Influences of LRT on travel behaviour: A retrospective study on movers in Minneapolis

Abstract

Chinese

Following scant evidence for the effects of proximity to rail transit on car use, we pinpoint the impacts of rail transit and neighbourhood characteristics on both transit and car use in the Minneapolis-St. Paul metropolitan area. We apply the structural equations modelling approach on 597 residents who moved into the Hiawatha Light Rail Transit (LRT) corridor after it opened. The analysis is based on a self-administrated survey where all attributes of the built environment and transit quality are perceived measures. Using a quasi-longitudinal design to compare the behaviour of movers into the Hiawatha and control corridors, we found that the Hiawatha LRT acts as both a catalyst and a magnet. Movers into the Hiawatha corridor experience transit improvement, which increases transit use and reduces car use. The LRT also enables transit-liking people who were unable to realise their preference previously to relocate near the LRT. However, the LRT has no significant effects on changes in car ownership.

Keywords

car ownership land use longitudinal design rail transit residential self-selection

Introduction

Developed countries have witnessed a considerable increase in private vehicle use. Total vehicle miles travelled (VMT) increased by 35% in the USA between 1990 and 2012 (EPA, 2014), although its growth stagnated after the 2007 peak (McCahill, 2014). This trend arouses the concern of policymakers over traffic jams at peak hours, energy consumption and adverse environmental impacts such as climate change and noise pollution. For example, 28% of the total US greenhouse gas emissions comes from the transportation sector, of which 43% is produced by private vehicles (EPA, 2014).

From the perspective of land use planning, manipulating the urban form, particularly siting residences in proximity to valued destinations, lowers travel distances and thereby reduces car travel (Ewing and Cervero, 2010). From the travel behaviour viewpoint, rail transit has been advocated as a way to promote transit ridership because the high accessibility to and by transit increases public transportation (Moniruzzaman and Páez, 2012). Hence, rail transit programmes and transit-oriented development (TOD) have grown throughout the past decades. It has imposed, however, a substantial financial burden on the government to improve the infrastructure and to implement strategies. The capital cost of Hiawatha Light Rail Transit (LRT) in Minneapolis, for instance, was more than US$710 million for an 11.9-mile trunk rail. The 14.9-mile LRT of North South Line in Salt Lake City cost around US$400 million (Cain et al., 2007). Hence, evaluating the costs and benefits of rail transit has become a hot issue in recent studies (Litman, 2005), and has been the cornerstone of transit network development research.

The underlying assumption of TOD programmes is that clusters of buildings near transit promote transit ridership and thus reduce auto travel. It is rooted deeply in a hypothesis that households near transit stations use private cars less frequently than those farther from transit (Cervero and Arrington, 2008). For example, Chatman (2013) studied households living within 2 miles of ten rail stations in northern New Jersey. He indicated each mile from a rail station is associated with a 74% increase in the odds of private car use. Many studies have sought to uncover the nexus between proximity to transit stations and transit ridership (Cervero, 2007; Dill, 2008). Subsequently, it has long been a mantra among transportation and urban planners that living near transit stations increases transit ridership. It remains, however, unclear whether the households near transit stations also own and use private vehicles less. Several studies have attempted to examine the impacts of living near transit stations on auto ownership and use (Chatman, 2013; Dueker and Bianco, 1999; Loo et al., 2010). However, they produced mixed outcomes.

This study strives to understand the impacts of rail transit and neighbourhood characteristics on both transit use and car use. Using a self-administered survey in the Minneapolis-St. Paul metropolitan area (Twin Cities) in May 2011, it employs quasi-longitudinal analyses to explore the effects of the Hiawatha LRT on travel behaviour of current residents who moved into the Hiawatha corridor after its opening. In this vein, the structure equations modelling approach is applied to disentangle the complex interaction among the Hiawatha LRT, transit attributes, neighbourhood characteristics and travel behaviour. The core questions are: (1) does moving into the Hiawatha corridor lead to an increase in transit use and a reduction in auto ownership and use? (2) Are neighbourhood characteristics around rail stations associated with changes in travel behaviour?

