Quantifying Transit’s Impact on GHG Emissions and Energy Use—The Land Use Component (2015)

Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

56 A P P E N D I X A This appendix contains a technical summary of the statistical models used, including datasets, model forms, and specifications of the best-fit models developed. It will give the reader an in-depth understanding of the elasticities developed for application in the TCRP Project H-46 research. A description of the longitudinal analysis of development patterns in Portland is also included. The level of detail here will be of interest to a general audience. For statisticians and modelers, further detail on the model design is provided in Appendix B: Statistical Models in Depth. Model Comparison At the highest level, two types of models using two entirely distinct datasets were constructed for this study: • Urbanized area models were constructed to analyze land use and transportation “ecosystems” over a large number of urban areas. These models use variables that are quantified at the level of FHWA-defined urbanized areas. The models draw on aggregate data that describe the total or average travel, land use, socioeconomic, and transit characteristics of a given region. While these data do not provide specific information about individual travelers or fine-grained information about the areas surrounding transit stations, these data are readily available from national data- sets, which enabled the research team to analyze relationships for the nation as a whole. • Neighborhood models were constructed to compare land use and transportation ecosystems in transit-accessible and non-transit-accessible areas within individual cities. These models use variables that are quantified at a very fine-grained level: travel patterns and transit access for individual households, land use patterns for specific parcels, and urban design characteristics of neighborhoods. Because this type of data requires much more effort to collect, fewer urban areas are included. Data availability and data quality are inherent constraints for any statistical modeling exercise. By using two entirely different datasets and modeling approaches, the research team was able to cross-validate results, a unique benefit of this study. The models have different ways of looking at key aspects of the land use effect of transit, as follows: • Density—Development density is quantified as gross population density (total population/ total land area) in the urbanized area model. In the neighborhood model, development den- sity includes both population and employment and is calculated at a finer level—the 1⁄2-mile radius around each household. • Land use mix—Land use mix is not considered in the urban area model. In the neighborhood model, land use mix is calculated in two ways: (1) the balance between jobs and population in the local area and (2) an entropy value that quantifies the representation of residential, office, retail, and institutional uses. Key Results from Statistical Models

Key Results from Statistical Models 57 • Urban design—Urban design is not considered in the urban area model. In the neighbor- hood model, urban design is represented by the density of street intersections in the local area. Places with denser networks of streets are generally more pedestrian friendly. • Destination accessibility—Destination accessibility is not considered in the urban area model. In the neighborhood model, access to regional destinations is quantified as the percentage of regional jobs accessible to each household within 20 minutes driving and within 30 minutes on transit. • Transit systems—In the urban area model, transit systems are represented by four unique vari- ables: total supply of light rail, total supply of heavy rail, route density of transit (route miles/ land area), and transit frequency (total revenue miles/total route miles). Bus transit is included in the latter two variables. Each variable represents the entire transit system of an urban area. In the neighborhood model, transit access is characterized for individual households by two variables. One variable indicates whether the household has access to rail transit within 1⁄2 mile. A second variable, as described above under “destination accessibility,” measures the percent- age of regional jobs accessible to each household within 30 minutes on transit. This variable acts as a proxy for both the number of transit routes and the frequency and speed of transit available in a given neighborhood, since households with access to more and better transit service will generally be able to reach a larger number of jobs via transit. Use of Models Two important factors not included in the models are public support and the strength of the market for land development. As described in Section 4.6.1 of this report, previous research has dis- cussed the importance of these factors to the land use effect (Cervero et al. 1995, ITDP 2013). Since these aspects of urban development are not captured in the datasets used in this study, the research team used contextual clues to interpret the importance of these variables to the model results. The following subsections describe the urbanized area models and the neighborhood models in turn. The basic process for each began with data collection and verification. Next, the research team constructed a best-fit model for each dataset using statistical analysis software. The best-fit model is the series of equations that best explains the relationships between variables within the dataset, based on widely accepted statistical goodness-of-fit measures. Finally, the research team interpreted results from each model in terms of the land use effect of transit. This step includes adjusting assumptions about individual transit systems, using the model equations constructed, to see how the land use effect of transit is impacted. Urban Area Model The urban area model was constructed with a statistical technique called structural equation modeling (SEM), as described in Appendix B: Statistical Models in Depth. SEM was an ideal approach for this analysis because it allowed the research team to analyze how multiple variables both influence and are influenced by each other and to isolate the effects of a given causal path- way in the transportation and land use ecosystem. Model Description Figure 31 illustrates how an SEM model creates equations for multiple relationships (illus- trated by multiple arrows) to examine the influence of transit and land use on VMT and dis- tinguishes between different types of variables. Note that the model diagram for an actual SEM model is more complex, because the model may include multiple transit, land use, or control variables that influence each other.

