Greenhouse Gas Emission
Impacts of Carsharing in North America
MTI Report 09- 11
MTI
Greenhouse Gas Emission Impacts of Carsharing in North America
MTI Report 09- 11
March 2010 The Norman Y. Mineta International Institute for Surface Transportation Policy Studies ( MTI) was established by Congress as part of the Intermodal Surface Transportation Efficiency Act of 1991. Reauthorized in 1998, MTI was selected by the U. S. Department of Transportation through a competitive process in 2002 as a national “ Center of Excellence.” The Institute is funded by Con­gress
through the United States Department of Transportation’s Research and Innovative Technology Administration, the Califor­nia
Legislature through the Department of Transportation ( Caltrans), and by private grants and donations.
The Institute receives oversight from an internationally respected Board of Trustees whose members represent all major surface transportation modes. MTI’s focus on policy and management resulted from a Board assessment of the industry’s unmet needs and led directly to the choice of the San José State University College of Business as the Institute’s home. The Board provides policy direction, assists with needs assessment, and connects the Institute and its programs with the international transportation community.
MTI’s transportation policy work is centered on three primary responsibilities:
MINETA TRANSPORTATION INSTITUTE
Research
MTI works to provide policy- oriented research for all levels of government and the private sector to foster the development of optimum surface transportation systems. Research areas include: transportation security; planning and policy develop­ment;
interrelationships among transportation, land use, and the environment; transportation finance; and collaborative labor- management relations. Certified Research Associates conduct the research. Certification requires an advanced degree, gener­ally
a Ph. D., a record of academic publications, and professional references. Research projects culminate in a peer- reviewed publication, available both in hardcopy and on TransWeb, the MTI website ( http:// transweb. sjsu. edu).
Education
The educational goal of the Institute is to provide graduate- level education to students seeking a career in the development and operation of surface transportation programs. MTI, through San José State University, offers an AACSB- accredited Master of Sci­ence
in Transportation Management and a graduate Certificate in Transportation Management that serve to prepare the nation’s transportation managers for the 21st century. The master’s de­gree
is the highest conferred by the California State University system. With the active assistance of the California Department of Transportation, MTI delivers its classes over a state- of- the- art videoconference network throughout the state of California and via webcasting beyond, allowing working transportation professionals to pursue an advanced degree regardless of their location. To meet the needs of employ­ers
seeking a diverse workforce, MTI’s education program promotes enrollment to under- represented groups.
Information and Technology Transfer
MTI promotes the availability of completed research to professional organizations and journals and works to integrate the research findings into the graduate education program. In addition to publishing the studies, the Institute also sponsors symposia to disseminate research results to transportation professionals and encourages Research As­sociates
to present their findings at conferences. The World in Motion, MTI’s quarterly newsletter, covers innovation in the Institute’s research and education programs. MTI’s extensive collection of transportation- related publications is integrated into San José State University’s world- class Martin Luther King, Jr. Library.
The contents of this report reflect the views of the authors, who are responsible for the facts and accuracy of the information presented here­in.
This document is disseminated under the sponsorship of the U. S. Department of Transportation, University Transportation Centers Program and the California Department of Transportation, in the interest of information exchange. This report does not necessarily reflect the official views or policies of the U. S. government, State of California, or the Mineta Transportation Institute, who assume no liability for the contents or use thereof. This report does not constitute a standard specification, design standard, or regulation.
DISCLAIMER
MTI Report 09- 11
Greenhouse Gas Emission
Impacts of Carsharing
in North America
June 2010
Elliot W. Martin, Ph. D.
Susan A. Shaheen, Ph. D.
a publication of the
Mineta Transportation Institute
College of Business
San José State University
San José, CA 95192- 0219
Created by Congress in 1991 T
echnical Report Documentation Page
R
eport No. 1.
CA- MTI- 10-- 2702
Government Accession No. 2.
R
ecipients Catalog No. 3.
T
itle and Subtitle4.
Greenhouse Gas Emission Impacts of
Carsharing in North America
R
eport Date5.
June 2010
P
erforming Organization Code6.
A
uthors 7.
Elliot W. Martin, Ph. D.
Susan A. Shaheen, Ph. D.
P
erforming Organization Report No. 8.
MTI Report 09- 11
P
erforming Organization Name and Address9.
Mineta Transportation Institute
College of Business
San José State University
San Jose, CA 95192- 0219
Work Unit No. 10.
C
ontract or Grant No. 11.
DTRT 07- G- 0054
S
ponsoring Agency Name and Address
12.
T
ype of Report and Period Covered13.
Final Report
S
ponsoring Agency Code14.
California Department of Transportation
Sacramento, CA 94273- 0001
U. S. Department of Transportation
Office of Research— MS42 Research & Special Programs Administration
P. O. Box 942873 400 7th Street, SW
Washington DC 20590- 0001
S
upplementary Notes15.
A
bstract16.
This report presents the results of a study evaluating the greenhouse gas ( GHG) emission changes that result from individuals participating in a carsharing organization. The principle of carsharing is simple: individuals gain the benefits of private vehicle use without the costs and responsibilities of ownership. Carsharing is most common in major urban areas where transportation alternatives are easily accessible. Individuals typically access vehicles by joining an organization that maintains a fleet of cars and light trucks deployed in lots located within neighborhoods, public transit stations, employment centers, and colleges/ universities. In this study, the authors conducted a survey of carsharing members across the country to develop a robust estimate of GHG emission impacts resulting from carsharing. The results illustrate the annualized change in GHG emissions among members within the largest carsharing organizations across Canada and the United States. GHG emissions from transportation are lower due to carsharing. The average change in emissions across all respondents is - 0.58 t GHG per household per year for the observed impact, and - 0.84 t GHG per household per year for the full impact. However, it is important that this result is understood in the context of the broad diversity of carsharing impacts. While carsharing does facilitate lower emissions, the reduction is not generalizable across all members or even a majority of members. Rather, carsharing as a system facilitates large reductions in the annual emissions of some households, which compensate for the collective small emission increases of other households. The results also show that respondent households exhibit significant reductions in vehicle ownership after joining carsharing.
Key Words17.
C
arbon dioxide ( CO2); Greenhouse gases; Market assessment; Market development; Vehicle miles of travel
Distribution Statement18.
No restrictions. This document is available to the public through
The National Technical Information Service, Springfield, VA 22161
S
ecurity Classif. ( of this report) 19.
Unclassified
Security Classifi. ( of 20. this page)
Unclassified
N
o. of 21. Pages
104
P
rice22.
$ 15.00
Form DOT F 1700.7 ( 8- 72)
C
opyright © 2010
by Mineta Transportation Institute
All rights reserved
Library of Congress Catalog Card Number: 2009943710
To order this publication, please contact the following:
Mineta Transportation Institute
College of Business
San José State University
San José, CA 95192- 0219
Tel ( 408) 924- 7560
Fax ( 408) 924- 7565
email: mti@ mti. sjsu. edu
http:// transweb. sjsu. edu A
cknowledgments
The Mineta Transportation Institute, the Transportation Sustainability Research Center ( TSRC) at the University of California ( UC), Berkeley, and the Honda Motor Company, through its endowment for new mobility studies at UC Davis, generously funded this research. The authors would like to thank the numerous carsharing programs in North America that have agreed to participate in this survey. Thanks also goes to Caroline Rodier, Adam Cohen, Denise Allen, Melissa Chung, Brenda Dix, Keith Brown, Josh Ma, Jarrett Bato, and Seth Contreras of TSRC and the Innovative Mobility Research group at UC Berkeley for their assistance with the literature review and survey development. The authors would also like to thank Neil Weiss of Arizona State University, as well as Alexander Gershenson and Asim Zia of San José State University. In addition, the authors thank Dave Brook, Clayton Lane ( formerly of PhillyCarShare), and Kevin McLaughlin of AutoShare for their assistance with survey development and report review. The contents of this report reflect the views of the authors and do not necessarily indicate acceptance by the sponsors.
The authors also thank MTI staff, including Research Director Karen Philbrick, Ph. D., Director of Communications and Special Projects Donna Maurillo, Research Support Manager Meg Fitts, Student Research Support Assistant Chris O’Dell, Student Publications Assistant Sahil Rahimi, Student Graphic Artists JP Flores and Vince Alindogan, and Student Webmaster Ruchi Arya. Additional editorial and publication support was provided by Editorial Associate Catherine Frazier.
Mineta Transportation Institute
i
T
able of Contents
E
xecutive Summary 1
INTRO
DUCTION 11
PASTPAST
RESEARCH
ON CARSHARING IMPACTSIMPACTS IN NORTH AMERICA AMERICA 15
FRAME
WORK FOR EVALUATINALUATING THE GREENHOUSE GAS IMPACTSIMPACTS OF CARSHARING 17
The Observed Impact and the Full Impact of Carsharing 18
Carsharing Impacts and Shifts in Travel Modes 19
SUR
VEY METHODOLOGY Y 21
Participating Organizations 21
The Survey Questionnaire 23
Personal Vehicle Driving and Carsharing Usage 24
Rental Vehicles and Taxi Usage 26
Supporting Data 26
Data Preparation 27
RESULTS RESULTS
31
Demographics 31
Carsharing Emissions Impacts 33
Sensitivity Analysis of Aggregate Emission Change 45
Carsharing Impacts by Urban Density 56
Impacts by Organization Type and Country 59
Impacts on Vehicle Holdings 63
The Aggregate Impacts of Carsharing 68
CONCLUSION
s AND POLICY IMPLICATIONS IMPLICATIONS 73
APPEN
DIX: SURVEY SAMPLE 75
E
ndnotes 91
A
bbreviations and Acronyms 95 Mineta Transportation Institute
Contents
ii
B
ibliography 97
A
bout the Authors 101
P
eer Review 103 Mineta Transportation Institute
iii
L
ist of Figures
Distribution of Annual Household GHG Emission Impact 1. 4
Distribution of Total Annual Personal Vehicle Miles Traveled by Household 2. 5
Profile of Cumulative Annual Change in GHG Emissions 3. 6
Age Distribution of Respondents 34. 2
Income and Education Distribution of Respondents 35. 3
Distribution of Annual Household GHG Emission Impact 36. 4
Distribution of Miles Driven by Carsharing Members 37. 5
Distribution of Total Annual Personal Vehicle Miles Traveled by Household 38. 6
Profile of Cumulative Annual Change in GHG Emissions 39. 7
Simulated Distribution of the Sample Mean of the Emissions Change 310. 8
Vehicle Stopped Working and Joined Carsharing 411. 0
Respondents Entering Carsharing Without a Vehicle 412. 1
Households Owning Vehicles but Avoiding Future Purchases 413. 2
Joined Carsharing and Shed Vehicles 414. 3
Distribution of Change in GHG Emissions From Local Taxi and Rental Car Use 415. 4
Sensitivity of Mean Impacts to PVMT Filter Threshold 416. 6
Sensitivity Analysis of Carsharing Impacts Given PVMT Ceiling 417. 8
Sensitivity of Impacts to PVMT Overestimation 518. 0
Sensitivity of Profile of Cumulative Annual Change in GHG Emissions to the 19. Activation of the Move Filter 52
Cumulative Annual GHG Emissions Change with No Filters Active 520. 3
Analysis of Impact by Membership Duration 521. 5
Average Observed Impact by Urban Density ( U. S. only) 522. 7 Mineta Transportation Institute
List of Figures
iv
Scatter Plot of Observed Impacts by Urban Density ( U. S. only) 523. 8
Profile Cumulative Annual Change in GHG Emissions by Respondent by
24. Organization Type ( Observed Impact) 62
Profile and Statistical Evaluation of the Change in Vehicle Holdings25. 64
Fuel Economy Distribution of Household Vehicles Shed/ Added and Carsharing 26. Vehicles Driven 66
Distribution of Vehicles Shed by Model Year ( Vehicle Age) 6727. Mineta Transportation Institute
v
L
ist of Tables
Participating Organizations 21.
Paired t- Test: Mean Difference from Zero 72.
Profile and Statistical Evaluation of the Change in Vehicle Holdings 73.
Transition of Household Vehicle Holding States Among Carsharing Households 84.
Participating Organizations 225.