Literature review

During the past decades, TODs have been in hot debates in planning practice and academic circles. TOD is centred on transit and combines six Ds, namely density, diversity, design, destination accessibility, distance to transit, and, in a few studies, demand management, which aims to encourage people to use transit services (Ewing and Cervero, 2010). Accordingly, it is hypothesised that TOD programmes increase transit ridership, and thereby reduce VMT. Both researchers and practitioners have attempted to scrutinise the effects of TODs on travel behaviour and auto ownership (Chatman, 2013; Renne, 2005). A comprehensive literature review focuses on the nexus between the built environment and travel behaviour (Ewing and Cervero, 2001, 2010). There is modest literature, however, examining travel behaviour of station area residents. This section, therefore, is limited to synthesise the impact of proximity to rail transit on travel behaviour, which is in line with the context of the current study.

Table 1 summarises studies with diverse geographic contexts, time spans and analysis methods. Exploring the effects of neighbourhood characteristics on automobile use is traced back to 1994 when Holtzclaw (1994) investigated the influence of four neighbourhood characteristics, namely transit accessibility, pedestrian accessibility, residential density and neighbourhood shopping on car use in California. Since then, a number of studies have shed light on the travel behaviour of station area residents. Previous studies unanimously concluded a positive correlation between living near rail stations and transit ridership (Cervero, 2007; Dill, 2008; Dueker and Bianco, 1999). A study in 2006 showed that residents in TOD housing units in California are up to five times more likely to take transit than non-TOD residents (Lund et al., 2006). Likewise, Renne (2005) found that TOD residents use transit to commute about 3.5 times as often as non-TOD residents, on average, by analysing 103 TODs across 12 US regions. Yet little evidence has been given to indicate whether station area residents own and use personal vehicles less.

Table 1.

Summary of previous studies.

Author	Location	Sample size	Mode			Data collection method	Method of analysis	Control group	Finding
			Transit	Car	Active
Cervero and Gorham (1995)	California	7 transit neighbourhood	×	×	×	1990 US Census	Descriptive and regression	7 car neighbourhood	• Lower car use
Dueker and Bianco (1999)	Portland	–	×			Census years 1980 and 1990	Before-after	A parallel bus corridor	• Lower car use and ownership in rail corridor
Lund et al. (2006)	California	624	×	×	×	Questionnaire survey	Descriptive	Movers	• Five times more likely to commute by transit
Cervero (2007)	San Francisco	10,968		×		2000 Bay Area travel survey	Nested Logit Model	Not TOD	• Low car ownership, but self-selection does more to explain this fact
Renne (2005)	Australia	848	×	×	×	Questionnaire survey	Descriptive	Average of the city	• No influence on car use
Chatman (2008)	California	1113	×	×	×	Telephone survey	Tobit regression	Similar demographic and built environment	• Higher car use
Dill (2008)	Portland	300	×	×	×	Questionnaire survey	Before-after	Movers	• 16% switched from non-transit to transit modes • Lower car use
Cervero and Arrington (2008)	Four regions in the USA	17 TOD projects		×		Vehicle count	Trip generation	ITE average rate	• Lower car use
Chatman (2013)	New Jersey	5193	×			Mail survey	OLS and logit regressions	Similar demographic and built environment	• Lower car use and ownership, but rail access does little to explain this fact
Nixon et al. (2015)	California	193	×	×	×	Mail survey	Before-after	Similar demographic and built environment	• Lower car use• Changes in bus service can negatively affect the overall transit ridership

Contradictory findings are reported on the impacts of rail transit on auto ownership and use. In a disaggregate study, Chatman (2008) showed that proximity to heavy rail has a positive correlation with VMT, after controlling for built environment variables including activity density, network load density and pedestrian connectivity. In other words, people residing near rail transit travel longer distances. Through an aggregate analysis, Loo et al. (2010) also concluded a positive correlation between auto ownership and rail transit ridership, presumably because of park-and-ride and kiss-and-ride activities. In contrast, several studies found a negative correlation. Dueker and Bianco (1999) studied the effects of LRT in Portland, Oregon, by employing a pre-post analysis. Using a parallel bus corridor as a control group, they found a slight negative effect on vehicle ownership. Further, the results show that households in the outer part of the rail corridor are more likely to use transit and be less auto-oriented compared with the control group. Dill (2008) conducted a survey near four LRT stations in Portland, Oregon, to explore whether residents of TODs drive less and use transit more. The results indicate that the net transit ridership increased by 16% in TOD corridors, of which 20% of the commuters switched from non-transit to transit modes. Further, a recent study within a two-mile radius of ten rail stations in New Jersey (Chatman, 2013) exhibited that car ownership and commuting are remarkably lower among households living in new housing near rail stations than those in new housing farther away. However, Chatman emphasised that access to rail transit plays an inconspicuous role in this behaviour because parking and neighbourhood characteristics are key drivers. Cervero and Arrington (2008) studied vehicle trips of 17 TOD multi-family housing projects near rail transit stations in four US metropolitan areas. Employing pneumatic-tube recorders, they revealed that the vehicle trips per dwelling unit are 44% lower than manual estimates.

