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.
91 B.1 Demographic Sector The stock variable is the number of people in a regional population of interest, and it is seg- mented into dimensions. These dimensions were selected for their strong relationship with travel behavior, based on the knowledge of the research team and Tasks 1 and 2 findings, documented in the project memoranda. Taken simultaneously, these dimensions represent the current state of a regional population at a given point in time. For all, 2000 Census data were used to derive the starting demographic estimates of marginal distributions within each region, and 2010 Census or 2005â2009 American Community Survey (ACS) data are used for validation. ⢠Age cohort: Six categories: 0â15, 16â29, 30â44, 45â59, 60â74, 75 and older. Each cohort is roughly 15 years in durationâshort enough to capture the main variations in life cycle and behavior, but long enough to avoid many different cohorts. ⢠Household structure: Four categories: single without children, couples with children, single with children, and couples without children. A âcoupleâ is defined as either married or cohabiting partners (but not, for example, two unrelated adults). ⢠Acculturation group: Three categories: foreign-born with less than 20 years in the United States, foreign born with 20 or more years in the United States, and native born. The threshold of 20 years in the United States for foreign born was selected to distinguish âacculturatedâ from ânonacculturatedâ residents. ⢠Race/ethnicity: Four categories: white/other, Asian, black, and Hispanic. ⢠Workforce status: Two categories: participating in the workforce and not participating in the workforce. This includes those who are employed or looking for employment. ⢠Household income: Three categories in 2009 dollars: lowest quartile ($0â$34,999), middle two quartiles ($35,000â$99,999), and highest quartile ($100,000 up). The middle two quartiles are grouped, as they tend to show fewer differences in behavior than the more extreme ones. ⢠Residence area type: Three categories: urban (central city), suburban, and rural areas. Our base condition for the area types in each of the regions is derived from 2000 Census dataâat the tract level. For each tract comprising the metropolitan statistical area (MSA), we identified the number of jobs per square mile and the number of residents per square mile. Urban areas are defined as having at least 4,000 jobs per square mile, or at least 10,000 residents per square mile inside the tract. Suburban areas are defined as having at least 500 jobs per square mile or 1,000 residents. Rural areas are defined as having less than 500 jobs or 1,000 residents per square mile. A P P E N D I X B Impacts 2050 Model Structure Documentation NOTE: These area type definitions were chosen to roughly match the Claritas PRIZM area type categories (Urban, Suburban and Second City, Town and Country) that are used for data sets, such as the National Household Travel Survey (NHTS).
92 The Effects of Socio-Demographics on Future Travel Demand Running the SD model for a specific region requires the initial distribution of the population along all these variables simultaneously. With the categories above, that requires values for 6 à 4 à 3 à 4 à 2 à 3 à 3 = 5,184 different combinations, or âcellsâ in a multidimensional matrix. We have kept the number of socio-demographic stock variables and dimensions as concise as pos- sible, to constrain the number of cells in order to facilitate rapid analyses of many scenarios. As noted in the objectives section of this report, âstrategic models are fast models.â Currently, the simulation of the population model runs in a matter of seconds. We did not have access to the necessary micro-data from the Census to be able to run the multi dimensional matrix for a given region (i.e., age à household structure à acculturation group à race/ethnicity à workforce status à household income à residence area type). So we applied an iterative proportional fitting (IPF) technique to derive a multidimensional matrix for each region. IPF is a procedure for adjusting a table of data cells, such that they add up to selected totals for columns and rows in the table. The data cells are referred to as the âseedâ cells, and the selected totals are referred to as the âmarginal totals.â First we developed a table that would serve as the âseedâ cells. We used the NHTS, 2009 (for which we did have access to the necessary micro data) to develop a national multidimensional matrix. In the toolâs Excel spreadsheet, in the tab âDemographic seed matrix,â this simultaneous distribution of the national population is displayed. However, to run the model for a given region, we needed to transform this national matrix into one that is representative for a specific region. We had the marginal totals for all of the socio-demographic variables in the model from the Census. These distributions can be viewed in the Excel spreadsheet, in the tab âDemographic initial values.â For example, the marginal distribution for Atlanta by age category based on the 2000 Census is shown here: ⢠Age 0â14 = 955,906 people ⢠Age 15â29 = 941,083 people ⢠Age 30â44 = 1,135,495 people ⢠Age 45â59 = 758,505 people ⢠Age 60â74 = 313,953 people ⢠Age 75+ = 143,038 people ⢠Total = 4,247,980 people B.1.1 Demographic Rates of Change The guts of the socio-demographic model are the assumptions that define how the people in a region will transition over time between the various categories of socio-demographic variables. The rates of change define how the population will transition from one âstateâ to another. Two of the rates are structural, as they depend only on the passing of time: ⢠Ageing: Transition from one age cohort to the next. This is completely structural and is not affected by other variables in the model (endogenous or exogenous). Our model assumption is that with 15-year cohorts, each year 1/15th of the people survive age transition to the next cohort. NOTE: Currently, the model ages the population in the aggregate from one age cohort to the next. It does not keep track of the age distribution within any age cohort. This could be an enhancement built into model at a later date, or could be addressed by making the age cohort duration shorterâ5 or 10 years instead of 15.
