Puget Sound Transportation Panel Research Articles

A High-Order Hidden Markov Model and Its Applications for Dynamic Car Ownership Analysis

This paper extends the dynamically formulated hidden Markov models to a high-order hidden Markov model (HO-HMM) formulation. In the HO-HMM, the Markovian assumption that the future states (interpreted as the states of preferences or attitudes) depend only on the current state has been relaxed. Instead, the HO-HMM generalizes that the future states will depend on a number of states occurring beforehand. This paper develops the theoretical formulation of a HO-HMM framework. A recursive algorithm of likelihood computation is derived for model estimation. The algorithm significantly reduces the complexity of estimation and ensures the applicability of high-order hidden Markov modeling. The proposed methodology is further demonstrated on a vehicle ownership choice application using Puget Sound Transportation Panel data coupled with a few supplementary data sources. Long-term life-cycle stage changes in households are used as proxies for the high-order Markov transitions in car ownership hidden states. Results indicate that the HO-HMM has superior explanatory power in fitting longitudinal data.

Triggers of behavioral change: Longitudinal analysis of travel behavior, household composition and spatial characteristics of the residence

Longitudinal data have the potential to reveal the causal mechanism underlying changes in observed behavior, can protect us from finding spurious relationships, study time precedence, and offer strength in helping discover association among variables. In this paper, we expand our previous analysis of the only longitudinal travel behavior database in the US that has ten waves (observation time points). The analysis here is using data from the two-day Puget Sound Transportation Panel (PSTP) travel diary and 230 households participating in all ten waves of the panel. The main objective of this analysis is to identify what triggers behavioral change in activity and travel and to quantify and compare the size of these triggering effects using different methods. Our search for triggers includes the birth of a child, child leaving the household, and the entry or exit from the labor force of a household member. In the land use changes we identify moves from high accessibility and diverse environments to lower accessibility and less diverse places and find that diversity is key in changing behavior. The method used to analyze the data is a longitudinal Mixed Markov Latent Class analysis that allows to not only account for behavioral heterogeneity at each time of observation but to also test for the existence of multiple latent trajectories of change and the role played by triggers in shaping these trajectories.

Hidden Markov Approach to Dynamically Modeling Car Ownership Behavior

It has become apparent to researchers in various domains that choice behavior occurs in a dynamic context and decision making involves strong temporal dependency, especially when it comes to car ownership decisions, because of consumers’ forward-looking behavior. However, a substantial portion of the literature focuses on static model formulations, and limitations exist, particularly in long-term travel demand forecasting. This study proposed a hidden Markov modeling (HMM) framework to analyze car ownership behavior dynamically. The dynamic model framework was applied to the 10-wave Puget Sound (Washington) Transportation Panel data. Two hidden states were identified in this study: State 1 tended to be land use entropy sensitive and vice versa for State 2. Empirical results reveal that households with preschool-age children are more likely to live in urbanized areas where they have easy access to various facilities. Also, one more licensed driver would lead to a 13.33% increase in owning two cars for State 1 households and a 28.45% increase in owning three or more cars for State 2 households. The comparison with both the multinomial logit model and the latent class model favors the study’s dynamic model framework with respect to model performance. The HMM approach offers insights on policy development for a target population and provides more accurate forecasting for long-term planning and policy analysis.

Behavioral Housing Search Choice Set Formation

The housing search process, a topic of interest to both practitioners and researchers, starts with an alternative formation and screening practice. Due to the limitation of cognitive capacity, household members at this level evaluate potential alternatives based on many factors, such as lifestyle, preferences, and so on, to form a manageable choice set. This article attempts to provide a detailed study of this screening and filtering practice to develop a modeling framework that can replicate the choice set formation process. In order to show the potential of the method, one prospective decision criteria—the average desired commute to work distance—is considered the potential attribute that the household evaluates for feasible housing alternatives. It is postulated that alternatives will only be included in the choice set if the average work distance satisfies the household distance threshold. This article explores the viability of using proportional hazard models in the housing search process. Some of the specifications of hazard-based models that are typically used on temporal data are examined on average work distance. Several household sociodemographic attributes from eight waves of the Seattle Metropolitan Area’s Puget Sound Transportation Panel (PSTP) are utilized for model estimation, along with built environment variables, characteristics of the supply side of the market, and several other economic indicators. The approach presented in this article provides a remedy for the large choice set problem typically faced in discrete choice modeling.

