Identification of Factors Contributing to Geographic Scale and Mode Choice of Long-Distance Travel by Benjamin Bacon A thesis submitted to the Graduate Faculty of Auburn University in partial fulfillment of the requirements for the Degree of Master of Science Auburn, Alabama August 02, 2014 Keywords: Long-Distance, Travel, Geographic Scale, Multinomial Logit Copyright 2014 by Benjamin Knight Bacon Approved by Jeffrey LaMondia, Chair, Assistant Professor of Civil Engineering Rod Turochy, Associate Professor of Civil Engineering Dheeraj Raju, Assistant Professor of Business ii ABSTRACT As long-distance travel increases it is emerging as an increasingly important consideration in transportation planning. Not only is this in response to an increase in magnitude but also in recognition that such trips possess a greater per person effect on the systems they utilize. However, as planning organizations begin to respond accordingly for such travel they have been confronted with the prospect of accounting for such travel with methodologies and data developed for daily travel. Unfortunately the behavior of long-distance travel differs drastically from that of daily travel and available data and procedures are often unable to provide the necessary means to develop meaningful models. Therefore, this paper develops a technique for long-distance travel analysis by developing a geographic scale and mode choice multinomial logit model. Use of this model, through the context of geographic scale, will provide a better understanding of the factors that influence long-distance trip making. Specifically, this study uses detailed long-distance trip diaries from the 2008 New Zealand Domestic Travel Survey to look at how household, travel party, trip characteristics, and scale affect where and how individuals make long-distance trips. Utilizing this approach, the transportation planning community will be able to better forecast travel trends and the demands placed on systems due to long-distance travel. iii ACKNOWLEDGEMENTS It is with the utmost appreciation that I would like to thank the following individuals, it is only through their support that this endeavor reached fruition. First and foremost, my loving wife April, her unwavering commitment provided me with the foundation upon which to achieve my dreams. It is through her that I found the strength and inspiration required. I would also like to thank the members of my committee, Dr. Rod Turochy and Dr. Dheeraj Raju for their dedication to the development of this project. Lastly I would like to thank my professor, Dr. Jeffrey LaMondia. His tireless guidance, patience, energy, and determination, provided an atmosphere built for success. It is to him that I owe the deepest of gratitude, fully acknowledging that without his trust I never would have had this opportunity. iv TABLE OF CONTENTS ABSTRACT ii ACKNOWLEDGEMENTS iii LIST OF TABLES vi LIST OF FIGURES vii LIST OF ABBREVIATIONS viii 1. INTRODUCTION 1 2. LITERATURE REVIEW 8 2.1. Characterizing long-distance Travel 8 2.2. Definition of Scale 10 2.3. Destination and Mode Choice Factors 17 3. DATA SOURCES 22 3.1. Defining Geographic Scale and Mode Choices 22 3.2. The 2008 New Zealand Domestic Travel Survey 25 3.3. Supplemental Regional Characteristics 28 3.3.1. Statistics New Zealand: 2006 Census 28 3.3.2. Statistics New Zealand: Survey of Family, Income and Employment 29 3.3.3. Ministry of Business, Innovation and Employment: CAM 30 v 3.3.4. Ministry of Business, Innovation and Employment: RTI 30 3.4. Compiling the Geographic Scale and Mode Choice Dataset 34 3.5. Characteristics of the Final Dataset 36 3.6. Transferability of New Zealand Data to the United States 40 4. METHODOLOGY 45 4.1. Multinomial Logistic Regression 45 5. MODEL ESTIMATION 50 5.1. Trip Characteristics 52 5.2. Party Characteristics 55 5.3. Household Characteristics 55 5.4. Characteristics for Zone of Origin 56 5.5. Characteristics for Adjacent Zones 59 5.6. Characteristics for Non-Adjacent Zones 63 6. SUMMARY AND CONCLUSIONS 66 7. REFERENCES 70 8. APPENDIX A: RTO DEVELOPMENT 78 vi LIST OF TABLES Table 1 Long-Distance Definition 2 Table 2 Average Person Trip Length 3 Table 3 Definition of Destination 16 Table 4 Geographies 23 Table 5 RTO Characteristics 32 Table 6 Mode by Scale Cross Tabulation 35 Table 7 Model Estimation Parameters 50 Table 8 Dataset Compilation 76 Table 9 Area Comparison 81 vii LIST OF FIGURES Figure 1 Regional Tourism Organizations 24 Figure 2 Mode by Distance 25 Figure 3 Ministry of Tourism Data 26 Figure 4 Questionnaire Major Components 27 Figure 5 Scale by Distance 28 Figure 6 National Education Levels 37 Figure 7 National Employment by Industry 38 Figure 8 Family Structure 39 Figure 9 National Income Distribution 36 Figure 10 NZ Population Distribution 40 Figure 11 NZ Infrastructure 41 Figure 12 NZ Domestic Air Routes 42 Figure 13 Model Structure 46 Figure 14 Sigmoid 49 Figure 15 RTO/ RC Comparison 80 viii LIST OF ABBREVIATIONS ATS American Travel Survey CAM Commercial Accommodation Monitor CATI Computerized Assistance CFD Cumulative Frequency Diagram CO2 Carbon Dioxide DCM Discrete Choice Model DOT Department of Transportation DTS Domestic Travel Survey EPA Environmental Protection Agency FTA Federal Transit Administration FHWA Federal Highway Administration GDP Gross Domestic Product HOV High Occupancy Vehicle IID Independent and Identically Distributed Km Kilometer ix LD Long-Distance MNL Multinomial Logistic MPO Metropolitan Planning Organization NHTS National Household Travel Survey NPTS Nationwide Personal Transportation Survey NTS National Travel Survey NZ New Zealand POV Personally Owned Vehicle RTO Regional Tourism Organization RTI Regional Tourism Indicators TA Territorial Authority TAZ Traffic Analysis Zone TMIP Travel Model Improvement Program UK United Kingdom US United States 1 1. INTRODUCTION Increasing economic prosperity, expanding infrastructure development, and integrated policy planning, have allowed populations of developed nations to enjoy growing mobility over the past few decades. The effect of such behavior has been tremendous. Set in a quid pro quo relationship with mobility, economic health is one area that has benefited directly, and many communities born due to proximity to a transportation network are testament to this fact. Yet such mobility comes at a cost; due to the vast scale of such travel, the supporting infrastructure requires continuous repair and construction, and environmental concerns have resulted in extensive financial investment to implement mitigation strategies. While transportation engineers have been able to account and prepare for most of the US? growing travel needs, much work has been focused on daily travel alone. Long-distance travel has also experienced tremendous growth in duration, extent, and frequency, but our understanding of this travel is still lacking to adequately inform our planning decisions and forecasts (Fr?ndberg and Vilhemson 2003). The impact of long-distance travel is becoming ever more pronounced. In fact, tourism, an industry heavily represented by long-distance travel, now represents the largest industry in the world (Lise and Tol 2002) and accounts for over 11% of the GDP in the European Union alone (LaMondia et al 2010). As a result, non-daily travel behavior is no longer able to be discounted as insignificant. Failure to account for it, especially in consideration that such growth is expected to continue, will result in inadequate, and gross mismanagement of, resources. 2 Response to shifting behaviors, from that of almost exclusive daily travel to substantial representation by long-distance travel, has been slow. Within transportation planning, daily travel has received extensive attention, as has the long-distance transportation of freight. Yet the endeavor to fully understand long-distance personal travel remains relatively young, only recently beginning to receive the attention it truly deserves. Without effective knowledge of such behavior, and an understanding of how to manage and implement this knowledge, the largest transportation network may also be the least effective. Investment of resources into the understanding of long-distance travel has become warranted for a number of reasons, all mentioned above. Yet, until such effects become fully recognized a true assessment for the impacts of such travel will remain underestimated. Hindering development of a thorough understanding of long-distance travel behavior is the lack of a unified approach to defining long-distance. Often, definitions are simply based upon past practice instead of an investigation of the specific context in question. Both the subjectivity used in deciding how to define and reliance upon past definitions are cause for concern, allowing for the introduction of less than appropriate thresholds. Current definitions differ in numerous ways, including distance variations approaching hundreds of miles, inclusion of mode components, and incorporation of temporal factors. Table 1 briefly outlines the current breadth of definitions used to characterize long-distance travel. Table 1- Long-Distance Definition (Zurich and Frei 2008) DATA SOURCE LONG-DISTANCE DEFINITION Census Switzerland 3 or more hours Italian National Travel Survey 20 km (12.4 miles) or more U.K./ U.S. National Travel Survey 50 miles or more French Nation Travel Survey 80 km or more (as the crow flies) Dateline (Europe) 100 km or more (as the crow flies) European Tourism Demand Survey One overnight stay 3 The majority of thresholds used a distance in the 50 to 100 mile range. The U.K. National Travel Survey definition of long-distance travel as 75 km (46 miles) or greater (Dargay 2010) was similar to American distance-based definitions, consistently hovering between 50 and 100 miles. Earlier American surveys, such as the NPTS and ATS, used a 100 mile threshold while more recent surveys used a shorter distance of 50 miles. One such example of this is the NHTS (DOT 2011). Yet as exhibited by Table 2, these distance-based definitions did not directly match to any long-distance personal travel patterns. Table 2: Average Person Trip Length (Bricka 2001) Geographic Level ATS (miles) NPTS (miles) NATIONAL 515 510 MASSACHUSETTS 610 525 NEW YORK 705 1,099 OKLAHOMA 392 220 Table 2 illustrates the variation present across regions in many of the surveys. These numbers represents data obtained in the both the NPTS and ATS, conducted in 1990 and 1995 respectively, in which long-distance travel was defined as 100 miles or above. It demonstrates that although trips considered were all above 100 miles, the average person trip length for long- distance travel in America was above 500 miles. Additionally, depending on the survey, vast differences were observed between trips originating at the national and state levels. For example, of all long-distance trips considered nationally the average length was 510 miles. However, for the same pool, trips originating from the state of New York more than doubled that average. This demonstrates not just the significant differences long-distance trips may contain in regards to distance thresholds, but especially as they relate to a threshold from an aggregated geographic level. 4 Despite such variations in long-distance thresholds, all are geared towards capturing a single component, a measure of scale. Whether it is through distance, temporal, or mode, all approaches attempt to measure some degree of proximity that differentiates between local travel behavior and long-distance travel behavior. Yet data collection efforts utilizing the above mentioned approaches, in addition to creating transferability barriers, fail to adequately account for the vast variations in impedance within, and between, regions. Alternatively, by using geographic regions as a measure of scale, researchers are able to avoid transferability based upon definition of scale and simultaneously capture impedance within a regional context. Furthermore, the contexts of geographic scale and mobility shift dramatically within the long-distance literature depending on the specific location and trip purpose being considered (Schwanen et al 2001). As such, a single definition of scale is not appropriate for all endeavors. For example, using a threshold applicable to a MPO for a major metropolitan area, interested in impact to local infrastructure regardless of purpose, may not necessarily be adequate for use by the tourism industry interested in the factors that will draw business based upon leisure. In other words, each trip is not created equal and impedance is different according to the context of the trip. Areas in which travel is conducted have different forms and degrees of travel barriers. Some are physical such as a lack of infrastructure or lodging. Some are specific to the trip-taker such not having access to a specific mode (Limtanakool 2006). For the consideration of travel behavior to be complete, geographic scale according to regional boundaries must be combined with mode and destination components. This is required, but often overlooked, in order to fully understand infrastructure and policy behavior response. According to the Bureau of Transportation Statistics, nine out of ten long-distance trips are conducted using a personal vehicle and almost all others are accomplished using air travel 5 (USDOT 2014). However, the behavior represented within each of these modes is drastically different. Without a proper understanding of these modes as they relate to the traveler and their region, any insight gained may be misleading. Similarly, results obtained without proper consideration of destination would also fail to properly convey travel behavior. Incorporation of destination considerations is critical as they allow a complete picture of what is drawing a visitor to an area. Without this, researchers may be able to ascertain a purpose, but such a determination is useless without understanding why that purpose motivated travel to that region. For example, planners attempting to account for tourism travel to a region are ill-equipped to act if the only knowledge they possess is that the trip was drawn to the region for the purpose of holiday travel. By expanding considerations to include measures of demographic characteristics, economic health, recreation resources, and accessibility, proper steps may be taken to plan policy according to the specific motivators influencing such behavior. Also important for proper consideration of scale, yet often overlooked, is the definition of destination. Different objectives dictate different requirements for geographic scale. For example, an understanding of behavior according to high speed rail policy and infrastructure may require a different definition of destination than an investigation into tourism behavior. Many approaches currently aim to accomplish this; however, from a practical planning perspective, it is more important to understand long distance trips according to scale, (i.e. within an area, in an adjacent area, or farther out). Whether it is in the interest of developing a larger long distance modeling framework or in support of specific urban area, understanding geographic scale is critical. Additionally, a proper understanding of how this scale interacts with trips and travelers according to different modes is also required for transportation planners to effectively develop policy and infrastructure. 