Impacts 2050 Model Structure Documentation 93 ⢠Acculturation transition: Peopleâs race/ethnicity and birthplace (foreign or native) do not change during their lifetime. Thus, the only transition that applies to acculturation is related to how long foreign-born people have lived in the United States. Our model assumption is that each year 1/20th of foreign-born population transitions to the âgreater than 20 yearsâ category. Rates of change other than aging and acculturation have been derived from the PSID. The very significant aspect of using the PSID data to derive the rates of change is that we were able to derive individual-level rates of change. Therefore, we were able to link specific rates of change to the indi- vidual categories comprising each socio-demographic variable. There is no other data set from which this information can be derived. To derive the rates of change used in the SD model, we focused on the PSIDâs three most recent pairs of wavesâ2003 versus 2005, 2005 versus 2007, and 2007 versus 2009âand tabulated the rates at which specific transitions were observed to take place over the two-year intervals, as a function of which categories the persons fit into in the prior year (age group, household type, and ethnicity). The resulting rates were divided by two (to transform from two-year intervals to rates per year), and were used to inform the rates used in the SD model. The resulting rates can be viewed in the Excel spreadsheet in the tab âDemographic transition rates.â ⢠Births: Births are generated from people in the cohorts of childbearing ages (16â29, 30â44). The model does not treat males and females as separate groups, so the rates used in the model are about half of what they would be for females only. In addition to age group, there is some variation in birth rates, depending on ethnicity/acculturation group and household type (prior to the birth). For example, birth rates are highest among the âforeign born/in U.S. less than 20 yearsâ group, and are substantially higher for those living as couples than for singles. Each new birth creates a new person for the simulation, and this person automatically enters the model into the â0â15â age cohort, the âwith childrenâ household structure, and the ânot in workforceâ workforce status. Household structure (single/couple), income group, and residence area type match those of the parent(s). Children of a foreign-born parent become U.S.-born/nonwhite or Hispanic ethnicity. ⢠Deaths: Deaths can occur in all age groups, although the death rates are very low for the lower age groups. No significant differences between ethnic groups or household types could be found from the PSID data, because the number of observed deaths in the sample was (fortunately) very small. However, there is evidence that seniors who are part of couples tend to live longer than those who are single, so the rates were adjusted to reflect that. There is also evidence that people in the lowest income quartile tend to have shorter life spans, and this can be reflected in the âScenario user inputsâ worksheet, in the row âLow-Income Effect on Death Rate.â ⢠Household structure transitions: Some transitions in household type occur automatically in the SD model due to births and deaths. Others, however, occur because of events, such as mar- riage, divorce, and/or children leaving the household. These rates vary along the categories of age cohort, current household structure, and ethnicity/acculturation group. The PSID data were used to estimate the following rates: â Marriage rate: fraction per year of single people getting married or starting cohabitation. â Divorce rate: fraction per year of people in couples getting divorced or separated (we do not count both as separate eventsâif people are already separated, the subsequent divorce is not included in the rates). â First child rate: fraction per year of people in households that transition from 0 children to 1+ children. (This is structurally related to births in 0-child households, so does not need a separate rate.) â Leave nest rate: fraction of children/young adults who leave the household of their parents to form a new householdâeither as a single person or as a couple. â Empty nest rate: fraction of parents in households that transition from 1+ children to 0 children. This is the result of some âleave nestâ events when the child/young adult leaves and there are no remaining children.
94 The Effects of Socio-Demographics on Future Travel Demand Note that the results of âmarriage,â âdivorce,â or âleave nestâ are not purely structural in terms of whether there will be any children in the resulting household. Singles who marry or young adults who âleave the nestâ may join a partner who already has children. Similarly, a âdivorceâ in a household with children may result in one or both parents retaining custody of children. ⢠Workforce status: The rates at which people enter and leave the workforce are derived from the PSID. (The number of people in the workforce who are employed versus unemployed is endogenous to the model, based on the size of the workforce relative to the supply of jobs, from the employment sector.) Although people most typically enter the workforce sometime in the 16â29 age group and leave in the 60â74 age group, there are many variations, since people can leave the workforce to raise children and/or become a âhousewife,â and they may enter the workforce again a later year. ⢠Household income group: Transitions in household income can be one of four types, each of which has a separate rate: (1) entering the low-income group from the middle-income group, (2) leaving the low-income group to the middle-income group, (3) entering the high-income group from the middle-income group, or (4) leaving the high-income group to the middle- income group. These rates are defined in terms of the percentage of people per year making each possible transition from the PSID data. In general, incomes tend to increase into the âmiddle yearsâ and decrease again in the senior years, although there are many variations in that pattern due to personal or societal economic circumstances. Figure B-1a shows some details of the socio-demographic transitions related to the rates of change that have been discussed thus far in the form of an SD flow diagram. Note that each rate is also affected by one or more exogenous variables that are predefined in the scenario worksheets in the SD model or that the user can set to define different future socio-demographic scenarios. Only two of the rates in Figure B-1aâbirth and deathâresult in people entering or leaving the simulation completely. The other rates simply shift the socio-demographic categories of people. We continue the discussion of the rates of change variables. ⢠Residence location transitions: This is an important aspect of the model because as popula- tions change (i.e., people age) they may make different residence location decisions, such as empty nesters leaving suburban homes for condos in the urban center. These types of deci- sions are less âmechanisticâ than most of the rates described above. In the model, we treat three types of location decisions (i.e., foreign migration, domestic mi- gration, and intra-regional migration) separately, since they affect different people and may have involved different decision processes. â Foreign migration: This refers to migration to or from other countries. â Domestic migration: This refers to migration to or from other U.S. regions. â Intra-regional migration: This refers to relocation between area types in the same region. The equations that define the various socio-demographic transition rates work somewhat differently for the different types of migration. Each has a âbase migration rateâ that is the fraction per year of the relevant population that tends to make a migration of the specific sort. The base rates have been derived for each region from the ACS and/or the Census. For migra- tion within the United States (domestic) and within the region (regional), these base rates (with no modifying influences) are assumed to be symmetric between coming and going. For inter- national (foreign) migration, however, the legal and practical processes for in-migration and out-migration are quite different, so different base rates are specified for both. The current base rates, which are in a table on the âDemographic transitions rateâ tab in the spreadsheet, are as shown in Table B-1 below. The equations for migration are in the following form: (1) a current population number, (2) a base migration rate that multiplies the current population, (3) a multiplier effect due to
Impacts 2050 Model Structure Documentation 95 p p (p ) q KEY FOR FLOW DIAGRAM: Rectangles are stock variables. In this diagram, the stock variable is population. The stacked triangles are âflowâ variables that determine the rate of change in the stock variables over time. The circles represent exogenous inputs or variables computed based on other variables. The clouds represent places where people can transition to a new demographic state (i.e., death, birth, divorce, income group change). The arrows re resent direct relationshi s that are arts of e uations in the model. Figure B-1a. Socio-demographic sector transitions (1). Base Migration Rates (fraction/year)a Foreign In-migration 0.10 Foreign Out-migration 0.04 Domestic Migration 0.04 Regional Migration 0.04 a Fraction/year is the fraction compared with the current relevant population. Table B-1. Base migration rates.