Categorical Circumstances of Respondent Membership 236.
Generic Vehicle Types and Assumed Fuel Efficiency Factors 267.
Balance of Circumstantial Responses Before and After Data Filters 308.
Paired t- Test: Mean Difference from Zero 399.
Average Observed Impact by Organization Type and Country 5910.
Average Full Impact by Organization Type and Country 6011.
Mean Comparison t- Test of Non- Profit and Profit Organizations Observed 12. Impacts in North America 61
Transition of Household Vehicle Holding States Among Carsharing Households 6513.
Sensitivity of Aggregate Carsharing Emissions Impacts 6914.
Sensitivity Analysis of Industrywide Carsharing Impacts on Vehicle Holdings 7115. Mineta Transportation Institute
List of Tables
vi Mineta Transportation Institute
1
E
xecutive Summary
This study evaluates the greenhouse gas ( GHG) emissions impact that results from the travel lifestyles changes exhibited by members of carsharing organizations. Carsharing ( short- term vehicle access) has been continuously operating in North America for about fifteen years. Just over ten years ago, carsharing emerged in select cities within the U. S. as a niche market alternative to offer members auto access without the costs of private vehicle ownership. Carsharing organizations operate by placing vehicles throughout urban neighborhoods, metropolitan centers, and colleges/ universities. The vehicles are accessible to members through a reservation that is booked in advance by phone or Internet. Members can pay for carsharing services in a variety of ways depending on the organization and pricing plan to which they subscribe. Most members pay a monthly or annual fee in some combination with per hour and per mile charges.
Carsharing influences emissions by allowing members access to a shared automobile on an as- needed basis. Carsharing members may use the shared vehicles to conduct trips that are more convenient with the flexibility of an automobile. However, the pricing structure of carsharing largely encourages the use of shared- vehicles for non- work trips ( outside of specialized business, campus, and governmental fleet packages). Commuting, as well as other short trips, are generally completed through walking, biking and public transit use. Carsharing can result in both increased and decreased emissions. Carsharing increases emissions by providing automotive access to people who were previously carless. These households drive more than before they joined carsharing. Carsharing also decreases emissions by permitting other people who were more reliant on personal vehicles to use automobiles in a more sparing and efficient manner. These households generally discard or shed one or more personal vehicles in substitute of a carsharing membership. These members adapt to a new travel lifestyle that is facilitated by carsharing. This lifestyle is usually characterized by a modal shift that generally leads to reduced emissions over the previous reliance on the personal vehicle owned by the household.
Because carsharing leads to emission increases in some households, and emission decreases in other households, a natural question arises pertaining to overall net impact of carsharing. This study explores this question on a large scale through a single survey of carsharing members within major organizations throughout North America. In cooperation with participating organizations, researchers surveyed carsharing members about their travel patterns during the year before they joined carsharing and at the time of the survey. This before- and- after analysis explores how the emissions of the household changed since joining carsharing. Researchers sent the Canadian and American respondents separate surveys due to the different distance and currency units used in the respective countries. The organizations that participated in the survey are listed in Table 1. Mineta Transportation Institute
Executive Summary
2
P
articipating Organizations Table 1
O
rganization
L
ocation
AutoShare
Toronto, Ontario, Canada
City CarShare
San Francisco/ Oakland, California
CityWheels
Cleveland, Ohio
Community Car Share of Bellingham
Bellingham, Washington
CommnuAuto
Montreal, Province of Quebec, Canada
Community Car
Madison, Wisconsin
Co- operative Auto Network/ The Company Car
Vancouver, British Columbia, Canada
IGo
Chicago, Illinois
PhillyCarShare
Philadelphia, Pennsylvania and Wilmington, Delware
VrtuCar
Ottawa, Ontario, Canada
Zipcar
United States and Canada
The organizations distributed the survey solicitations to their members through their own email lists. The email that the organizations sent out included a link to the survey at a third- party site. Two reminders were sent out via each organization, and the survey closed on November 7, 2008. Most organizations, which are located in a single city, distributed survey solicitations to all of their members. Because of Zipcar’s size and geographic distribution, the solicitation was capped at a total of 30,000 randomly selected Zipcar members within specific markets. This included 5,000 each within New York City, New York; Boston, Massachusetts; Washington DC; Portland, Oregon; and Seattle, Washington. An additional 2,500 each in Canadian cities Vancouver and Toronto also received survey solicitations. In aggregate, the authors estimate that nearly 100,000 carsharing members received the survey solicitation. Based on the coverage, size, and selection of this population, the authors consider it to be random and representative of the carsharing population within North America. In total, 9,635 surveys were completed, constituting a response rate of about 10%.
The unit of analysis of this study is the entire household of the carsharing member, as an individual’s carsharing use can affect the travel emissions of all household members. For example, an individual may join carsharing and shed ( gets rid of) their personal vehicle that they used exclusively. But another member of the household retains his or her vehicle, which is subsequently shared with the carsharing member when it is available. The vehicle belonging to the non- member within the household is driven more than previously because two people are using it.
The survey calculated the GHG impacts that result from the change in annual overall automotive use. This consisted of the annual personal and carsharing automotive emissions of the household at the time of the survey minus the annual personal automotive emissions of the household during the year before joining carsharing. The result is a change in the annual rate of household emissions before and after carsharing. The population of study in this survey includes households that use carsharing within the neighborhood business Mineta Transportation Institute
Executive Summary
3
model. The neighborhood business model places vehicles within urban residential neighborhoods and downtowns that are accessible to any and all members. This market is the predominate market within the carsharing industry and comprises the vast majority of members. The survey excludes members that use carsharing strictly within a business application and university students using carsharing within a college setting. These cohorts constituted 2% and 6% of the sample, respectively. The analysis also filtered respondents that indicated a move of home or work that significantly altered their overall driving. In addition, respondents that indicated that they did not use carsharing at all were filtered as “ inactive” users. Inactive users are a cohort of carsharing members that do not use the service but retain their membership. Because their travel lifestyles are conducted without carsharing, they are assigned a zero impact in this study. Further discussion of data processing and respondent filtering is presented in the complete report. The influence of these cohorts on the overall results are also explored in a sensitivity analysis.
This study explores the GHG emission change through two distinct but related metrics. One impact is termed the “ observed impact,” which describes the emission change that actually occurred. The observed impact considers the total household driving before the member joined carsharing and the total household driving at the time of the survey. A second impact is termed the “ full impact,” which includes the observed impact but also an additional component of avoided emissions. To explain further, carsharing gives people who are considering purchasing a vehicle an alternative means in which to achieve “ automobility.” As a result, some people who would have bought a car choose to join carsharing instead. The driving of the forgone personal vehicle would have resulted in some emissions that never then occur. The survey explores this dynamic with relevant respondents and estimates the additional emissions that were avoided due to forgone vehicles that were never acquired and driven. These avoided emissions, when added to the same emissions covered by the observed impact, constitute the full impact of carsharing. Because the full impact introduces an additional component of abstraction and measurement uncertainty, it is reported separately alongside the observed impact throughout the report.
The results show that overall net annual emissions of households joining carsharing are lower than they were before they joined carsharing. Across the 6,281 respondents that were applied in the final analysis, carsharing facilitates a decrease in annual emissions for some members and an increase in annual emissions among other members. The authors found that on balance, net carsharing emissions are negative and statistically significant for both the observed impact and full impact. Hence, GHG emissions from transportation are lower due to carsharing. The average change in emissions across all respondents is - 0.58 t GHG per household per year for the observed impact, and - 0.84 t GHG per household per year for the full impact. However, it is very important that the “ how and why” of this result is understood in the context of the broad diversity of carsharing impacts. While carsharing does facilitate lower emissions, the reduction is not generalizable across all members or even a majority of members. Rather, carsharing as a system facilitates large reductions in the annual emissions of some households, which compensate for the collective small emission increases of other households. This dynamic is important for the construction of sound policy, which can encourage carsharing growth in a manner that provides mobility benefits and continued emission reductions within urban and suburban environments. Mineta Transportation Institute
4 Executive Summary
Exploring the data in more detail, the results show that a majority of households are increasing their emissions through carsharing— but the degree to which these households are increasing their emissions is very small. In contrast, the minority of households reducing their emissions are exhibiting changes that are of larger magnitude and greater variance. Figure 1 shows a histogram that illustrates the distribution of impacts by respondent count for both the observed and full impact.
Distribution of Annual Household GHG Emission ImpactFigure 1
Distribution of Annual Household GHG Emission Impact
For both the observed and full impact, the distribution shows the large number of respondents increasing their emissions. This is evident with the high number of respondents that exhibit an increase in annualized emissions within the bounds of 0 and 0.25 t GHG/ yr. The distribution of members lowering their emissions is far more evenly spread for both the observed and full impact. In total, 4,456 ( 71%) of respondents have a positive observed impact, while 1,825 ( 29%) have a negative observed impact. For the full impact, the balance is more evenly distributed by respondent frequency, as 3,281 respondents ( 53%) have a positive full impact while 2,953 respondents ( 47%) have a negative full impact. Mineta Transportation Institute
Executive Summary
5
The difference between the number of respondents decreasing their emissions in the observed impact and the full impact highlights the importance of considering the avoided emissions. The resulting shift of the full impact reduces the number of members with impacts greater than zero. Absent any consideration of avoided mileage, these respondents would appear to be increasing their net emissions through carsharing.
Most members drive carsharing vehicles very short distances over the course of a year. For example, 30% of all households report placing less than 250 miles per year on carsharing vehicles. An additional 16% reported driving between 250 and 500 miles, and 19% placed between 500 and 1,000 miles annually. In total, more than 80% of all households in the sample drive less than 2,000 miles per year on carsharing vehicles. In contrast, households decreasing their emissions were driving much longer annual distances in personal vehicles before adapting to a carsharing lifestyle. Figure 2 shows the distribution of personal vehicle miles traveled ( PVMT) of the sample both before and after joining carsharing.
Distribution of Total Annual Personal Vehicle Miles Traveled by Figure 2 Household
The distribution within Figure 2 shows the overall shift of households toward lower personal vehicle driving. The “ before- and- after” shift in the PVMT distribution shows a significant gain in the number of carless households, an increase of nearly 30%. The distribution of annual household PVMT distances shows a general decline of households driving all distances. This does not mean that no households reported an increase in household PVMT, some did. But most households lowered mileage by eliminating at least one vehicle. Mineta Transportation Institute
6 Executive Summary
When added together, the result of these collective movements provides a clear picture of the shape of the overall impact of carsharing. Figure 3 presents the same aggregate distribution of emissions change as Figure 1. But Figure 3 shows the impact as weighted by the annual emissions change for each respondent within the categorical bin. In other words, each categorical bin of the horizontal axis contains the summation of the annual change in respondent emissions. The result is a distribution that illustrates the cumulative net annual change in emissions for all survey respondents. The top graph in Figure 3 illustrates this distribution for the observed impact, and the bottom graph shows the full impact.
Profile of Cumulative Annual Change in GHG EmissionsFigure 3
For both the observed and full impact, Figure 3 makes it visually apparent that the area constituting emission reductions is larger than the area constituting emission increases. Thus, while the majority of respondent households are increasing annual emissions, the cumulative annual emissions change is negative and thus so is the average. The statistical significance of the average change in annual emissions is shown in Table 2 as given by the paired t- test. Mineta Transportation Institute
Executive Summary
7
This overall result that carsharing lowers emissions is robust to a variety of assumptions and key input modifications to the data. A sensitivity analysis given in the full report shows how the average and distribution of emission impacts will change given an alteration of key assumptions. For example, the sensitivity analysis illustrates how the emissions would change if the maximum annual PVMT value given by respondents is constrained with an upper bound that is gradually lowered to zero. In addition, the sensitivity analysis illustrates how the results change with the re- admission of filtered respondents, including movers, students, business users and inactive members. Overall, the inclusion of these cohorts increases the variance of the impacts, but they do not change the overall mean to a significant degree. Thus, by examining the data from several perspectives, the sensitivity analysis illustrates how the mean and statistical significance of the aggregate impacts vary with changes to key assumptions and data.