6 Therefore, this thesis develops a unique region- and ringed-based discrete choice model to predict households? preferred joint geographic scale and mode choice for different types of long-distance travel. It examines the ability of geographic scale combined with mode considerations, rather than a linear distance-based definition, to more accurately incorporate long-distance travel into regional transportation planning and forecasting analyses. This work considers the impact of household and travel party characteristics, along with readily available census data for origin-destination pairs, to predict magnitude of long-distance trips at a household level. Models in which destination/mode choice pairs have been analyzed as a product of trip, household, travel-party, and geographical based characteristics, or a combination thereof are not new (Yang et al 2010). However, when these models have been developed, they have lacked proper consideration of destination attraction considerations (Bhat et al 1998). In consideration of this, the following work develops a joint long-distance geographic scale and mode choice multinomial logit (MNL) model to understand the factors that not only influence long-distance trip making, but also the travel that falls within different geographic limits. The following thesis begins with a discussion of past work through a literature review. In this section, past work concerning transportation planning is explored with a specific focus on long-distance travel, geographic scale, mode choice, and destination definitions. Following the literature review the methodology utilized for this project is introduced. In this section, the reader is provided information on the model type and structure utilized and the reasons supporting such a decision. The methodology section transitions into model estimation in which model results are presented. In this section, independent variables are discussed as they relate to the dependent variable, that is, the scale/ mode choice represented, and the meaning of such results. In closing, 7 the research is summarized and resulting conclusions presented. It is in this section that the specific findings are presented and discussed in detail. 8 2. LITERATURE REVIEW 2.1. Characterizing long-distance Travel Long-distance travel is still underrepresented as a comprehensive consideration in transportation modeling within the United States, especially long-distance modeling that incorporates geographic scale and mode based considerations. However, a developing recognition of the importance of such behavior has been spurring the emergence of related research. Due to the infancy of such endeavors, when it has been attempted, the effort to model such travel has been met with substantial obstacles. Primarily amongst these are the limited data sources available to account for long-distance travel. Additionally, when they are available they often characterize long-distance travel according to a distance based threshold. Occasionally purpose (such as excluding commute-trips) (United States 2013), temporal (Andreas 2008), or mode-based definitions (such as automatic long-distance categorization for air travel) (Dargay 2010), are utilized in conjunction with distance for defining long-distance travel. Yet, almost non-existent are thresholds according to geographic scale. The dominance of such distance based data is understandable, it is relatively easy to acquire, explain, and in many instances provides adequate information for simple analysis. However, Yang et al (2010) points out that developing long-distance models structured according to distance alone does not address topics such as inter and intra-regional trips, both significant and unique in their characteristics, especially as they relate to rapidly changing regions. Some agencies have responded to such concerns and defined 9 trips according to the crossing of boundaries (Mokhtarian 2001) but these have been relatively limited. Not surprisingly, many of the mathematical models used to study personal daily travel can be applied to understanding long-distance travel behavior. However, models of geographic scale are not as common in the transportation literature, more often found in migration and animal-behavior studies. Utilizing methodologies developed for individuals of a species, such models have also proven accurate in predicting the scale of travel for human travel. In such models the probability of length frequency for geographic scale to a certain point is able to be predicted. However, these models generally are better suited for shorter distances and obviously lack many of the considerations applicable to human behavior such as increased ability to travel further and more often due to freedoms allowed through technology and infrastructure (Brockmann et al 2006). Additionally limiting was that implementation of this model was accomplished using tracking of individual currency bills. A unique, and subsequently proven, approach but one that lacks the ability to directly account for trip, household, and person based characteristics. Such an approach also fails to consider the transaction of bills between individuals between location identification periods. The work was able to provide valuable information on general flows of bills, but failed to directly connect the dispersion to individual trip-takers and trips. However, the study did account for limited regional characteristics in which the bill first appeared in the tracking system allowing for consideration of some region specific traits. From this research two thresholds were observed, 10km (6.2 mi) and 800 km (497.1 mi). These thresholds corresponded to a positive linear relationship between the probability and distances prior to 10 km and a negative linear relationship between probability and distance above 10 km. 10 2.2. Definition of Scale Much of the research examined in which travel was defined according to geographic scale of trips was often limited to constrained areas, resulting in more localized considerations and the exclusions of many travelers and trips into or out of a region. For example, in an effort to model traffic demands within a city, distance-based definitions often are structured relative to the extent of a city boundary. Consequently, in regards to use of boundaries, there is debate on how a geographic scale should be defined. Currently, arguments exist for defining long-distance travel according to both political boundaries, such as state, and behavioral boundaries, such as tourism regions. This is especially true in consideration of travel impedance. The concept, as it relates to long-distance travel, does not apply equally to all individuals or trips. Characteristics such as gender, age, and income, all change the manner in which a long-distance trip is defined. Men tend to travel farther than women; age negative corresponds to travel time; and a higher income corresponds to longer trips (Yang 2010). However, it is also observed that despite extensive insight into many of the relationships between individual socio-demographics and complete destination choice behavior, many remain relatively unstudied. Of these, a lack of knowledge exists on the effect of trip-taker occupation on destination choice, rather than simply work destination choice. The majority of research that was found to include geographic scale as a dependent variable was done so at the most basic of levels: domestic and international trips. As observed in Table 3, categorization of geographic scale in identified research was able to be accomplished according to three labels: ?By Border?, ?By Type?, and ?By Structure?. Well over half of these fell within the ?By Border? classification with all but two research endeavors defining scale 11 according to a national boundary. Lang et al (1997) examined destination choices of travelers according to country and an ?Asia-Pacific?. The research did prove productive in realizing the effect of a number of trip and traveler factors yet did so at an aggregated level without consideration of destination characteristics. Bieger et al (2002) sought to differentiate domestic pleasure travelers from international according to tourism sector utilized. Lise et al (2002) sought to utilize linear regression to predict the number of foreign travelers to a country based upon the size of the country and climatic data. Nicolau et al (2005) developed a nested model in which geographic scale was the second nest below the most basic of decisions, to take a vacation. In this approach, if the trip taker did decide to take a vacation then the model would determine if the vacation was foreign or domestic in nature. Within the scale component node frequency was predicted. In his work The Role of Routines In The Vacation Decision-Making Process of Dutch Vacationers, Bargeman et al (2005) examines the effect normalization of vacation travel has on the domestic/ international travel engaged in. For example, if travelers have developed a pattern of routine travel are they more inclined to engage in domestic travel or international travel? Beerli et al (2007) utilized a more empirical approach attempting to predict domestic versus international travel according to differences between a destination?s image and the travelers own perception of the destination. Deviations from the dominance of national boundaries to define geographical scale according to boundaries are McFadden (1978) and Boarnett (2005). McFadden?s research (1978) attempts to predict housing location according to self-defined residential zones, however, the zones were determined to be too specific for effective calculations with available resources. Boarnet et al (1995) attempted to test the relationship between local (county level) infrastructure development and economic health according to county and state based thresholds. Four projects were identified in which scale was predicted 12 according to type. Handy et al (2002) sought to predict travel behavior according to three unique scales based upon mode: human scale, automobile scale, and airplane scale. The remaining research within this classification, Yang et al (2010), Bhat et al (2008), and Jie et al (2014) all deviated from the multi-scale approach by utilizing a single scale the varied according to characteristics as the dependent variable. For example Yang et al attempted to predict destination according to traffic analysis zones but did so according to the type of zoning code such as residential or commercial. This allowed for extensive level of detail but limited consideration of scale. Bhat et al and Jie et al also utilized small scale zones according to type but, again, in doing so eliminated proper consideration of scale. Lastly, Lew (2002) attempted to predict geographic scale according to the type of destination, according to whether it was singular or a gateway in nature. Using these thresholds allowed for compilation of minute regions according to singular, or extensive chained regions if multiple gateways were determined. Such an approach, while unique, due to the vast array of possible outcomes did not provide adequate structure to fully incorporate destination based characteristics. Inter-regional and intra-regional travels tend to demonstrate different behaviors as a result of the constraints and freedoms particular to each (LaMondia and Bhat, 2008, Jansen- Verbeke 1995). Inter-regional travel, for example, is primarily motivated by activities unique enough to warrant the additional cost and time required to travel further. In other words, activities such as holiday travel allows for greater inter-regional travel for the purpose of visiting attractions (Pearce et al 1993). Alternatively, intra-regional travel tends to include activities that fit within the time and spatial constraints of a typical day. Despite the unique travel behaviors associated with each scale, efforts to identify models in which interregional or intraregional personal trips are estimated yielded minimal or outdated results (Dredge 1999). McFadden was 13 on the forefront of this effort, when in 1978 he applied economic theory of choice behavior into analysis of residential zones and while more current work does exist, it often does so in the shadow of other endeavors. Many current efforts to model this behavior rely on procedures developed for daily travel and, as discussed above, often using distance or some combination with mode or time thereof as the definition of a long-distance trip. At first glance this appears to be an obvious decision, however long-distance trips inherently have unique demands on the trip-maker and a distance threshold definition may not translate to the same amount of resource required by each user, be they financial, temporal, or accessibility. This makes it difficult to properly consider factors unique to long-distance travel (Van Nostrand et al 2013, Rohr et al 2013, Yang et al 2013) using mainstream practices. By ignoring geographical scale based considerations, defining travel by zonal considerations, models severely limit their consideration of location-based characteristics and region-based applications. In support of this, the 1995 TMIP, under the guidance of the FHWA, FTA, EPA, recommended that long-distance travel be studied relative to discrete geographic scales rather than a continuous measure of distance (Waddell 2002). Since then, much work in the field of freight forecasting has focused on geographic-scale based forecasts of demand (Hesse et al 2004, Boarnet 1995), but the research pertaining to long-distance geographic scales in personal travel is rather limited (Dellaert et al 1998; Handy et al 2002). Zhang et al (2013) furthered the argument, demonstrating utilizing a geographic scale based definition, structured according to individual behaviors, would enable a reduction in false reads within discrete choice models. This would be enabled through an added degree of selectiveness through individual patterns interacting with boundaries. Furthermore, as the use of geographic scale increases, so too does the potential for improved planning beyond the context of self- 14 selected boundaries. Transportation organizations responsible for planning may gain a better understanding of behavior within their jurisdiction as considerations are expanded to include not only geographical areas within the area of responsibility but other geographical areas that may influence the area under consideration. The Michigan Department of Transportation 2009 Household Travel Program provides an example of alternative distance-based long-distance definition with a geographic scale component. This survey defines long-distance trips as approximately 100 miles or greater. Additionally, within the long-distance category trips were classified according to a basic geographic scale: ?Within Michigan?, ?Neighboring States?, and ?Other States or International? (Michigan 2009). Utilizing these classifications during consecutive surveys allowed researchers to identify an increase in ?Within Michigan? trips at the expense of ?Other States or International? trips and created an opportunity to develop a model that accounted for an increase in inter region travel. This trend was a continuation on the part of the State to include geographic scale in model development stemming from previous success in such endeavors. As early as the 1970 Michigan?s Statewide Travel Demand Model incorporated 2,307 in state zones and 85 outstate zones (Nellett 1996) to improved long-distance components of forecasting models. The use of boundaries as scale components in models, incorporated with measures of the characteristics contained within such boundaries would capture physical constraints and other influences affecting mode choices specific to a location. For example, Moeckel et al (2013) distinguished between areas by incorporating zone identities in model considerations. In considering zonal effects, models were able to successfully account for unspecified attributes in mode selection, such as a regional presence of HOV lanes. This approach is feasible in the United States and other developed countries. Yang demonstrated in consideration of TAZs that 15 those with TAZ purpose with highest employment per are unit also had the highest draw from surrounding areas (2010). The U.S. Census (U.S. Department of Commerce 2010) for example provides extensive characteristics of specific areas according to Region, Division, or Tract. However, limited surveys directly account for such areas. Of the geographic boundaries that are accounted for in data collection and research efforts, as observed in Table 3, much diversity is present. All definitions do adhere to Yang?s observation that scale based definitions are developed using five factors: size, location, structure, physical barriers (such as coastline), and distance in relation to origin (2010). Interestingly, size assignments in select cases were determined based on trip purpose. For example, for trips conducted for the purpose of employment, zonal employment was used to denote size; shopping trips determine zone sized based upon retail employment. Additionally, zonal structure was often defined by concentric, regions. Impedance due to factors specific to the trip taker must also be accounted for when considering scale. Yang reinforced the belief that gender, age, and income correlated to impedance, age negatively and income positively. 