96 The Effects of Socio-Demographics on Future Travel Demand the attractiveness of the region/area type for residents (see below for more information on this), and (4) an exogenous modifier effect that the user can define for different scenarios (relating to the scenario variable pertaining to immigration policy). B.1.1.1 Attractiveness Function Effects on Base Residence Location Transition Rates The main modifying effects on the base rates come from the multipliers due to residence attractiveness. This is in turn a function of relative demand and supply for jobs, residential space, and road capacity, as shown in Figure B-1b, the second part of the SD flow diagram for this sector. The overall attractiveness function is a weighted sum of three separate values for jobs, hous- ing, and congestion. The relative weights can vary by area type and by migration type. The values for those weights are found in the âMigration Attraction FunctionsâWeights on Inputsâ table on the âDemographic transitions rateâ tab in the Excel model. These values were derived from our teamâs analysis of Census and ACS data. The analysis indicates that international migration is most highly weighted toward job availability, while intra-regional migration has a heavier KEY FOR FLOW DIAGRAM: Rectangles are stock variables. In this diagram, the stock variable is population. The stacked triangles are âflowâ variables that determine the rate of change in the stock variables over time. The circles represent exogenous inputs or variables computed based on other variables. The clouds represent sinks or sources that are outside the scope of the model. The arrows represent direct relationships that are (parts of) equations in the model. Figure B-1b. Socio-demographic sector transitions (2).
Impacts 2050 Model Structure Documentation 97 weight on housing availability and traffic congestion (especially since someone can move within the region but keep the same job). Once the attractiveness multiplier is calculated for a given area type/migration type, its use depends on the type of migration. For intra-regional migration, the attractiveness of each pair of area types is compared with the net migration going from the one with the lowest attractiveness toward the one with the highest attractiveness. For domestic migration, the attractiveness of competing U.S. regions is exogenous to the model. The net domestic migration to/from the region is then based on the relative magnitudes of the internal and external attractiveness multipliers. For foreign in-migration and out-migration, there is no way to explicitly represent the attractiveness of other countries, so the attractiveness multiplier is used directly, without comparison with another region or area type. The socio-demographic variables are summarized in Table B-2. Stock Variables Categories Rates of Change Variables Transitions Age Cohort 0â15 16â29 30â44 45â59 60â74 75+ Aging Transition from one age cohort to the next. Structural based on cohort duration. Household Structure Single without Child Single with Child Couple without Child Couple with Child Birth rate Death rate Marriage rate Divorce rate Empty nest rate Leave nest rate New births enter model in âwith childrenâ category. Deaths can impact both presence of children and single/couple status. Marriage and divorce rates mainly affect single/couple status. Empty nest rate affects presence of children. Children leave nest to form various types of households. Acculturation Group Foreign born, In U.S. <20 years Foreign born, in U.S. 20 years U.S. born Acculturation For foreign born, transition from one acculturation group to the otherâstructural, 20 years. Race/Ethnicity White, other Asian Black Hispanic N/A At the individual level, this is structural. People are born as a particular race/ethnicity, and this does not change. Workforce Status In workforce Not in workforce Enter workforce Leave workforce Rates Transitions between the two workforce states. Household Income $0â$34,999 $35,000â$99,999 $100,000+ Enter low income Leave low income Enter high income Leave high income Transition from middle-income quartiles to/from high- or low- income brackets. Residence Area Type Urban Suburban Rural Foreign in-migration Foreign out-migration Domestic in-migration Domestic out- migration Regional migration Moving between the region and area type and (1) other countries, (2) other regions of the U.S., (3) other area types in the same region. Table B-2. Socio-demographic variables: stock and rates of change.
98 The Effects of Socio-Demographics on Future Travel Demand B.2 Travel Behavior Subsector The models are applied separately for every combination of socio-demographic characteris- tics in the model. They are applied in the following order: 1. The Car Ownership Model splits the person into three groups, effectively adding another dimension onto the socio-demographic breakdown: a. Own car: The person lives in a household where the number of cars is equal to (or greater than) the number of driving-age adults, so that each person can drive his or her âownâ vehicle. b. Share car: The person lives in a household that has one or more cars, but fewer cars than the number of driving-age adults, so that at least two adults may need to share a vehicle. c. No car: The person lives in a household that has no vehicles. 2. The Trip Rate Models indicate the number of trips per day made by the people in each socio- demographic/car ownership category, for two types of trips: a. Work trips: to or from work, work-related or business activities. b. Nonwork trips: all other trips. 3. The Mode Choice Models split the trips in each socio-demographic/car ownership/trip purpose category into four modes: a. Car driver. b. Car passenger. c. Transit. d. Walk/bike. 4. The Trip Distance Models give the number of miles traveled per day in each socio-demographic/ car ownership/trip purpose/mode category (except for walk/bike trips, for which the model does not need a measure of distance). The data used to estimate all models described below are the full sample of the 2009 NHTS, including all add-on subsamples. Modeling was done at the person level and at the trip level, to match how the resulting equations are applied in the SD model. The NHTS sample contains 308,901 person records (from roughly 140,000 households). B.2.1 Car Ownership Model ⢠Own car: The person lives in a household where the number of cars is equal to (or greater than) the number of driving-age adults, so that each person can drive his or her âownâ vehicle. ⢠Share car: The person lives in a household that has one or more cars, but fewer cars than the number of driving-age adults, so that at least two adults may need to share a vehicle. ⢠No car: The person lives in a household that has no vehicles. The dependent variable is percentage share for each of these three alternatives. The indepen- dent variables are age, household structure, acculturation, ethnicity, work status, household income, residence location type, and region. The model estimation results are shown in Table B-3. âOwn carâ was selected as the base category and âshare carâ and âno carâ are interpreted relative to the base category. The model esti- mates that 22 percent of people were in share-car households and 6 percent in no-car households, leaving 72 percent in âown carâ households. The base categories apply to the independent variables as well. The base categories are the variables not found: (1) 30â44 age group, (2) single-person households, (3) white, non-Hispanic ethnicity, (4) not employed, (5) $35â100K income, (6) living in suburban area, and (7) living outside all of the five selected MSA regions. As with the dependent variable, the other categories are interpreted relative to the base category.