Although the literature has grown in recent years, it leaves several gaps. First, while previous studies have concluded a positive association between rail transit and transit use, they have yet to answer the following question: Does rail transit catalyse transit use of station area residents? Or does rail transit attract frequent transit users? Previous studies were built on cross-sectional data that are less able to tease out the connections between travel behaviour and residential self-selection among station area residents. Since overlooking residential self-selection may misestimate the results, longitudinal analysis is in need. Second, together with rail transit, station area built environment not only persuades people already living in transit corridors to change their behaviour, but also attracts other people, or movers, to choose to live near transit stations. In light of the residential self-selection notion, movers and non-movers to rail station areas tend to behave heterogeneously (Cao and Schoner, 2014). However, to the best of our knowledge, few studies have attempted to understand the behaviour of movers. Third, previous studies have suffered from weak controlled corridors. Ipso facto, many studies measured travel mode share in corridors where people dwell within quarter- or half-mile distance rings of rail stations. The share, then, is compared with that of either the whole region or the outer parts of TOD corridors (Chatman, 2013; Lund et al., 2006; Renne, 2005). The consequence is that it overstates the impacts of transit, since people living in rail transit corridors tend to have higher transit share than those in the region before the introduction of rail transit (Cao and Schoner, 2014). Lastly, although rail transit aims to diminish the car-oriented lifestyle by promoting transit use, few studies have examined the nexus among rail transit, transit use and car ownership and use in a simultaneous framework.

This study attempts to fill the gaps from the following two perspectives: (1) choosing comparable corridors to examine travel behaviour of residents moving into the Hiawatha LRT corridor after its opening, (2) providing an insight into the mechanism that the LRT influences transit use, car ownership and car use. To achieve these goals, a quasi-longitudinal analysis is used for two main reasons: (1) the survey of the target population before the launching of LRT does not exist; (2) the pre-survey of movers is almost impossible.

Methodology

Conceptual model

This study attempts to explore the impacts of the Hiawatha LRT on travel behaviour of station area residents. Since the Hiawatha LRT commenced in 2004, a before-after test is impossible. Instead, we opted to compare residents who moved into the LRT corridor after its opening with those who moved into other corridors without LRT. We expect that residents who relocated into the Hiawatha corridor would experience transit improvement (Figure 1). Based on the literature and our informed knowledge, we further hypothesise that the improvement is positively correlated to a reduction in car ownership, which is in turn associated with an increase in transit use and a decrease in car use. Transit improvement is expected to have a negative association with change in car use and a positive association with change in transit use. We assume that the Hiawatha corridor also has associations with car ownership and travel behaviour and the associations may result from both observed and unobserved factors (such as changes in attitudes).

Figure 1.

Conceptual model.

Overall, the conceptual model illustrated in Figure 1 presents complex relationships among the Hiawatha LRT, transit improvement and travel behaviour. We need to employ the structure equations modelling approach (Mueller, 1996) to uncover the connections. In Figure 1, moving into the Hiawatha corridor is an exogenous variable and other variables are endogenous. Simply put, an exogenous variable impacts other variables but will not be affected by other variables whereas an endogenous variable can serve as a mediating variable or the dependent variable in an equation.

Data and variables

To test the conceptual model in Figure 1, we mailed a self-administered survey to households in five corridors in the Twin Cities in May 2011. It is worth noting that a data limitation is all measures of the built environment and transit quality are perceived and not objective measures. The description of data and variables is heavily borrowed from Cao and Cao (2014), who explore LRT effect on car ownership. The Hiawatha LRT line totals 12 miles, has 19 stations, and runs between downtown Minneapolis and the Mall of America in Bloomington, and through the Minneapolis-Saint Paul International Airport. Five stations are located around downtown Minneapolis and six stations mainly serve the airport and the Mall of America (http://metrotransit.org/light-rail). These station areas mostly accommodate commercial or institutional land uses, whereas the middle section of the LRT line is mainly for industrial and residential uses. The downtown Minneapolis, LRT line, and LRT stations along with the Mall of America are depicted in Figure 2. In this study, the Hiawatha corridor means the middle section: the corridor within a half mile of the Hiawatha LRT between Lake Street Station and 50th Street Station in South Minneapolis. The corridor is shaded by light grey in Figure 2.