58 Quantifying Transit’s Impact on GHG Emissions and Energy Use—The Land Use Component The primary purpose of the urbanized area model is to examine differences in travel behavior between urbanized regions that have experienced different levels and types of transit investment. The urbanized area models enabled the research team to answer the following research questions: • What is the total land use effect of an urban area’s existing transit system? • What is the likely additional land use effect within the urban area of incremental improve- ments in the transit system? Table 5 contains descriptions of variables and data sources of variables included in the final urban area model. The research team collected all variables for 315 urbanized areas for years Figure 31. Example SEM model of the effects of transit and land use on VMT. Category Variable Name Variable Definition Source Outcome variable vmt Daily VMT per capita FHWA Highway Statistics Transit variables tfreq Transit service frequency (annual revenue miles/route miles) National Transit Database rtden Transit route density per square mile (route miles/land area) National Transit Database tpm Annual transit passenger miles per capita National Transit Database hrt Directional route miles of heavy-rail lines per 100,000 population National Transit Database lrt Directional route miles of light-rail lines per 100,000 population National Transit Database Urban form variables popden Gross population density (in persons per square mile), excluding rural census tracts with fewer than 100 persons per square mile U.S. Census Control variables pop Population (in thousands) U.S. Census inc Annual per capita income American Community Survey flm Freeway lane miles per 1,000 population FHWA Highway Statistics olm Other street lane miles per 1,000 population FHWA Highway Statistics & NAVTEQ fuel Metropolitan average fuel price (in 1982 dollars) Oil Price Information Service Table 5. Variables and data sources of variables included in the urban area model.

Key Results from Statistical Models 59 2000 and 2010. Variables tested but not ultimately included were average transit fares and vehi- cle revenue miles, both derived from the National Transit Database. The latter is incorporated in the transit frequency variable. Figure 32 shows the best-fit model that illustrates the relationships among the variables pre- sented in Table 5. Causal pathways associated with the land use effect are highlighted with the solid blue line. Higher route densities and higher transit frequencies are associated with higher population density, and higher population density is in turn associated with lower VMT per capita. Causal pathways associated with the ridership effect of transit are highlighted with a dashed green line. Higher route densities and higher transit frequencies are also associated with higher transit passenger miles, which is in turn associated with lower VMT. Other model forms were tested but ultimately rejected because they did not produce as good a fit with the model data. The research team also tested models using only a subset of urban areas to determine whether relationships among key variables were different in cities that have rail versus cities that do not and in urban areas of different sizes. These models were rejected for having sample sizes that were too small. Figure 32. Best-fit model for the relationships among transit, land use, and VMT in urbanized areas. Land use effect Ridership effect