16 Table 3: Definition of Destination ENTITY DEFINITION OF DESTINATION CHOICE o BY BORDER Lang et al (1997) Biegeer et. Al (2002) Lise et al. (2002) Nicolau et al (2005) Bargeman et al. (2006) Beerli et al. (2007) o Country McFadden (1978) ? Residential Zone Boarnet (1995) o State ? County BY TYPE Handy et al. (2002) o Human Scale o Automobile Scale o Airplane Scale Yang et al. (2010) o TAZ by purpose o Education TAZ o Administration TAZ o Commercial TAZ o Recreation TAZ o Sport TAZ o Medical TAZ o Industry TAZ o Park TAZ Bhat et al. (2008) Zones built upon spatial clusters Jie et al. (2014) o Neighborhood Type o Urban Elite o Non-Hispanic Urban Black o Low-Income City Minority o Suburban Young o Suburban Retired o Middle-Income Suburban Working Class o Suburban Wealthy o Non-Black Hispanic o Rural o Natural Scenic BY STRUCTURE Lew (2002) o Single Destination o Gateway Destination o Egress Destination o Touring Destination o Hub Destination 17 Geographic scale of trip, for the purpose of this study, is defined as trip extent, ranging from trip entirety remaining in local area to extending beyond, or remaining within, multi-level regions defined upon New Zealand tourism and political boundaries. This definition is more in line with those found in studies conducted by Pearce (1995) during analyses of tourism boundaries to investigate travel behavior. Despite these diverse approaches, a deficit continues to exist in incorporating scale-based measures into planning considerations (Hesse et al 2004). Furthermore, the use of boundaries as scale components in models, in a tiered structure such as division, state, and county, would capture not only the effect of scale on travel but also physical constraints and other influences affecting mode choice. For example, Moeckel et al (2013) distinguished between areas by incorporating zone identities in model considerations. In considering zonal effects, models were able to successfully account for unspecified attributes in mode selection, such as a regional presence of HOV lanes. 2.3. Destination and Mode Choice Factors By incorporating mode choice into destination choice considerations, accessibility and impedance are able to be considered. This is important, as studies have demonstrated (e.g. Yang 2010), this allows for a more behaviorally-accurate representation of the true decision of destination choice and ease of mobility. Additional support for the inclusion of mode choice consideration in conjunction with scale improving resulting models is Ben-Akiva?s et al 1985 book titled Discrete Choice Analysis: theory and application to travel demand. In this work with Lerman, Ben-Akiv develops a joint mode destination model in accordance with sample of alternatives theory. 18 Past research has demonstrated that decision makers utilizing flexible modes of travel (i.e. motorcycle and car) are more sensitive to impedance than those using inflexible modes such as train and air (Yan 2010). Such observations equate to flexible modes correlating to shorter travel distances over the entire spectra of purpose, and trip-taker characteristics. In such research it has been found that mode choices are highly correlated with travel distances (Hagen-Zanker 2013). Additionally, de Lapparent et al (2013) argue that for linked long-distance trips, relative to short-distance trips, traveler?s mode choice decisions are more sensitive to travel times to main-modes of travel than on the main modes themselves. This specific example indicates a greater willingness to pay more for specific mode choices during long-distance travel as opposed to short-distance travel. Explicit attempts to model mode choice commonly employ logit discrete choice models (Hagen-Zanker 2013, Whalen 2013), multinomial probit discrete choice models (Can 2012), or linear regression models (Friedrichmeier 2013). In all observed mode choice models distance was a variable considered during model development. In addition to distance, Lehto et al (2004) demonstrate mode choices are significantly influenced by prior experience. Additionally, travel decisions are largely ?reasoned decisions?, affected by planning and policy decisions (Bamber et al 2003). This attribute of travel behavior indicates that, when considered in planning, future geographical scale and mode choices of personal travel may be better guided by planners aware of individual motivators much as they are by past habits. As such, it is important to consider these decisions in tandem. Among trip characteristics identified in existing research, trip purpose was presented most often. Yang et al (2010) examined it according to shopping, recreation, and business. Notably absent from this list but included in almost all others was travel for the purpose of visiting, friends and relatives. Although not specified it is possible this may have been included 19 under the classification of recreation. However, other research examined noted the differences between visiting friends and relatives and other recreation travel. Also commonly considered as was destination type, Whitten et al (2011) explored this concept thoroughly in her examination of destination and travel modes. However, unlike many other endeavors, in her research Dr. Whitten also incorporated the day of the week travel occurred in determination of scale and mode. While such a consideration was uncommon, it did prove significant in her research by capturing cyclical travel behaviors occurring on a weekly basis. In other words, in the context of her work, examined behavior adhered to trends established by weekly schedules. Party characteristic exhibited much less variation between examined research than that of trip characteristics. These characteristics tending to capture four basic measures: travel party size, age distribution, gender distribution, and relationship. While all contribute to the scale of travel to some extent, party size corresponds to scale in the most direct manner, decreasing the likelihood for increased scale as party size increases (Yang 2013). Age, while not shown to directly correspond to scale, was shown to possess affinity to specific regions (Lin2008) (Bhat 1998). Sex, remarkably, did demonstrate significance on travel scale in multiple instances. In studies in which gender was considered it was shown males tended to travel more often and farther than their female counterparts (Bhat 1998) (Whitten 2011) Relationships of travel parties were also shown to effect scale. LaMondia et al (2010) demonstrated that trips taken alone or accompanied by young children were more likely to remain in close proximity to the trip origin (LaMondia 2010) Household characteristics included income (Yang 2013) (LaMondia 2010), size, vehicle count, and tenure (Lin 2008). In his research on traveler behavior in the context of vacation travel Dr. LaMondia (2010) demonstrated income directly correlates to mode choice. Lower 20 incomes, those in the bottom quartile, were more likely to conduct travel using surface transportation. In contrast those in the top quartile of household incomes were more likely to utilize air travel. Past research has also demonstrated that the more vehicles a household owned the greater the frequency and extent of travel that will be undertaken. Of characteristics utilized to capture regional measures, Yang et al (2013) demonstrated significance in employment according to occupation, population density, and land use. All three were shown to significantly impacted destination choice of travel. Additionally, in research conducted by Lin (2008), population density, housing density, road density, transit, intersection density also proved effective in prediction of travel to geographic areas. In regards to employment characteristics, Bhat et al (1998) demonstrated regional employment positively correlate to travel to a region. The following research develops a unique region- and ringed-based discrete choice model to predict households? preferred joint geographic scale and mode choice for different types of long-distance travel. The geographic scales considered in this study are smaller than all past research: travel within an origin, to an adjacent region, and to a non-adjacent region. This allows for two outcomes: first, regions are structured according to geographic ?rings? allowing for direct incorporation of scale, second, destination characteristics may be developed. Second, these choices are predicted according to characteristics based upon household, travel party, trip, and regional characteristics. The specified methodology improves upon past research in four areas. First, structuring geographic zones according to concentric regions is unique and allows for direct consideration of both scale and zonal characteristics. Second, the scale of travel behavior presented in this research is smaller than that provided in other endeavors. This allows for greater level of detail and a more complete understanding of travel behavior. Third, an extensive use of 21 policy variables, enable through zonal structure, allows for a better understanding of the factors that motivate visits. Lastly, the research effectively and thoroughly considers trip, traveler, household, origin, and destination characteristics simultaneously in the determination of scale and mode choices. 22 3. DATA SOURCES This thesis utilizes a unique location-based long-distance household travel dataset from New Zealand (along with supporting policy-oriented regional characteristics) to analyze long- distance trip geographic scale and mode choices. While some long-distance travel datasets exist within the United States and Europe, they are limited for such an analysis for a number of reasons, including the lack of detailed geographic origin and destination identification, either small or too large geographic extent of trips cataloged, and relative age of the datasets. This section describes the data collection process and summarizes the travel data used in the geographic and mode choice modeling analysis. 3.1. Defining Geographic Scale and Mode Choices Geographic scale, as mentioned in the literature review, can be defined in various ways. Table 4 presents the geographic scales available for analysis within New Zealand?s travel surveys. For information of the characteristics contained within specific datasets please refer to Appendix A. 23 Table 4- New Zealand Geographies COMPONENT GEOGRAPHIC IDENTIFIER Domestic Travel Survey Place (Point) Territorial Authority Regional Tourism Organization Regional Council Census Place Meshblock Area Unit Territorial Authority Regional Council Other Statistical Data Place Tourism Flows Area Territorial Authority Regional Tourism Organization Regional Council For this thesis, geographic scale is defined by travel within and between New Zealand?s tourism regions, characterized by regional tourism organizations (RTOs). This was selected for the geographic scale because they represent one of the smallest planning decision-making scales, both in New Zealand and in the US, that have variable land uses and policies. In New Zealand, transportation planning for the country is primarily done with respect to the RTOs, which are similar in size and governance to the US? metropolitan planning organization regions. The country, as observed in Figure 1, is separated into 29 RTOs, each roughly the size and operation of a large US MPO. RTOs comprise a self-funded organization dedicated to the market of their respective regions to both domestic and international visitors. Other than acting as a link between the tourism community and local governments, RTOs vary greatly in structure, funding, size, and focus. For example, Auckland RTO, the largest RTO, is established as a charitable trust; Gisborne RTO is a non-profit incorporated society; while Hawke?s Bay RTO is a integrated RTO funded by administered fees. However, regardless of the structure, some degree 24 of funding is often provided by local government or through membership fees administered to constituents within an RTO. Figure 1: Regional Tourism Organizations 25 Mode choices are limited to driving, rail and air travel based upon analysis presented in Figure 2. In this cumulative frequency diagram (CFD) personal vehicles and air travel are shown to account for the vast majority of travel in the Domestic Travel Survey (DTS). Figure 2: Mode by Distance 3.2. The 2008 New Zealand Domestic Travel Survey The New Zealand Domestic Travel Survey is an annual survey conducted since 1981 with the exception of the 1991 to 1998 time period. Prior to 1991 the survey only accounted for overnight trips, however in 1999 the Foundation for Research and Science Technology (FRST) funded additional development of the survey resulting in expansion to include day trips of 40 km or more away from the home. The ongoing survey receives funding from the Government of New Zealand. Currently managed by the NZ Ministry of Tourism (previously the Office of Tourism and Sport), with data collection portions outsourced, the survey captures both leisure 26 and work-oriented long-distance trips with domestic trip origins. The survey is conducted as part of the core dataset the Ministry of Tourism maintains in order to forecast travel demands. The dataset, as shown in Figure 2, also includes regional, international and commercial surveys. Figure 3: Ministry of Tourism Data 27 Figure 4: Questionnaire Major Components The DTS reports quarterly on the characteristics, types, behavior, and expenditure of domestic travel. In addition to the behavior of domestic travelers, including expenditure, activity participation, demographics, and resources utilized. Quarterly domestic tourist expenditure reports and domestic trip reports are generated from this data. Responses regarding those trips completed over the past month (i.e. 4 weeks) are collected using telephone surveys with computer aided telephone interviews (CATI). The DTS was selected for this study because it was a recent comprehensive and detailed long-distance survey that covered a large geography with many similarities with other countries, especially those in the developed world. Of paramount importance to this research was the determination of geographic scale, and specifically the fact that this dataset included very detailed identification of origins and destinations of the non-daily trips. Ultimately, upon crosstab examination of characteristics according to the scale, cumulative distribution figures (as 28 observed in Figures 4 and 5), consideration of the scope and intent of the research, and the use of scale by the Ministry of Tourism for describing different planning jurisdictions and modeling travel patterns, the Regional Tourism Organization (RTO) was selected for definition of scale. Figure 5: Scale by Distance 3.3. Supplemental Regional Characteristics 3.3.1. Statistics New Zealand: 2006 Census The New Zealand Census is administered once every five years in order to acquire accurate data for government, non-government, and community groups to effectively plan and develop policy. Statistics New Zealand, a government department, is responsible for the collection, processing, and distribution of official statistics. New Zealand Census practices provide a paper form to every dwelling. However, the 2006 Census was the first census in which 29 an online option was available to complete the census. As a result seven percent of the 3,160,371 responses were completed online. Although more recent census data was available, 2006 data was utilized as it was closer to the time frame captured in the 2008 New Zealand Domestic Travel Survey. This provided more accurate representation of characteristics during the behavior captured in the travel survey. The census provided gender, income, age, and education characteristics according to regions and territorial authorities. This data was readily available on Statistics New Zealand?s online database. Other than restructuring according to desired geographic scale (RTO then consolidated adjacent and non-adjacent zones for each trip destination) data processing was minimal. Data was able to be assigned to RTOs as these areas, for the most part, followed territorial authority (TA) boundaries. In cases in which this was not the case it was due to RTOs containing multiple TAs. In these cases the TA that comprised the majority of the RTO was selected as the controlling area. Following TA assignment to RTO classifications zones were developed. 