|UFDC Home||myUFDC Home | Help|
This item has the following downloads:
1 TOURISM DEMAND: A BASE MODEL FOR FLORIDA By MAYRA VILLACIS A THESIS PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLOR IDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF SCIENCE UNIVERSITY OF FLORIDA 2007
2 2007 Mayra Villacis
3 To my Family
4 ACKNOWLEDGMENTS I would like to express my sin cere admiration and gratitude to Dr. Ronald W. Ward, chair of my supervisory committee for his invalu able assistance and academic guidance. His continuous support and encouragement helped to ma ke this work achievable. His example as a professional and human being will remain in my memory as an inspiration for excellence. I would also like to thank to the Food and Resource Economics Department for giving me this lifetime opportunity of being part of this community. Its friendly environment, combined with the knowledge and skills acquired during my time here, made this experience remarkable. Special mention to Jessica Herman, who is one of the most helpful people I have ever known and Dr. Jeffrey Burkhart for opening the doors of this community to foreign students like me. Finally, I want to thank my family for their love and support throughout my whole life, to my friends who have been a blessing in this journey, and to my God for everything.
5 TABLE OF CONTENTS page ACKNOWLEDGMENTS...............................................................................................................4 LIST OF TABLES................................................................................................................. ..........7 LIST OF FIGURES................................................................................................................ .........8 ABSTRACT....................................................................................................................... ............10 CHAPTER 1 INTRODUCTION..................................................................................................................12 Overview....................................................................................................................... ..........12 Tourism Demand: The Study..................................................................................................16 Research Problem............................................................................................................... ....17 Research Objectives............................................................................................................ ....18 Research Hypotheses............................................................................................................ ..18 Scope of the Analysis.......................................................................................................... ...18 Methodology.................................................................................................................... .......19 2 LITERATURE REVIEW.......................................................................................................21 Demand of Tourism.............................................................................................................. ..21 Floridas Profile.............................................................................................................. ........22 Forecasting Tourism Demand.................................................................................................24 Introduction to the Probit..................................................................................................... ...28 Introduction to the Multinomial Logit Model........................................................................29 3 DATA DESCRIPTION..........................................................................................................32 4 MODEL SETTING AND ESTIMATIONS...........................................................................47 Probit Model Estimates......................................................................................................... ..47 Multinomial Logit Model Estimates.......................................................................................51 5 SIMULATION ANALYSIS...................................................................................................61 Ranking the Probabilities of Traveling for Pleasure...............................................................62 Probabilities of Travelin g Across Each Variable...................................................................63 Destination Probabilities...................................................................................................... ...69 6 SUMMARY AND CONCLUSIONS.....................................................................................85
6 APPENDIX A SIMULATED PROBABILITIES ..........................................................................................88 B SIMULATED PROBABILITIES ..........................................................................................90 LIST OF REFERENCES............................................................................................................. ..92 BIOGRAPHICAL SKETCH.........................................................................................................94
7 LIST OF TABLES Table page 2-1 Domestic and foreign travel in the US: 2000-2004...........................................................21 2-2 Calendar year visitor numbers for 2006.............................................................................23 2-3 Total tourism spending (Tourism/R ecreation Taxable Sales) 1999-2006.........................23 2-4 Top origin states by percentage of total domestic visitors in 2005.................................24 2-5 Seasonality of visitation to Florida by quarter in 2005......................................................24 3-1 Variables description.........................................................................................................44 3-2 Distributions of states by regions.......................................................................................46 4-1 Probit estimates for taking a pleasure trip..........................................................................55 4-2 Probit taking a pleasure trip fo r those who took only one trip...........................................56 4-3 Supporting statistics for Logit Model taki ng a pleasure trip to any of the 6 destinations, TNUM=1, including IMR.............................................................................57 4-4 Multinomial Logit estimation (Logit taking a pl easure trip to California and Florida, TNUM=1, including IMR)................................................................................................58 4-5 Multinomial Logit estimation (Logit taking a pleasure trip to New York and Pennsylvania, TNUM=1, including IMR).........................................................................59 4-6 Multinomial Logit estimation (Logit taking a pleasure trip to Texas and Ohio, TNUM=1, including IMR)................................................................................................60