60 Quantifying Transit’s Impact on GHG Emissions and Energy Use—The Land Use Component Results Table 6 includes elasticity values for the key variable relationships making up the land use effect and the ridership effect. An elasticity represents the percentage change in one variable associated with a percentage change in another variable in the model. For example, the elasticity of VMT per capita with respect to population density is -0.238. This means that a 1% increase in population density is associated with a 0.24% decrease in VMT per capita. This result is consistent with the literature on the topic, given that other urban form variables that have an impact on VMT (land use mixing, urban design, and destination accessibility) are not accounted for in the model (Ewing et al. 2008). A 1% increase in transit passenger miles per capita is associated with a 0.02% decrease in VMT per capita. This result makes intuitive sense when one considers the scale of transit travel relative to car travel. Only 4% of all trips are made by transit in the United States. In contrast, 84% of trips are made by driving or riding as a passenger in a private vehicle.14 The quantity of 1% of transit passenger miles is thus far smaller than 1% of VMT. The elasticity can be interpreted as indicating that roughly one out of every two or three trips made on transit replaces a car trip. Table 6 shows the land use effects of the two transit variables or, in other words, the elasticity of VMT per capita with respect to the transit variables, following the land use effect pathways. The final values are derived by multiplying the elasticities along each pathway. The land use effect of a 1% increase in route density is a 0.047% decrease in VMT per capita. The land use effect of a 1% increase in transit frequency is nearly the same, a 0.045% decrease in VMT per capita. The model presented shown in Figure 32 is based on “logged” versions of key variables, mean- ing that the natural log of each variable was the model input. This type of model best answers the question of how incremental improvements in transit systems will change the land use effect. To examine the land use effects of existing transit systems, a similar model was constructed using variables that were not log transformed. The specifics of that model are provided in Appendix B: Statistical Models in Depth. The urban area model provides strong evidence of a land use effect of transit at the regional scale, based on the regional characteristics of more than 300 urban areas. Both expanding the transit network and increasing transit service frequencies are associated with higher overall gross regional densities and therefore with lower VMT per capita. However, densities can vary sub- stantially within a region. In order to examine the land use effect of transit at a finer scale, models using more detailed datasets are required. Neighborhood Model A neighborhood model was constructed in order to examine the land use effect of transit at a finer scale. Whereas the urban area model was constructed by comparing whole regions to one another, the neighborhood model was constructed by comparing neighborhoods to one Table 6. Land use effect elasticities derived from the urban area model. Transit Variable Land Use Effect (Elasticity of VMT) Route Density −0.0469 Transit Frequency −0.0445 14 2009 NHTS. Includes all buses, trains, streetcar, and trolleys. Excludes taxicabs.

Key Results from Statistical Models 61 another, where neighborhoods are distinguished primarily by their level of access to transit. This modeling exercise required collecting highly detailed data on neighborhoods and households within a handful of cities. Model Description The statistical analysis technique used is called multilevel modeling (MLM) or hierarchical modeling (HM) and is explained in detail in Appendix B: Statistical Models in Depth. To con- struct the model, the research team needed to compare the land use patterns and transportation characteristics of neighborhoods with varying levels of transit service to one another. In order to compile a sufficient sample of neighborhoods, data from multiple cities had to be used. How- ever, comparing a neighborhood in one city to a neighborhood in a different city introduces complications, since each city has its own unique regional transportation and land use charac- teristics that impact its neighborhoods. Neighborhoods located within the same region are more likely to have similar travel patterns. And a transit-oriented neighborhood in the Washington, D.C., area, which has an extensive regional transit system, may have higher transit ridership than the identical neighborhood would if it were located in greater Houston, which does not have such an extensive transit network. MLM allows the analyst to separately analyze sources of varia- tion both between regions and within regions. In the case of this research, the research team was most interested in the sources of variation within regions, from neighborhood to neighborhood. Controlling for regional sources of variation makes it possible to use data from multiple regions to inform the comparison of neighborhoods to one another. Figure 33 illustrates this relationship. Without MLM, there appears to be no relationship between the urban environment of a household and its VMT. With MLM, households are grouped into regions. In this hypothetical scenario, MLM reveals that different regions tend to have higher per capita VMT than others and that the relationship between urban environment and VMT at the neighborhood level is in fact relatively constant, as shown by the identical slopes of the black lines representing regions. The neighborhood model augments the results of the urbanized area model described in the previous section in several key ways. It incorporates local variations in land use patterns and Source: Adapted from “Introduction to Multilevel Modelling” (University of Bristol Centre for Multilevel Modelling 2011). Figure 33. Example illustration of how MLM applies to the neighborhood level model.