3.3.2. Statistics New Zealand: Survey of Family, Income and Employment The Family, Income, and Employment Survey, also collected by Statistics New Zealand contained information on employment, household, and income characteristics. Although updated annually as a longitudinal survey, the 2004 update was utilized for this research as surveys conducted closer to the 2008 time period covered in the Domestic Travel Survey were not complete. That is, they were missing significant characteristics that were available in the 2004 data. Although representing a smaller sample size than the census (hundreds of thousands rather than millions) it did provide a greater level of detail on the change experienced over time within characteristics. As with the 2006 Census data all datasets were available to the public on the 30 Statistics New Zealand online database. All zone and zone characteristics development was accomplished in the same manner as that used for census data. 3.3.3. Ministry of Business, Innovation and Employment: CAM The Commercial Accommodation Monitor contained measures of accommodation usage according to Regional Tourism Organizations or Territorial Authorities. As this research developed scale based upon RTO definitions, TA data was not considered in analysis. The CAM dataset provided monthly measures of capacity, occupancy, and nights stayed according to lodging type (i.e. motel, hotel, or hostel). Datasets reflecting accommodation for 6,375 lodging establishments were surveyed across the entire country. All data was publicly available on the Statistics Ministry of Business, Innovation and Employment online database. Zone development did not require RTO assignment as these assignments were already in the data, however, zone characteristics did require development and was accomplished in the same manner as that used for census and Survey of Family, Income and Employment data. 3.3.4. Ministry of Business, Innovation and Employment: RTI The Regional Tourism Indicator database provided expenditure amounts according to region for both domestic and international travel. As the scope of this study only included domestic travel, only the domestic data was downloaded from the publicly accessible Ministry of Business, Innovation and Employment website. Although not ideal due to the data reflecting a time not included in the travel survey, only 2012 data was available, dictating its use. Additionally, this survey is updated monthly and reflects the past four years. Therefore the 2012 database actual reflected a time period immediately following the behaviors collected in the travel survey and reflects trends that were already in place during the survey. As with the 31 Commercial Accommodation Monitor, RTOs were used for regional definitions, eliminating any need to assign such labels. These indicators simply reflected economic growth according to region but provided tremendous insight into the health and performance of an area. No modification was required for this data other than consolidating it with the other regional characteristics for addition to a master file. In addition to trip, party, and household characteristics for each trip provided by the 2008 New Zealand DTS, Regional Tourism Organization characteristics were attained from multiple government agencies. This was conducted in order to enable identification and examination of RTO based characteristics influencing travel within Origin RTO, to neighboring RTO, and to a RTO beyond a neighboring RTO. Located through a collective database managed by the Department of Internal Affairs (New Zealand 2013), many of the characteristics located and utilized, highlighted in Table 5 were already aggregated to the RTO level. 32 Table 5: RTO Characteristics AGENCY SOURCE CHARACTERISTIC POPULATION AND SOCIETY Statistics New Zealand 2006 Census ? Age ? Education ? Household Compositiona ? Household Structureb ? Employment by Industry ? Employment by Typec Survey of Family, Income, and Employment ? Income (HH) TOURISM Ministry of Business, Innovation, and Employment Commercial Accommodation Monitor ? Nights stayed by lodging type Domestic Travel Survey ? Number of trips by purpose ? Number of trips by lengthd ? Financial transactions by typee Regional Tourism Indicators ? Expendituref a number of individuals in a household by role b relationships present in the household respectively c employment status (i.e. full-time, part-time, seasonal, or unemployed) d categorized as either day trip or overnight trip e categorized for retail, accommodation, food, tourism, and other transportation f traveler change in expenditure between 2013 and 2008 For resources utilized that did not directly provide RTO level data, characteristics were calculated from component territorial authorities. In these cases the calculations were determined using mean values. Following determination of RTO characteristics, additional processing was conducted in order to allow for such measures to correspond to identified destination choices. This was accomplished by determining the neighboring RTOs, and the RTOs beyond the neighboring RTOs for every Regional Tourism Organization in New Zealand. In addition to this 33 allowing for each trip to be classified according to scale, by comparing the origin RTO with destination RTO, it also allowed for a series of neighboring and beyond neighboring zones to be established relative to every origin RTO. Characteristics for each of these zones were determined using the median value for characteristics of contained RTOs. Categories utilized were selected based upon measures proven to be significant in past work or otherwise shown as lacking in previous endeavors on individual decision maker choices. However, other considerations were also included such as the manner in which the characteristic related to mobility. The regional data, maintained by the Government of New Zealand, spanned a number of years ranging from 1996 to 2013. Additionally, as may be observed in Table 5, the collection and maintenance of the data pertinent to this study spanned two broad government agencies. Statistics New Zealand, New Zealand?s official statistics office, consists of approximately 1000 employees effectively executing continuous operations for over 120 years. Its primary goal to ?collect, compile, analyze, and communicate? numerous measures spanning the economy, environment, people and communities, and government activity (Statistics New Zealand 2014). The 2006 Census is part of a five-year census program conducted by Statistics New Zealand. The Ministry of Business, Innovation, and Employment operates in order to assist economic development in New Zealand. It pursues this endeavor through assisting domestic business?s grow productivity and become more competitive against foreign counterparts. Originally four separate agencies; the Department of Building and Housing, the Ministry of Economic Development, the Department of Labour (sic), and the Ministry of Science and Innovation; the Ministry collects, maintains, and analyses data in support of developing economic competitiveness (Ministry of Business 2014). 34 3.4. Compiling the Geographic Scale and Mode Choice Dataset Final dataset development was accomplished by combining all datasets into a single master dataset. This original dataset included long-distance trips organized across three files describing households, trips, and visits (i.e. intermediate stops and side trips). As highlighted by Table 6, the final dataset was heavily representative of trips extending beyond RTOs boundaries and those utilizing personal vehicles as the primary mode. Despite the appearance of data skewed according to boundary, it is important to note these classifications are established by the New Zealand Government to manage tourism specific to those regions and instead reflects a perspective as it relates to influential factors within those boundaries. For example, the lopsided representation may be indicative more of attraction to new experiences rather than boundary definitions as they relate to scale. Alternative-specific variables were calculated separately and merged into the trip dataset. Each trip?s origin and destination RTOs were recorded in the survey, and the decision of travel ?within origin region?, ?to adjacent region?, or ?to a beyond region? was identified from a master list of every possible combination of adjacent RTOs. Additionally, 29 sets of alternative- specific variables were generated for each of the origin RTOs. Once the adjacent and non- adjacent RTOs were identified for each origin RTO, the median characteristics of land use, economic, and demographic characteristics were taken to describe each set. Ultimately, each trip was identified according to scale (within origin, adjacent, or non-adjacent) and had geographic characteristics assigned that reflected the characteristics of the RTO of origin and destination according to the median values of the RTOs comprising that destination. 35 Table 6: Mode by Scale Cross Tabulation Main Mode by Geographic Scale Cross Tabulation Geographic Scale Within RTO Neighboring RTO Beyond Neighboring RTO Mode Mode Choice Distribution Mode Choice Distribution Mode Choice Distribution Personal Vehicle 5248 9658 10609 Air or Train 58 166 3217 The final dataset was developed with each case representing a single household trip (of which a single household may have completed multiple trips). Starting with 58,172 cases, 17,199 were removed from consideration as they did not meet the Survey?s classification of 40 km for non-daily long-distance travel. An additional 12,017 cases were removed in order remove cases in which data was incomplete resulting in a final dataset contained valid responses for all contained variables. Of these 12,017 data was not ascertained for one of the following measures: ? Number of person trips conducted during period ? Housing Density ? Percent Renter Occupied ? Population Density ? Respondent Age ? Respondent Employment Status ? Worker Density ? Number of Adults in Household 36 The final dataset used in the analysis was comprised of 28,956 completed long-distance trips, with characteristics describing each household, home RTO, neighboring RTO and all other RTOs. 3.5. Characteristics of the Final Dataset Figure 6 displays the national educational distribution of New Zealand according to the 2006 Census. ?None?, representing those that did not finish high school or a trade school seems exceptionally high. However, it must be noted, that culturally the importance of a high school diploma does not contain the same stigma, or loss of employment opportunities, as other areas of the world. NZ has a robust apprenticeship program. In this program, in lieu of their final two years of high school, students may transfer to an apprenticeship program and begin training for a career in a trade industry. Also notable is 40% of the population reporting at least some education past a high school level. This is well under the 58% reported for the same statistics for Americans by the United States Census Bureau (2013). This too may be a result of the apprenticeship program, largely vacant from the American workforce. 37 Figure 6: National Education Levels (New Zealand 2006) Figure 7 represents the distribution of the New Zealand workforce according to industry. This figure demonstrates the heavy representation of administrative personnel in the workforce. Managers, clerical, and administrative employment represents approximately 30% of the entire workforce. Additionally, professionals represent 20%, which, in addition to technical employment, may represent some other administrative employment. 38 Figure 7: National Employment by Industry (New Zealand 2006) Figure 8 represents the structure of all New Zealand families. It indicates that, of all the families, the vast majority (82%) are couples. Similarly, the majority (60%) have children. This leaves a sizeable 18% of all families conforming to a one parent with child(ren) structure. This is meaningful in the context of this paper in terms of the additional constraints this adds to long- distance travel. Couples without children are, independent of all other considerations, by far the best able to engage in long-distance trips. 39 Figure 8: Family Structure (New Zealand 2006) Figure 9 displays the income distribution for all New Zealand households. The distribution indicated demonstrates that, while a considerable group of households earn minimal incomes, the majority of NZ households earn a significant amount. In fact, two peaks may be observed in the distribution, one at $10,001 to $15,000 and the other at $50,001 to $70,000. Although it would be expected that higher incomes would lead to greater long-distance trip activity, this has not always been the case. Evidence seems to indicate this is due to higher incomes corresponding also corresponding with higher ages that are less likely to engage in long- distance travel. 40 Figure 9: National Income Distribution (New Zealand 2006) 3.6. Transferability of New Zealand Data to the United States Although New Zealand, a relatively isolated island nation in the southwest of the Pacific Ocean, differs from many other nations in regards to proximity to other countries (the closest neighbor to New Zealand is Australia, 900 miles to the northwest), it is similar to many developed nations with respect to personal mobility. Infrastructure, wealth, education, and culture, are all comparable to many western nations. Another benefit of examining the factors contributing to scale in the context of New Zealand is that, although consisting of numerous islands, the main land mass of New Zealand is primarily comprised of only two large islands, the North Island and the South Island. This allows research to inherently incorporate impedance due to the nature of such aquatic boundaries while not limiting the scope of work due to them. Also beneficial to the scope of this research is the population density of New Zealand. The nation?s 41 population is approximately 4.5 million and, as observed in Figure 10, contains a full spectrum of population densities within its borders, ranging from dense urban communities to sparse rural areas. The majority of the population resides within, or in close proximity to, one of the 16 main urban areas. Over half reside in the cities of Auckland, Christchurch, Wellington, and Hamilton (Statistics New Zealand 2013). Combined with a land mass of 271,000 km2 (104,000 mi.2) the population density is 16 persons/km2 (41 persons/mi.2) (Statistics New Zealand 2009). Figure 10: NZ Population Distribution (New Zealand 2006) 42 Despite only six metropolitan centers contributing to over half of the nation?s population, Figure 11 demonstrates infrastructure provides access to all regions and rail networks extend the entire length of the nation. Figure 11: NZ Infrastructure (National Infrastructure Unit 2014) 43 As observed in Figure 12, domestic air service provides air access to 26 domestic locations within the nation (Air New Zealand 2014). This level of service provides direct air access to all but the most minute of geographic scales. Figure 12: NZ Domestic Air Routes (Airline Network News and Analysis 2014) 44 Both the nation?s geography and wildlife are varied, and combined with a mild and temperate climate, contribute to a robust tourism industry that draws both domestic and foreign visitors.With a gross domestic product of $166 Billion (NZD) the country invests over $36 Billion (NZD) on roads and public transit (NZ Ministry of Transport 2014. Adding to the breadth of knowledge available in understanding NZ travel is the nation?s geography and wildlife. Both are varied and, combined with a mild and temperate climate, contribute to a robust tourism industry that draws both domestic and foreign visitors. 45 4. METHODOLOGY 4.1 The Multinomial Logistic Regression Model Destination and mode choice models are frequently estimated jointly using discrete choice multinomial logistic (MNL) regression models, as opposed to gravity models (LaMondia et al 2010). This is done for a number of reasons, including the ability to evaluate the impact of many factors on these decisions beyond the few used in a gravity model (e.g. distance or time), the ability to consider multiple unique alternatives (as well as evaluate factors differently for each alternative), and the relationship between mode and destination decisions. Specifically, this thesis utilizes a MNL regression to examine the factors influencing the joint dependent decision- variable geographic scale and mode choice, of which there were four possible outcomes (Figure 13). This simplified set of alternatives was selected due to the fact that the majority of within zone and to adjacent zone trips were done with personal autos, as seen in Table 6. 