Impacts 2050 Model Structure Documentation 99 The model utility coefficients are shown, along with the related t-statistic. âOwn carâ has an implicit utility of 0, and separate utility functions were estimated for the other two alternatives. In general, a t-statistic of 1.9 or greater means that a coefficient estimate is statistically different from 0, with 95 percent certainty. Nearly every estimate in the table appears to be statistically sig- nificant. The coefficients can be interpreted according to their sign and relative size. For exam- ple, the largest positive coefficient for both alternatives is for the low-income group, meaning that having a lower income is the main factor related to living in a share-car or no-car household instead of an own-car household. Conversely, the high-income group coefficients are strongly negative, meaning those people are less likely to be in either of these low-/no-car households. Children are less likely to live in low-/no-car households, while seniors age 75+ are more likely to live in low-/no-car households. People who live in households with couples (versus single adults), and those age 16â29 are more likely to share a car, but are less likely to be in a no-car household. People in nonwhite or Hispanic ethnic groups are more likely to be in low-car households, and this effect is even stronger if they were born outside the United States, and stronger still if they have been in the United States for less than 20 years (these three effects are additive). Note that Variables Alternative Share Car (22% share) No Car (6% share) Coefficient T-statistic Coefficient T-statistic Constant â1.811 â87.7 â2.599 â74.4 Age group 0â15 â0.526 â27.7 â0.838 â24.0 Age group 16â29 0.543 37.2 â0.076 â2.6 Age group 45â59 â0.131 -8.4 â0.121 â4.2 Age group 60â74 â0.405 â21.2 â0.324 â10.1 Age group 75 up 0.189 8.2 0.202 5.7 Couple in household 0.834 56.5 â1.058 â38.9 1+ children in household â0.452 â32.5 â0.433 â12.1 Single with children 0.670 32.0 0.190 4.8 Ethnicity nonwhite or Hispanic 0.592 49.7 0.999 48.6 Born outside of U.S. 0.292 13.8 â0.309 â8.9 Born outside U.S., <20 years in U.S. 0.369 15.0 0.435 10.4 Worker â0.597 â51.2 â1.133 â53.5 Low-income group 1.111 101.3 1.661 82.0 High-income group â0.701 â45.2 â0.635 â14.5 Urban residence area type 0.690 49.9 1.757 85.5 Rural residence area type â0.528 â46.1 â0.599 â24.4 Atlanta MSA region 0.106 2.9 â0.232 â3.0 Boston MSA region 0.135 3.8 â0.042 â0.7 Detroit MSA region 0.650 20.0 â0.133 â1.8 Houston MSA region 0.172 4.8 â0.213 â2.8 Seattle MSA region â.0235 â5.1 0.362 4.6 Note: Although the sample person expansion weights were used in estimation to account for nonrepresentative sampling, the weights were first normalized to a mean of 1.0, so that the sum of weights is equal to the number of observations (to avoid inflated measures of statistical significance). Table B-3. Car ownership model coefficients.
100 The Effects of Socio-Demographics on Future Travel Demand this effect is over and above the income effects that are simultaneously included in the model. A worker effect is also included simultaneously, and has strong negative coefficients, indicating that workers are more likely to have their âown car.â The urban and rural variables also have strong, expected effects, with people living in urban areas most likely to live in low-/no-car households, and those in rural areas most likely to be in âown carâ households. The region-specific effects are relatively minor, and indicate the effect of the region over and above all of the other variables in the model. This is a promising sign that the area type catego- rization (e.g., urban, rural) worked to capture a good deal of the land-use-related variation that exists in reality. The model goodness-of-fit measure, McFaddenâs Rho-squared (somewhat analogous to R-squared for regression models) is 0.198, which is a typical magnitude for this type of model. B.2.2 Trip Rate Models Log-linear regression models were estimated for the number of work trips and nonwork trips per person-day, with work trips classified as all trips with the purpose at either (or both) trip ends coded as work, work-related, or business, and all other trips classified as nonwork trips. The dependent variable for both models is LOG (#trips + 1), the 1 included to avoid taking the log of 0 for those with 0 trips. Note that the NHTS data include all days of the week throughout the year (i.e., including weekends and holidays), so it is truly an âaverage dayâ in the sense that multiplying by 365 would give an annual expected trip rate. The model results are shown in Table B-4. The independent variables are the same as in the previous model, except that two new variablesââno carâ and âshare carââare added, to rep- resent the effect of car ownership on trip rates, relative to the base group, âown car.â Also, the Work Trip Rate model was only estimated for people who are workers, so the age group 0â15 and worker variables were not included. The work trip rate model contains relatively few significant effects, since the fact that some- body is a worker already explains most of the variation in the population, and the rest of the vari- ables try to explain who tend to go to work on fewer days per week or make more work-related trips, such as non-home-based work trips. There are age effects, as workers over age 60 tend to make fewer work trips, as do, to a lesser extent, workers under age 30. Also, workers with children in the household tend to make fewer trips, either working part time, or having to stay home with sick children periodically. Nonwhite and Hispanic workers make slightly fewer work trips, but this is offset by a positive additive coefficient for those born outside the United States. Those in low-income groups and in urban areas make somewhat fewer work trips, and those in low-/no-car households make fewer trips as well, particularly those in no-car households. The only region-specific effects that are fairly strong are for fewer work trips in Boston and Detroit. The nonwork trip rate model shows stronger effects, with the strongest negative effect being for workers, who, presumably, make fewer nonwork trips because they are busy working. There is also an age effect, with nonwork trip rates increasing with age up until age 75. People in house- holds with children also make substantially more nonwork trips (many of those for school and/ or taking children to school). There is not a strong influence of ethnicity, except that people born outside the United States tend to make fewer nonwork trips, all else being equal. Nonwork trip rates increase with income and decrease in rural areas, where distances are longer and people tend to group more activities into each trip. (One home-based trip that visits two different nonwork destinations requires three trips, whereas visiting them on two separate home-based tours would require four trips.) People in low-car, and especially no-car households, also make
Impacts 2050 Model Structure Documentation 101 fewer nonwork trips. All of these effects are typically found in travel demand models. Again, the regional variables only explain any effects over and above other independent variables. B.2.3 Mode Choice Models Three separate mode choice models were estimated: one for work trips, one for nonwork trips made by people of driving age (16+), and another for children under age 16 who do not have the option of driving a car. Four modes are considered: car driver, car passenger, transit, and walk/ bike. Any NHTS survey trips made by other modes, such as taxi or paratransit, were excluded from the estimation. B.2.3.1 Work Trips The work trip mode choice model, shown in Table B-5, was estimated on a sample of about 244,000 work trips. As before, the data were weighted using the NHTS trip expansion weights, nor- malized so that the mean weight is 1.0. The mode shares in the sample are 82.5 percent car driver (the base alternative), 7.6 percent car passenger, 3.3 percent transit, and 6.6 percent walk/bike. Variables Alternative Work Trips Nonwork Trips Coefficient T-statistic Coefficient T-statistic Constant 0.825 155.1 1.597 310.4 Age group 0â15 n/a â0.209 â45.0 Age group 16â29 â0.031 â6.7 â0.069 â18.4 Age group 45â59 0.0 0.1 0.018 5.1 Age group 60â74 â0.076 â12.4 0.058 12.5 Age group 75 up â0.153 â9.3 â0.004 â0.6 Couple in household â0.007 â1.5 â0.028 â7.6 1+ children in household â0.040 â9.4 0.115 33.3 Single with children â0.008 â1.0 0.010 1.8 Ethnicity nonwhite or Hispanic â0.027 â6.1 -0.009 â2.8 Born outside of U.S. 0.035 4.6 â0.073 â12.7 Born outside U.S., <20 years in U.S. 0.020 2.3 â0.017 â2.4 Worker n/a â0.473 â153.3 Low-income group â0.016 â3.6 â0.031 â10.1 High-income group â0.008 â1.9 0.065 20.7 Urban residence area type â0.013 â2.5 0.008 2.2 Rural residence area type â0.001 â0.2 â0.052 â19.5 Atlanta MSA region 0.014 1.1 â0.027 â2.9 Boston MSA region â0.084 â7.5 0.0 0.0. Detroit MSA region â0.077 â6.5 0.127 14.6 Houston MSA region 0.005 0.4 â0.056 â5.8 Seattle MSA region â0.009 â0.6 â0.083 â8.0 No-car household â0.078 â8.0 â0.110 â18.6 Share-car household â0.021 â4.6 â0.016 â5.1 Table B-4. Daily trip rate models.