Attitudes, mode switching behavior, and the built environment: A longitudinal study in the Puget Sound Region

Carpooling in the US has a storied history. After experiencing a peak 20% mode share in 1980, the current share of carpooling for work trips is about 10% and the majority of these carpooling trips are made by intra-household members. Casting the choice between SOV and carpool as a social dilemma in which SOV is a noncooperative choice and carpool is a cooperative one, we propose to test two hypotheses. First, the switch from SOV to carpool and the reverse choice are attributed to different factors—structural factors, or those factors altering the objective features of a decision scenario such as travel time and travel cost, play a dominant role in the switch from carpool to SOV while psychosocial factors (attitudes and beliefs) play a critical role in the switch from SOV to carpool. Second, the two choices are underlay by different behavioral mechanisms. In particular, we expect self-justification by carpool-to-SOV switchers—after they switch from carpool to SOV, they adjusted their attitudes toward carpool accordingly to match their behavior. The analysis of the first three waves of the Puget Sound Transportation Panel supports these two hypotheses. Our study results recommend developing programs and policies that aim at influencing people’s subjective assessments of carpooling, in addition to the existing ones that mostly focus on incentivizing carpooling, and differentiating between programs seeking to encourage SOV users to switch to carpool and those aiming to maintain existing carpoolers.

Intra-regional residential movement of the elderly: testing a suburban-to-urban migration hypothesis

The majority of the elderly and soon-to-be retired baby boomers live in suburbs in the US. This paper empirically tested an intra-regional residential movement of elderly households, which is postulated based on life course theory. Using data from the Puget Sound Transportation Panel survey in Washington State in the US, this study found that elderly households are less likely to change their residential environments than younger households upon intra-regional residential movement. This study also found that elderly households tend to move from less urbanized neighborhoods to slightly more urbanized neighborhoods, while younger households tend to move from quite urban neighborhoods to substantially less urbanized neighborhoods.

A structural equation model of activity participation and travel behavior using longitudinal data

This paper presents a dynamic structural equation model (SEM) that explicitly addresses complicated causal relationships among socio-demographics, activity participation, and travel behavior. The model assumes that activity participation and travel patterns in the current year are affected by those in previous years. Using the longitudinal dataset collected from Puget sound transportation panel ‘wave 3’ and ‘wave 4,’ these assumptions are tested with suggested SEMs. Within each wave, the model is structured to have a three-level causal relationship that describes interactions among endogenous variables under time-budget constraints. The resulting coefficients representing the activity durations indicate that people tend to allocate their time according to the importance and the obligation of the activity level. Results from the dynamic SEM confirm the fact that people's current activity and travel behavior do have effects on those in the future. The resulting model also shows that activity participation and travel behavior in ‘wave 3’ are closely related to those in ‘wave 4.’ These explicit explanations of relationships among variables could provide important perspectives in the activity-based approach which becomes recognized as a better analytical tool for the transportation planning and policy making process.

Commuting time stability: A test of a co-location hypothesis

During the past few decades, the populations of many US and European cities have shown the same residence and workplace mobility patterns: Annually, approximately 10% of the population changed residences and approximately 20% of employed workers changed workplaces within the same metropolitan area. Even though the Seattle metropolitan region experienced a substantial amount of residential and workplace mobility and a boom in employment and population in the 1990s, the morning commute time and distance was roughly constant. To explain this situation, researchers have proposed a co-location hypothesis, that is, residents and workers will change their residence or workplace or both adapt to worsening congestion. This research attempted to shed light on the mechanism of the co-location hypothesis using the Puget Sound Transportation Panel data consisting of seven waves of two consecutive years between 1989 and 1997 conducted by the Puget Sound Regional Council. Because most studies used cross-sectional work trip data to study location and commuting, the underlying relationship between location and commuting was limited. This study attempted to understand commuting patterns by residential and workplace changers. The study found that the commuting patterns of residence and workplace location changers were the same and this was reinforced by existing gravity model. Particularly, when workers change their locations, they prefer similar commuting zone (i.e., time and distance) compared to their previous commuting zone. These behaviors caused the average commute time and distance to be stable, regardless of high residence and workplace mobility, and the rapid growth of employment and population.