46 Figure 13: Model Structure This model was selected as it supported the nominal nature of the dependent variables under examination while, allowing for the relaxation of the independent and independent and identically distributed (IID) assumption between some sets of alternatives (in this case between modes within a single geographic scale). The relaxation of these was determined to be desirable as random variables for the data set could not be determined to have equal probability distributions. In fact, it was suspected, that due to traveler preference, probability distributions were not equal. The dependent variables supported by this model are nominal, that is, not linear or ordinal but conforming to categorical data with no relative value between categories. The multinomial logit model output is structured by utility and probability. Each possible choice within the dependent variable is assigned a utility function, shown in Equation 1. Subsequently, with the utility for each choice in a particular alternative (trip), the probability function, shown in Equation 3, is used to determine the relative probability of each choice for that alternative. The utility function for each alternative choice i and decision maker j is defined as: 47 Where is the utility of alternative j, is a singular matrix of alternative specific characteristics, and is an error term following a generalized extreme value distribution, independent and identically distributed with a variance of , that incorporates unobserved utility. By setting the variance to this value the model can be normalized to the scale of the utility. The incorporation of the unobserved utility is accomplished in this type of model through approximation of an integral through the determination of the probability of some unobserved value ( . The error term allows the model to specify the difference between predicted outcomes and actual outcomes. In other words, if ?, and subsequently all model considerations, could be observed, then the utility equation could be perfectly predicted. However, this is obviously not the case; all influencing factors are never able to achieve full consideration in a specified model. The multinomial logit model is able to account for this through consideration of the probability of given the observed portion , where , of Equation 1. This is accomplished through the use of simulation or the calculation of an average over some density. Simulation may be accomplished using a number of different techniques, however, for multinomial logit the most common is an inverse cumulative technique. In this method any density is transformed to an associated cumulative density through the use of random variable selection and transformations according to a determined distribution. In this technique any value for the 48 associated mean output provided by the model may have the probability of that value or less occurring. Simulation with the use of an indicator function can be developed as shown in Equation 3. (3) In turn, Equation 3 allows the probability of such that . Equation 4 is then used to calculate a complete closed form, shown in Equation 5, for the logistic in which action is taken only if . Consequently leading to the choice probability model shown in Equation 6. 49 Utilizing this model provides a number of appealing aspects. First, each choice is assigned a probability between 0 and 1. In this scenario all choice probabilities will add to one. It must be noted that in this model structure one of the choice alternatives must be utilized as a base for the other alternatives to be compare to. Additionally, as with most discrete choice models, the probability representation of a choices utility is sigmoid, as shown in Figure 14. Figure 14: Sigmoid (Train 2009) This is appealing as it directly reflects the utility of a choice against other choices. For example, in the context of this report, if the trips utility for a specific choice increases, the magnitude of change for the probability of the same choice will also be dependent on the magnitude of the choices utility in relation to the other choice utility. In other words, if a choices utility is small compared to the alternative utilities a significant increase in the utility may not results in a similar increase in the probability. As such the greatest changes in choice probability are observed when the probability is close to 50%. 50 5. MODEL ESTIMATION The model estimation results can be found in Table 7. The multinomial logit structure proved to be a preferred model for estimating these choices. The likelihood ratio test value of 37,961 was significantly greater than the critical chi-squared test at any confidence level, indicating that this model is statistically more effective at forecasting joint destination and mode choices than a constants-only model. Table 7: Model Estimation Parameters TRIP SCALE Adjacent Zone with Auto Non-Adjacent Zone with Auto Non-Adjacent Zone with Plane or Train VARIABLE Coeff. t-stat Coeff. t-stat Coeff. t-stat TRIP CHARACTERISTICS PRIMARY TRIP PURPOSE IS? ? Business 1.134 13.34 ? Holiday -0.852 -9.43 ? Visit Friends and Relatives 0.248 4.57 0.320 5.52 PRIMARY TRIP ACTIVITY IS? ? Sports -0.167 -4.42 -0.344 -8.79 -0.507 -8.96 ? Cultural Attraction -0.073 -1.77 0.480 7.56 ? Dining 0.256 5.51 ? Shopping 0.253 4.14 TRIP STRUCTURE IS? ? Day trip 0.320 8.72 -0.483 -5.22 ?Overnight trip with only one destination but multiple excursions TRIP LENGTH 0.381 3.30 Number of stops exceeding one hour 0.342 11.89 -0.728 -10.58 Number of legs in trip -0.278 -8.81 -0.582 -17.56 -1.283 -30.70 Round trip distance (KM) 0.014 110.50 0.018 157.54 0.021 169.48 PARTY CHARACTERISTICS Number of Children on the Trip -0.472 -15.67 HOUSEHOLD CHARACTERISTICS HH INCOME IS? ? $19,999 or less -0.247 -2.36 ? $20,000 to $39,999 -0.306 -3.62 ? $40,000 to $69,999 AGE OF RESPONDENT IS? -0.223 -2.86 ? Age of Respondent 25 to 34 -0.200 -5.48 COMPOSITION Number of Adults -0.168 -5.39 ORIGIN CHARACTERISTICS Percent of population with higher than a Bachelors Degree 0.043 12.63 -0.007 -13.89 HOUSEHOLDS Percent of households earning $10,000 to $19,999 -0.003 -20.09 Percent of household earning $9,999 or less 0.004 24.34 0.002 12.45 Number of single parent households 0.007 22.08 51 MODEL ESTIMATION PARAMETERS CONTINUED EMPLOYMENT Percent of population employed as clerks -0.061 -13.39 0.163 19.54 Percent of population with full time employment -0.028 -21.00 LODGING Number of guest nights reported from hostels -0.007 -29.43 -0.003 -37.13 Number of guest nights reported from hotels or motels 0.001 28.22 TRIP TYPES Number of overnight trips for business -0.008 -20.47 -0.004 -26.81 Number of day trips for holiday 0.003 -15.94 Number of overnight trips for holiday 0.003 22.87 Percent population 65 years or older -0.004 -8.97 NEIGHBORING RTO CHARACTERISTICS HOUSEHOLDS Percent of households that are multi- person -0.110 -12.54 Household income $9,999 or less 0.009 10.65 EMPLOYMENT Percent employed as equipment operators 0.200 14.17 FINANCIAL Count of financial transactions for accommodation -0.213 -11.88 Count of financial transactions for food 0.121 25.61 0.084 13.68 Count of financial transactions for fuel -0.123 -14.20 0.040 11.06 Count of financial transactions not related to tourism -0.306 -27.14 Count of financial transactions for retail 0.101 22.15 -0.029 -10.85 -0.094 -18.42 Count of financial transactions not related to retail 0.001 15.31 LODGING Number of guest nights at hostel -0.002 -5.22 0.003 26.62 0.001 8.47 Number of guest nights at motel or hotel 0.008 26.66 -0.001 -17.66 TRIP TYPES Number of overnight business trips -0.015 -18.21 0.002 11.63 0.006 10.38 Number of overnight holiday trips -0.002 -29.85 -0.003 -32.85 Count of day trips for visiting -0.004 -25.22 -0.001 -12.30 -0.002 -9.81 Number of overnight trips for visiting 0.005 6.92 BEYOND NEIGHBORING RTO CHARACTERISTICS HOUSEHOLDS Percent of household that are one parent -0.002 -5.60 LODGING Guest nights at hostel -0.009 -13.58 Guest nights at motel or hotel 0.004 16.24 0.007 11.99 TRIP TYPES Day business trips -0.002 -10.53 0.006 8.99 Number of overnight trips for holidays -0.001 -2.30 Number of day trips for visiting -0.008 -18.12 52 5.1. Trip Characteristics The first category of variables included characteristics specific to the recorded trip. Trips conducted for the purpose of business, relative to traveling within the origin, to an adjacent zone, or to a non-adjacent zone using an automobile, had an increased utility of 1.134 to travel to a non-adjacent zone plane or a train. This increased utility was also relative to all other trip purposes and is likely reflective of the fact that business trips are often funded by entities other than the traveler and, as compared to personal trips, demonstrate a higher value of time than on financial cost. In contrast to this, trips conducted for the holiday travel, compared to all other purposes and relative to the same zones as business travel, had a decreased utility of 0.852 to travel to a non-adjacent zone by plane or a train. Similar to the reasons supporting business travel to this scale and mode, this was likely due to the fact that personal trips are often more sensitive to cost and less sensitive to time. This supports the negative coefficients reflecting a negative utility for a longer travel distance and a more expensive mode. Relative to the travel within the origin zone and non-adjacent zone using air or rail, trips conducted for the purpose of visiting friends and relatives had an increased utility of 0.2248 for travel to adjacent zones and 0.320 for travel to non-adjacent zones by automobile. This is most likely due to visiting of acquaintances within the origin zone reflecting more daily travel considerations than other long-distance travel. Furthermore when long-distance travel is conducted in order to visit friends and family it is often done so with in an environment in which an automobile is more convenient, such as travel with children. Additionally this type of travel adheres to personal travel characteristics in which the traveler is more sensitive to financial cost than time. 53 Trips conducted for sporting activities, compared to all other primary activities and relative to traveling within the origin, had a decreased utility of 0.167 to travel to an adjacent zone, a decreased utility of 0.344 to travel to a non-adjacent zone by auto, and a decreased utility of 0.507 to travel to a non-adjacent zone by air or train. This is likely due to sporting activities reflecting local competition, inherently not requiring long travel distances on the part of the participant. Trips conducted to experience a cultural attraction, also relative to all other activities and travel within the origin and to an adjacent zone, had a decreased utility of 0.073 to travel to a non-adjacent zone by automobile but an increased utility of 0.480 in travel to a non-adjacent zone using air or train. This is likely tied to travel behavior in which travelers desire unique experiences and are willing to travel further distances in order to accomplish as much. The increased utility of air and train most likely reflect the greater distances trip-takers travel in pursuit of such an endeavor. Relative to the travel within the origin, to an adjacent zone, and to a non-adjacent zone using an automobile, travel conducted for participation in a dining activity, also relative to all other activities, had an increased utility of 0.256 for travel to non-adjacent zones by air or train. This was most likely reflective of long-distance travel for dining representing a unique event in which travel mode also reflected the unique nature of the travel involved. Relative to the travel within the origin, to an adjacent zone, and to a non-adjacent zone using automobile, shopping activities had an increased utility of 0.253 for travel to non-adjacent zones by air or train. As with dining this was most likely due to long-distance travel for shopping representing a unique, and extravagant, event in which financial resources supported more expensive travel. Increased financial ability translated to more sensitivity to time rather than financial cost. 54 For trips in which stops exceeding one hour were present, relative to traveling within the origin and to adjacent zones, an increased utility of 0.342 was present for travel to a non-adjacent zone using an automobile and a decreased utility of 0.728 in travel to a non-adjacent zone by air or train. This was likely due to the fact that trains and airplane travel are not conducive to stops exceeding one hour. The more stops the traveler was likely to make the greater the ability of the automobile over a train or airplane to conform. For trips in which multiple legs were present, for every additional leg within a trip a decreased utility was observed. For travel to an adjacent zone a decreased utility of 0.278, a decreased utility of 0.582 in travel to a non-adjacent zone using a car, and a decreased utility of 1.283 in travel to a non-adjacent zone using air or rail. This was likely due to the fact that, as the number of legs in a trip increased, so too did the likelihood the trip conformed to a chained format. In this format more attention is focused on multiple activities, usually in closer proximity to an origin, than one destination farther away. If a trip was a day trip, relative to trips extending overnight and traveling within the origin or to a non- adjacent zone by automobile, both an increased utility of 0.320 for travel to an adjacent zone and a decreased utility of 0.483 in travel to a non-adjacent zone using air or rail were observed. This was likely due to the restriction of day trips to time constraints and conformance to those restraints by closer regions requiring less time for travel. Trips conducted that conformed to an overnight structure with only one destination but multiple excursions, relative to all other trip structures and travel within the origin, to an adjacent zone, or a non-adjacent zone using an automobile, had an increased utility of 0.381 to travel to a non-adjacent zone by plane or a train. This was likely due to the diminished freedom of movement provided by automobile at the destination outweighed by the convenience of air and train for a trip that contained no deviations. For trip distance, for every additional kilometer traveled round trip, an increased utility of 0.014 55 to travel to an adjacent zone, an increased utility of 0.018 in travel to a non-adjacent zone using a car, and an increased utility of 0.021 in travel to a non-adjacent zone using air or rail was observed. This agreed with the condition that travel to more distant regions naturally requires farther distances be traversed. 5.2. Party Characteristics Travel parties with children present, relative to traveling within the origin, to an adjacent zone, or to a non-adjacent zone using an automobile, had a decreased utility of 0.472 to travel to a non-adjacent zone by plane or a train for every additional child in the party. This was likely due to the presence of children adding higher cost as modes charged by individual. This was in contrast to automobile in which additional children did not directly increase cost. Additionally, the presence of children also increased the amount of time and resources that would be required to be absorbed by those responsible for the children. By remaining in the origin, adjacent zone, or travel to a non-adjacent zone by automobile, many of these demands would be more easily met. 5.3. Household Characteristics Household characteristics observed to have a significant impact on analyzed travel scale mode choice combinations were household income and household composition. Households earning $19,999 or less annually, relative all other income levels, travel within the origin, to an adjacent zone, or to a non-adjacent zone using an automobile, had a decreased utility of 0.247 to travel to a non-adjacent zone by plane or a train. This is likely due to travel to further destinations by these modes require more financial investment. Similarly, households earning 56 between $20,000 and $39,999, relative to all other income levels, also had a decreased utility of to travel to a non-adjacent by plane or a train. This negative utility was of the magnitude 0.306. Again this was likely due to households with this level of annual income representing travelers that are more sensitive to mode cost than their counterparts. Additionally, households earning between $40,000 and $69,999, had a decreased utility of 0.223 to travel to a non-adjacent zone by plane or a train. This level of income may no longer be considered a direct factor in decreased utility tied to the higher cost of airplane and train. Instead these cases are most likely due to a higher age bracket in which travel by air or train is not desired due to time or convenience preferences. 