102 The Effects of Socio-Demographics on Future Travel Demand We again used the same set of variables as for the preceding models. However, we did have one additional variableâthe price of gasoline. Variables Alternative Car Passenger Transit Walk/Bike Coefficient T-statistic Coefficient T-statistic Coefficient T-statistic Constant â3.287 â89.4 â4.190 â71.4 â2.853 -74.8 Age group 0â15 n/a n/a n/a Age group 16â29 0.638 31.1 0.079 2.2 â0.111 â4.5 Age group 45â59 â0.136 â6.2 â0.092 â2.8 â0.206 â9.5 Age group 60â74 0.003 0.1 0.154 3.1 â0.264 â7.9 Age group 75 up â0.607 â5.4 0.109 0.7 â0.364 â3.7 Couple in household 0.275 11.5 â0.633 â17.1 â0.356 â15.0 1+ children in household â0.053 â2.6 0.273 7.2 0.004 0.2 Single with children 0.125 3.5 â0.327 â6.0 â0.756 â17.5 Ethnicity nonwhite or Hispanic 0.107 5.0 0.243 7.1 â0.366 â14.7 Born outside of U.S. 0.176 5.0 0.199 4.3 0.035 0.9 Born outside U.S., <20 years in U.S. â0.012 â0.3 0.256 5.3 0.195 4.5 Worker n/a n/a n/a Low-income group 0.318 16.0 â0.199 â6.2 â0.215 â8.8 High-income group â0.039 â1.8 0.406 11.8 0.439 21.3 Urban residence area type â0.012 â0.5 1.366 46.5 0.982 44.4 Rural residence area type 0.198 11.2 â1.252 â23.7 â0.363 â16.4 Atlanta MSA region 0.252 4.7 â0.783 â4.9 â0.027 â0.4 Boston MSA region â0.176 â2.8 â0.006 â0.1 â0.511 â7.3 Detroit MSA region â0.197 â3.2 â0.957 â7.4 â0.296 â4.2 Houston MSA region â0.095 â1.5 â0.720 â6.1 â0.334 â4.4 Seattle MSA region â0.011 â0.1 1.026 10.9 â0.251 â2.8 No-car household 3.176 59.2 5.157 94.7 3.964 79.8 Share-car household 1.333 71.4 1.934 60.3 1.179 55.2 Fuel price (per dollar) 0.005 0.6 â0.050 â3.6 0.102 11.1 Table B-5. Work trip mode choice model. Note that this variable could not be used for the car ownership and trip rate models described earlier, because it is only available in NHTS for people who actually made trips. The results in Table B-5 show that age has strong effects on work trip mode choice, with workers under age 30 more likely to go as car passengers and the older age groups over 45 are less likely to bike or walk. The age effects for transit use are not strong. Workers who are part of a couple (or live with a couple) are more likely to rideshare, but less likely to use transit or walk/bike; the same pattern is found for low-income workers (and the opposite pattern for high-income workers). Nonwhite and Hispanic workers are more likely to rideshare and use transit for the work trip, but less likely
Impacts 2050 Model Structure Documentation 103 to walk; this effect is even stronger for workers born outside the United States. The effects by area type are also strong, as workers in urban areas are most likely to use transit or walk/bike for their work trip, while those in rural areas are less likely to use those modes and more likely to rideshare. As one may expect, the most important variables are related to car ownership, with workers in no-car households more likely to use all the alternatives to driving, particularly transit. Finally, as gas price increases the walk/bike share increases somewhat, counter-intuitively, the transit share seems to decrease somewhat. B.2.3.2 Nonwork Trips A mode choice model with the same multinomial logit specification was estimated for non- work trips made by people age 16+ (old enough to be a car driver). The sample is roughly 750,000 weighted trips, with 65.5 percent by car driver, 20.1 percent by car passenger, 21.0 per- cent by transit, and 12.3 percent by walk/bike. The model fit (McFadden Rho-squared) is 0.107, and the results are shown in Table B-6. Variables Alternative Car Passenger Transit Walk/Bike Coefficient T-statistic Coefficient T-statistic Coefficient T-statistic Constant â1.767 â111.4 â5.187 â104.5 â2.115 â110.6 Age group 0â15 n/a n/a n/a Age group 16â29 0.682 79.1 0.628 24.8 0.218 20.0 Age group 45â59 â0.075 â8.1 â0.020 â0.7 â0.039 â3.7 Age group 60â74 â0.013 â1.2 â0.342 â10.1 â0.311 â22.8 Age group 75 up 0.230 16.2 â0.519 â11.9 â0.625 â30.8 Couple in household 0.533 57.2 â0.347 â12.8 â0.102 â9.3 1+ children in household â0.146 â17.6 0.018 0.6 â0.109 â10.2 Single with children 0.312 22.2 â0.318 â8.2 â0.021 â1.2 Ethnicity nonwhite or Hispanic 0.053 6.6 0.378 16.0 â0.213 â20.6 Born outside of U.S. â0.009 â0.7 â0.050 â1.5 0.152 8.9 Born outside U.S., <20 years in U.S. 0.087 5.0 0.137 3.6 0.215 11.0 Worker â0.523 â78.3 â0.526 â25.9 â0.248 â29.4 Low-income group 0.039 5.1 0.008 0.4 â0.031 â3.2 High-income group 0.048 5.9 0.176 5.5 0.198 19.9 Urban residence area type â0.098 â9.5 1.501 69.9 0.780 76.5 Rural residence area type 0.075 11.0 â0.884 â22.8 â0.186 â19.9 Atlanta MSA region â0.026 â1.0 0.402 5.2 â0.057 â1.8 Boston MSA region 0.171 7.7 0.444 9.0 0.264 10.7 Detroit MSA region â0.131 â6.2 â0.809 â9.1 â0.163 â5.9 Houston MSA region 0.013 0.5 â0.363 â4.1 â0.089 â2.8 Seattle MSA region 0.063 2.1 0.525 6.3 0.460 14.5 No-car household 2.653 109.7 5.261 154.4 3.641 156.0 Share-car household 0.645 85.4 1.524 56.6 0.575 59.6 Fuel price (per dollar) 0.041 12.6 0.124 12.7 0.124 30.8 Table B-6. Nonwork trip mode choice modelâage 16î±.