P
aired t- Test: Mean Difference from ZeroTable 2
The emissions impacts described above are in large part driven by households shedding vehicles upon joining carsharing. As part of the survey, respondents were asked to provide the make, model, and year of each vehicle owned by the household before and after joining carsharing. These data permitted an analysis of the change in household vehicle holdings within the sample, which is presented in Table 3.
Profile and Statistical Evaluation of the Change in Vehicle HoldingsTable 3
Table 3 illustrates how households with different quantities of vehicles before joining carsharing adjusted their vehicle holdings. When changing vehicle holdings, there are four possible actions that a household can take: the household can shed, retain, add, or replace
   Mineta Transportation Institute
8 Executive Summary
a vehicle. Vehicle replacement involves the shedding and adding of a vehicle within the same household. For instance, in a household that sheds two vehicles and adds one, the added vehicle is counted as a replacement. Similarly, in a household that sheds one vehicle and adds two, one of the new vehicles is a replacement, and the other is an added vehicle. The results show that the sample of 6,281 households shed a total of 1,461 vehicles, which amounts to a statistically significant reduction in the average vehicles per household.
Further insights with respect to vehicle shedding are presented within a matrix that shows how households transitioned from different states of vehicle holdings before and after joining carsharing. Table 4 presents a cross- tabulation of household vehicle holdings “ before” and “ after” joining carsharing and shows how households within the sample transitioned to new vehicle holding states.
T
ransition of Household Vehicle Holding States Among Carsharing Table 4 Households
The column on the far right (“ Total”) illustrates the distribution of household vehicle holdings before joining carsharing while the bottom row (“ Total”) illustrates the distribution of vehicle holdings after joining carsharing. The cells within the table show the counts at each transition. As evident from the upper- left cell ( the zero- car household to zero- car household transition), most households ( 3686) joining carsharing were carless and remained carless. The second largest count is within the cell immediately below, in which one- car households became carless households. Overall, the transition matrix shows that most of the changes in vehicle holdings were the result of a household shedding a single car.
In summary, this study completed a survey of members of carsharing organizations across the United States and Canada. The results of the data show that in aggregate, transportation emissions of households that join carsharing are lower after they join. The average change in annual emissions is consequently negative and statistically significant. The results also show that carsharing households lower their average vehicle holdings by a degree that is also statistically significant. The shedding of vehicles that were driven before household members joined carsharing plays a major role in driving the emission reductions.
After Joining Mineta Transportation Institute
Executive Summary
9
It is important to recognize that in the context of carsharing, the “ average” emissions change is not the same as the “ typical” emission change. Carsharing provides mobility benefits to many members that come from carless households. These mobility benefits accrue directly to the member and offer their own internal advantages. But strictly from an emission perspective, carless households that drive more through a carsharing membership are increasing emissions. These households constitute a majority of the carsharing membership, but their contributions to emissions are small because carsharing vehicles are generally not driven long distances by members. Instead, carsharing vehicles are predominantly used for short non- work trips or the occasional long- distance day trip. Households that reduce their emissions through carsharing generally do so by shedding personal vehicles and placing far fewer emissions on carsharing vehicles. The combination of this dichotomous process results in an overall net reduction of emissions. This result is robust to a variety of assumptions and data modifications as conducted in a broad sensitivity analysis.
This study contributes to mounting evidence that carsharing is lowering GHG emissions by providing people with automotive access on an as- needed basis. The scope of the impacts evaluated is restricted to the household travel- based emissions. The sample population constitutes carsharing members that use the neighborhood business model of carsharing. No emission impacts from vehicle holding reductions or land- use changes are considered. The results and scope of the study have important implications for policy design. Carsharing systems provide environmental benefits. However, caution regarding the caveats of this study in any policy design and emission crediting is necessary. It is clear from the data that not all members reduce emissions. In addition, not all members of carsharing organizations are active members. Carsharing organizations contain some number of inactive members. These members use carsharing very infrequently and are only members for occasional events and emergencies. Carsharing provides a supplement to their lifestyle, but it may not influence or facilitate it in a major way. The share of these members within an organization could vary over time based on industry pricing plans as well as general economic conditions. The diversity of impacts across members suggests that credits for carsharing impacts should be certifiable in some form. Future studies should continue to evaluate carsharing trends, as they will likely evolve. Based on these results, as long as carsharing continues to thrive economically, its benefits are likely to grow, as more carholding households find carsharing to be an established and stable option for meeting automotive travel needs within North American cities.
Mineta Transportation Institute
10 Executive Summary
Mineta Transportation Institute
11
INTRO
DUCTION
Mounting evidence of climate change and increasing energy costs are motivating many state and local governments to explore policy options that can simultaneously reduce petroleum consumption and greenhouse gas ( GHG) emissions. Within the United States, transportation activity accounts for close to 30% of all anthropogenic carbon dioxide ( CO2)- equivalent GHG emissions and nearly 70% of all petroleum consumption. As a sector, transportation is almost exclusively petroleum dependent, as roughly 96% of all energy consumed in the U. S. is comprised of either gasoline or diesel. 1 Furthermore, a longstanding dependence on the private automobile for urban transportation has placed the U. S., and to a lesser extent Canada, in uniquely difficult positions to adjust travel in ways that mitigate the impacts of higher energy costs, air pollution, and global warming.
This study evaluates the GHG emission impact that results from changes in travel when households join a carsharing organization. Carsharing ( short- term vehicle access) has been continuously operating in North America for about fifteen years. Just over ten years ago, carsharing emerged in select cities within the U. S. as a niche market alternative to offer members auto access without the costs of private vehicle ownership. Carsharing organizations operate by placing vehicles throughout urban neighborhoods, metropolitan centers, and colleges/ universities. The vehicles are accessible to members through a reservation that is booked in advance by phone or Internet. Members can pay for carsharing services in a variety of ways depending on the organization and pricing plan to which they subscribe. Most members pay a monthly or annual fee in some combination with per hour and per mile charges. 2
Since its inception, carsharing has grown rapidly under both non- profit and for- profit business models. Today, the industry is comprised of 42 organizations within North America, most of which have primarily focused on serving a single metropolitan region. As of July 1, 2009, there were 16 active programs in Canada and 26 in the U. S., with an estimated 378,000 carsharing members sharing approximately 9,818 vehicles in North America. In addition, 30% of the operators in the U. S. were for- profit ( 8 of 26), accounting for 86% and 88% of the members and vehicles, respectively. In Canada, 38% of Canadian carsharing operators were for- profit ( 6 of the 16) and represented 87% of members and 85% of the total fleet deployed. 3
The consumer appeal of carsharing is fundamentally economic. Owning a car requires a considerable outlay of recurring fixed expenses, regardless of how much the vehicle is driven. In urban areas, fixed ownership costs are typically higher than the national average, while driving distances are typically lower than average. This dynamic makes transit rich urban areas among the most viable carsharing markets. Individuals who occasionally require a car for shopping can use a carsharing service, paying only for the time and distance that they need to travel. 4 Meanwhile, they avoid vehicle purchase/ lease, gasoline, insurance, and storage costs, which are regularly associated with ownership.
In addition to the private economic benefits gained by consumers, past research has suggested that carsharing may offer considerable environmental and social benefits. 5 These benefits include GHG emission reductions and greater use of alternative modes, Mineta Transportation Institute
Introduction
12
such as public transit, walking, and cycling. In the industry today, carsharing vehicles are newer relative to the average personal vehicle and generally have higher than average fuel economy. 6 Long- term land- use benefits may also arise as carsharing permits a single car to satisfy the mobility needs of multiple individuals. Among the most consistent findings of past research is that many users reduce or eliminate their household’s vehicle holdings, reducing the total number of vehicles that need to be parked within an urban environment. 7 Thus, carsharing has been considered a promising transportation demand management tool capable of displacing gasoline consumption that would otherwise occur in its absence.
While past research suggests a link between carsharing and vehicle miles/ kilometers traveled ( VMT/ VKT) and/ or GHG emission reduction, many of the studies have evaluated this association using different methodologies and metrics that are difficult to compare. Defining a consistent system boundary that characterizes the bulk of measureable environmental impacts from carsharing remains a challenge. Furthermore, most studies have focused their evaluations on a single organization. While these past efforts are extremely valuable in contributing to the public knowledge, no study has applied a standard methodology for assessing the impacts of members across organizations or metropolitan regions. Past research exhibits a general consensus that carsharing results in lower VMT/ VKT, private auto ownership, and lower emissions, but there is little agreement regarding the magnitude of those impacts. One important factor that has not been considered in any study to date is the potential link between a member’s carsharing organization type and VMT/ VKT reductions. There is variation within the industry, as profit and non- profit organizations operate carsharing organizations differently. These differences exist with respect to the design of pricing plans, the mix of vehicle fleets, and the distribution of vehicle networks. 8 The pricing plan determines the nature of the marginal cost to the consumer and likely influences their VMT/ VKT.
This report presents the results of a survey of carsharing members across the North American continent. The objective of the study was to evaluate the change in GHG emissions that result from household members joining carsharing. The hypothesis of this study is that across all members, the net impact of carsharing is a reduction in emissions. The focus of this evaluation is the impact of the neighborhood model of carsharing on the transportation emissions of working households. That is, this study does not evaluate the GHG impacts of carsharing on members who are part of the college submarket or the business- use submarket. Explorations of these smaller submarkets require a separate survey design. The survey was conducted online in October and November 2008, with all of the major carsharing organizations in the U. S. and Canada. The survey asked about past and current vehicle holdings as well as travel patterns to estimate GHG changes that result from people joining carsharing.
This report proceeds with five main chapters. First, the authors present a review of earlier studies and surveys assessing the environmental impacts of carsharing, with an emphasis on North America, in “ Past Research on Carsharing Impacts in North America.” The next chapter, “ Framework for Evaluating the Greenhouse Gas Effects of Carsharing,” provides a theoretical framework to describe how GHG impacts are assessed within this study. This includes an overview of the dynamics that govern how carsharing can alter member emissions. The following chapter, “ Survey Methodology,” presents the methodological Mineta Transportation Institute
Introduction
13
approach for this analysis, including an overview of the study instruments and participating organizations. This follows with a presentation of the analytical results in ” Results.” The results characterize the emission impacts of carsharing across several dimensions, including circumstances of joining, urban density, and organization type. In addition, the results section contains a series of sensitivity analyses that illustrate the robustness of the findings under a variety of circumstances. Following the sensitivity analysis, the impacts of carsharing on vehicle holdings is presented. The last subsection of the results applies the factors computed for both vehicles and emissions to an aggregate analysis. The last chapter of this report, “ Conclusions and Policy Implications,” provide a dissemination of the information gleaned from the data and recommendations for carsharing agencies in the United States and Canada. Mineta Transportation Institute
14 Introduction
Mineta Transportation Institute
15
PASTPAST
RESEARCH
ON CARSHARING IMPACTSIMPACTS IN NORTH AMERICA
Among the most consistent findings of past research is that carsharing reduces car ownership. The first demonstration of carsharing in North America started in San Francisco with the Short Term Auto Rental ( STAR) program. Established in 1983, STAR was a 55- vehicle pilot designed to operate for three years but terminated after 18 months of operation. In the STAR evaluation, Walb and Loudon ( 1986) reported on changes in car ownership and travel among members. They found that 17% of members sold a vehicle, while 43% postponed a vehicle purchase. However, their assessment of travel changes raised doubts as to whether carsharing would result in more efficient travel as members reported increasing their travel slightly. 9 While the STAR program did not gain traction, lessons learned from that effort were used to inform and improve the launch of CarSharing Portland more than a decade later. 10 Similar to STAR, an early study of CarSharing Portland’s impacts found that 26% of members sold a car, while 53% avoided a purchase. 11 The study also reported members using public transit, biking, and walking more. But similar to STAR, the early study found little change in VMT/ VKT among members. 12 For a more extensive review on the history of the carsharing industry, see Shaheen et al., ( 2007) and Shaheen et al., ( 1998). 13
Similar results from evaluations of carsharing programs persisted through the early years of this decade. Carsharing returned to San Francisco with the launch of City CarShare in March 2001. Cervero ( 2003) initiated a before- and- after study to evaluate the impacts of City CarShare of both member and nonmember travel behavior three months before the launch and nine months after. 14 A profile of the early members indicated that they were in their early 30s, college graduates, and worked in professional fields. Most significantly, two thirds of members came from zero- car households, while 20% came from one- car households. This early study found that mean daily VMT/ VKT dropped for both members and nonmembers, but changes for both groups were not statistically significant. In addition, shares of walking and biking fell. Cervero’s early results of City CarShare were consistent with past work in North America; they found similar demographics among members and that changes in VMT/ VKT were not substantial. The early carsharing adopters were those who were primarily carless and used carsharing as a means to augment their mobility. 15
Cervero’s early work was soon followed by Lane ( 2005), which evaluated the first- year impacts of PhillyCarShare, a non- profit organization operating in Philadelphia since November 2002. One year after PhillyCarShare’s launch, Lane administered a 500 member online and mail- in survey in November 2003. Roughly 60% of members who joined were from households with zero cars. Members were otherwise demographically similar to the early adopters of City CarShare. Lane evaluated vehicles sold as a result of membership as well as vehicles not acquired. He reported that each PhillyCarShare vehicle removed roughly 23 cars from the road. Finally, Lane discussed VMT/ VKT drops among members, while acknowledging uncertainty in his estimate. He concluded that a typical reduction would amount to a couple hundred miles per month for members who gave up a car, but that there is considerable variance in his estimate. 16 Mineta Transportation Institute
Past Research on Carsharing Impacts on North America
16
As carsharing evolved, research began to discern more pronounced effects on VMT/ VKT. Cervero and Tsai ( 2004) and Cervero et al. ( 2007) revisited City CarShare impacts. 17 By the 2007 study, VMT/ VKT reductions attributable to carsharing were becoming more evident as member VMT/ VKT was found to decrease relative to nonmember VMT/ VKT. VMT/ VKT reductions among carsharing members appeared to occur during the first two years, but large variations existed within the group. Overall, mean mode- adjusted VMT/ VKT, which accounted for occupancy levels, dropped 67% for carsharing members in contrast to a 24% increase among nonmembers. 18
As carsharing has matured in North America, emerging evidence suggests the presence of considerable reductions in VMT/ VKT among members. This trend may continue as carsharing continues to draw new members from households that fit the more traditional American profile of higher vehicle ownership and driving.