5.4. Characteristics for Zone of Origin Origin characteristics observed to have a significant impact on scale of travel and mode choice for trips generated were household characteristics, employment characteristics, lodging characteristics, and trip type. Additionally, for origin bound trips, education also demonstrated a significant impact on travel. For every additional percentage of the population with an education above a bachelor?s degree, relative to traveling within the origin and to an adjacent zone, an increased utility of 0.043 was observed to travel to an adjacent zone by automobile and a decreased utility of 0.007 for travel to a non-adjacent zone by air or train. This was likely due to higher educational levels providing increased motivation and ability to travel. However, the preference to travel by air or train is not supported indicating higher education does not, by itself, provide sufficient financial or time advantages to develop these as preferred mode choices. For every additional percentage of households earning between $10,000 and $19,999, relative to traveling within the origin, to an adjacent zone, and to a non-adjacent zone by automobile, a 57 decreased utility of 0.003 to travel to a non-adjacent zone by air or train was observed. This was most likely reflective of this income level being more sensitive to mode cost than other income levels may be. This statement is supported by the findings for the representation of $9,999 and less households. For every additional percentage of households earning less than $9,999, relative to travel within the origin, and to travel to non-adjacent zones by air or train, both travel to an adjacent zone and travel to a non-adjacent zone by automobile observed an increased utility of 0.004. Again, as with the proceeding income variable, this was likely representative of households with this level of income indicating fewer resources to devote to travel. For every additional household with only one parent, relative to traveling within the origin, to a non- adjacent zone by automobile, and to a non-adjacent zone by air or train, an increased utility of 0.007 for travel to an adjacent zone was observed. This was likely a result of fewer resources available within a household to support long-distance travel. In regards to the effect of zone of origin on long-distance trips two employment measures proved significant: percent of origin zone employed as a clerk, and percent of origin employed full time. For every additional percentage of workers employed as a clerk, relative to traveling within the origin and to an adjacent zone, a decreased utility of 0.061 to travel to a non-adjacent zone by automobile was observed. Also an increased utility of 0.163 for travel to a non-adjacent zone by air or train was observed. These utilities were likely due to the employment characteristics of the origin and the associated involvement of business activities for travel. For every additional percentage of workers employed full-time, relative to traveling within the origin, to a non-adjacent zone by automobile, and to a non-adjacent zone by air or train, a decreased utility of 0.028 to travel to an adjacent zone was observed. This was likely due to an 58 increased employment within the origin generating an environment of greater financial freedom to engage in trips at the non-adjacent zone scale. In regards to the effect of origin lodging characteristics on long-distance trips, two lodging measures proved significant: number of guest nights reported by hostels, and number of guest nights reported by either motels or hotels. For every additional guest night reported by hostels, relative to traveling within the origin and to a non-adjacent zone by air or train, a decreased utility of 0.007 for travel to an adjacent zone was observed. Additionally, a decreased utility of 0.003 for travel to a non-adjacent zone by automobile was observed. This was likely due to hostels supporting individuals without use of an automobile. For every additional guest night reported by hotels or motels, relative to traveling within the origin and to a non-adjacent zone by air or train, an increased utility of 0.001 for travel to an adjacent zone was observed. This was likely representative of hotel activity involving automobiles. Factors within the origin demonstrated to have significance in determining scale and mode choice were found to be number overnight trips for business, number day trips for holiday travel, and number of overnight trips for holiday travel. For every additional overnight trip for business generated by the zone of origin, relative to traveling within the origin, and to a non- adjacent zone by air or train, a decreased utility of 0.008 for travel to an adjacent zone, and a decreased utility of 0.004 for travel to a non-adjacent zone by automobile were observed. This was likely due to business trips conducted by automobile being less likely to take longer than one day as opposed to business trips in which transportation by air or train was involved. For every additional day trip generated by the origin for holiday travel, relative to traveling within the origin, to a non-adjacent zone by automobile, and to a non-adjacent zone by air or train, an increased utility of 0.003 for travel to an adjacent zone was observed. This was likely 59 representative of holiday travel reflecting repeating travel, or traditions. Additionally this was better enabled with the convenience of regions more easily accessible than those non-adjacent. Similarly, for every additional overnight trip for holiday travel generated by the origin, relative to traveling within the origin, to a non-adjacent zone by automobile, and to a non-adjacent zone by air or train, an increased utility of 0.003 for travel to an adjacent zone was observed. Again, as with day holiday travel, this was likely due to holiday travel reflecting repeating travel, or traditions, better enabled with the convenience of regions more easily accessible than those non- adjacent. 5.5. Characteristics for Adjacent Zones Adjacent zone characteristics observed to have a significant impact on scale of travel and mode choice for trips generated were household characteristics, employment characteristics, financial characteristics, lodging characteristics, and trip type. Additionally, for adjacent zones, age demonstrated an impact on travel. For every additional percentage of the population above 65 years old demonstrated by the median value for adjacent zones, relative to traveling within the origin, to an adjacent zone, and to a non-adjacent zone by automobile, a decreased utility of 0.004 for travel to a non-adjacent zone by air or train was observed. This was likely due to higher age levels of the area reducing the involvement of air and train travel. In regards to the effect of adjacent zone household characteristics on long-distance trips two household measures proved significant: median percent of households that are multi-person and median percent of households that have an income of $9,999 or less. For every median percent increase in households that are multi-person for the adjacent zones, relative to traveling within the origin, to an adjacent zone, and to a non-adjacent zone by automobile, a decreased 60 utility of 0.110 for travel to a non-adjacent zone by air and train was observed. This was likely due multi-person households reflecting communities in which families with children are present and the associated logistical challenges that are inherent with air and rail travel. For every median percent increase in households that have an income of $9,999 or less for the adjacent zones, relative to traveling within the origin, to a non-adjacent zone by automobile, and to a non- adjacent zone by air or train, an increased utility of 0.009 for travel to an adjacent zone was observed. This was likely due to the cost of living in these regions providing financial incentive for them to become trip destinations. For employment characteristics of adjacent zones, every median percent increase in workers employed as equipment operators, relative to traveling within the origin, to an adjacent zone, and to a non-adjacent zone by automobile, an increased utility of 0.200 for travel to a non- adjacent zone by air and train was observed. This was likely due these employment characteristics reflecting the industry (and conversely attractions) present in surrounding areas that would motivation travel to non-adjacent zones by air or train. For financial characteristics of adjacent zones, every median increase in the count of financial transactions for accommodation, relative to traveling within the origin, to a non- adjacent zone by automobile, and to a non-adjacent zone by air or train, a decreased utility of 0.213 for travel to an adjacent zone was observed. This was likely due to travelers finding more utility in remaining at home rather than in an adjacent zone when high demand for lodging (and increased prices) are present. Every median increase in the count of financial transactions for food in the adjacent zones, relative to traveling within the origin, and to a non-adjacent zone by automobile, an increased utility of 0.121 for travel to adjacent zones, and an increased utility of 0.084 for travel to non-adjacent zones by air or train was observed. This may be due to 61 conditions in which zone residents were compelled to travel to adjacent zones, most likely to engage in a more robust economy. The increase in utility for travel to an non-adjacent zone by air or train may be reflective of greater financial ability reflected by residents in regions with more purchasing activity. For every median increase in the count of financial transactions for fuel in the adjacent zones, relative to traveling within the origin, and to a non-adjacent zone by air or train, a decreased utility of 0.123 for travel to an adjacent zone, and an increased utility of 0.040 for travel to a non-adjacent zone by automobile were observed. This was probably due to increased fuel purchases reflecting an increased presence of daily life rather than experiences of novelty. This is supported by the increased utility of travel to a non-adjacent zone, reflecting travelers bypassing the higher fuel consumption areas to visit other areas. For every median increase in the count of financial transactions not related to tourism in adjacent zones, relative to traveling within the origin, to a non-adjacent zone by automobile, and to a non-adjacent zone by air or train, a decreased utility of 0.306 for travel to an adjacent zone was observed. This was likely due to tourism being a significant driver of travel to an adjacent zone. For every median increase in the count of financial transactions for retail in adjacent zones, relative to travel within the origin, an increased utility of 0.101 for travel to an adjacent zone, a decreased utility of 0.029 for travel to a non-adjacent zone by automobile, and a decreased utility of 0.094 for travel to a non-adjacent zone by air or train was observed. This was likely due to shopping activities in the adjacent zones inherently providing more convenience and cost-effectiveness than those in non- adjacent zones. For every median increase in the count of financial transactions for other than retail in adjacent zones, relative to traveling within the origin, to an adjacent zone, and to a non- adjacent zone by air or train, an increased utility of 0.001 for travel to a non-adjacent zone by 62 automobile was observed. This was likely a result of increased business activities, associated with non-retail purchasing, increasing motivation to travel beyond the adjacent zone. In regards to the effect of adjacent zone lodging characteristics on long-distance trips two lodging measures proved significant: number of guest nights reported by hostels, and number of guest nights reported by either motels or hotels. For every additional guest night reported by hostels, relative to traveling within the zone of origin, a decreased utility of 0.002 for travel to an adjacent zone, an increased utility of 0.003 for travel to a non-adjacent zone by automobile, and an increased utility of 0.001 for travel to a non-adjacent zone with air or train were observed. This was probably due to hostels supporting individuals desiring new experiences and that are willing to travel to gain those experiences. For every additional guest night reported by hotels or motels, relative to traveling within the origin, and to an adjacent zone, an increased utility of 0.008 for travel to a non-adjacent zone by automobile, and a decreased utility of 0.001 for travel to a non-adjacent zone by air or train was observed. This is likely due to travelers in a zone engaging in travel activity more often. Trip types to adjacent zones demonstrated to have significance in determining scale and mode choice of trips were found to be number overnight trips for business, number of overnight trips for holiday travel, number of day trips for visiting friends and relatives , and number of overnight trips for visiting friends and relatives. For every additional overnight trip for business generated by an adjacent zone, relative to traveling within the origin, a decreased utility of 0.015 for travel to an adjacent zone, an increased utility of 0.002 for travel to a non-adjacent zone by automobile, and an increased utility of 0.006 for travel to a non-adjacent zone by air and train, were observed. This was likely due to the fact that business trips conducted in a region generate additional travel to non-adjacent zone. For every additional overnight trip for holiday travel 63 generated by the adjacent zone, relative to traveling within the origin, and travel to a non- adjacent zone by air or train, a decreased utility of 0.002 for travel to an adjacent zone, and a decreased utility of 0.003 were observed. This was likely due to the fact that areas with higher holiday travel may experience less automobile trips to non-adjacent zones because they are better able to provide the financial resources for travel by plane. For every additional day trip for visiting generated by the adjacent zone, relative to traveling within the origin, a decreased utility of 0.004 for travel to an adjacent zone, a decreased utility of 0.001 for travel to a non-adjacent zone by automobile, and a decreased utility of 0.002 for travel to a non-adjacent zone by air and train, were observed. This is likely due to an increase in day trips for visiting in an adjacent zone reflects a regional characteristic meaning the travelers in the origin are also engaging in day travel, trips limited by time from extending to further destinations. For every additional overnight trip for visiting generated by the adjacent zone, relative to traveling within the origin, adjacent zone, and to a non-adjacent zone by automobile, an increased utility of 0.005 for travel to a non-adjacent zone by air and train were observed. This was likely due to an increased regional participation in overnight travel increasing the likelihood of travel by air or train. 