Figure 2.

Locations of corridors.

We employed a case-control observational design and chose two sets of control corridors. As shown in Figure 2, the corridors along Nicollet Avenue and Bloomington Avenue in South Minneapolis were chosen as urban control corridors to resemble the Hiawatha corridor in terms of location context, built environment elements and demographics. The urban corridors have similar transit services but are not served by LRT. Suburban control corridors are chosen from Coon Rapids and Burnsville. These two corridors are located directly north (14 miles) and south (17 miles) of downtown Minneapolis. Their key demographic characteristics (such as household income and size) are relatively similar to those of the Hiawatha corridor. The suburban corridors were mainly developed in the 1970s. Compared with the Hiawatha corridor, they have different street networks and poor transit service (Figure 3). Table 2 illustrates the information of transit service in five corridors.

Figure 3.

Map of corridors.

Table 2.

Transit service in five corridors in 2010.

Corridors	Service hours	Vehicle trips
Weekday
Hiawatha	1329	1620
Nicollet	1278	1603
Bloomington	1062	1125
Coon Rapids	158	200
Burnsville	222	291
Weekend
Hiawatha	1851	2318
Nicollet	1188	1479
Bloomington	1467	1572
Coon Rapids	45	50
Burnsville	50	55

We ordered a database of ‘movers’ and a database of ‘non-movers’ for each corridor from a commercial data provider, AccuData Integrated Marketing (http://www.accudata.com). In this study, we use the movers, who included all residents who had relocated to the corridors after 2004 when the Hiawatha LRT commenced. We asked the provider to randomly draw a sample of approximately 1000 residents in the Hiawatha corridor and approximately 500 residents in each of the Nicollet, Bloomington, Coon Rapids and Burnsville corridors.

We invited students and staff members of our School, neighbours and friends of the principal investigator to pre-test the survey. We revised the survey based on pre-testers’ feedback. Two reminder postcards were mailed one and two weeks after we posted the survey. As an incentive for participation, ten US$50 gift cards were provided through a lottery. The original database of movers consisted of 3040 addresses; however only 2951 were valid. The number of responses totaled 597, equivalent to a 20.2% response rate based on the valid addresses only. This is a good response rate for a long (ten-page) survey, because the typical rate for the general population survey ranges from 10% to 40% (Sommer and Sommer, 1997). Table 3 compares the characteristics of movers for different corridors. Urban control corridors have the highest population density and four-way intersection density, followed by the Hiawatha corridor and then suburban control corridors. There are no significant differences in demographics between Hiawatha movers and urban movers. However, movers in suburban corridors tend to be older and less educated than movers in the Hiawatha and urban corridors. Employment rate is also lower in suburban corridors.

Table 3.

Sample characteristics of movers.

	Hiawatha movers	Urban movers	Suburban movers
Built environment
Population density	9.2	13.9	5.0
Land use entropy index	0.33	0.30	0.31
Four-way intersection density	0.16	0.17	0.05
Demographics
Household size	2.22	2.28	2.52
Income	5.65	5.73	5.16
Education	4.25	4.40	3.77
Age	40.6	41.1	45.8
Share of workers	0.84	0.89	0.71
Share of renters	0.27	0.28	0.31
Share of female	0.56	0.52	0.56

Notes: Bold variables are significantly different between suburban movers and urban (Hiawatha) movers at the 0.05 level (Bonferroni tests of analysis of variance). Income and education were measured in ordinal scales. Categories for education are 1: some grade school, 2: high school diploma, 3: some college, 4: college degree, 5: some graduate school, and 6: graduate degree. Categories for income are 1: US$ 0–14,999, 2: US$15,000–24,999, 3: US$25,000–34,999, 4: US$35,000–44,999, 5: US$45,000–59,999, 6: US$60,000–74,999, 7: US$75,000–99,999, 8: US$100,000–124,999, and 9: US$125,000 or more.