8 LIST OF FIGURES Figure page 1-1 Principal recreation activities for US residents..................................................................14 1-2 Principal tourist activities in Flor ida. (Source: Visitflorida.com)......................................15 2-1 Map of Florida................................................................................................................. ..23 2-2 Classic Probit model..........................................................................................................29 3-1 Respondents that traveled more than 50 miles..................................................................35 3-2 Distribution of distance traveled........................................................................................36 3-3 Distribution of origin region of respondents......................................................................36 3-4 Distribution of destinations among interviewed households.............................................37 3-5 Number of trips taken during the assigned period. A) U.S. popul ation. B) Florida..........38 3-6 Seasonality distribution. A) U. S. population. B) Florida...................................................39 3-7 Pre and Post 9/11 distribution............................................................................................40 3-8 Income distribution............................................................................................................40 3-9 Race distribution. A) U.S. population. B) Florida.............................................................41 3-10 Age distribution.......................................................................................................... .......42 3-11 Education distribution.................................................................................................... ....42 3-12 Purpose distribution...................................................................................................... .....43 3-13 Distributions. A) Empl oyment B) Urban/rural..................................................................43 5-1 Ranking of the effects of va riables on the probability of taking a pleasure trip................62 5-2 Proportion of variation in rela tion to the average by income............................................64 5-3 Proportion of variation in rela tion to the average by Age.................................................64 5-4 Proportion of variation in rela tion to the average by Race................................................65 5-5 Proportion of variation in relation to th e average by Education and Employment...........66 5-6 Proportion of variation in relation to the average by Location and 9-11 event.................67
9 5-7 Proportion of variation in rela tion to the average by Season.............................................67 5-8 Proportion of variation in rela tion to the average by Region.............................................68 5-9 Average probability of taking a pleasure trip by destination.............................................69 5-10 Effect of Income Level on the probability of taking a pleasure trip to any of the six top destinations............................................................................................................... ...70 5-11 Effect of Age on the probability of taking a pleasure trip to any of the six top destinations................................................................................................................... .....71 5-12 Effect of Race on the probability of taking a pleasure trip to any of the six top destinations................................................................................................................... .....72 5-13 Effect of Education and Empl oyment on the probability of taking a pleasure trip to any of the six top destinations............................................................................................73 5-14 Effect of Urban/Rural origin and 9-11 Ev ent on the probability of taking a pleasure trip to any of the six top destinations.................................................................................74 5-15 Effect of Seasonality on the Pr obability of taking a pleasure trip to any of the six top destinations................................................................................................................... .....75 5-16 Effect of Region on the probability of taki ng a pleasure trip to any of the six top destinations................................................................................................................... .....76 5-17 Ranking of the effects of vari ables on the probability of taking a pleasure trip (FL)........77 5-18. Proportion of variation in relation to the average by income (for Florida).......................79 5-19 Proportion of variation in relation to the average by Age (for Florida).............................80 5-20 Proportion of variation in relation to the average by Race (for Florida)...........................80 5-21 Proportion of variation in relation to the average by Education and Employment (for Florida) ...................................................................................................................... ........81 5-22 Proportion of variation in relation to th e average by Urban/Rural origin and 9-11 event (for Florida)............................................................................................................ ..82 5-23 Proportion of variation in relation to the average by Season (for Florida)........................83 5-24 Proportion of variation in relation to the average by Region (for Florida)........................83
10 Abstract of Thesis Presen ted to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Master of Science TOURISM DEMAND: A BASE MODEL FOR FLORIDA By Mayra Villacis August 2007 Chair: Ronald W. Ward Major: Food and Resource Economics Floridas weather, natural resources, parks a nd attractions are major reasons why tourists visit Florida. Demand for resources needed to support tourism often competes with the agricultural use of many of these same resources. Yet touris m adds to the demand for Floridas agricultural products. Clearly, t ourism, natural resources, and the agricultural sectors are interrelated. Furthermore, both generic and brand promotions of tourism are important factors expected to impact tourism demand. A first st ep in understanding tourism was to have good quantitative measures of tourism demand, driven by diverse factors nationally and for Florida. Hence, we examined one approach to gaining better analytical analyses of tourism demand. Using a National Household Travel Survey da tabase covering the months from March 2001 through May 2002, models were estimated that predicted the probability of making long distance trips and, specifically, trip s to Florida. This national da tabase included details travel activities by individuals as well as demographic information about the travelers. Each individual was assigned one month within the time span to r ecord all of his or her long distance travels, including considerable details about the destinatio n and travel days. Given the recorded travel for any one individual was limited to the assigne d period, we only knew the number of trips for that period and not the potential other trips that could have been taken in non-assigned months.
11 Implementing Probit and Multinomial Logit mode ls one could estimate the probabilities for long distance travel while explicitly accounting for househol d demographics; pre and post 9/11; seasonal demand; geographical locations of household from Florida; competing traveling opportunities; and frequency of tr aveling. Impacts of these demand drivers were ranked based on their significance and re lative impacts on the likelihood of trav eling to Florida. Since the data covered the months before and after 9/11, this was an opportunity to quantitatively show the impact on tourism demand as well as insight into the recovery. Once the models were estimated, simulations a nd sensitivity methods were used to show the relative effects as each demand driver. This had important implications for identifying the benefit of potential tourism based on demogr aphics such income, age, education, race, employment, family size, and profession. Likewi se, comparing variations across demographics should help one develop and adopt targeted marketing plans w ith the expressed purpose of enhancing tourism demand.