104 The Effects of Socio-Demographics on Future Travel Demand Again, car ownership is very important, with those in low-car and especially no-car house- holds much more likely to choose any of the alternatives to being a car driver. It is interesting that once car ownership has been taken into account, those in high-income groups are also more likely to choose alternatives to driving. Nonwhite or Hispanic workers are more likely to rideshare or use transit, but less likely to walk/bike. Being born outside the United States does not have a strong influence, except being more likely to walk or bike, especially if living less than 20 years in the United States. Workers appear more likely to drive and less likely to use any alternatives. (This is not a tour-based model, so it is likely that some of these trips are between home and nonwork stops that car drivers make as part of work tours.) The area type once again shows significant effects, with workers in core urban areas less likely to choose to be car passengers and more likely to go by transit and walk/bike, and those in rural areas choosing the opposite. There are a few significant region-specific effects, but they are generally much less significant than the area-type effects, suggesting that the model would also do fairly well at explaining mode choice in other regions. Finally, the fuel price variable has the expected effects, with a higher fuel price related to higher uses of all the alternatives to car driver, especially transit and walk/bike. B.2.3.3 Childrenâs Travel The final mode choice model, shown in Table B-7, was estimated only for children younger than 16. (Because children under 5 do not have travel diaries in NHTS, the age group for estimation is 5â15.) This model includes about 135,000 weighted trips, with mode shares 79.6 percent for car passenger, 1.6 percent for transit, and 18.7 percent for walk/bike. Because car driver is not a valid alternative for this age group, the base alternative is car passenger, and equations were estimated for the transit and walk/bike modes. Also, many of the variables could not be included in the model, because they applied either to all cases or to no cases. Those variables are indicated by n/a (not applicable) in Table B-7. The model fit (McFadden Rho-squared) is 0.86. As in the other mode choice models, the car ownership effects are the strongest. Nonwhite or Hispanic children are more likely to use transit or walk/bike, somewhat more so if born outside the United States. Children in low-income households are also more likely to use non-car alternatives, especially walk/bike. The area-type variables also show strong effects in the usual direction, again stronger than the region-specific effects (with the exception of Detroit, which had no transit choices in the data set). Fuel price also shows the expected effects, with higher fuel prices meaning that children are less likely to get a ride (from their parents). B.2.4 Trip Distance Models The final set of models, shown in Table B-8, is used to convert demand for car driver, car passenger and transit from number of trips to number of person miles travelled (PMT). For car drivers, this also gives a direct value of VMT. The SD model does not need to know miles traveled for walk or bike trips, so there is no model for those trips. The process used to calculate VMT from the trip distance model is described next. The car ownership model splits the demographic categories further into subcategories by car ownership type. Then, within that demographic/car ownership group: VMT work trips work trip car driver mode share work trip car driver trip distance nonwork trips nonwork trip car driver mode share nonwork trip car driver trip distance ( ) ( ) = Ã Ã + Ã Ã
Impacts 2050 Model Structure Documentation 105 Six different models get the six different inputs to the VMT equation. This is VMT per person per day. The same equation can be applied for car passenger miles traveled and transit passenger miles traveled. The models were estimated using log-linear regression, with the dependent variable being the log of 1.0 plus the number of miles from the trip origin to the destination (a variable provided on the NHTS trip records). To avoid outlier effects, a small number of trips with distances greater than 100 miles were excluded. The strongest effect in the models is the âwork trip purposeâ variable, which indicates that trips made for work tend to be longer than nonwork trips, particularly for car drivers. Trip distances by all modes also tend to increase with income but decrease with age. People in house- holds with children tend to make shorter car trips (many of them chauffeur-type trips). Non- white and Hispanic workers tend to make longer car driver trips, while workers born outside the United States tend to make longer car passenger and transit trips. There are strong area-type effects, with urban dwellers making shorter trips and rural dwellers making longer trips by all modes, as expected. Again, the region-specific effects are small after Variables Alternative Transit Walk/Bike Coefficient T-statistic Coefficient T-statistic Constant â5.783 â54.5 â1.898 â64.3 Age group 0â15 n/a n/a Age group 16â29 n/a n/a Age group 45â59 n/a n/a Age group 60â74 n/a n/a Age group 75 up n/a n/a Couple in household â0.235 â4.5 â0.007 â0.4 1+ children in household n/a n/a Single with children n/a n/a Ethnicity nonwhite or Hispanic 0.896 14.6 0.102 6.0 Born outside of U.S. n/a n/a Born outside U.S., <20 years in U.S. 0.271 3.3 0.107 2.9 Worker n/a n/a Low-income group 0.166 2.8 0.248 13.0 High-income group 1.067 15.7 â0.026 â1.4 Urban residence area type 1.631 31.5 0.377 17.9 Rural residence area type â1.592 â13.0 â0.359 â20.7 Atlanta MSA region 0.960 7.8 â0.325 â5.3 Boston MSA region â0.991 â4.4 0.325 6.4 Detroit MSA region â20.000 n/a â0.661 â10.9 Houston MSA region â0.911 â4.1 â0.401 â6.5 Seattle MSA region 0.191 0.8 0.710 14.2 No-car household 3.258 47.0 1.828 49.8 Share-car household 1.112 18.0 0.387 19.5 Fuel price (per dollar) 0.090 3.9 0.089 11.7 Table B-7. Nonwork trip mode choice modelâage under 16.