Research to date has yet to standardize the evaluation of GHG impacts due to carsharing. In addition, there are many factors influencing carsharing use that have not been explored, including the impact as categorized by members of different organization types. Furthermore, as carsharing networks expand into more diverse residential environments, the potential for VMT/ VKT reductions may be greater. Lower density environments, where carsharing typically struggles economically, may offer greater gains as they enter markets with higher levels of car ownership and VMT/ VKT. This research aims to address the magnitude and distribution of GHG emission change that are exhibited by members of carsharing organizations. In the following chapter, the authors present a conceptual framework for evaluating the GHG impacts of carsharing. Mineta Transportation Institute
17
FRAME
WORK FOR EVALUATINALUATING THE GREENHOUSE GAS IMPACTSIMPACTS OF CARSHARING
The scope of this study is focused on evaluating how members of carsharing change their travel behavior. A change in travel behavior is the most direct and observable short- term impact that occurs when a household joins a carsharing organization. It is important to acknowledge that there are two other ways in which carsharing can impact GHG emissions. They include changes in vehicle ownership and changes in local land use. The change in vehicle ownership observed among members joining carsharing is evaluated in this study, but the analysis does not tie impacts from changes in vehicle ownership to GHG emissions. Such changes do occur, as the life- cycle impacts of vehicle production cause additional emissions to be released at the plant and upstream. In the long run, reduced personal vehicle demand would lower vehicle production and hence emissions, but tying such impacts to vehicles shed by carsharing households is subject to considerable uncertainty. Therefore, in the analysis presented here, changes in vehicle ownership are presented, but zero credit is given for changes in GHG emissions from reduced vehicle production.
The third impact that carsharing could have on GHG emissions relates to land use, which is subject to even greater uncertainty. As carsharing reduces the need personal vehicles, some land use effect may exist over time. This effect could be manifested in the form of reduced construction of parking and more compact urban environments. But the broad uncertainties and confluence of factors required to bring about land use change make an evaluation of GHG emissions with the instruments applied here infeasible. Therefore, it is appropriate to note that changes in GHG emissions resulting from changes in vehicle ownership and land use could occur. But because these impacts are very uncertain and manifested over a long- time horizon, they are given zero credit in this research and left to future study.
As this study is focused on the GHG impacts of changes in travel behavior, the authors now discuss the units by which this change is measured. The operating statistic of this study is the change in annual emissions that result from a household joining carsharing. This statistic describes the “ change in annual GHG emissions” of the carsharing household. We discuss this measurement in units of metric tons of GHG per year ( t GHG/ yr).
This unit is chosen because it offers an intuitive illustration of the change in “ state of travel” that carsharing facilitates among its member households. Members enter carsharing with a travel lifestyle suitable to them in the absence of carsharing. This initial travel lifestyle may have involved driving a personal vehicle or living as a carless household. Upon joining carsharing, members transition into a new travel lifestyle. This lifestyle might exhibit reduced driving for those households that join carsharing and discard or shed vehicles. Households may also transition into a state of increased driving, as happens with carless households that gain vehicle access through carsharing.
This unit is used both as a matter of simplicity and practicality in generating respondent information. A year is a natural time frame in which people think about travel and due to the practical limitations of the one- time survey, the researchers could not expect Mineta Transportation Institute
Framework for Evaluating the Greenhouse Gas Impacts of Carsharing
18
respondents to construct a cumulative year- by- year assessment of their travel behavior since joining carsharing. Such a survey would take an inordinate amount of respondent time. Furthermore, the change in annual emissions is a metric normalized by time that permits comparisons across organization types and regions. In addition, previous research indicates that adjustment in travel behavior that results from carsharing often occur rather quickly and remain stable. 19 Cervero et al. ( 2007) finalized their longitudinal study of City CarShare in San Francisco and found that most of the impacts on VMT occurred soon after respondents joined City CarShare. Intermediate and long- term effects occurred in increments that were less substantial. 20 This suggests that capturing the change in the annual emission rates provides an effective proxy for near- term changes facilitated by carsharing. The influence of member tenure within the organization on carsharing impact is further explored among other elements in a sensitivity analysis.
THE
OBSERVED IMPACTIMPACT AND THE FULL IMPACTIMPACT OF CARSHARING
In this chapter, the authors explore two distinct classifications of impact by which we evaluate carsharing. The two classifications are measured in the same units but differ in the system boundary of impacts that they consider. The classifications are separated by the degree to which they consider emissions that would have occurred in the absence of carsharing. Carsharing facilitates people to change their travel lifestyles in ways that both increase and decrease emissions. Changes that are “ observed” include decreases in emissions that result from a household that sheds a car and drives less overall, as well as increases in emissions that result from a carless household driving more due to the additional vehicle access offered by carsharing. These impacts constitute changes that actually happened and are directly measureable. Through the remainder of the report, the authors call this the “ observed impact.”
Carsharing also provides an alternative to households that may substitute for actions that would occur otherwise in its absence. For example, a car owning household may join carsharing in substitute of acquiring an additional car. The vehicle that would have been acquired would have inevitably been driven some annual amount of miles for its forgone purpose. But a member of a household joins carsharing instead, which prevents this car from being acquired. Those miles and emissions never occur in the private vehicle because it is never purchased. Instead, miles to achieve the same purpose are placed on carsharing vehicles, and this alternative driving could be more or less than what would have happened, if carsharing were not available.
To consider impacts not manifested due to carsharing requires an additional level of abstraction. If a household joins carsharing and drives 1,000 miles a year instead of acquiring a private vehicle, and this vehicle would have also been driven 1,000 miles a year, then the net effect in terms of travel emissions would be close to zero ( a function of the different fuel efficiencies). The only change is the reduction in vehicle ownership that is now satisfied by a shared vehicle. Alternatively, if the household drives 1,000 miles a year in a carsharing vehicle, but would have driven 2,000 miles a year in a private vehicle, then the availability of carsharing prevents 1,000 miles from being driven and the corresponding fuel consumption from occurring. 21 Mineta Transportation Institute
Framework for Evaluating the Greenhouse Gas Impacts of Carsharing
19
The full impact accounts for new emissions that would have happened but do not because carsharing is available. Questions within the survey capture respondent estimates of this impact. The consideration of these additional non- manifested impacts, taken in sum with the observed impact is described in this report as the “ full impact.” It should be understood that although the full impact is a real impact associated with carsharing, it will always be subject to a greater degree of uncertainty. The full impact ascertains what “ would have happened otherwise” in carsharing’s absence. Respondents are asked to give a speculative answer with respect to the vehicles that they would acquire and the miles that they would drive on them. There is an elevated level of uncertainty associated with such stated responses. However, they are not entirely hypothetical either, as most people do have prior experience with driving distances based on previous travel patterns. For these reasons, the observed impact should be considered closer to a lower bound of carsharing emissions impact, whereas the full the impact is closer to the true impact. Throughout the report, the observed impact and the full impact are always presented separately, as there will always be a larger degree of uncertainty with respect to the measurement and precision of the full impact.
CARSHARIN
G IMPACTSIMPACTS AND SHIFTS IN TRAVEL MODES
A household that joins carsharing may use other modes more or less than before joining carsharing. Naturally, the household that joins carsharing and sheds a car will shift some of their travel to carsharing and may increase their use of public transit, biking, and walking for transportation. But the carless household that joins carsharing will drive more and use a car for trips that were previously accomplished with alternative modes.
Given these diverse shifts in travel behavior, it is important to consider how shifts to and from other modes would impact net GHG emissions. Some cases are simple. For instance, shifts to non- motorized modes, such as walking and biking, exhibit no increase in GHG emissions. With respect to public transit, the impact on GHG emissions is more complicated. Fixed rail and bus routes operate regardless of capacity utilization. Energy conservation does dictate that a single additional person switching to public transit has to increase GHG emissions by some marginal amount. As a person steps onto a bus or train, the transit vehicle must exert more energy than otherwise to move that person to his or her destination. However, because public transportation is traveling regardless of the presence of the additional passenger, a rider is only responsible for the marginal emissions caused by his or her presence on the bus or train. To provide some perspective, a typical empty bus in North America weighs about 40,000 pounds; hence, an additional 200 pound person increases the machine’s weight by only 0.5%. 22 The ratio is even smaller for a train. Because the contribution of an additional passenger contributes a small amount of marginal energy use, this study counts emission impacts of marginal public transit shifts to be negligible. Furthermore, if a trip has to be made within an urban region ( e. g., a commute), and non- motorized travel is infeasible for such a trip, traveling by public transit on an established network is the most efficient decision an individual can make from an energy and emissions perspective. There are circumstances that could arise in which a new route might be added to handle excess capacity. But the complexity of forecasting these long- term dynamics is outside the scope of this study. Mineta Transportation Institute
20 Framework for Evaluating the Greenhouse Gas Impacts of Carsharing
With emissions from motorized public transit minimal at the margin, the evaluation of GHG emission impacts attributable to carsharing is determined by the change in mileage traveled by private vehicles and carsharing vehicles. Prior to a member joining carsharing, this consists primarily of private vehicle mileage, but it also includes some local usage of rental cars ( as opposed to vehicles rented for travel in a distant city) and local taxis, if any. After joining carsharing, motor vehicle use is more complicated, consisting of personal autos that still remain in the household ( if any), carsharing vehicles, local rental vehicles, and local taxi trips.