Catching the Next Big Wave

The aging of American baby boomers creates various new policy contexts and problems. Depending on institutional preparedness and baby boomers’ behavior, the changing demand for transportation services by this group may be positive or negative. This potential change in demand is described through an analysis of individual longitudinal histories over a long period (1989 to 2003) that explores the effects of personal changes (e.g., entry to and exit from the labor force), household changes (e.g., relocation and dissolution), and changes in land use characteristics. The Puget Sound Transportation Panel (PSTP)—a record of approximately 20,000 travel diaries of Seattle residents, each of whom provided 2-day reports for 10 repeated contacts (waves)—was used. Within-household dynamics and the effect of within-household change on individual and household behavior are studied. Focus groups are used to extract behavioral themes, latent class cluster analysis is used to identify groups of behavior, and an array of regression models of change is used to identify the key determinants that underlie behavioral dynamics. Findings include a need to focus on employment, heterogeneity in land use impact, and the significant effect of household composition. Together, the findings imply the need for models that can handle more diverse behavior and the need to accommodate employment status and within-household demographics in forecasting models.

Guidelines for samplers: measuring a change in behaviour from before and after surveys

This paper addresses the issue of using before and after surveys to evaluate behavioural changes in response to transport policies and investments, a procedure that, we argue is done far too rarely in this profession. Further, it demonstrates very clearly that, in almost all conceivable cases, there are considerable economies to be obtained by using a panel (again, under-utilised in our profession) to undertake evaluation, rather than successive independent cross-sectional surveys. The paper also addresses the critical issue of sample size requirements for measuring changes of a relatively small magnitude in travel behaviour; i.e., to state, with 95% confidence, that if there is a ∂ percent change in behaviour for the sample, there is a ∂ percent ± e percent change in the behaviour of the population, where e is the sampling error. In this paper, we present a method for calculating such sample size requirements from first principles and demonstrate the applicability both hypothetically and then empirically using data from the Puget Sound Transportation Panel. The formulation enables designers of before and after surveys to investigate the trade-offs between the statistical accuracy of their predictions and the sample size requirements systematically, without the need to specify ∂ a priori. This latter point is crucial, we argue, because we have limited information on ∂, yet, as we explain here, it drives the sample size requirements using alternative, well-cited approaches for calculating sample sizes to assess behavioural change. The results have important ramifications both for those implementing transport policies intended to produce behavioural change, especially when a cost-benefit evaluation of the policy is desired, and for those interpreting the results reported in previous studies.

Multilevel Structural Equation Model for Activity Participation and Travel Behavior: Data from the Puget Sound Transportation Panel

A multilevel structural equation model (SEM) is applied to handle the hierarchical nature of the data and explain complex relationships among socioeconomic factors for individuals and households, activity participation, and travel behavior by using data from Wave 3 of the Puget Sound Transportation Panel. On the basis of the model structure, variations in individual activity participation and travel behavior can be divided into two categories: within-household variation and between-household variation. Empirical results show that the interdependency index of leisure activity duration for household members within a household has a value of 0.33, indicating high interdependency. Total travel time duration likewise has a value of 0.26, indicating high similarity. These results suggest that the multilevel SEM approach is an appropriate modeling methodology and can provide additional information on activity and travel behavior. The effects from exogenous variables on endogenous variables developed in this study also provide meaningful and interpretable results, which are consistent with results of previous studies. Most personal and household characteristics influence activity participation and travel behavior within a household as well as between households.

TRAVEL BEHAVIOR CHANGES: EVIDENCE FROM A LONGITUDINAL TRAVEL SURVEY

Changes in travel behavior and in the factors affecting travel behavior can provide useful input to policy definitions, and also have important implications for the design of decision support systems for transport policy analysis. The primary source of behavioral changes and models for decision support systems is a panel survey; that is, repeated observations by the same persons over time. This paper analyzes data from 9 waves of the Puget Sound Transportation Panel in order to study travel behavior change and the sources of this travel behavior change. The frequency of trips and the proportion for each mode are studied by means of a series of Poisson and linear regression models. A complex pattern of relationships emerged from this analysis with some findings confirming prior research, while others indicated inconsistent correlations in time.