62 Quantifying Transit’s Impact on GHG Emissions and Energy Use—The Land Use Component travel patterns and includes both population and employment densities. The neighborhood model also explicitly considers more land use characteristics: land use mixing, pedestrian envi- ronment, and job accessibility. As a direct result of these advantages, the data collection burden for an individual region in the neighborhood model is orders of magnitude higher than that for an individual region in the urbanized area model. For this research, data for the neighborhood model were collected from nine regions. To be incorporated in the modeling exercise, each region required a household travel survey and, from the same year as that survey, parcel-level land use data, a detailed model of the transit network, and travel time skims from the regional travel demand model. The research team was able to gather data from nine regions (listed in Table 7). The regions are diverse in their travel and land use characteristics. Average daily household VMT ranges from 21 in Boston and Eugene to 40 in Sacramento. The average activity density in the 1⁄2-mile area surrounding each household ranges from a low of 2,500 in Kansas City to a high of 23,000 in Boston. Table 8 provides the full list of variables used in the neighborhood model. Using the neighborhood dataset, the research team constructed a series of interrelated models to explain the relationship among transit access, land uses, and travel patterns at the neighbor- hood level. One model explains the impact that transit service has on local densities. Other models explain the relationship that local densities and urban form have on travel patterns. Linking these models together allowed the research team to quantify the land use benefits of transit. A complete description of the model specifications is provided in Appendix B: Statistical Models in Depth. The conceptual framework used in the neighborhood model is very similar to that of the urbanized area model, although the specific variables used are different. Figure 34 illustrates the model theory. Causal pathways associated with the land use effect are highlighted with a solid blue line. Rail access and higher employment accessibility by transit are associated with higher population density, and higher population density is in turn associated with lower VMT per capita. Causal pathways associated with the ridership effect of transit are highlighted with the dashed green line. Rail access and higher employment accessibility by transit are also associ- ated with higher transit passenger miles, which is in turn associated with lower VMT. Other Average Daily Household VMT Average Activity Density (Jobs + Population per Square Mile) Austin 37 8,678 Boston 21 22,966 Eugene 21 5,009 Houston 39 5,549 Kansas City 27 2,451 Portland 27 4,364 Sacramento 40 7,321 Salt Lake City 23 7,637 Seattle 30 8,745 *Averages are for the metropolitan-planning-organization-designated modeling region Table 7. Travel and land use characteristics of cities used to derive the neighborhood model.*

Key Results from Statistical Models 63 Category Symbol Definition Level Primary outcome variable vmt Household daily VMT Household Intermediate outcome variables ttrips Household daily transit trips Household actden Activity density within 1/2 mile (sum of population and employment divided by gross land area in square miles) Household Exogenous transit variables emp30t Proportion of regional employment accessible within 30-minute travel time via transit (in-vehicle time only) Household rail Rail station within 1/2 mile (dummy variable; yes=1, no=0) Household Exogenous built environmental variables jobpop Job-population balance within 1/2 mile of a household (index ranging from 0, where only jobs or residents are present within 1/4 mile, to 1, where there is one job per five residents) Household entropy Land use mix within 1/2 mile of a household (entropy index based on net acreage in different land use categories that ranges from 0, where all developed land is in one use, to 1, where developed land is evenly divided among uses) Household intden Intersection density within 1/2 mile of a household (number of intersections divided by gross land area in square miles) Household int4way Percentage of four-way intersections with 1/2 mile of a household (four-way intersections or intersections where more than four streets meet divided by total intersections) Household emp20a Percentage of regional employment accessible within a 20-minute travel time via automobile Household emp30a Percentage of regional employment accessible within a 30-minute travel time via automobile Household Household control variables hhsize Number of household members Household employed Number of household members employed Household income Household income (in 1,000s of 2012 dollars) Household Regional control variables rpop Total regional population (in 1,000s) Regional remp Total regional employment (in 1,000s) Regional ract Total regional activity (sum of population and employment in 1,000s) Regional rind Regional compactness index (index measuring compactness vs. sprawl based on a combination of four factors that measure density, land use mix, degree of centering, and street accessibility); higher values signify great compactnessa Regional aFor more information on the regional sprawl index and how it is calculated, see Measuring Sprawl and Its Impact (Ewing, Pendall, and Chen 2002). Table 8. Category, definition, and scale of variables included in the neighborhood level model. “D” variables15 including intersection densities, job-population balance, and land use mixing have measurable effects on both activity densities and transit ridership. Results Table 9 provides the elasticities of VMT with respect to urban form variables often studied in the literature, as determined by the best-fit neighborhood model. A 1% increase in activity 15 Density, diversity of land uses, design, destination accessibility, and distance to transit. See Footnote 4 or Ewing and Cervero (2010) for more information.