This study collects vehicle VMT/ VKT measurements pertaining to automotive travel. The measurements are segregated by vehicle such that appropriate fuel economy factors can be applied to determine the gallons of gasoline consumed by each vehicle driven by household members. Once the total gallons of gasoline consumed by the household is known, the GHG emissions are computed using a standard methodology published by the U. S. Environmental Protection Agency ( EPA). 23 The EPA methodology was published to help establish a standardization of GHG analysis within the United States. The methodology accounts for the CO2 generated from gasoline combustion as well as trace emissions from other more potent GHG emissions, such as methane ( CH4), nitrous oxides ( N2O), and hydrofluorocarbons ( HFCs) from leaking air conditioners. The simplified estimation method assumes that these trace emissions account for 5% of the global warming potential produced by the combustion of a gallon of gasoline. This assumption includes the adjustment for the increased potency of these pollutants. 24 The EPA assumes that the average amount of CO2 produced by a gallon of gasoline is 8.8 kg ( 19.4 lbs.). 25 The total GHG potential from a gallon of gasoline is adjusted to account for other pollutants by multiplying CO2 emissions by a factor of 100/ 95. The adjusted GHG potential of a gallon of gasoline computed in this study is 9.3 kg ( 20.4 lbs.) CO2- e/ gallon. The CO2- e ( GHG) emission change that results from carsharing within a household is the difference between the annual travel emissions exhibited by the household during the year before joining carsharing and the annual travel emissions exhibited by the household at the time of the survey. Mineta Transportation Institute
21
SUR
VEY METHODOLOGY
The authors generated the study data from an online survey sent to carsharing members within organizations across the United States and Canada. There were two primary objectives pursued in the survey design. First, researchers needed the survey to collect enough data from the respondents such that GHG emission changes could be evaluated for the respondent households. Second, the survey design had to efficiently capture this information from carsharing members and ask questions that the respondents could reasonably answer, so as to maximize response rates and stay within the time tolerances of as many participants as possible. The survey took on average 15 minutes to complete.
The unit of analysis in the survey was the household, as an individual’s carsharing use can affect the travel decisions of all household members. There are several reasons why a household level analysis is more complete and appropriate than an individual level analysis, even if only one member of the household is a carsharing member. For example, an individual may join carsharing and shed their personal vehicle that they used exclusively. But another member of the household retains his or her vehicle, which is subsequently shared with the carsharing member when it is available. The vehicle belonging to the non- member within the household is driven more than previously because two people are using it. Another example could occur with vehicle switching. Consider a situation in which two working spouses each have their own vehicle. One spouse works in a downtown region, joins carsharing and switches to public transit for the commute. But because this spouse regularly drives the newer of the two vehicles, that vehicle is retained within the household and transferred to the other spouse, who requires a car to commute to work. The vehicle normally driven by the other spouse is shed, even though this person does not join carsharing. These and other situational permutations are plausible and require that the travel behavior of the entire household is assessed in order to evaluate how carsharing is influencing overall emissions. In addition, many organizations permit members of the same household to share a joint account. Joint membership plans permit multiple members of a household to use the same credit card, but they have unique membership IDs and otherwise operate independently. In addition, growth in carsharing business accounts adds an additional complication, as employers may cover a range of employee carsharing usage costs.
PARTICIPATINPARTICIPATINPARTICIPATINPARTICIPATIN
G
OR
GANIZATIONSATIONS
Researchers sent the Canadian and American respondents separate surveys due to the different distance and currency units used in the respective countries. As an incentive, each respondent was entered into a drawing for a $ 100 U. S./ Canadian credit to a member’s carsharing account. At least one member from each organization was selected as a winner. Additional incentives were drawn from the total respondent pool. A total of $ 2,200 credits were dispersed. The organizations that participated in the survey and are listed in Table 5. Mineta Transportation Institute
Survey Methodology
22
T
able 5 Participating Organizations
O
rganization
L
ocation
AutoShare
Toronto, Ontario, Canada
City CarShare
San Francisco/ Oakland, California
CityWheels
Cleveland, Ohio
Community Car Share of Bellingham
Bellingham, Washington
CommnuAuto
Montreal, Province of Quebec, Canada
Community Car
Madison, Wisconsin
Co- operative Auto Network/ The Company Car
Vancouver, British Columbia, Canada
IGo
Chicago, Illinois
PhillyCarShare
Philadelphia, Pennsylvania and Wilmington, Delware
VrtuCar
Ottawa, Ontario, Canada
Zipcar
United States and Canada
The organizations distributed the survey solicitations to their members through their own email lists. The email that the organizations sent out included the survey link. A third- party online survey program hosted the survey. Two reminders were sent out via each organization, and the survey closed on November 7, 2008. Most organizations, which are located in a single city, distributed survey solicitations to all of their members. Because of Zipcar’s size and geographic distribution, the solicitation was capped at a total of 30,000 randomly selected Zipcar members within specific markets. This included 5,000 each within New York City, New York; Boston, Massachusetts; Washington DC; Portland, Oregon; and Seattle, Washington. An additional 2,500 each in the Canadian cities of Vancouver and Toronto also received survey solicitations. In aggregate, the authors estimate that nearly 100,000 carsharing members received the survey solicitation. Based on the coverage, size and selection of this population, the authors consider it to be random and representative of the carsharing population within North America. The size of the membership base of each individual organization is proprietary information and cannot be reported. For similar reasons, it is not possible to compare demographics of respondents with demographics of the organizations. As with all surveys ( including the U. S. Census), respondents must consent to being surveyed and take the time to be surveyed. This injects some self- selection into the sample. However, in the case of this study, this self- selection applies to the propensity of the respondent to take an online survey. Among regular carsharing users, how this propensity is distributed is considered to be random. However, there is a cohort within the population that are carsharing members, but they do not use the service on a regular basis. This cohort, which the authors term “ inactive users,” are less likely to take a survey about a carsharing service that they use infrequently. As explained in more detail later, this cohort exhibits zero impact from carsharing, but their share of the sample is likely an underrepresentation. This has implications for the aggregate results that will be addressed in more detail within the sections that follow. In total, 9,635 surveys were completed, constituting a response rate of approximately 10%. Mineta Transportation Institute
Survey Methodology
23
THE
SURVEY QUESTIONNAIRE
The questionnaire began by soliciting basic parameters of the respondent’s membership. See “ Appendix” for the complete questionnaire. The survey asked for the year and month the member joined carsharing; this revealed the respondent’s membership tenure at the time of the survey. The survey also collected the pricing plan to which the member subscribed within their organization, as this determines their marginal cost of carsharing vehicle use. Following the collection of these basic parameters, the respondent was asked to characterize the circumstances in which they joined carsharing. These circumstances play a critical role in defining the nature of GHG impacts that would be expected from carsharing participation. The question and the circumstances listed in the survey appear in Table 6
T
able 6 Categorical Circumstances of Respondent Membership
Question: Please select the statement that best characterizes the household circumstances under which you joined carsharing.
A car of mine stopped working, and instead of replacing it I joined carsharing.•
I am in college, and I joined carsharing to gain access to a vehicle while in college.•
I live in an apartment bulding with a designated carsharing vehicle, and I joined through • its membership arrangement.
My employer joined carsharing, and I joined through my employer.•
My household did not have a car, but changes in life required a car and I joined • carsharing instead.
My household did not have a car, but joined carsharing to gain additional personal • freedom.
Owned at least one car, but needed an additional car for greater flexibility, and joined • carsharing instead of acquiring an additional car.
Owned more than on car. Got rid of at least one car and joined carsharing.•
Owned one car, but I joined carsharing and got rid of the car. •
I joined carsharing for reasons other than those listed above. Please explain:•
These circumstances are reflective of the transportation lifestyle that the respondent was leading prior to joining carsharing. They are succinct sentences that describe a specific situation pertaining to the role that carsharing serves for the household. These circumstances also capture the personal motivations for joining, which exist independent of personal demographics. Understanding member circumstances is important because carsharing can facilitate new travel patterns that fit with a household’s travel needs. For example, two households living in the same neighborhood could appear demographically identical with the household’s wage earners holding the same occupations. However, their travel patterns are dictated by their employment locations, which may require different transportation needs. Carsharing may effectively fit into the transportation lifestyle of one of the households, with commuters working in an area well served by public transit. Yet, the other household may have travel needs that cannot be effectively served by carsharing because an automobile is required to commute to one or more work locations. For this reason, the circumstances of joining carsharing are very important for classifying carsharing’s household impact. Mineta Transportation Institute
24 Survey Methodology
PERSONAL
VEHICLE DRIVING AND CARSHARING USAGE
Next, respondents were asked about the vehicles owned by their household. Two questions addressed personal driving. The first question asked about the number of vehicles owned prior to joining carsharing. Specifically, the question asked about the vehicles owned by the household during the year prior to joining carsharing. Researchers solicited the vehicle make, model, and year, along with an estimate of how many miles the vehicle was driven during the year immediately prior to joining. In a second question, researchers asked for the same information but pertaining to their current driving ( at the time of the survey). For all questions in which distance was relevant, American respondents were asked to think and respond in terms of miles, and Canadian respondents were asked to think and respond in terms of kilometers. For simplicity, the remaining methodological discussion is given in terms of miles.
To aid respondents in computing the annual mileage driven on each car, researchers provided descriptive text to walk the respondent through a rudimentary calculation that would produce a reasonable estimate. Respondents were given the option of following the calculation, if the annual mileage for a household vehicle was not a value immediately known ( see Appendix). The text also reinforced the idea that annual mileage on each vehicle was the desired response in contrast to odometer readings. Most respondents rounded their answers to the nearest thousand.
The make, model, and year of each vehicle were used to determine the fuel economy of the vehicle, which is required to estimate the gallons of gasoline consumed as result of a given mileage. Each vehicle dating back to 1978 was linked to an appropriate entry in the EPA fuel economy database. When a vehicle model had trims with two different engines sizes, the fuel economy of the smaller engine was applied. The combined fuel economy rating for each vehicle entry was applied to compute the gallons consumed, which could then be converted to GHG emissions. A small minority of vehicle entries was incomplete, as not all respondents knew the model name of the vehicle within their household. Typically such cases were accompanied with the year and vehicle make, absent the model name. For these entries, the average fuel economy for all passenger cars within the given year was applied as a proxy. Vehicles older than 1978 are not listed in the EPA’s fuel economy database; these vehicles were given a standard combined fuel economy of 15 miles per gallon. Motorcycles and scooters were also requested to ensure that all motor vehicle travel was accounted for; however, no public database currently holds certifiable fuel economy numbers for each model over time. There is an additional complication associated with the emissions of motorized two- wheeled vehicles. Scooters exhibit a wide range of environmental impacts. While scooters are often touted as fuel efficient (~ 90 mpg), the proliferation of two- cycle engines within leading scooter brands can result in a considerable degradation of emissions quality. 26 While four- cycle scooter models are growing in number, at the time of the survey, leading brands of new scooter vehicles could still be purchased with two- cycle engines. Motorcycles present similar emission problems in spite of elevated fuel efficiency relative to most automobiles. 27 Because of these issues with two- wheeled motor vehicle emissions, it is not representative of the true GHG impact to apply the nameplate fuel efficiency factors. As an adjustment, scooter vehicles and motorcycles were assigned a fuel economy factor of 30 miles per gallon as a proxy to Mineta Transportation Institute
Survey Methodology
25
account for the degraded emissions per gallon. This factor is close to the fuel economy implied by the CO2- e emission factor of motorcycles used for the EPA to generate the annual U. S. Greenhouse Gas Inventory Report. 28 While these vehicles received special consideration in the assignment of factors, they account for a small share (~ 5%) of all unique vehicles held by respondent households.
Following completion of personal driving questions, the survey asked respondents about their carsharing usage. Many carsharing organizations supply their members with monthly billing statements that provide miles driven, so the survey framed the carsharing questions to solicit information on monthly driving. To gauge usage, reservations per month and miles per month were solicited for all household members. Carsharing permits members to use a diversity of vehicles, and many members take advantage of this variety by using different vehicles throughout the year. However, many members will gravitate toward specific vehicles, often governed by the convenience of the “ point of departure” ( or pod) location that they access most frequently. Researchers asked respondents about the carsharing vehicle that they drive most often. This vehicle was used as a proxy factor for the efficiency of miles driven in carsharing vehicles. Specific efficiency factors were applied for the given make and model, but researchers did not expect the respondent to know the year of the carsharing vehicle that they drove most often. Most carsharing vehicles are relatively new, and fuel economy varies little from year to year for the same model. Hence, the year 2007 was assumed as a proxy for the carsharing vehicle model. Exceptions were made for vehicles that did not exist in 2007, such as the Toyota Echo used by a carsharing organization in Montreal. For these vehicles, the last year of production ( 2004 for the Echo) was applied as a proxy.