The variables used in this study include four categories: travel behaviour, demographics, neighbourhood characteristics and residential preferences. In the survey we asked respondents to indicate how many personal vehicles (cars, SUVs, vans, small trucks and motorcycles) their households currently have and how many they had just before they moved. The difference in the two numbers is the change in the number of cars after residential relocation. Respondents were also asked to report ‘How much do you drive now, compared to when you lived at your previous residence?’ on a five-point ordinal scale from ‘a lot less now’ (1) to ‘a lot more now’ (5). A similar question asked their use of transit including bus and LRT. The two variables represent change in car use and change in travel use after residential relocation. In addition to common demographic characteristics, the survey measured changeable demographics such as household income and household size, which help explain movers’ changes in travel behaviour.

We measured perceived neighbourhood characteristics in the survey. In particular, we asked respondents to indicate how true 30 characteristics are for their current and previous neighbourhoods on a four-point scale from ‘not at all true’ (1) to ‘entirely true’ (4). They are associated with living units, land use and transportation systems, social environment, safety and so on. In terms of residential preferences, we measured the importance of the 30 characteristics to respondents when they were looking for a new place to live on a four-point scale from ‘not at all important’ (1) to ‘extremely important’ (4). Because some of the characteristics are highly correlated, we conducted a joint factor analysis to reduce them (after dropping some) to eight dimensions: transit, attractiveness, spaciousness, safety, quietness, safety, socialising and physical activity infrastructure (Table 4). Each of current neighbourhood characteristics, previous neighbourhood characteristics, and residential preferences included the same eight factors. Changes in neighbourhood characteristics were computed as the difference in the factors between movers’ current and previous neighbourhoods.

Table 4.

Pattern matrix for perceived and preferred neighbourhood characteristics.

	Attractiveness	Spaciousness	Physical activity infrastructure	Transit	Accessibility	Quietness	Safety	Socialising
High quality living unit	0.605
Good investment potential	0.391
Attractive appearance of neighbourhood	0.673
High level of upkeep in neighbourhood	0.609
Large back yards		0.640
Lots of off-street parking (garages or driveways)		0.481
Sidewalks throughout the neighbourhood			0.452
Good bicycle routes beyond the neighbourhood			0.839
Parks and open spaces nearby			0.377
Good public transit service (bus or rail)				0.964
Easy access to transit stop/station				0.921
Shopping areas within walking distance					0.393
Easy access to a regional shopping mall					0.720
Easy access to downtown					0.375
Religious or civic buildings nearby					0.507
Low level of car traffic on neighbourhood streets						0.874
Quiet neighbourhood						0.735
Low crime rate within neighbourhood							0.426
Safe neighbourhood for walking							0.815
Safe neighbourhood for kids to play outdoors							0.540
Lots of interaction among neighbours								0.784
Lots of people out and about within the neighbourhood								0.752

Note: The method was Principal axis factoring with Oblimin with Kaiser Normalisation. Loadings smaller than 0.300 were suppressed.

Table 5 illustrates that transit improvement and transit preference differ among the three corridors. Bonferroni tests further indicate that residents moving into the Hiawatha LRT corridor tend to experience a higher transit improvement than those moving into the urban and suburban control corridors and Hiawatha movers prefer transit more strongly than movers in the control corridors (results not shown). Thus, transit-preferring individuals did self-select into the Hiawatha corridor. Moreover, Hiawatha movers are more likely to increase transit use than their counterparts in the control corridors whereas there are no significant differences in change in car ownership and use among the three corridors.

Table 5.

Key variables by corridor.

	Transit improvement	Transit preference	Chang in # of cars	Change in transit use^a	Change in car use^a
Hiawatha	0.944	0.170	0.153	3.445	2.747
Urban	0.516	−0.178	0.107	3.106	2.679
Suburban	0.220	−1.031	0.069	2.598	2.699
ANOVA p-values	0.000	0.000	0.588	0.000	0.860

Note: ^aValues larger than 3 indicate an increase.