Analysis of Lifestyle Choices: Neighborhood Type, Travel Patterns, and Activity Participation

Activity-based travel modeling has begun to make significant progress toward a more behavioral framework for simulating household travel behavior. A significant challenge remains in the need to address the interaction of daily activity and travel patterns with longer-term household choices of vehicle ownership, residential location, and employment location. The choices often depend on one another and jointly define the lifestyle of the household. These choices are likely to evolve over the course of the life cycle as households are formed; as children are born, raised, and ultimately depart to form their own households; and as retirement and old age change patterns of residence, work, and travel. A framework is developed for analyzing household choices relating to three dimensions of lifestyle: travel patterns (including vehicle ownership), activity participation, and residential location (neighborhood type). With cluster analysis on data from the Puget Sound Transportation Panel, nine classifications of lifestyle are uncovered. These clusters demonstrate empirically how decisions of residential location reinforce and affect daily decisions related to travel patterns and activity participation. The applicability of these lifestyle clusters for land use transportation planning is discussed.

Analysis of activity duration using the Puget sound transportation panel

In this paper, we extend previous analyses of temporal effects by examining the factors associated with activity generation and duration over four waves of the Puget sound transportation panel survey (PSTPS). A Cox proportional hazards model was specified for each of five activities: visiting, appointment, free time, personal business, and shopping. For each activity, the duration times are modeled with an emphasis on examining important higher order interactions. The results suggest that activity durations have changed significantly over the survey period. Many of the differences in activity durations over time were significant, and often associated with increasing numbers of children in the household and higher order interactions between sex and the sequencing of activities.

Linking land use with household vehicle emissions in the central puget sound: methodological framework and findings

A leading cause of air pollution in many urban regions is mobile source emissions that are largely attributable to household vehicle travel. While household travel patterns have been previously related with land use in the literature (Crane, R., 1996. Journal of the American Planning Association 62 (1, Winter); Cervero, R. and Kockelman, C., 1997. Transportation Research Part D 2 (3), 199–219), little work has been conducted that effectively extends this relationship to vehicle emissions. This paper describes a methodology for quantifying relationships between land use, travel choices, and vehicle emissions within the Seattle, Washington region. Our analysis incorporates land use measures of density and mix which affect the proximity of trip origins to destinations; a measure of connectivity which impacts the directness and completeness of pedestrian and motorized linkages; vehicle trip generation by operating mode; vehicle miles/h of travel and speed; and estimated household vehicle emissions of nitrogen oxides, volatile organic compounds, and carbon monoxide. The data used for this project consists of the Puget Sound Transportation Panel Travel Survey, the 1990 US Census, employment density data from the Washington State Employment Security Office, and information on Seattle’s vehicle fleet mix and climatological attributes provided by the Washington State Department of Ecology. Analyses are based on a cross-sectional research design in which comparisons are made of variations in household travel demand and emissions across alternative urban form typologies. Base emission rates from MOBILE5a and separate engine start rates are used to calculate total vehicle emissions in grams accounting for fleet characteristics and other inputs reflecting adopted transportation control measures. Emissions per trip are based on the network distance of each trip, average travel speed, and a multi-stage engine operating mode (cold start, hot start, and stabilized) function.

Period Effects and Cohort Effects in Life Cycles: Preliminary Analysis

Travel behavioral data from five successive waves of the Puget Sound Transportation Panel were examined to determine whether period effects or cohort effects have a significant effect in life-cycle behavior. It was found that period and cohort effects may have a greater influence in household life-cycle models than previously believed. The results bring into question current methods using cross-sectional data analysis for life-cycle models, since predicted changes in activity behavior were not observed in households that made life-cycle transitions. The results are not conclusive, since other variables influencing activity behavior were not accounted for in the analysis.