64 Quantifying Transit’s Impact on GHG Emissions and Energy Use—The Land Use Component density is associated with a 0.11% decrease in VMT. This value is somewhat higher than typi- cal values from the literature, but lower than the elasticity of VMT with respect to population density found in the urbanized area model. The latter discrepancy makes sense, given that other “D” variables are controlled for in the neighborhood model that are not controlled for in the urbanized area model. Since “D” variables such as population or employment density, land use mixing, and intersection density are often correlated, we would expect the elasticity of popula- tion or employment density to be higher when other “D” variables are not explicitly incorpo- rated into a model. Thus, the elasticities of VMT with respect to density from the urban area and neighborhood models are roughly consistent. The elasticities of the two land use mixing variables, job-population balance and entropy, are in the range of -0.03 to -0.04. The elasticities of the street network variables are in the range of -0.09 to -0.10. The elasticity of regional employment accessibility is -0.10. Since higher activity densities are typically associated with denser street networks, better land use mixing, and better employment accessibility, it makes sense to use a higher elasticity of VMT with respect to density. In order to be consistent with the urban area model, an elasticity of -0.24 was used. This puts the effect of land use on VMT within the range of values used in the literature. Figure 34. Conceptual model for the relationships among transit, land use, and VMT in neighborhoods. Urban Form Variable Elasticity of VMT with respect to variable Activity density (population and employment divided by land area) −0.112 Job-population balance −0.037 Entropy −0.032 Intersection density −0.102 Percentage of four-way intersections −0.088 Percentage of regional employment accessible within 20 minutes by automobile −0.104 Table 9. Elasticities of VMT with respect to key urban form variables.

Key Results from Statistical Models 65 The best-fit neighborhood model finds two key transit variables that impact the land use effect: rail access and employment accessibility via transit. Both are intuitive components of the land use effect. Rail access is associated with higher activity densities, as can be observed in many urban areas where rail stations are surrounded by dense development. Transit employ- ment accessibility measures the percentage of regional employment that is accessible within 30 minutes from the closest transit station (excluding access times). The best-fit model finds that activity densities tend to be higher in places that have better employment access. This phenom- enon is born out in studies of newer transit-oriented developments, which find that proximity to downtown and other job markets has an important impact on the ability of transit station areas to attract development (CTOD 2011). Table 10 shows land use effects of transit predicted by the neighborhood model. Adding a rail station to a neighborhood that does not currently have rail accessible within 1⁄2 mile is associated with a density increase of 9% and a drop in VMT due to the land use effect of 2%. Increasing transit employment accessibility by 50% (for example, increasing the percentage of regional jobs accessible within 30 minutes from 20% to 30%) is associated with a density increase of 32% and a drop in VMT due to the land use effect of 8%. To validate these results, the research team applied the findings from the urbanized area model to evaluate the impact of adding a single rail station to a given urban area. For each urban area that currently has a rail system, the research team increased transit directional route mileage by 4. The research team made the following assumptions: rail stations are spaced 2 miles apart and thus the ratio of directional route miles to stations is 4:1; new service on the route would be equivalent to 60 trains per day in each direction (4 trains per hour for 10 hours and 2 trains per hour for 10 hours); and all density changes in the region as a result of the new transit service would occur within the immediate area of influence of the rail station, defined as a 1-mile catchment area around the rail station. On average, for all cities that currently have rail systems, the urbanized area model predicts a 17% increase in population density around the new rail station.16 This result provides a strong cross-validation of the urbanized area model and the neighbor- hood model, since it is expected that the urbanized area model captures broader regional density changes than the neighborhood model, which only captures density changes within 1 mile of a rail station. If all density changes due to the new rail station are confined within the 1-mile catchment, the amount of new population and number of jobs would be about double that usu- ally seen within the 1-mile catchment (17%/9% = 1.9). This suggests that the immediate station area accounts for approximately half of the expected regional increase in population and jobs. Longitudinal Analysis of a Portland Light-Rail Line Similar to the urbanized area model discussed above, the main neighborhood model con- structed is a cross-sectional model. That is, the model explains variation in land use patterns Transit Variable Land Use Effect (Activity Density Increase) Land Use Effect (VMT Decrease) Rail Station Accessible within ½ mile 9% 2% Transit Employment Accessibility Increases by 50% 32% 8% Table 10. Land use effects derived from the neighborhood model. 16 This is the population-weighted average for all cities that currently have rail.