Modelling and results

Since the data includes respondents from three types of corridors, we create two corridor dummy variables (Hiawatha and suburban) with urban corridors as the reference category. Similar to the Hiawatha corridor in Figure 1, we assume that suburban control corridors are associated with transit improvements and changes in travel behaviour. However, the associations with suburban control corridors are expected to have opposite signs to those with the Hiawatha corridor. In the conceptual model, change in the number of cars was treated as an endogenous variable. However, none of our key variables (including the corridor dummies and transit improvement) had direct effects on it. This is consistent with Cao and Cao (2014), who analyse change in car ownership using the same data. So it was relaxed as an exogenous variable. The structural equations model (SEM) approach was employed using the maximum likelihood estimation in Stata 12.0. Because the modelling approach requires the data to follow multivariate normal distribution and our data may not meet the requirement, we adopted the bootstrap method to produce the variance-covariance matrix of the estimates. We adopted an incremental modelling approach and developed two SEMs, with and without accounting for the influences of additional confounding factors. When estimating models, the exogenous variables that were insignificant at the 0.10 level were manually dropped to obtain parsimonious models.

We use Root Mean Square Error of Approximation (RMSEA), Comparative Fit Index (CFI) and Standardised Root Mean Squared Residual (SRMR) to evaluate the goodness of fit for the SEMs. According to the Stata manual (http://www.stata.com/manuals13/semestatgof.pdf, accessed on 17 May 2015), RMSEA values less than 0.05 indicate a good fit; CFI values closer to 1 indicate a good fit; and SRMR values less than 0.08 indicate an acceptable fit. As shown in Figures 4 and 5, both models have adequate measures of goodness of fit.

Figure 4.

Structural equations model without controlling for confounding factors.

Figure 5.

Structural equations model while controlling for confounding factors.

Figure 4 shows the SEM without controlling for confounding factors. The error terms of change in transit use and change in car use are significantly and negatively correlated. This indicates that unobserved factors influence the two variables in an opposite way. This is reasonable because transit and driving are two competing means of transport. Moving into the Hiawatha corridor is positively associated with transit improvement whereas the latter has a negative association with moving into suburban corridors. Transit improvement is positively correlated with change in transit use but has a negative connection with change in car use. That is, the Hiawatha LRT affects transit use and car use through its impact on transit improvement. Additionally, Hiawatha and suburban dummy variables are directly associated with change in transit use. However, their direct links with change in car use are insignificant. Furthermore, change in the number of cars is positively associated with change in car use but has no significant influence on change in transit use. Overall, all of the observed significant effects are consistent with our assumptions.

Figure 5 considers the influences of potential confounding factors including changes in neighbourhood characteristics, changes in demographics and residential preference for transit. After controlling for the confounding factors, all of the variables significant in Figure 4 are also significant in Figure 5.

In terms of confounding factors, changes in other neighbourhood characteristics may bring about changes in travel behaviour. We found that change in social attributes (the socialising factor) is negatively associated with change in car use. Therefore, the presence of people and interactions with neighbours, or a sense of community, help reduce driving. This finding is consistent with Aditjandra et al. (2012). Other neighbourhood characteristics do not significantly affect changes in travel use and car use. Changes in demographics may also influence changes in travel behaviour. We tested change in household income and household size and found that only change in household size has a positive association with change in car use. Furthermore, residential preference for transit has a positive association with change in transit use but is insignificant for change in car use.

Overall, transit improvement (in terms of service quality and access to stops) is positively correlated to environment-friendly travel behaviour (either an increase in transit use or a decrease in driving). This study controls for confounding factors, particularly residential preference for transit, and presumably eliminates the rival hypothesis that individuals preferring transit intentionally choose to live near the LRT. Further, respondents reported changes in travel behaviour after moving into new neighbourhoods with perceived better transit service. Accordingly, this study seems to establish the three prerequisites for a causal inference (Singleton and Straits, 2005): association, non-spuriousness and time precedence. Thus, it offers stronger evidence on the causal influence of transit service on travel behaviour than do previous studies.

Moving into the Hiawatha corridor leads to an increase in transit use directly and indirectly through transit improvement. Its total effect is positive and significant. This result is contradictory to Cao and Schoner (2014), who found transit use of movers into the Hiawatha corridor and urban corridors is similar. Since we only measured the change in the level of transit use, we hypothesise that before their residential relocation, movers into urban corridors used transit more frequently than their counterparts in the Hiawatha corridor. Although they have similar transit use now, the former tended to have a smaller change in transit use than the latter. This implies the process of residential self-selection: movers relocated into urban corridors to match their preference for travel, leading to a similar level of transit use. Therefore, the LRT is a magnet that attracts potential transit users. Further, this study adopts a quasi-longitudinal design. Modelling changes in variables is able to cancel out the influences of time-invariant variables. This is better than the cross-sectional design used by Cao and Schoner (2014). Although they adopted the propensity score matching approach to adjust observed confounding factors, the unobserved confounders may still bias the results.