Dynamic Analysis of Traveler Attitudes and Perceptions Using Panel Data

A dynamic analysis of travelers’ attitudes, preferences, and values was carried out using three waves of the Puget Sound Transportation Panel survey to investigate dynamics in traveler attitudes and perceptions. An in-depth descriptive analysis was performed to examine the variations in attitudinal ratings over time. A traditional one-way analysis of variance (ANOVA) method was used to explore for similarities and differences in traveler attitudinal ratings across different waves of the panel survey. A similar analysis was performed on the stayer sample (i.e., the respondents who participated in all three waves of the panel used in this study). The ANOVA results show significant differences in mean attitudinal ratings across the three waves of the panel survey. The differences in traveler attitudes and perceptions among stayers, dropouts (respondents who leave the panel survey), and refreshments (respondents who are newly recruited as the panel survey proceeds) were also captured. Finally, differences in traveler attitudes and preferences across different modal market segments were examined. The results indicate the need for greater consideration of attitudinal dynamics in transportation planning and policy analysis.

Longitudinal analysis of activity and travel pattern dynamics using generalized mixed Markov latent class models

Understanding the dynamics of time allocation by households and their household members is becoming increasingly important for travel demand forecasting. A unique opportunity to understand day-to-day and year-to-year behavioral change, is provided by data from multi-day travel diaries combined with yearly observation of the same individuals over time (panel surveys). In fact, the “repeated” nature of the data allows to distinguish units that over time change their behavior from those that are not and to uncover the underlying stochastic behavior generating the data. In this paper data from the Puget Sound Transportation Panel (PSTP) are analyzed to identify change in the patterns of time allocation by the panel participants (i.e., patterns of activity participation and travel). The data analyzed are sequences of states in categorical data from reported individuals' daily activity participation and travel indicators. This is done separately for activity participation and trip making using probabilistic models that generalize the restrictive Markov chain models by incorporating unobserved variables of change. The PSTP data analysis here suggests the more likely presence of multiple paths of change for time allocation to activities, non-stationary switching of activity participation from one year to the next, and day-to-day stationarity in activity participation pattern switching. Travel pattern change is best explained by a single path of change with stationary day-to-day pattern transition probabilities that are different from their year-to-year counterparts.

Evaluating Effects of Transportation Demand Management Strategies on Trip Generation by Using Poisson and Negative Binomial Regression

Most trip generation models are insensitive to the effects of transportation demand management (TDM) strategies. To evaluate the potential effectiveness of TDM solutions, transportation professionals must rely largely on the results of case studies, which cannot be generalized for all urban areas. To evaluate TDM strategies in a context that is sensitive to the unique characteristics of each urban area, TDM strategies should be incorporated into regional travel demand models. Five TDM strategies affecting trip generation rates are examined: telecommunications, alternative work schedules, on-site amenities, pricing strategies, and land use strategies. To analyze these strategies, data from the Puget Sound Transportation Panel (PSTP) were used. Variables derived from the PSTP data that may help explain the impacts of these strategies were evaluated for significance in trip generation models for several home-based and non-home-based trip purposes. The trip generation models were specified using Poisson and negative binomial regression techniques. After the models were estimated, the significance of the variables representing the impacts of TDM strategies was analyzed. Many of the TDM variables were indeed significant in the trip generation models; however, in some cases, the significance of the variables can be attributed to factors such as trip chaining, which does not describe the effects of TDM strategies. Additional research is needed to fully determine the effects of trip chaining on the variables examined in this study, and additional data could enable the development of variables that more accurately describe the effects of TDM strategies.

Exploratory Longitudinal Analysis of Solo and Joint Trip Making Using the Puget Sound Transportation Panel

The Puget Sound Transportation Panel is used to study the propensity of people to make solo and joint trips. The analysis consists of trip-based models in which solo versus joint trip making is explained in terms of person and household sociodemographics, daily activity and travel patterns, dwelling unit and workplace level of service and land use characteristics, and trip attributes. The analysis is repeated for all the days and all the waves (years) of the panel at hand. The analysis reveals that the major factor that determines joint trip making is the life-cycle stage of the household. It is also observed consistently across all the waves that, as the age of the person increases, the number of joint trips he or she also increases. Other factors that consistently affect joint trip making are household size, age of the household members, number of vehicles in the household, daily activity and travel patterns, and certain accessibility measures such as the access from transit to auto.