Not all respondents were comfortable providing the name of the vehicle that they used most often, and they were given an option to indicate this as a response. As a backup, these respondents were diverted to a follow- up question that asked about the general type of vehicle that they used most. General categories of vehicles were given as available responses, and an appropriate combined fuel economy factor was applied in the case of each possible answer. Table 7 illustrates the efficiency factors that were applied for each generic vehicle type. Mineta Transportation Institute
26 Survey Methodology
Table 7 Generic Vehicle Types and Assumed Fuel Efficiency Factors
RENTALRENTAL
V
EHICLES
AND TAXITAXI USAGE
Carsharing member use of rental vehicles and taxis could also contribute to GHG emissions, and carsharing can impact the degree to which a member uses either mode. In assessing carsharing impacts, only local trips in rental cars and taxis are important. Travel by these modes, which is initiated away from the carsharing member’s city of residence ( for example, in a distant city to which a person would have to fly), is outside the scope of carsharing impacts because such travel would occur regardless of a person’s carsharing membership in their hometown. Generating information for these two vehicle modes, however, posed unique challenges for the survey and the respondent. While carsharing and personal vehicle use is governed by annual lifestyle routines and regular travel, local rental car and taxi use is far more erratic. This makes recollection and accuracy more challenging for the respondent. Thus, researchers hypothesized that the overall net impact of changes for these two modes would be small. At the same time, researchers were also concerned about respondent survey fatigue because such questions can tax the respondent for small analytical gain. To address these concerns, a subsample of respondents was asked questions about their taxi and rental car use before and after joining carsharing. About 20% of each sample opted out of the question, stating that they did not know the mileage of one or both modes during the year before they joined carsharing or currently. Those that did offer complete responses provided researchers with a subsample to evaluate the range and distribution of mileage changes that occurred after carsharing. The results, presented later, show that the net changes in rental car and taxi use are very small and make an insignificant overall contribution to emission change among carsharing users.
SUPPORTIN
G DATAATAATA
Supporting data collected by the survey permitted researchers to characterize carsharing impacts in richer detail. Researchers collected demographic information at the end of the questionnaire, including location information ( e. g., home zip code in the U. S. and Canadian postal code). The location information permits an analysis of carsharing impacts by urban density.
Not surprisingly, a change in work or home location can seriously disrupt the imputed
  Mineta Transportation Institute
Survey Methodology
27
results from previous responses, and moving often coincides with many important life events. Nevertheless, some moves exhibit trivial impacts on overall automotive travel needs, while other moves induce significant impacts that are either positive or negative. Respondents that moved a home or work location were asked a follow- up question that prompted them to self- assess the degree to which their driving mileage change was a result of the move or due to carsharing. Specifically, respondents were asked: “ What would you say has contributed more to your overall change in driving? The move ( of home or work) OR the availability of carsharing?” There were five possible responses. Respondents who stated “ Mostly carsharing” or “ More carsharing than the move” were retained for the emission analysis. While respondents stating “ Equally carsharing and the move,” “ More the move than carsharing,” or “ Mostly the move” were dropped from the final analysis because their move to a new home or work played a significant part in the driving change. Due to the complexity of travel changes that can be induced by a significant move, the survey did not attempt to collect information to correct for the isolated impact of the move. Because many people are mobile in both home and work, the follow- up question was designed to preserve as many respondents as possible from being removed from the analysis as a result of this important confounding factor. A section detailing how the main results would differ had all movers been included or extracted is presented in a sensitivity analysis of the results.
DATAATAATA PREPARATIONPREPARATIONPREPARATION
Overall, the respondent was given a fair degree of freedom to compose responses within the survey. The data required careful attention to ensure that each survey was complete. Due to the University of California, Berkeley’s Human Subjects regulations, the survey was not permitted to force any answer of the respondent before proceeding. Respondents were free to skip answers to any question but still complete the survey. The data were filtered of records with extreme outliers or missing responses of key questions that would make individual calculations impossible. Responses filtered for any of these reasons are not included in the final analysis. In total, respondents completed 9,635 surveys across all organizations, and 6,281 are applied in the final analysis. The filtering of the data is discussed in this section, detailing who was removed and why.
The most prominent cause for respondent filtering was due to a household move. As explained earlier, a move can have significant impacts on overall mileage and many people move home locations or change jobs. The main motivation of this filter was to prevent GHG impacts that result primarily from a move of home or work to be attributed to the carsharing impacts. Respondents were asked whether they had moved their home or work location during their time with carsharing. If they had, they were asked a follow- up question regarding the nature of the move’s impact on driving mileage. Those indicating that the move had an equal or greater share of the responsibility than carsharing for mileage changes were dropped from the analysis. Among the 3,484 who indicated either a home or work move, 1,572 respondents were exclusively filtered from the analysis for indicating that the move was a prominent factor in altering their mileage driven.
The second most prominent cause for respondent filtering was due to carsharing use. The survey revealed that some respondents use carsharing very infrequently. A sizeable share of respondents clearly indicated that they use carsharing as a back- up travel option as Mineta Transportation Institute
28 Survey Methodology
opposed to a necessary component of their travel lifestyle. These members are referred to as “ inactive members,” which can exist in carsharing organizations with membership plans that have small or zero fixed annual cost. As such, households can hold memberships in case a spare car is needed, and low fixed- cost plans allow them to do this with little penalty. While carsharing provides them with a benefit in this respect, it would be challenging to argue that such members reduce their emissions due to carsharing because their travel lifestyle is manageable without it. 29 Researchers filtered a total of 488 respondents from the final analysis exclusively because they indicated no use of carsharing even though they were members.
A critical question asked of respondents pertained to household vehicle holdings and annual driving distances for each vehicle. Because of this question’s importance in evaluating the overall change in household GHG emissions, the survey offered guidance in advising respondents on how to calculate a good estimate of annual vehicle miles for a vehicle. If respondents did not already know the annual miles placed on their vehicles, they could follow the textual guidance to develop an estimate. 30 Under this design, a vast majority of respondents answered the question appropriately. Even so, some inevitably reported mileage numbers that were clearly odometer readings for the vehicle. Researchers removed these records from consideration in the analysis by establishing an upper bound on annual mileage. A conservative cutoff was chosen to implement the filter. Any respondent that reported an annual mileage larger than 30,000 miles per year for any vehicle was filtered from the analysis. This threshold was suggested by the data and by practical limits on annual driving. Annual driving distances greater than 30,000 miles per year are feasible but extraordinary. For example, the average annual distance driven by an American is 12,300 miles per year, and the average in Canada is 8,800 miles per year. 31 In total, researchers filtered 192 respondents ( 2% of all completed surveys) exclusively for stating annual driving distances that exceeded this established threshold. Because many of these high mileage drivers were driving such distances before they joined as opposed to after, their exclusion lowers the potential emission reduction exhibited by carsharing. To illustrate the impact of this cut- off on the results, a sensitivity analysis is later presented that explores the influence of this threshold.
As mentioned earlier, the focus of this study is on the impact that the neighborhood carsharing model on the GHG emissions of working households. There are two other submarkets in the carsharing industry that constitute smaller shares of the carsharing market. This includes the college submarket and the business use submarket.
A total of 632 university/ college students took the survey of which 349 were filtered exclusively because they were college students. The remainder also had other filters apply. The college market is not addressed in this study because the survey was not designed to simultaneously handle all of the nuances associated with college life.
University life is a dynamic time of frequent moving, as well as changes in roommates, employment, course schedules, and vehicles. University students often live in different cities and households during different times of the year as they go home for a break. It is a time when social objectives and travel lifestyle can be very different from one year to the next. Because of all the confounding variables associated with university/ college life, Mineta Transportation Institute
Survey Methodology
29
researchers did not design the survey to isolate these impacts. A separate study that is focused on this changing market is recommended.
Strict business use is another submarket of carsharing that was not addressable through the existing survey design. This filter was applied to respondents that used carsharing exclusively for business use. Respondents that used carsharing for both home and business use were retained because the neighborhood model still applied, and separate questions sorted respondents that were strict business users from home and business users. A total of 100 respondents were filtered exclusively for using carsharing solely for work- related trips.
As shown in Table 8 which lists the circumstantial categories that respondents could choose, an “ Other” category was provided in which respondents could write out the circumstances of their carsharing membership, if one of the given categories did not fit. With the “ Other” response, respondents could explain their circumstances, as appropriate. A total of 481 respondents that were not filtered for any other reason provided an “ Other” response. Researchers reviewed each of these responses, and most of them generally fell into the other categories provided. Relatively few ( 21) provided responses that suggested that they should not be included in the analysis. One common reason for removal was the circumstance in which the respondent actually lived in a city far from carsharing services. Many of these respondents were carsharing members so that they could use the service when they were in a city that they visited frequently ( such as when visiting a son or daughter).
Other exclusive reasons for filtering respondents had small effects on the usable sample size. This included 34 respondents that were filtered because they indicated that they did not know how far they drove in a carsharing vehicle and declined to give any estimate. Researchers eliminated another six responses because their estimate of carsharing mileage was far outside reasonable distances that would be traveled in any vehicle. The authors also designed the survey with particular questions to detect duplicate or redundant responses from households. This would occur if two members of a joint account took the same survey, duplicating the household activities. Several questions were used to construct a unique eight- digit ID that would match across household members but no one else. Researchers filtered a total of 16 responses because they were duplicated by two different people from the same household that took the survey.
Finally, the numbers discussed thus far describe respondents that were filtered for only a single reason. But a fair number of respondents were filtered due to some combination of reasons, including moving, non- use, outlier personal mileage or carsharing data, and unavailable carsharing use estimates. That is, if one filter was not active, then another filter would still have removed these respondents from the analysis. Researchers filtered a total of 576 respondents for some combination of reasons. The collective impact of the filters reduced the initial dataset for 9,635 to a core of 6,281 households. Table 8 illustrates how
the filter altered the balance of circumstantial responses by respondent share for both the complete and core sample. Mineta Transportation Institute
30 Survey Methodology
T
able 8 Balance of Circumstantial Responses Before and After Data Filters
For most circumstantial categories, the balance of respondents changes very little. The largest change in sample share is Category 4 in Table 8, which includes people who did not have a car and joined carsharing to gain additional personal freedom. This shift is in fact unfavorable for finding a reduction in GHG emissions for carsharing because this category consists of people who can only increase their “ observed” emissions as they were not driving prior to joining carsharing. Overall, the comparison shows that the data filtering process does not shift the circumstantial balance of respondents in other significant ways. Further discussion follows in the next chapter ” Results,” showing similar comparative results among the demographics of the complete and final dataset. A sensitivity analysis within the results section illustrates how the results vary according to key assumptions and respondent inputs, including an analysis of how the balance of results would change had certain filters not been active.
  Mineta Transportation Institute
31
RESULTSRESULTS
The survey results illustrate how carsharing interacts with different households in different ways, and the aggregate results show that carsharing generates a wide distribution of impact on personal annual GHG emissions. Across all respondents, carsharing facilitates decreases in annual emissions for some members and increases in annual emissions among other members. The authors found that on balance across all survey respondents, the net carsharing emissions are negative and statistically significant for both the observed impact and the full impact. GHG emissions from transportation are lower due to carsharing. The average change in emissions across all respondents is - 0.58 t GHG/ yr for the observed impact, and - 0.84 t GHG/ yr for the full impact. However, it is very important that the “ how and why” of this result is understood in the context of the broad diversity of carsharing impacts. While carsharing does facilitate lower emissions, this result is not generalizable across all members or even a majority of members. Rather carsharing as a system facilitates large changes in the annual emissions of some households, which compensate for the collective small emission increases of other households. This dynamic is important for the construction of sound policy, which can encourage carsharing growth in a manner that provides mobility benefits and continued emission reductions within urban and suburban regions.