The effect of the Hiawatha dummy may capture the influences of many factors in the Hiawatha corridor. Besides its influence through transit improvement, the Hiawatha dummy variable is directly and positively associated with change in transit use. Therefore, some unobserved elements of the LRT corridor affect transit use. Without separating transit service from other influential elements, we are unable to disentangle the impact of the LRT. Future studies should also attempt to identify other elements with a true longitudinal analysis.

This study has a few limitations. It used retrospective measures. Because of potential recall bias, we focused on recall accuracy instead of measurement precision. Changes in travel behaviour were measured in an ordinal scale. Moreover, perceived neighbourhood characteristics may be subject to reporting bias. For example, residents who use transit frequently may be more aware of transit attributes than those who do not use transit, and thus it may inflate the impact of neighbourhood characteristics on transit use. However, this will not affect the impact of the Hiawatha corridor (see Table 4 and Figure 5) because changes in neighbourhood characteristics are mediating variables. On the other hand, a true longitudinal study is essential to obtain precise estimates of the influence of neighbourhood characteristics on travel behaviour, as Giles-Corti et al. (2013) did. Scholars could also design a companion study: a longitudinal design compared with a quasi-longitudinal design. This will help quantify recall and reporting biases and their impacts on travel behaviour. Nevertheless, this study offers insights on the influences of LRT on changes in car ownership, transit use and car use simultaneously.

Conclusions

This study employs the structural equations modelling approach to examine the nexus among rail transit, transit use, car ownership and car use in a simultaneous framework. It chooses comparable corridors to contrast with the Hiawatha LRT corridor, and compares the residents who relocated into the Hiawatha corridor and control corridors after the LRT started revenue service. The study adopts a quasi-longitudinal design by retrospectively asking respondents to report their changes in travel behaviour and neighbourhood characteristics.

The Hiawatha LRT acts as both a catalyst and a magnet. As a magnet, it attracts movers who already are predisposed to using transit. In this case, the movers are self-selecting into the Hiawatha corridor because of the proximity to the LRT. As a catalyst, residents who move into the Hiawatha corridor for other reasons experience transit improvements, which result in an increase in transit use and a reduction in driving. Empirically, residents moving into the Hiawatha LRT corridor do experience transit improvement, which increases transit use and reduces car use. Since this study adopts a quasi-longitudinal design and controls for transit preference, we are more confident to conclude that the Hiawatha LRT affects travel behaviour by improving transit service. This illustrates the mechanism that the LRT influences travel behaviour. The capacity of the Hiawatha LRT to reduce car use is particularly prominent because few studies have addressed the issue. On the other hand, residents who prefer transit move into the Hiawatha corridor and thus the LRT attracts transit-liking people. This self-selection does not mean that the LRT does not have benefits. The LRT enables transit-liking people who were unable to realise their travel preference for using transit with greater frequency. That is, they can self-select and increase transit use (Levine, 1999; Næss, 2009). Further, about 1000 new housing units have been constructed near the Hiawatha LRT since it opened. The residents in these units would have had to live in other urban neighbourhoods or suburban neighbourhoods if the LRT were not built. Regardless of whether the LRT is a magnet or catalyst, the increase in transit use and reduction in car use associated with these housing units are non-trivial, particularly compared with those people living in suburban neighbourhoods.

There is no difference in changes in car ownership among the three corridors. This suggests that the accessibility benefits of a single LRT line are inadequate to enable many residents to shed cars. Choice riders tend to use transit for commute trips and may still rely on personal vehicles for non-work trips.

Footnotes

Funding

Funding was received from the Transitway Impact Research Program in the Twin Cities.

References

Aditjandra

Cao

Mulley

(2012) Understanding neighbourhood design impact on travel behaviour: An application of structural equations model to a British metropolitan data. Transportation Research Part A: Policy and Practice 46: 22–32.