106 The Effects of Socio-Demographics on Future Travel Demand taking into consideration other variables, particularly area type. People in low-car and no-car households tend to make shorter trips by carâperhaps because it is more difficult to find a ride to farther destinations. The fuel price effects are fairly small, and not in the expected (negative) direction for car driver or transit. This result is different from the finding in the mode choice models, and may indicate that people tend to change modes for their shorter trips, but still make the longer trips by car (in which case, overall VMT would still decrease). B.3 Land-Use Sector The stock variable is the amount of space (in square miles) by area type (urban, suburban, local) and by use type (residential, nonresidential, developable, and protected). This stock variable is segmented into: ⢠Urban, suburban, rural residential space ⢠Urban, suburban, rural nonresidential space Variables Trip Type: Dependent = LN (Distance + 1) Car Driver Trips Car Passenger Trips Transit Trips Coefficient T-statistic Coefficient T-statistic Coefficient T-statistic Constant 1.538 305.8 1.884 182.0 1.959 61.4 Age group 0â15 0.218 6.6 â0.211 â36.2 â0.401 â15.9 Age group 16â29 0.044 13.4 â0.070 â11.1 â0.008 â0.5 Age group 45â59 â0.051 â17.5 â0.027 â3.8 â0.100 â5.8 Age group 60â74 â0.083 â22.0 â0.088 â10.7 â0.154 â6.8 Age group 75 up â0.178 â30.6 â0.204 â19.7 â0.140 â4.4 Couple in household 0.042 13.3 â0.061 â8.3 0.008 0.4 1+ children in household â0.050 â16.8 â0.059 â9.8 0.035 1.8 Single with children 0.039 7.9 â0.058 â6.7 0.047 1.9 Ethnicity nonwhite or Hispanic .052 17.7 â0.003 â0.6 0.001 0.1 Born outside of U.S. 0.022 4.3 0.097 9.1 0.240 10.1 Born outside U.S., <20 years in U.S. â0.008 â1.3 â0.044 â3.5 â0.239 â9.2 Worker 0.434 179.8 0.190 25.9 0.277 20.4 Low-income group â0.074 â25.7 â0.067 â13.9 â0.119 â7.9 High-income group 0.043 15.5 0.047 10.1 0.073 3.9 Urban residence area type â0.105 â27.8 â0.108 â17.2 â0.229 â16.7 Rural residence area type 0.229 95.4 0.295 73.7 0.270 10.0 Atlanta MSA region 0.131 15.4 0.087 6.1 0.678 13.0 Boston MSA region 0.116 14.5 â0.121 â8.8 â.0025 â0.7 Detroit MSA region 0.010 1.3 0.021 1.7 0.478 8.1 Houston MSA region 0.097 10.8 0.002 0.2 0.191 3.4 Seattle MSA region 0.150 14.6 0.108 6.6 0.061 1.3 No-car household 0.101 6.1 â0.274 â25.1 â0.294 â16.3 Share-car household â0.026 â8.1 â0.055 â11.6 â.0171 â9.9 Fuel price (per dollar) 0.008 7.2 â0.005 â2.8 0.027 4.3 Table B-8. Trip distance models.
Impacts 2050 Model Structure Documentation 107 ⢠Urban, suburban rural developable space ⢠Urban, suburban, rural protected space B.3.1 Land-Use Rates of Change Figure B-2 presents a flow diagram for the land-use sector. The rates of change that are rel- evant to this sector are: ⢠Development of Residential Space and Release of Residential Space: Converting land from developable space to use for housing. ⢠Development of Nonresidential Space: Converting land from developable space to use for employment, industry, and other commercial uses. ⢠Release of Nonresidential Space: Converting land back to developable spaceâthrough rezoning, redevelopment, demolition, etc. ⢠Release of Protected Space: Redesignating land from protected to developable. (Note that this can be negative, which would signify the case of more land put under protection.) KEY FOR FLOW DIAGRAM: Rectangles are stock variables. In this diagram, the stock variable is the amount of space. The stacked triangles are âflowâ variables that determine the rate of change in the stock variables over time. The circles represent exogenous inputs or variables computed based on other variables. The clouds represent sinks or sources that are outside the scope of the model. The arrows represent direct relationships that are (parts of) equations in the model. Figure B-2. Land-use sector flow diagram.
108 The Effects of Socio-Demographics on Future Travel Demand The equations that comprise the rates for the first three bullets have four main components: (1) the existing stock of space in the use that would be converted out of; (2) the amount of new space needed for the relevant land use (that can be zero or negative); (3) an effect of relative demand and supply for developable land, which can moderate land prices and the amount of development undertaken by the market; and (4) the market delay time needed to create new development or release land for new development. The delay times are exogenous, and can reflect, for example, programs or tax policies to spur new housing or commercial development. By comparison, the equation for the release of protected space is based totally on exogenous scenario inputs, as it is assumed that such actions are the result of nonmarket decisions. There are a number of things to note for the rate equations: ⢠The key sector inputs of demand for residential space and demand for nonresidential space come from the employment and demographic sectors, respectively. (The model does not explicitly account for land space needed for road infrastructure, as that is not likely to vary enough under the scenarios to significantly affect the land-use marketâalthough the supply of new roads can indirectly spur the demand for land by helping to attract new residents and businesses.) ⢠The real estate market is modeled here as being reactive to demand, rather than predictive of what demand may be in the more distant future. Although purely speculative development is fairly rare in reality (e.g., not much construction is happening in the current recession), it would be possible to represent it in the model by including exogenous scenario variables for new development that is not dependent on market demand. B.4 Employment Sector The stock variable is the number of jobs (employment) by area type (urban, suburban, local) and by employment type (retail, service, other). This stock variable is segmented in the model into: ⢠Urban, suburban, rural retail jobs ⢠Urban, suburban, rural service jobs ⢠Urban, suburban, rural other jobs The source data for these variables were the Longitudinal Employer-Household Dynamics data from the U.S. Census Bureau, 2002 and 2010. B.4.1 Employment Rates of Change Figure B-3 presents a flow diagram for the employment sector. The rates of change that are relevant to this sector are: ⢠Job creation: Jobs created in the region by companies moving to the region, new companies starting up, or existing companies adding jobs. (For the purposes of this model, it is not important to model those separately.) ⢠Job loss: Jobs lost in the region by companies leaving the region, going out of business, or downsizing. ⢠Job migration: Jobs changing location within the region, such as moving from the city center to the suburbs.