DEMOGRAPHICS
Researchers logged a total of 9,635 completed surveys across the U. S. ( NUS = 6,895) and Canada ( NCAN = 2,740). Basic demographics of the respondent pool illustrate a diverse population using carsharing. Carsharing serves a wide diversity of household incomes, education, and age groups. In the following discussion, the authors present sample sizes ( N) within the figures to describe the demographics of both the complete and cleaned data. These will vary and be slightly less than the total survey population, as some respondents inevitably skipped or declined to respond to certain demographic questions. The demographics figures show the complete dataset ( Ncomplete = 9,635) as well as the final cleaned dataset ( Ncleaned = 6,281), which includes only those respondents who remained after all filters were applied. The purpose of the comparison is to show that the filter applications did not significantly alter the demographic mix of the dataset. The main differences include a slight shift toward older populations and commensurately a slight shift toward higher incomes.
The respondent age distribution shows that carsharing still remains relatively more popular with younger adults between the ages 25 and 40. The average age of all respondents was 36.6 years, with a median of 33 and mode of 28. Figure 4 illustrates the distribution of age groups among respondents. Mineta Transportation Institute
Results
32
F
igure 4 Age Distribution of Respondents
While the distribution shows that carsharing members are skewed toward the young adult demographic, there is considerable representation among older respondents. Both datasets show that at least a third of respondents are over 40 years old.
The income and education of respondents illustrates a similar level of diversity. Respondents were asked to provide their 2007 household income within $ 10,000 intervals denominated in their respective home currency. The intervals of $ 30,000 to $ 40,000 and $ 40,000 to $ 50,000 were selected with near equal frequency, but the remaining responses varied across a wide range of household income levels. Figure 5 illustrates the distribution of income and education levels among the respondents that answered the question.
The income response of all respondents in the U. S. and Canada are listed together. During much of 2007, the currencies of the two countries traded at near parity within a $. 20 range around 1, ( 1 USD = {. 95 to 1.15} CAD). Incomes during this time between the two countries were close to nominal equivalence. The median interval is $ 50,000 to $ 60,000, which indicates that nearly 50% of the respondents had household incomes greater than $ 60,000. Thus, carsharing is a service that is shared by a wide range of household incomes. In terms of education, the respondent distribution is skewed toward higher education levels. More than 80% of respondents hold at least a bachelor’s degree, and nearly 40% had completed some form an advanced graduate degree. Mineta Transportation Institute
Results
33
F
igure 5 Income and Education Distribution of Respondents
The size of respondent households tend to be smaller than average. The average household size in the U. S. is 2.6, whereas the average among all respondents was 1.9 persons. 32 This difference is in part driven by the fact that cities have smaller household sizes. The mode of household size is one, while the median is two. The gender balance of respondents is slightly dominated by females at 57% to 43% males.
CARSHARIN
G EMISSIONS IMPACTSIMPACTS
The respondent distribution for the change in annual household GHG emissions shows the wide diversity of GHG impacts exhibited by carsharing members. Carsharing members both increase and decrease their annual emissions, and the distribution shows that a majority of carsharing members are increasing their annual emissions. But across all 6,281 respondents, the results show that carsharing’s net effect in North America is a reduction in annual GHG emissions. As mentioned earlier, this average is - 0.58 t GHG/ yr for the observed impact, and - 0.84 t GHG/ yr for the full impact. The discussion that follows presents the dynamics of this result in more detail.
Figure 6 presents the distribution of annual emission impacts by respondent frequency for both the observed and full impact of carsharing. The horizontal axis define “ bins” of annual GHG change in metric tons of GHG per year ( t GHG/ yr), while the vertical axis defines the count of respondents within each bin. Mineta Transportation Institute
34 Results
 Figure 6 Distribution of Annual Household GHG Emission Impact
A striking feature of the distribution is the high number of respondents that exhibit an increase in annualized emissions within the bounds of 0 and 0.25 t GHG/ yr. The spike is evident within both the observed impact and the full impact. Members increasing their annual emissions by some amount under 0.25 t GHG/ yr outnumber the frequency of any other bin along the horizontal axis. Another notable feature is the distribution of members increasing their emissions, which follows an exponential trend of respondent frequency decline as the rate of annual emissions increases. This decline is far faster to the right of zero than it is to the left. The decline is rapid enough such that the frequency of respondents exhibiting a change of 1.25 to 1.5 t GHG/ yr ( n = 58) is smaller than the frequency of respondents altering their annual emissions by - 1.25 to - 1.5 t GHG/ yr ( n = 78) and for all bins extending to positive and negative infinity. The distribution of members lowering their emissions is far more evenly spread for both the observed and full impact. In total, 4,456 ( 71%) of respondents have a positive observed impact, while 1,825 ( 29%) have a negative observed impact. For the full impact, the balance is more evenly distributed by respondent frequency, as 3,281 respondents ( 53%) have a positive full impact while 2,953 respondents ( 47%) have a negative full impact.
The difference between the number of respondents decreasing their emissions in the observed impact and the full impact highlights the importance of considering the avoided emissions. These are emissions that would have occurred in the absence of carsharing but do not because carsharing is available. The resulting shift of the full impact reduces the number of members with impacts greater than zero. Absent any consideration of avoided Mineta Transportation Institute
Results
35
mileage, these respondents would appear to be increasing their net emissions through carsharing. Simply put, there exist some members of carsharing who would acquire a car and drive it some distance, but instead join carsharing. Because these emissions on the acquired car are never manifested, the observed impact calculation only shows an increase in emissions for this type of respondent. The full impact takes into account the offset of what would have happened otherwise.
The exponential drop in annual emissions to the right of zero suggests that those joining carsharing for access to automotive mobility do not drive much. To illustrate this point in more detail, Figure 7 presents the distribution of the annual miles driven by carsharing members for all respondents of the survey.
F
igure 7 Distribution of Miles Driven by Carsharing Members
Figure 7 shows that most households place very low annual mileage on carsharing vehicles. In theory, this suggests that households that transition from driving more typical distances in private vehicles into carsharing have the potential to impose considerable reductions in annual GHG emissions. The miles placed on carsharing vehicles by households are generally small. Nearly 30% of all households report placing less than 250 miles per year on carsharing vehicles. An additional 16% reported driving between 250 and 500 miles, and about 19% placed between 500 and 1,000 miles annually. In total, more than 80% of all households drive less than 2,000 miles per year on carsharing vehicles. Figure 7 shows that the potential increase in driving by carless households is generally small. The change in the distribution of personal vehicle miles traveled ( PVMT) illustrates how carsharing Mineta Transportation Institute
36 Results
simultaneously shifts overall driving in private vehicles. Figure 8 presents the distribution of the annual mileage placed on personal vehicles by households before joining carsharing and at the time of the survey. The mileage shown in Figure 8 is the total mileage across all vehicles held by the household during the given period.
 Figure 8 Distribution of Total Annual Personal Vehicle Miles Traveled by Household
Figure 8 shows that the majority of households joining carsharing drove zero personal miles. These are essentially carless households, and the only miles they drive are on carsharing vehicles. The “ before- and- after” shift in the PVMT distribution shows a significant gain in the number of carless households, an increase of nearly 30%. The distribution of annual household PVMT distances shows a general decline of households driving all distances. This does not mean that were no households reporting an increase in household PVMT, some did. But most households achieved the shift in mileage by eliminating at least one vehicle. From Figures 7 and 8 the derivatives of the unique shape of Figure 6 begin to become apparent. The large number of carless households that joined carsharing are now driving a little more, giving rise to the shape of the distribution to the right of zero in Figure 6. Households reducing their driving from a range of annual PVMT distances and vehicles create the long tail to the left of zero.
Figure 6 illustrates the GHG impact on the horizontal axis and the respondent count on the vertical axis; the majority of respondents are increasing their emissions in the full and observed impact categories. But the net impact of carsharing remains unclear, as the long tail of respondents reducing their emissions exhibits greater reductions with Mineta Transportation Institute
Results
37
greater distance from zero. Figure 9 presents the same aggregate distribution weighted by the annual emission change for respondents. Each categorical bin of the horizontal axis contains the summation of the annual change in respondent emissions. The result is a distribution that illustrates the cumulative net annual change in emissions for all survey respondents. The top graph in Figure 9 illustrates this distribution for the observed impact, and the bottom graph shows the full impact.
Figure 9 Profile of Cumulative Annual Change in GHG Emissions
The horizontal axis of Figure 9 is in the same units of Figure 6, and the respondents represented within each bin are exactly the same for both figures. The difference between Figure 9 and Figure 6 is that the vertical axis is the sum of the annual change in emissions ( in t GHG/ yr) of each respondent within each bin. Figure 9 graphically shows a clear perspective on the overall net change in annual emissions observed among all respondents. For both the observed and full impact, it is visually apparent that the area constituting emission reductions is larger than the area constituting emission increases. Thus, the results show that while the majority of respondents are increasing annual emissions, the cumulative emissions change for carsharing is negative.
It follows that the average emissions change across all respondents is also negative. The distribution of the sample population is not normal. The respondent distribution exhibits high kurtosis and is negatively skewed. However, the Central Limit Theorem ( CLT) and the large sample size establish the appropriate conditions for a paired t- test to evaluate the statistical Mineta Transportation Institute
38 Results
significance of the aggregate mean impacts. The CLT establishes that as sample sizes become large, the distribution of the sample mean converges to a normal distribution. 33 This permits the application of parametric statistical tests, such as the t- test, to determine the mean significance. This point can be illustrated with a technique called the bootstrap method. The bootstrap method applies computer simulation to replicate distributions of specific statistical moments when an analytical approach is difficult or intractable. For evaluating the mean, the bootstrap method simply draws a large set of respondents from the sample, computes the mean, stores the value, and repeats this many times. The stored mean values then constitute a simulated sample mean distribution. At a high number of draws, the simulated distribution converges to the actual distribution. Figure 10 shows an implementation of the bootstrap method using 6,000 draws from the sample of this study to compute the sample mean distribution using MATLAB. Figure 10( a) on the left, shows the simulated distribution of the sample mean for the observed impact; 10( b) on the right shows the same distribution for the full impact. Both distributions can be seen to resemble the shape of the normal distribution.
F
igure 10 Simulated Distribution of the Sample Mean of the Emissions Change
It can be seen from Figure 10 that both mean impacts are negative and statistically significant from zero. The results of a paired t- test of the aggregate mean impact is presented in Table 9. The null hypothesis is that the mean change in emissions is zero.
T
able 9 Paired t- Test: Mean Difference from Zero
 
While carsharing members are shown to have both positive and negative changes of annual household GHG emissions, the observed impact across all respondents is an average of - 0.58 t GHG/ yr/ household. The average full impact of - 0.84 t GHG/ yr is naturally further
  Mineta Transportation Institute
Results
39
away from zero, as it includes avoided emissions. In both cases, the collective magnitude of reductions by those decreasing their emissions outweighs the collective magnitude of those increasing emissions.
Table 9 shows that the mean impact of carsharing is statistically significant. The observed impact is contained within a 99% confidence interval of - 0.50 to - 0.65 t GHG/ yr, while the full impact is contained between - 0.76 to - 0.91 t GHG/ yr. Thus, the overall survey results indicate that carsharing has facilitated a net reduction in the annual rate of GHG emissions of members across North America.
Distributions of Subsamples by Circumstances of Membership
The aggregate carsharing impact is the composition of a far more complex and diverse set of relationships governing how individual households alter their emissions under carsharing. GHG emission changes arise from members joining under different circumstances and taking unique actions as they adjust to a lifestyle that uses carsharing.
The nuances within the aggregate distribution Figure 6 and Figure 9 become more apparent with a disaggregate analysis that illustrates the distribution of respondent subpopulations. Interestingly, the overall trends governing the aggregate responses are very apparent within the subcategories that describe the circumstances in which a respondent’s household joined carsharing. As outlined in Table 6, respondents were asked early in the survey to characterize as best as possible the circumstances in which their household joined carsharing. These circumstantial categories offer important insights as to which subpopulations drive the overall emissions change that is observed in aggregate. Figure 11 presents the distribution of emissions change for respondents who joined carsharing when a household vehicle stopped working. The units of the axes of Figure 11 and all subsequent figures in this section are the same as in Figure 6, with respondent counts on the vertical axis. The exact response selected by the respondent in the survey is listed at the top of each graph. Mineta Transportation Institute
40 Results
F
igure 11 Vehicle Stopped Working and Joined Carsharing
Figure 11 shows that a large share of respondents within this circumstantial category ( 86%) report reductions in annual GHG emissions. The reduction range is large, although a majority of respondents are reducing emissions between 0 and 6 t GHG/ yr. To put this range in perspective, a 25- mpg vehicle driven 15,000 miles per year would produce 5.6 t GHG/ yr. Reductions larger than 6 t GHG/ yr come from a minority of households that drove further distances or shed multiple vehicles. It is important to note that respondents within this category exhibit equal observed and full impacts. This is a function of the methodological calculation to prevent the full impact from being overstated. As respondents in this category are already shedding vehicles, the application of avoided driving factors stated by respondents would constitute a previous driving replacement. Thus, the application of avoided emissions to members of these circumstances would be double counting. For this and other categories in which a vehicle was shed, similar computational rules are followed.
