studied
Name	Co-Variance Deviation	ZIPs	Census Tracts
San Francisco Area
Nob-Rus-NoBea		94108, 94111, 94133	101, 103 - 108, 112 - 119, 121
San Francisco		941xx	all San Francisco
centr Berkeley	$ < Rock	94702, 94703	4218, 4219, 4222, 4223, 4230, 4231, 4233 -4235, 4240
Daly City	MU < Rock	94014, 94015	6004 - 6009, 6011 - 6015, 6016.01 -6016.03
Alameda	Tr < Rock	94501	4271 - 4286
Rockridge		94618, 94705	4001 - 4003, 4042, 4043, 4236 - 4239
Walnut Creek		94596, 94598	3382.01, 3382.02, 3383.01, 3383.02, 3390, 3400.01, 3400.02, 3430.01 - 3430.03, 3461.01, 3553.02
San Ramon		94583	3451.01 - 3451.04, 3451.08, 3451.09, 3452.02
Morgan Hill	"Rural" Ala	95037	5123.03, 5123.04, 5123.98
Lafayette	$ > SR	94549	3470, 3480, 3490, 3500, 3512
LAltos-LAH	Ped > Laf	94022, 94024	5100.01, 5100.02, 5101 - 5105, 5117.01 - 5117.03
Los Angeles Area
s Long Beach	Densest	90802 90813	5759, 5760, 5761, 5762, 5765, 5766 5728, 5729, 5752, 5753, 5754, 5755, 5756, 5758, 5763, 5764
s Santa Mon	Hi/med dens	90401 90405	7015.02, 7017.02, 7019 7020, 7021, 7022, 7023
sw Bev Hills	$ > Alh	90212	7009.02, 7010
sc Pasadena	Hi/med dens	91101 91106	4622, 4636 4623, 4627, 4634, 4635, 4640
Alhambra	Med dens	91801 91803	4808.02, 4818, 4819.01. 4819.02 4803, 4804, 4808.01, 4810, 4815, 4816
c Downey	Tr < Alh	90241	5508, 5509, 5510, 5513, 5514
n Riverside	Rural Alh	92501	301, 302, 303, 423
Moreno Valley	Lo dens	92387 92388	422.04, 424, 425.01, 425.02, 425.03, 426.01
San Diego
Upto-MisH-Hil	Med dens	92103 92116	1, 2, 3, 4, 6, 7, 60, 61 5, 10, 11, 12, 17, 18, 19, 20.01, 21
Clairemont	Med/lo dens	92117	85.01, 85.02, 85.03, 85.04, 85.05, 85.06, 85.07, 91.01, 91.02
La Costa	$ > Bos	92009	178.05, 178.07, 178.08, 200.11, 200.12
Escondido	Med dens	92025 92027	202.02, 202.98, 204.03, 205, 206.01, 206.98, 207.01 201.01, 202.04, 202.05, 202.97, 207.03, 207.05, 207.06
Bos-Cr-Fl Sp-BV	Lo dens	92021	155.01, 155.02, 163, 164.01, 164.02, 165.01, 165.02, 167.01, 168.02, 168.06, 168.09
Sacramento
Cent City Med	dens	95814 95816	5 - 12, 20, 21, 53 4, 13, 14, 19
ESac-nLndPk	Med/lo dens	95819 95818	1 - 3, 15 - 17, 52.01 22 - 24
S Sac	Lo dens	95823	45, 49.03 - 49.06, 96.02
Note: n, e, s, w, c = north, east, south, west, central			John Holtzclaw