Impacts 2050 Model Structure Documentation 109 The equations that comprise this sector have three main components: (1) the existing stock of jobs, (2) the indicated change in the stock of jobs, and (3) the market delay time needed to reach the indicated level. The delay time is exogenous, and could reflect, for example, job cre- ation programs or tax policies to remove barriers to creating new jobs. The rates of change in job creation, job loss, and job migration across area types is modeled from trends analysis by the team using the Longitudinal Employer-Household Dynamics data from the U.S. Census Bureau, 2002 and 2010. The meat of the sector dynamics is in the rate functions used to determine the indicated change in the stock. Each of these in turn has four main inputs: (1) an exogenous scenario effect, reflecting, for example, the health of the economy for creating new jobs; (2) the balance KEY FOR FLOW DIAGRAM: Rectangles are stock variables. In this diagram, the stock variable is the number of jobs. The stacked triangles are âflowâ variables that determine the rate of change in the stock variables over time. The circles represent exogenous inputs or variables computed based on other variables. The clouds represent sinks or sources outside the scope of the model. The arrows represent direct relationships that are (parts of) equations in the model. Figure B-3. Employment sector flow diagram.
110 The Effects of Socio-Demographics on Future Travel Demand of supply and demand for jobs, reflecting the availability of labor; (3) the balance of supply and demand for commercial space, reflecting the availability of land; and (4) the balance of supply and demand for road capacity, reflecting traffic congestion levels for commuting. There are a number of things to note for each of the demand-and-supply relationships that enters the rate functions: ⢠While the supply of jobs and the demand for commercial space are endogenous to this sector, the other key inputs come from other model sectors, as shown in capital letters at the bottom of Figure B-3. ⢠The same three demand-and-supply ratios enter all of the functions, but the user can give them different weights in the different functions. For example, companies within the region may know more about traffic congestion than companies from outside the region, so that will tend to have a higher weight (relative influence) for job migration than for job creation or job loss. ⢠The word ârelativeâ is used in each of the demandâsupply variables because what is important to reflect is how this region is doing relative to other regions in the country, particularly for jobs coming from or going to other regions (reflected as part of job creation and job loss). Since those other regions are exogenous to the model, those external trends are exogenous scenario inputs to the model. The attractiveness multiplier equations for this sector work in a similar way to the resident attractiveness multiplier equations described in Chapter 4. In this case, the weights vary by area type and employment type. B.5 Transport Supply Sector The stock variable is the number of road lane miles by area type (urban, suburban, local) and by road type (freeway, major arterial, other), and number of transit route miles by area type (urban, suburban, local) and by transit type (rail and bus). The stock variable is segmented in the model into capacity measures in terms of: ⢠Urban, suburban, rural freeway lane miles ⢠Urban, suburban, rural arterial lane miles ⢠Urban, suburban, rural other lane miles ⢠Urban, suburban, rural rail transit route miles ⢠Urban, suburban, rural nonrail transit route miles The sources for these data were the National Transit Database of the FTA, 2002 and 2010. B.5.1 Transport Sector Rates of Change Figure B-4 presents a flow diagram for the transport sector. The rates of change that are rel- evant to this sector are: ⢠Addition to road lane miles: Construction of new road capacity (which could include widen- ing of existing roads). ⢠Retirement of road lane miles: Reflects road capacity being closed or becoming unusable due to lack of maintenance. ⢠Addition to transit route miles: Opening of new transit services, or addition of routes and/ or increase of frequency on existing services. ⢠Retirement of transit route miles: Reflects the closing of transit services or routes, or reduc- tion of frequencies.
Impacts 2050 Model Structure Documentation 111 The demand for peak-hour road supply is a function of the number and distance of work and nonwork trips by car drivers made by people living in each area type, and depending on a number of other user inputs, including the mix of road type/area type combinations used by commuters for each type of flow (e.g., suburbanâurban commutes) and the peak-hour fraction of daily trips assumed for work and nonwork trips. Multiplied together, these determine the peak-hour demand for lane miles for each road type within each area type, which can be com- pared with the road supply to determine the extent to which the existing road supply can meet the indicated demand. Because the model does not use explicit network assignment, it approxi- mates traffic demand growth relative to supply over time. The rate equations for the transport sector are relatively simple, and mainly rely on user scenario inputs. This reflects the fact that transportation capital, operations, and mainte- nance funds mainly use public funds, with relative little influence from private market forces. Figure B-4. Transport sector flow diagram. KEY FOR FLOW DIAGRAM: Rectangles are stock variables. In this diagram, the stock variable is amount of road lane miles and amount of transit route miles. The stacked triangles are âflowâ variables that determine the rate of change in the stock variables over time. The circles represent exogenous inputs or variables computed based on other variables. The clouds represent places where people can transition to a new demographic state (e.g., death, birth, divorce, income group change). The arrows represent direct relationships that are (parts of) equations in the model.
112 The Effects of Socio-Demographics on Future Travel Demand For addition of road and transit capacity, the main indicator is the amount of new capac- ity needed to meet passenger demand, if any, based on the demand calculations described above. The exogenous scenario inputs represent (1) the amount of that capacity that is to be provided (i.e., the government could decide to try to meet mobility demand by other means than increasing capacities); and (2) the delay before new capacity will be provided (largely a function of the availability of public funds). For retirement of existing capacity, there are again two types of exogenous inputs: (1) the fraction of existing capacity to be retired (for example, the decision to discontinue certain transit services); and (2) the rate in time at which existing capacity is to be retired. Currently, transit route miles are treated simpler than road lane miles, and the relative demand and supply for transit do not feed back to the other sectors of the model. Modeling of transit in this sector is an aspect that could be improved in a future area-specific case study.