5.6. Characteristics for Non-Adjacent Zones Non-adjacent zone characteristics observed to have a significant impact on scale of travel and mode choice for trips generated were household characteristics, lodging characteristics, and trip type. In regards to the effect of non-adjacent zone household characteristics on long-distance trips one household measures proved significant: median percent of households that are one parent. For every median percent increase in households that are one parent for the non-adjacent 64 zones, relative to traveling within the origin, to an adjacent zone, and to a non-adjacent zone by automobile, a decreased utility of 0.002 for travel to a non-adjacent zone by air and train was observed. This is likely due one parent households reflecting less affluent communities that have less appeal to those that would travel by air or train. In regards to the effect of non-adjacent zone lodging characteristics on long-distance trips two lodging measures proved significant: number of guest nights reported by hostels, and number of guest nights reported by either motels or hotels. For every additional guest night reported by hostels, relative to traveling within the origin, to an adjacent zone, and to a non- adjacent zone by auto, a decreased utility of 0.009 for travel to a non-adjacent zone with air or train was observed. This was likely due to travel to areas with many hostels would not often be conducted by those that spend the money to travel by air or train. For every additional guest night reported by hotels or motels, relative to traveling within the origin, and to an adjacent zone, an increased utility of 0.004 for travel to a non-adjacent zone by automobile, and an increased utility of 0.001 for travel to a non-adjacent zone by air or train were observed. This was likely due to travelers in being drawn to the activities in the areas that support the hotel and motels in the area. Trip types to non-adjacent zone demonstrated to have significance in determining scale and mode choice of trips were found to be number of day business trips, number of overnight trips for holiday travel, and median number of day trips for visiting friends and relatives. For every additional day trip for business generated by the non-adjacent zone, relative to traveling within the origin, and travel to non-adjacent zone by air or train, a decreased utility of 0.002 for travel to an adjacent zone, and an increased utility of 0.006 for travel to a non-adjacent zone by automobile were observed. This likely represented the business trips in a zone drawing travel to 65 the zone. For every additional median overnight trip for holiday travel generated by the non- adjacent zone, relative to traveling within the zone, travel to an adjacent zone, and travel to a non-adjacent zone by air or train, a decreased utility of 0.001 for travel to a non-adjacent zone by auto was observed. This was likely due to attractions that draw overnight holiday travel make travel to a non-adjacent zone by plane and train worth the added financial resources for travels. For every additional median day trip for visiting generated by non-adjacent zones, relative to traveling within the origin, travel to an adjacent zone, and travel to a non-adjacent zone by air or train, a decreased utility of 0.008 for travel to a non-adjacent zone by auto was observed. This was likely due to the fact that travel to visit friends and relatives warrants the cost in order to save time. 66 6. SUMMARY AND CONCLUSIONS This study examined the influence of household, travel party, trip, and RTO (in conjunction with mode) characteristics on discrete geographic scale choices of long-distance personal travel. This process was performed through an examination of outcomes from a multinomial logit discrete choice model. However, unlike mainstream practices that rely heavily upon distance thresholds to define scale (or simply forego any consideration of scale once the determination of long-distance has been made), this study utilized discrete geographical boundaries to develop zones. This successfully allowed for the identification of numerous factors contributing to long-distance trip making and enabled a better understanding of how such trips are completed. In this approach the selection of a MNL model for use, initially considered for its ability to efficiently compute utilities of multiple unordered discrete choices that lacked correlation, proved to be a viable option for consideration of geographic scale/ mode choice combinations. Initial concerns did arise about the models ability to process the large number of variables. However, running the model in separate iterations for each category of characteristics until a suitable number remained to enable all remaining to be considered in a single model proved to be an easy and effective means to overcome such a concern. Ultimately the model identified 51 variables that proved both significant and thorough in determining not only the scale of travel but also mode of travel for trips to non-adjacent zones. 67 Use of zones, structured as concentric regions based upon proximity of each trips origin, proved advantageous over definitions of scale based upon continuous measures of distance for a number of reasons. First, results indicate geographic zones are deeply ingrained in long-distance travel behavior. Results proved zones effectively capture the different factors that contribute to scale of long-distance trips. Each of the zones considered for this research were shown to be effected by highly unique sets of independent variables that were both significant and able to develop an effective consideration of household, travel party, trip, and regional demographic components examined. Second, such a structure allowed for researchers to overcome subjectivity associated with distance based approaches while allowing for geographic factors, both physical and perceptual, to be incorporated directly into the definitions of scale. Furthermore, it must be noted that relationships between these scales also became apparent through this research. For example, in cases in which the trip characteristics category coefficients were identified for variables in adjacent and non-adjacent zones, almost all of the variables that had a negative (or positive) coefficient for the utility of an adjacent zone with an automobile would also have a negative (or positive) coefficient for the utility of the non-adjacent zones. This indicated that when a variable had a specific effect on the utility of travel to an adjacent zone by car, relative to travel within the zone of origin, it would tend to have a negative effective on the utility of travel to non-adjacent zones also. However, within the category capturing demographic characteristics of RTOs neighboring the RTO of origin this trend was observed to be greatly reduced indicating that, within this category, variables did not tend to influence travel to non-adjacent zones in the same manner as travel to adjacent zones. Model results indicated significant influence on selected scale and mode of travel by a diverse set of variables spanning household, travel party, trip, and regional demographic 68 characteristics. Of these, variables within the trip characteristic category possessed coefficient with the greatest magnitude. Part of this may be attributed to the diminished scale of other variables such as the numerous variables in the zonal characteristics that reflect percentages rather than the predominately binary variables in trip characteristics. However, it must be noted that these variables are by no means the only binary variables represented and, of the variables representing values based upon percentages or other continuous values, minimal variability is often observed within them. This indicates that these values are comparable and magnitudes are meaningful comparisons. The fact that the trip characteristics selected for inclusion in the final model possess a larger magnitude is an indication that this category tends to have a greater impact on the scale and mode choice of observed trips. Also of note is every variable representing household characteristics had no effect on the utility of travel to an adjacent zone but a negative effect on travel to non-adjacent zones as compared to travel within the zone of origin. Development of this methodology provides the greatest use to organizations committed to the development of policy regarding travel behavior, especially those supporting the interests of tourism and other sectors highly susceptible to long-distance travel. Such agencies will benefit through a better understanding of the factors that contribute to demands on infrastructure, both transportation related and that tied to support of tourism activities, and visitor motivation for travel. Such benefits will be realized through vastly improved efficiency in allocation of resources, both fiscal and otherwise. However, although counterintuitive, secondary benefits may also be provided to community planners as they may better understand the local resources that affect the travel within their region from an external perspective. Tourism organizations will be able to better analyze travel to areas as a consequence of attraction/ repulsion factors particular to 69 an area and surrounding regions. Additionally, utilizing this approach, the transportation planning community will be able to better forecast travel trends and the demands placed on systems due to long-distance travel, independent of a subjective definition of scale. Benefits of this technique may also extend to broader areas influenced by transportation. For example, utilizing this approach to analyze travel behavior governmental agencies may better assess economic development and quality of life that has become so dependent on accessibility and equity of mobility. Logically, this methodology would benefit from being further explored in other areas, with different geographies. Not only would this validate existing methodologies, it could also unveil variables significant to specific cultures, geographies, infrastructure, and a better understanding of the effect of policy. This may be accomplished through the consideration of other datasets that incorporate broader regions that are comprised of distinctly different populations. Through this approach, development of a larger dataset with a greater number of cases would be enabled that would also serve to increase statistical confidence. Accomplishing this would allow for a model better able to consider choices according to broader geographic contexts. 70 7. REFERENCES Airline Network News and Analysis. (2014). Retrieved from http://www.anna.aero/wp- content/uploads/2010/01/airnewzealand-network-map.jpg Air New Zealand: New Zealand Domestic Travel. (2014). Retrieved from http://www.airnewzealand.com/domestic-travel Andreas F., (2008). "Survey Issues in Long-Distance Travel". Conference paper 8th Swiss Transport Research Conference. Bamberg, S., Ajzen, I., Schmidt, P., (2003). ?Choice of Travel Mode in the Theory of Planned Behavior: The Roles of Past Behavior, Habit, and Reasoned Action?. Basic and Applied Social Psychology. Volume 25. Issue 3. Bargeman, B., & van der Poel, H. (2006). The role of routines in the vacation decision-making process of Dutch vacationers. Tourism Management, 27(4), 707-720. Beerli, A., Meneses, G. D., & Gil, S. M. (2007). Self-congruity and destination choice. Annals of Tourism Research, 34(3), 571-587. Ben-Akiva, M. E., & Lerman, S. R. (1985). ?Discrete Choice Analysis: Theory and Application to Travel Demand?. (Vol. 9). MIT press. Bhat, C., Govindarajan, A., & Pulugurta, V. (1998). ?Disaggregate Attraction-End Choice Modeling: Formulation and Empirical Analysis?. Transportation Research Record: Journal of the Transportation Research Board, 1645(1), 60-68. Bieger, T., & Laesser, C. (2002). Market segmentation by motivation: The case of Switzerland. Journal of Travel Research, 41(1), 68-76. Boarnet, M.G., (1995). ?Transportation Infrastructure, Economic Productivity, and Geographic Scale: Aggregate Growth versus Spatial Redistribution?. UC Berkeley: University of California Transportation Center. < http://escholarship.org/uc/item/6sj276z4>. Borooah, V. K. (2002). Logit and Probit: Ordered and Multinomial Models. Thousand Oaks: Sage Publications. Bricka , S. (2001). ?Variations in Long-Distance Travel?. Personal Travel: The Long and Short of It. Retrieved from http://trid.trb.org/view/2001/C/686693 71 Brockmann, D., Hufnagel, L., Geisel, T., (2006). ?The Scaling Laws of Human Travel.? Nature. Volume 439. Issue 7075. Pages 462-465. Br?g, W., Erl, E., Sammer, G., Schulze, B., (2003). "Dateline?Design and Application of a Travel Survey for Long-Distance Trips Based on an International Network of Expertise? Concept and Methodology." Conference paper: 10TH International Conference on Travel Behavior Research. Lucerne. Bryan, H., (1977). ?Leisure Value Systems and Recreational Specialization: The Case of Trout Fisherman?. Journal of Leisure Research. Volume 9. Pages 74-187. Can, Vo Van., (2013). "Estimation of Travel Mode Choice for Domestic Tourists to Nha Trang using the Multinomial Probit Model.". Transportation Research Part A: Policy and Practice. Volume 49. Pages 149-159. Cassel, S., Macuchova, Z., Rudholm, N., Rydell, A., (2012). ?Willingness to Commute Long- Distance Among Job Seekers in Dalarna, Sweden?. Journal of Transport Geography. Volume 28. Issue 0. Pages 49-55. Cervero, R., (2002). ?Built Environments and Mode Choice: Toward a Normative Framework?. Transportation Research Part D: Transport and Environment. Volume 7. Issue 4. Pages 265-284. Dargay, J., (2010). "The Prospects for Longer Distance Domestic Coach, Rail, Air and Car Travel in Britain." Report by Institute of Transport Studies, Leeds, to the Independent Transport Commission. Dellaert, B., Ettema, D., Lindh, C., (1998). ?Multi-Faceted Tourist Travel Decisions: A Constraint-Based Conceptual Framework to Describe Tourists' Sequential Choices of Travel Components?. Tourism Management. Volume 19, Issue 4. Denver Council of Regional Governments. (2010). Front Range Travel Counts. Web. . Dredge,D., (1999). ?Destination Place Planning and Design?. Annals of Tourism Research. Volume 26. Issue 4. Pages 772-791. Erhardt, G., Freedman, J., Stryker, A., Fujioka, H. and Anderson, R. (2007). ?The Ohio Long-Distance Travel Model?. Transportation Research Record, Washington, D.C. European Commission. (2013). Standard Eurobarometer 79. Eurobarometer Vacation Travel Survey. Web. < http://ec.europa.eu/public_opinion/archives/eb/eb79/eb79_en.htm>. 72 Fr?ndberg, L., Vilhelmson B., (2003). ?Personal Mobility: A Corporeal Dimension of Trans- Nationalisation. The Case of Long-Distance Travel from Sweden?. Environment and Planning. A Volume 35. Issue 10. Pages 1751 ? 1768. Friedrichsmeier, T., Matthies, E., Kl?ckner, C., (2013). "Explaining Stability in Travel Mode Choice: An Empirical Comparison of Two Concepts of Habit." Transportation Research Part F: Traffic Psychology and Behavior. Volume 16. Pages 1-13. G?ssling, S., (2002). ?Global Environmental Consequences of Tourism?. Global Environmental Change. Volume 12. Issue 4. Pages 283-302. Government of New Zealand, National Infrastructure Unit. (2014). Transport. Retrieved from website: http://www.infrastructure.govt.nz/plan/2011/20.htm Greene, W., Hensher, D., Rose, J., (2006) "Accounting for Heterogeneity in The Variance of Unobserved Effects in Mixed Logit Models". Transportation Research Part B: Methodological. Volume 40.Issue 1. Pages 75-92. Hagen-Zanker, A., Jin, Y., (2013). ?Adaptive Zoning for Transport Mode Choice Modeling?. Transactions in GIS. Ham, H., John Kim, T., Boyce, D., (2005). ?Implementation and Estimation of a Combined Model of Interregional, Multimodal Commodity Shipments and Transportation Network Flows?. Transportation Research Part B: Methodological. Volume 39. Issue 1. Pages 65- 79. Handy, S., Boarnet, M., Ewing, R, Killingsworth, R., (2002). ?How the Built Environment Affects Physical Activity: Views from Urban Planning?. American Journal of Preventive Medicine. Volume 23, Issue 2, Supplement 1. Hesse, M, Rodrigue,J., (2004). ?The Transport Geography of Logistics and Freight Distribution?. Journal of Transport Geography, Volume 12, Issue 3. Hunecke, M., Bl?baum, A., Matthies, E., H?ger, R., (2001). ?Responsibility and Environment: Ecological Norm Orientation and External Factors in the Domain of Travel Mode Choice Behavior?. Environment and Behavior. Volume 33: 830-852. Jansen-Verbeke, M., Spee, R., (1995). ?A Regional Analysis of Tourist Flows within Europe?. Tourism Management. Volume 16. Issue 1. Pages 73-80. Jones, R., (1998). ?Remittances and Inequality: A Question of Migration Stage and Geographic Scale?. Economic Geography. Volume 74. Issue 1. de Jong, G., Gunn, H., Ben-Akiva, M. (2004). "A Meta-Model for Passenger and Freight Transport in Europe." Transport Policy. Volume11. Issue 4 Pages 329-344. 