Further examination of other circumstantial subsamples reveals more detailed insight into the nature of emission impacts exhibited by households that enter carsharing without vehicles. Figure 12 presents the graphs of two such categories in which households were carless prior to joining. The avoided emissions that generate the full impact are applicable for both categories as both respondent subsamples have no prior personal vehicle emissions to replace. Mineta Transportation Institute
Results
41
F
igure 12 Respondents Entering Carsharing Without a Vehicle
The shift in the distributions of annual change in GHG emissions illustrates the importance of capturing the latent effects. Nearly 35% of respondents using carsharing as an explicit substitute for vehicle acquisition would report higher emissions in the absence of carsharing. Similarly, for the broader population of members that joined carsharing for greater mobility, 26% suggest that carsharing is resulting in lower emissions than would otherwise occur. While it is clear that carless households joining carsharing are by- in- large increasing emissions as a result of their membership, the avoided emission impact that would occur otherwise is an important component offsetting this increase. Another key distinction of both distributions is the range of emissions change observed on both sides of zero. The changes exhibited by households that enter carsharing without a history of personal vehicle holdings are contained within a small range relative to the aggregate data. More than 90% of observed and full impacts are contained with +/- 2 t GHG/ yr, thus emphasizing that emission increments generated by carless households are small.
As a related circumstance of membership, carsharing can also serve as a means for car- owning households to avoid the acquisition of an additional vehicle that may become necessary within the household. Figure 13 illustrates the distribution of annual emissions impact for a circumstantial category in which the households may be both vehicle shedding and avoiding the acquisition of additional vehicles. Mineta Transportation Institute
42 Results
F
igure 13 Households Owning Vehicles but Avoiding Future Purchases
The distribution illustrates the combined effects of some vehicle shedding as well as the shift from the avoided impact. As with the aggregate distribution, a majority of respondents ( 59%) in this category are increasing their emissions according to the baseline impact. While their impact is overwhelmingly contained within the range of a 0 to 2.5 t GHG/ yr increase, the tail of negative emission changes extends much further. As indicated by the circumstances of the respondents, the avoided impact shifts emissions considerably, and the balance of change decreases emissions for a majority of respondents ( 57%).
Finally, Figure 14 illustrates the distribution of changes in emissions yielded by respondents that entered carsharing with vehicles that they subsequently shed. Both graphs within Figure 14 illustrate how households that drop vehicles after entering carsharing can exhibit large GHG emission changes per year. These changes, along with those in Figure 11 and Figure 13, drive much of the net reduction observed in the aggregate distribution.
Both distributions in Figure 14 are characterized by a significant majority of respondents reducing annual GHG emissions. Among multi- vehicle households shedding cars, 88% of respondents reduced emissions. Similarly, among single- vehicle households shedding cars, 93% exhibited a negative emission change. Figure 14 illustrates how a large majority of respondents reduced emissions by an amount less than 5 t GHG/ yr. A total of 73% of all vehicles shed were driven 10,000 miles or less. An additional 17% of all vehicles shed were driven between 10,000 miles and 15,000 miles per year. Thus, 90% of all vehicles shed were driven 15,000 miles per year or less. Although shed vehicles are not the only source of impact, the distribution of GHG impacts largely reflect the mileage distribution of shed vehicles as most of the respondents in both categories report reductions between 0 and 5 t GHG/ yr. Mineta Transportation Institute
Results
43
F
igure 14 Joined Carsharing and Shed Vehicles
The disaggregation of key categories within the aggregate distribution illustrates the underlying circumstances that drive carsharing impacts. Households that are reducing their observed emissions through carsharing are outnumbered seven to three, but those households are reducing their emissions by magnitudes that far outweigh the small increases in emissions that are incurred when carless households join carsharing.
I
mpacts from Changes in Local Taxi and Rental Car Use
As discussed in the methodology, the authors asked a respondent subsample questions regarding their local car rental and taxi use. The motivation for pursuing a subsample of respondents was based on concerns regarding respondent fatigue and limitations in respondent knowledge. The subsample results confirmed the hypothesis that local rental car and taxi use changes do not influence aggregate carsharing impacts. This is not to say that some people do not change their local taxi or rental car use due to carsharing. However, the average change is statistically indistinct from zero or negligible. As expected, a sizeable proportion ( 20%) of respondents within the subsample could not recall their local car rental or taxi use in the past. For those that could, the distribution of change in t GHG/ year is within a tight range close to zero. Figure 15 illustrates the impact distribution with respect to local taxi and rental car use. The majority of respondents reported zero change for both modes. Mineta Transportation Institute
44 Results
F
igure 15 Distribution of Change in GHG Emissions From Local Taxi and Rental Car Use
Figure 15 illustrates two distributions that reflect the reported change in mileage from using local taxi vehicles and rental cars before and after carsharing. The distributions show that the average impacts are small. The average GHG change from local rental cars is less than 0.01 t GHG/ yr and is statistically insignificant. The average GHG change from taxi use is - 0.0097 t GHG/ yr and is statistically significant. However, this is negative, suggesting that taxi use tends to fall after people join carsharing. Because the magnitude of these impacts is close to zero, they are negligible in influencing the overall carsharing impact.
SENSITI
VITY ANALYSISANALYSIS OF AGGREGATEATE EMISSION CHANGE
The results of the aggregate analysis are striking in that the mean observed and full carsharing impact are negative and statistically significant in spite of the fact that a majority of respondents technically increased their emissions due to carsharing. It is natural to wonder how much this result depends on the presence of households reporting very significant emission decreases. To show how this result varies with assumptions and data, this section presents a sensitivity analysis to illustrate how the mean and statistical significance of impacts change when the most influential observations are adjusted to dampen their impact on the mean. This section also explores how the results change if key filters, such as the removal of respondents that had moved, are altered. Mineta Transportation Institute
Results
45
T
he Passenger Vehicle Miles Traveled Filter
As mentioned earlier, the threshold of 30,000 PVMT per year was chosen as a benchmark for the upper bound of PVMT responses. Any respondent that indicated that a personal mileage exceeding 30,000 miles per year on any vehicle was filtered from the analysis. While this benchmark is somewhat arbitrary, it was chosen to prevent very high PVMT responses ( true or not) from drastically shifting the mean impact. This would result in a small number of respondents playing an outsized role in characterizing the average carsharing impact. But what if the maximum permitted PVMT value was higher, how would this affect the mean impacts? Figure 16 shows a sensitivity analysis of the observed and full impact mean were the maximum PVMT raised to 100,000 annual miles traveled.
F
igure 16 Sensitivity of Mean Impacts to PVMT Filter Threshold
The trend in Figure 16 shows how the results would vary had the PVMT filter been set higher. The error bars indicate the 99% confidence interval about the mean and the leftmost data points present the baseline analysis averages. The N values above indicate what the sample size would have been with the adjusted threshold. That is, as the threshold is raised, more respondents would be added to the sample indicating PVMT values at or below the threshold. This trend illustrates that carsharing would be evaluated as more effective in reducing emissions with a higher PVMT filter threshold. The filter does not discriminate between before or after responses of PVMT. If a respondent indicates a PVMT value above the threshold for any vehicle in the household before or after joining carsharing, the filter is activated. The trend with higher PVMT values is downward because the majority of newly included respondents were shedding cars. Thus, Figure 16 shows that the 30,000 PVMT Mineta Transportation Institute
46 Results
filter is conservative with the data. There are of course people who could driver longer distances legitimately. However with higher PVMT, it becomes more difficult to verify whether the respondent was accurate, mistakenly indicated an odometer value, or simply offered a gross overestimate. The problem with such ambiguities at high PVMT values as opposed to low PVMT values is that that they can shift the result more drastically. Thus, the conservative PVMT threshold of 30,000 mitigates this effect and prevents a small number of respondents with higher PVMT values ( true or not) from shifting the average carsharing impact significantly.
A
n Upper Bound on Personal Miles Traveled Responses
A similar sensitivity analysis evaluates the potential impact of overestimation of PVMT values on the aggregate results. In this section, we evaluate how the average aggregate impacts would change if the maximum allowable PVMT response ( a PVMT ceiling) is gradually lowered to zero. In this analysis, any respondent that indicates a particular vehicle was driven more than this upper limit has the value truncated to match the limit. For example, if the established limit is 20,000, then all responses within the final data set containing PVMT values higher than 20,000 are subsequently reset to 20,000. The aggregate observed and full impact is evaluated with these modified terms. Unlike the previous sensitivity analysis, the sample size remains the same at 6,281, as no additional respondents are trimmed from the data. This approach permits the acknowledgement of a respondent’s direction of emission change, but the magnitude of change is dampened as the PVMT ceiling is lowered. Simply put, the sensitivity analysis states to a respondent that: “ while you claim that you drove some Y annual miles prior to joining carsharing, we assume that you could have driven no more than X miles ( with X < Y), and we will evaluate your contribution to the aggregate impact under this assumption.” The sensitivity analysis incrementally adjusts the X of this statement and evaluates the resulting mean and statistical significance of the observed and full impact. As is apparent in Figure 6 on page 38, the spread of those reducing emissions is far wider than the spread of those increasing their emissions. This method of truncation mitigates the impact of those respondents reducing their emissions far more than it mitigates the impact of those increasing their emissions. Figure 17 illustrates this sensitivity analysis with the PVMT ceiling given on the horizontal axis, and the value of the mean impact given on the vertical axis. The blue and red X marks indicate the point of the mean at each max- PVMT, while the bar passing through the X indicates the 99% confidence intervals surrounding the given mean. Mineta Transportation Institute
Results
47
F
igure 17 Sensitivity Analysis of Carsharing Impacts Given PVMT Ceiling
The trends shown in Figure 17 offer some insights about the robustness of the aggregate carsharing impact. The shallow slope of the trends from 30,000 miles to 20,000 miles indicates that the respondents stating PVMT distances above 20,000 are not influential on the magnitude of the aggregate impacts. The mean aggregate impacts increase only gradually, and the confidence intervals of the means within this range overlap. From 20,000 to 10,000, the slope of the aggregate impact trend starts to increase more rapidly as a larger number of respondents have their PVMT levels reduced.
The mean observed impact is - 0.41 t GHG / yr at the 10,000 PVMT limit and is statistically different from zero. That is, if all respondents reducing their emissions by joining carsharing were permitted to drive no more than 10,000 miles per year prior to joining carsharing, the impact would still constitute an emission reduction that is statistically significant. In a more extreme case, the observed carsharing impact is still negative and statistically significant even if the PVMT responses of all respondents are restricted to be no larger than 4,000 miles per year. At a restriction of 3,000 miles per year, the mean of the observed impact turns positive but statistically indistinguishable from zero. When PVMT responses are restricted to 2,000 miles per year or less, those joining carsharing from carless households begin to dominate, and the observed impact becomes positive and statistically significant. The mean full impact is always negative and statistically significant regardless of the restriction.
The importance of this result from the sensitivity analysis merits further discussion. The driving patterns of carsharing members prior to joining are a critical input into the overall carsharing impact. If carsharing was entirely populated by people who were not driving Mineta Transportation Institute
48 Results
prior to joining, then the observed impact could only be positive, as carsharing would provide additional automotive access to people who were not driving before. Under such a hypothetical case, carsharing could only reduce emissions through the full impact ( i. e., where potentially higher emissions that would have occurred are displaced). However, the sensitivity analysis shows that even when hypothetical but significant restrictions are placed on the magnitude of emission reductions of respondents when joining