Cain

Darido

Baltes

. (2007) Applicability of Transmilenio Bus Rapid Transit System of Bogotá, Colombia, to the United States. Transportation Research Record: Journal Of The Transportation Research Board 2034: 45–54.

Cao

(2014) The impacts of LRT, neighbourhood characteristics, and self-selection on auto ownership: Evidence from Minneapolis-St. Paul. Urban Studies 51: 2068–2087.

Cao

Schoner

(2014) The influence of light rail transit on transit use: An exploration of station area residents along the Hiawatha line in Minneapolis. Transportation Research Part A: Policy and Practice 59: 134–143.

Cervero

(2007) Transit-oriented development’s ridership bonus: A product of self-selection and public policies. Environment and Planning A 39: 2068–2085.

Cervero

Arrington

(2008) Vehicle trip reduction impacts of transit-oriented housing. Journal of Public Transportation 11: 1–17.

Cervero

Gorham

(1995) Commuting in transit versus automobile neighborhoods. Journal of the American Planning Association 61(2): 210–225.

Chatman

(2008) Deconstructing development density: Quality, quantity and price effects on household non-work travel. Transportation Research Part A: Policy and Practice 42: 1008–1030.

Chatman

(2013) Does TOD need the T? On the importance of factors other than rail access. Journal of the American Planning Association 79: 17–31.

10.

Dill

(2008) Transit use at transit-oriented developments in Portland, Oregon, area. Transportation Research Record: Journal of the Transportation Research Board 2063: 159–167.

11.

Dueker

Bianco

(1999) Light-rail-transit impacts in Portland: The first ten years. Transportation Research Record: Journal of the Transportation Research Board 1685: 171–180.

12.

Environmental Protection Agency (2014) Inventory of U.S. Greenhouse Gas Emissions and Sinks: 1990–2012. Washington, DC: US Environmental Protection Agency.

13.

Ewing

Cervero

(2001) Travel and the built environment: A synthesis. Transportation Research Record: Journal of the Transportation Research Board 1780: 87–114.

14.

Ewing

Cervero

(2010) Travel and the built environment – A meta-analysis. Journal of the American Planning Association 76: 265–294.

15.

Giles-Corti

Bull

Knuiman

. (2013) The influence of urban design on neighbourhood walking following residential relocation: Longitudinal results from the reside study. Social Science & Medicine 77: 20–30.

16.

Holtzclaw

(1994) Using Residential Patterns and Transit to Decrease Auto Dependence and Costs. San Francisco, CA: Natural Resources Defense Council.

17.

Levine

(1999) Access to choice. Access 14: 16–19.

18.

Litman

(2005) Impacts of rail transit on the performance of a transportation system. Transportation Research Record: Journal of the Transportation Research Board 1930: 21–29.

19.

Loo

BPY

Chen

Chan

ETH

(2010) Rail-based transit-oriented development: Lessons from New York City and Hong Kong. Landscape and Urban Planning 97: 202–212.

20.

Lund

Cervero

Willson

(2006) A re-evaluation of travel behavior in California’s TODs. Journal of Architecture and Planning Research 23: 247–263.

21.

McCahill

(2014) Per capita VMT drops for ninth straight year. Dots Taking Notice. Madison, WI: State Smart Transportation Initiative.

22.

Moniruzzaman

Páez

(2012) Accessibility to transit, by transit, and mode share: Application of a logistic model with spatial filters. Journal Of Transport Geography 24: 198–205.

23.

Mueller

(1996) Basic Principles of Structural Equation Modeling: An Introduction to Lisrel and Eqs. New York: Springer.

24.

Næss

(2009) Residential self-selection and appropriate control variables in land use: Travel studies. Transport Reviews 29: 293–324.

25.

Nixon

Boarnet

Houston

. (2015) Changes in Transit Use and Service and Associated Changes in Driving Near a New Light Rail Transit Line. San Jose, CA: Mineta Transportation Institute.

26.

Renne

(2005) Transit Oriented Development: Measuring Benefits, Analyzing Trends, and Evaluating Policy. New Brunswick: Urban Planning and Policy Development Program in the Bloustein School of Planning and Public Policy, Rutgers University.

27.

Singleton

Straits

(2005) Approaches to Social Research. 4th edition. New York: Oxford University Press.

28.

Sommer

(1997) A Practical Guide to Behavioral Research: Tools and Techniques. 4th edition. New York: Oxford University Press.