73 Kodrzycki, Y., (2001). ?Migration of Recent College Graduates: Evidence from the National Longitudinal Survey of Youth?. New England Economic Review. Pages 13-34. Koordinates Data and Maps, (2009). ?2006 Census.New Zealand Area Units?. Statistics New Zealand. Web. 20 May 2013. . Koppelman F., Sethi V., (2005). ?Incorporating Variance and Covariance Heterogeneity in the Generalized Nested Logit model: An Application to Modeling Long-Distance Travel Choice Behavior?. Transportation Research Part B: Methodological. Volume 39. Issue 9. Pages 825-853. Krieger, H., Fernandez, E., (2006). ?Too Much or Too Little Long-Distance Mobility in Europe? EU Policies to Promote and Restrict Mobility?. Seminars of the European Foundation for the Improvement of Living and Working Conditions on Worker Mobility in Latvia and Lithuania. Kuhnimhof, T., J?rg L., (2009). "The Path to Better Long-Distance Travel Data in Europe?The Potential of Combining Established Household Survey Instruments and Methodological Innovations." First International Conference on the Measurement and Economic Analysis of Regional Tourism in San Sebastian, Spain. Kwigizzile, V., & Teng, H. (2009). ?Comparison of Methods for Defining Geographical Connectivity for Variables of Trip Generation Models?. Journal of Transportation Engineering, 454-466. LaMondia, J., Snell, T., & Bhat, C. R. (2010). ?Traveler Behavior and Values Analysis in the Context of Vacation Destination and Travel Mode Choices?. Transportation Research Record: Journal of the Transportation Research Board, 2156(1), 140-149. Lang, C. T., O'leary, J. T., & Morrison, A. M. (1997). Distinguishing the destination choices of pleasure travelers from Taiwan. Journal of Travel & Tourism Marketing, 6(1), 21-40. de Lapparent, M., Axhausen , K.W., Frei, A. (2013). ?Long-Distance Mode Choice and Distributions of Values of Travel Time Savings in Three European Countries?. European Transport. Trasporti Europei. Volume 53. Paper 1. Lehto, X., O?Leary, J., and Morrison A., (2004) ?The Effect of Prior Experience on Vacation Behavior?. Annals of Tourism Research. Volume 31. Issue 4. Pages 801-818. Limtanakool, N., Dijst, M., Schwanen, T., (2006). ?The Influence of Socioeconomic Characteristics, Land Use and Travel Time Considerations on Mode Choice for Medium- and Longer-Distance Trips?. Journal of Transport Geography. Volume 14. Issue 5. 2006. Pages 327-341. Lise, W., & Tol, R. S. (2002). Impact of Climate on Tourist Demand. Climatic Change, 55(4), 429-449. 74 Mallett, W. (n.d.). ?Long-Distance Travel by Low-Income Households?. TRB Transportation Research Circular E-C026?Personal Travel: The Long and Short of It. Pages 169-177. Menard, S. (2001). Applied Logistic Regression Analysis. Thousand Oaks: Sage Publications. Metz, D., (2013). "Mobility, Access and Choice: A New Source of Evidence." Journal of Transport and Land Use. Volume 6. Issue 2. Michigan. Department of Transportation, (2010). ?2009 Comprehensive Household Travel Data Collection Program?. MI Travel Counts II Comparison Report. New York: Abt SRBI. Ministry of Business, Innovation, and Employment. (2014). ?About Us?. Retrieved from http://www.mbie.govt.nz/about-us. Mokhtarian, P., Ilan S., (2001). "How Derived is the Demand for Travel? Some Conceptual and Measurement Considerations." Transportation Research Part A: Policy and Practice. Volume 35.Issue Pages 695-719. Moeckel, R., Fussell, R., & Donnelly, R. (2013). ?Mode Choice Modeling for Long-Distance Travel?. Transportation Research Board 92nd Annual Meeting (No. 13-1796). Murphy, P. E. (2013). Tourism: A Community Approach. Routledge. Nellet, C., et al. "Michigan?s Statewide Travel Demand Model." 75th Transportation Research Board Annual Meeting. Washington, DC. 1996. New Zealand, (2014). ?Connecting New Zealand- Land Transport Infrastructure and Services?. (2014). Retrieved from http://www.transport.govt.nz/ourwork/keystrategiesandplans/connectingnewzealand/cnzl andtransport/ New Zealand, Ministry for the Environment. (2006). Figure 2.2: Population distribution by region . Retrieved from website: http://www.mfe.govt.nz/publications/ser/enz07- dec07/html/chapter2-environment/figure-2-2.html New Zealand, (2006). Statistics New Zealand. ?2006 Census?. Web. . New Zealand, Department of Internal Affairs. (2013).Government Datasets Online. Retrieved from website: https://data.govt.nz/catalog New Zealand, Ministry of Business, Innovation, and Employment. (2013). CAM Regional Pivot Tables. Retrieved from website: http://www.med.govt.nz/sectors- industries/tourism/tourism-research-data/commercial-accommodation-monitor-data/cam- regional-pivot-tables 75 New Zealand, Ministry of Business, Innovation, and Employment. (2013). RTI Detailed Data: Domestic RTI. Retrieved from website: http://www.med.govt.nz/sectors- industries/tourism/tourism-research-data/regional-tourism-indicators/detailed-data New Zealand, (2008). Ministry of Tourism. ?Domestic Travel Survey Technical Description?. New Zealand, (2013). Regional Tourism Organisation. ?RTO Location Map?. . New Zealand, Statistics New Zealand. (2006). Regional Summary Tables. Retrieved from website: http://www.stats.govt.nz/Census/about-2006-census/regional-summary- tables.aspx New Zealand, Statistics New Zealand. (2006).Quickstats about New Zealand. Retrieved from website: http://www.stats.govt.nz/Census/2006CensusHomePage/QuickStats/AboutAPlace/SnapS hot.aspx?type=region&tab=Work&id=9999999 New Zealand, Statistics New Zealand. (2009).Subnational Population Estimates: At 30 June 2009. Retrieved from website: http://www.stats.govt.nz/browse_for_stats/population/estimates_and_projections/Subnati onalPopulationEstimates_HOTP30Jun09.aspx New Zealand, Statistics New Zealand. (2013). Survey of Family, Income, and Employment. Retrieved from website: http://www.stats.govt.nz/browse_for_stats/income-and- work/Income/survey-of-family-income-and-employment-info-releases.aspx New Zealand, Statistics New Zealand. (2013).Subnational Population Estimates Tables. Retrieved from website: http://www.stats.govt.nz/browse_for_stats/population/estimates_and_projections/subnati onal-pop-estimates-tables.aspx Nicolau, J. L., & M?s, F. J. (2005). Stochastic modeling: a three-stage tourist choice process. Annals of Tourism Research, 32(1), 49-69. Pearce, A., Douglas, G., (1993). ?Domestic Tourist Travel Patterns in New Zealand?. Volume 29. Issue 3. Pages 225-232. Pearce, A., Douglas, G. (1995). Tourism Today: A Geographical Analysis (No. Ed. 2). Longman scientific & technical. Rodrigue, J., Claude C., and Brian S., (2009). ?Geography of Transport Systems?. Web. . 76 Rohr, C., Fox, J., Daly, A., Patruni, B., Patil, S., & Tsang, F. (2013). "Modeling Long-Distance Travel in the United Kingdom." Transportation Research Board 92nd Annual Meeting. No. 13-3497. Schwanen, T., Dieleman, F. M., & Dijst, M. (2001). ?Travel Behaviour in Dutch Monocentric and Policentric Urban Systems?. Journal of Transport Geography,9(3), 173-186. Southworth, F. (1981). ?Calibration of Multinomial Logit Models of Mode and Destination Choice?. Transportation Research Part A: General, 15(4), 315-325. Statistics New Zealand (2014). ?About Us?. Retrieved from http://www.stats.govt.nz/about_us.aspx United States. Bureau of Transportation Statistics. Research and Innovative Technology Administration. Web. . United States Department of Transportation. (2011). "2009 National Household Travel Survey UsersGuide."Our Nations Travel. Federal Highway Administration, Web. 29 Jul 2013. . U.S. Department of Commerce, U.S. Census Bureau. (2010). ?Geography: 2010 Census Tallies of Census Tracts, Block Groups & Blocks?. Retrieved from website: https://www.census.gov/geo/maps-data/data/tallies/tractblock.html Van Nostrand, C., Vijayaraghavan S., and Abdul Rawoof, P., (2013). "Analysis of Long- Distance Vacation Travel Demand in the United States: A Multiple Discrete?Continuous Choice Framework." Transportation. Volume 40. Issue 1. Pages 151-171. Verhoeven, M., Arentze, T., Timmermans, H., van der Waerden, P., (2005). ?Modeling the Impact of Key Events on Long-Term Transport Mode Choice: Decision Network Approach Using Event History Data?. Transportation Research Record: Journal of the Transportation Research Board. Volume 1926. Issue 1. Waddell, P., (2002). ?UrbanSim: Modeling Urban Development for Land Use, Transportation, and Environmental Planning?. Journal of the American Planning Association. Volume 68. Issue 3. Woodside, A., and MacDonald, R.,(1994). ?General System Framework of Customer Choice Processes of Tourism Services?. Gasser and K. Weiermair (cds), Spoilt lop Choice. KulturVerlag, Austria. 77 Whalen,K., P?ez, A., Carrasco, J., (2013). ?Mode Choice of University Students Commuting To School and the Role of Active Travel?. Journal of Transport Geography. Volume 31. Pages 132-142. Yang, L., Zheng, G., Zhu, X., (2013). ?Cross-Nested Logit Model for The Joint Choice of Residential Location, Travel Mode, and Departure Time?. Habitat International. Volume 38. Issue 0. Pages 157-166. Yang, M., Wang, W., Chen, X., Wang, W., Xu, R., & Gu, T. (2010). ?Modeling Destination Choice Behavior Incorporating Spatial Factors, Individual Sociodemographics, and Travel Mode?. Journal of Transportation Engineering, 800-810. Zhang, L., Krause, C., (2013). ?An Activity Space Based Approach for Capturing Long-Distance Travel Using Longitudinal GPS Survey Data?. Transportation Research Board 92nd Annual Meeting. No. 13-3079. Z?rich , E, and A Frei. "Survey Issues in Long-Distance Travel." Swiss Transport Research Conference. (2008): n. page. Web. 18 Mar. 2014. http://www.strc.ch/conferences/2008/2008_Frei_SurveyIssuesLongDistanceTravel.pdf 78 8. APPENDIX A: RTO DEVELOPMENT Table 8: Dataset Compilation Component Title Dataset RTO Component Comments Tourism Tourism Forecasts 2012-2018 Sector Outlook No Focused on origin of international visitors, examines characteristics by country of origin no data on where visited within NZ Regional Tourism Indicators International Data Tables Yes Focused on Destination RTOs. RTI is percentage of spending above 2008 (100 indicates equal spending) Domestic Data Tables Yes Focused on Destination RTOs. RTI is percentage of spending above 2008 (100 indicates equal spending) Detailed Data Yes Focused on Destination RTOs. RTI is percentage of spending above 2008 (100 indicates equal spending). Provides intra- RTO RTIs and RTO RTI by Industry. Tourism Satellite Account Tourism Satellite Account No Reports expenditure amount. National Tourism Forecast Pivot Tables 2011 - 2016 Tourism Domain Plan No Plan to ensure data collected meets needs of the tourism industry International Travel No Interface reporting country of origin, age, gender, purpose, port of arrival, and length of stay of 79 international visitors. Some measures may be filtered within each other. International Visitor Survey No Determines country of origin and purpose. Web states regional data available but unable to find any. Regional Tourism Indicators Yes Duplicated in Regional Tourism Indicator Section Regional Tourism Estimates Yes Reports expenditure amount by County of Origin, Industry, and RTO (and also TA and Regional Council) Domestic Travel survey Yes Reports domestic travel trips. Includes trip distributions by destination RTOs Commercial Accommodation Monitor Yes Includes PDF data by RTO by month on number of guest nights, accommodation type, occupancy rates, and guest origin (domestic or international) Tourism Satellite Account No Duplicated in Tourism Satellite Account Sector Tourism Forecasts No Reports trends and expenditures from a domestic and international perspective Other Research and Reports Yes (Tourism Flows Model) Graphically reports expenditure by RTO, nights by RTO, and air and road flows. Yes (Regional Visitor Monitor) Extensive data but limited to 6 RTOs and only data available is summary of results in the form of a satisfaction index Population and Society Enrolment and Voting Statistics No Voting statistics organized by electoral region 80 from the General Election and Referendum on New Zealand's Voting System held on 26 November 2011 National Family and Household Projections No Projected families by type New Zealand General Social Survey Yes "Happiness" survey, only includes the 6 most populated RTOs Subnational Population Projections Yes Projected population, age structure, and components of change National Population Projections No By age structure and component of change National Population Estimates No By age, gender, and year National Ethnic Population Projections No By ethnicity Family Income and Employment Survey No Family structure, earnings, and number in HH employed Time Use Survey: 2009/10 No Time use by demographics Social Services Contracting Map Yes RTO apportioned amount 2010 Annual Yes Duplicated in Meshblock data 81 Areas File Recorded Crime Tables No Apprehensions and crimes listed by police districts Household Economic Survey Yes Only HH data for four RTOs explicitly reported Regional Family and Household Projections Yes Projected number of families and households by family type and household type Marriage, Civil Union, and Divorce statistics No By year for all of New Zealand Births and Deaths statistics Yes Births and Deaths by Year by Region The Social Report and Regional Indicators 2009 Yes Suicides, life expectancy, smokers, voter turnout, physical activities, road casualties by Region Regional Tourism Organizations provided the best level of detail for the purpose of the research. However, only the Domestic Travel Survey and a number of other statistical survey?s provided by the Ministry of Tourism actually included this geographic definition. In order to overcome this it was observed that RTOs were comprised of Territorial Authorities, and were able to have aggregated characteristics developed according to the TLAs they were comprised of. Through this approach Regional Councils could also be linked to the TLAs they were comprised of. 82 Figure 15: RTO/ RC Comparison 83 Table 9- Area Comparison RTO TA RC Northland Far North Northland Region Northland Kaipara Northland Region Northland Whangarei Northland Region Auckland Auckland City Auckland Region Auckland Franklin Auckland Region Auckland Manukau City Auckland Region Auckland North Shore City Auckland Region Auckland Papakura Auckland Region Auckland Rodney Auckland Region Auckland Waitakere City Auckland Region Coromandel Hauraki Waikato Region Coromandel Thames-Coromandel Waikato Region Waikato Hamilton City Waikato Region Waikato Matamata-Piako Waikato Region Waikato Otorohanga Waikato Region Waikato South Waikato Waikato Region Waikato Waikato Waikato Region Waikato Waipa Waikato Region Waikato Waitomo Waikato Region Bay of Plenty Tauranga City Bay of Plenty Region Bay of Plenty Western Bay of Plenty Bay of Plenty Region Rotorua Rotorua Bay of Plenty Region Lake Taupo Taupo Waikato Region Lake Taupo Taupo Bay of Plenty Region Lake Taupo Taupo Hawke's Bay Region Lake Taupo Taupo Manawatu-Wanganui Region Kawerau-Whakatane Kawerau Bay of Plenty Region Kawerau-Whakatane Whakatane Bay of Plenty Region Gisborne Gisborne Gisborne Region Gisborne Opotiki Bay of Plenty Region Taranaki New Plymouth Taranaki Region Taranaki South Taranaki Taranaki Region Taranaki Stratford Taranaki Region Taranaki Stratford District Manawatu-Wanganui Region Hawke's Bay Central Hawkes Bay Hawke's Bay Region Hawke's Bay Hastings Hawke's Bay Region Hawke's Bay Napier Hawke's Bay Region Hawke's Bay Wairoa Hawke's Bay Region Ruapehu Ruapehu Manawatu-Wanganui Region Manawatu Manawatu Manawatu-Wanganui Region Manawatu Palmerston North City Manawatu-Wanganui Region Manawatu Rangitikei Manawatu-Wanganui Region 84 Manawatu Tararua Manawatu-Wanganui Region Wanganui Wanganui Manawatu-Wanganui Region Wairarapa Carterton Wellington Region Wairarapa Masterton Wellington Region Wairarapa South Wairarapa Wellington Region Kapiti-Horowhenua Horowhenua Manawatu-Wanganui Region Wellington Kapiti Coast Wellington Region Wellington Lower Hutt City Wellington Region Wellington Porirua City Wellington Region Wellington Upper Hutt City Wellington Region Wellington Wellington City Wellington Region Marlborough Marlborough Marlborough Region Nelson Tasman Nelson City Nelson Region Nelson Tasman Tasman Tasman Region Canterbury Ashburton Canterbury Region Canterbury Christchurch City Canterbury Region Canterbury Hurunui Canterbury Region Canterbury Kaikoura Canterbury Region Canterbury Selwyn Canterbury Region Canterbury Waimakariri Canterbury Region Canterbury Waimate Canterbury Region Timaru Timaru Canterbury Region MacKenzie Mackenzie Canterbury Region Waitaki Waitaki Canterbury Region Waitaki Waitaki Otago Region West Coast Grey West Coast Region West Coast Westland West Coast Region West Coast Buller West Coast Region Lake Wanaka Queenstown-Lakes Otago Region Queenstown Central Otago Otago Region Central Otago Dunedin City Otago Region Dunedin Dunedin Otago Region Clutha Clutha Otago Region Fiordland Gore Southland Region Southland Invercargill City Southland Region Southland Southland Southland Region