66 Quantifying Transit’s Impact on GHG Emissions and Energy Use—The Land Use Component according to key transportation factors using a snapshot in time. The model is agnostic on the subject of the time it takes for development patterns to change in response to transit networks. The datasets used in the neighborhood model provided an opportunity to conduct a parallel longitudinal analysis for a single city, Portland, where the research team was able to obtain datasets for two different years, 1994 and 2011. The 2 years of data reveal empirically observed changes in land use patterns over a 17-year period. Using models similar to those constructed in the cross- sectional analysis, the research team could study the relationship of changes in the transit network to changes in land use patterns. A quasi-experimental pretest-posttest study design was used. This research design required the research team to select a specific corridor that received a transit investment between the pretest year (1994) and the posttest year (2011). A control corridor that had comparable land use, transportation, and demographic patterns in the pretest year was also selected. Changes observed in the transit investment corridor and the control corridor during the study period were then compared. The research team selected the Westside LRT line (western portion of the Blue Line) as the transit investment corridor. The portion of interest starts west of downtown Portland and extends through Beaverton out to Hillsboro. The 15-mile section, with 17 stations, opened in 1998, after the first study year survey and well before the second. Much of the alignment is through land that was ripe for development or redevelopment. Station areas have had many years to densify and thereby affect travel behavior. The control corridor is another corridor heading southwest from downtown Portland to Tigard and beyond. This is a highway corridor, in contrast to the treated corridor, running along the SW Pacific Highway and (for the first few miles) Interstate 5. This portion of the corridor is 12.5 miles long and has 14 interchanges or major intersections. In order to capture sufficient households to generate statistically valid results, the research team analyzed households living within 2 miles of the new Blue Line stations and households living within 3 miles of the major intersections in the control corridor. Density changes were measured within a 1⁄2-mile radius of each household. The effective geography analyzed for each corridor was therefore a 2.5-mile catchment area. Further detail about the study corridors and the experimental techniques applied is provided in Appendix B: Statistical Models in Depth. With the comparison highway corridor as a baseline, Portland’s Westside LRT extension is associated with an increase in activity densities within the 2.5-mile catchment area of 24% and an increase in average daily transit trips per household of 60%. These changes correspond to a 6% household VMT reduction due to the land use effect and an additional 8% VMT reduction due to the ridership effect. The research team validated these results in comparison to changes in density predicted by the urbanized area model. Adding 30 new directional route miles with service of approximately 60 trains per day in each direction to the Portland region is expected to increase total regional population density by 0.4%. If the population growth is confined to the 2.5-mile catchment area around the transit corridor, densities in the corridor area would increase by 6%. The observed increase in activity densities of 24% demonstrates the high degree of variation in the land use effect of individual transit investments. The 6% estimate from the urbanized area model rep- resents an average response in land use patterns without regard to key determinants, including public support and land potential. The Westside LRT corridor identified for this test had both many sites ripe for redevelopment and one of the highest levels of government support for TOD of any city in the country. The result of these factors was an increase in densities four times that of the average seen in U.S. cities.