12 CHAPTER 1 INTRODUCTION Overview Tourism activities facilitate individuals, fam ilies, and groups escaping the daily routine of work and normal lifestyles. These activities range from intense educational experiences to absolute isolation. Activities are met th rough experiencing both natural and man-made attractions. Sometimes the tourism destination fu lfills a multitude of escape needs and heightens the desire for repeat experiences in the same se tting. For example, a summer vacation to Florida beaches can be relaxing, educational, and intensiv e. Other experiences may be just once in a lifetime, depending on the special event and cost. Obviously, the range of activities fitting into tourist options are many since t ourism offerings must satisfy an almost countless number of individual preferences and need s for diversity. Preferences should be closely tied to demographics and economics with economics poin ting to the ability to afford a variety of tourism options. Implicit in most tourism opti ons is an underlying infr astructure of supports such as housing, food services, tr ansportation, safety and secur ity, shopping, personal services, and timing. Disruptions and/or fa ilures between one or more of these can greatly affect the tourism experience and substantially impact the likelihood of repeat visits. The instant dramatic negative effects of a terrorist and calamitous act on tourism have ample anecdotal evidence (Taylor, 2006). The tragedy of Se ptember 11 (i.e., 9/11) is the pe rfect example of an event that profoundly impacted tourism because of it effect s on the services and, more importantly, the consumers concern for security. Uncertainty is a frightening concep t that can have both immediate and prolonged negative e ffects on tourism. The extent of the effects depends on the individual and how the tourism industry responds to any particular event.
13 Among the wide number of tourism options availa ble to those seeking to escape, Florida is unique among the states in offering a diverse set of options, both natural and man-made. Year round good weather, hundreds of miles of seacoas t and thousands of lakes, along with major theme parks are all within reach of most consumers. With this diversity of options, one would expect seasonal variation in demand depending on the specific tourism(s) goal. Understanding this demand and how it is influenced by natural and non-natural events is essential to providing continued services and for identifying needed innovations. A good example of the need of innovation because of changes in the Florida at traction industry can be seen with the long established Florida Marineland Park that existed for decades. Attendance declined precipitately with the competition from SeaWorld in Orlando Through innovation and infusion of resources, the offerings from Marineland have been completely revamped. As seen with this example, the need to be dynamic and responsive is essentia l for success and measuring tourism demand is one essential component to that success. Methods for explaining and for ecasting tourism demand are important and that was the focus of this res earch effort with the major emphasis being on Florida. The majority of studies for forecasting t ourist demand are concerned with econometric modeling and the most appropriate explanatory vari ables. It is important to emphasize that no single forecasting method performs consistently across different situations (Witt and Witt, 1995) and hence the same variables may have di fferent impacts depending on location, time and tourism offerings. The allocation of scarce resources to satisfy consumers demand for tourism and its impact on microeconomic levels are a concern for the economics of tourism. The demand and supply side characteristics of tourism provide an appr oach to understanding the impact of tourism, from
14 measuring the size of tourism to quantifying the impact of change in any of the factors that determine tourism behavior (Sinclair et al, 1989). Among the components for the economics of tourism we find: direct tourism related employ ment and compensations, direct and indirect benefits of travel (spending and sales) and tax revenues to name some of them. Tourism generates employment, directly in the sector where the expenditure occurs and also between industries. Tourism is one the largest employe rs in the United States, employing directly 7.9 million people and indirectly 9.4 million people for a total of 18 million people, generating travel related payrolls of $174 billion. Also tourism is one the Americas larges t retail sales industries generating about $537 billion in tota l expenditures; $98.7 billion in tax revenue for federal, state and local governments; $1.5 billion a day, and a bout $1 million a minute in travel and tourism (Tourism Works for America 2002 Report, Trav el Industry Association of America). The most popular activities of U.S. resident s when taking a trip to any destination are shown in Figure 1-1 (Tourism Works for Ameri ca 2002 Report, Travel In dustry Association of America). 0% 5% 10% 15% 20% 25% 30% 35% 40%Shoppi ng His to r i cal P l ac es/M us eums Beache s Cultural Ev en ts National/State Parks Them e/ Am us ement P arks Nightli f e/Dancing Gam bl ing S po r t s Events Figure 1-1. Principal recreati on activities for US residents
15 The role of national characteristics in the be havior of tourists ha s been an area of investigation. Respondents who travel a few hours dr iving distance to their destination are more likely to travel on a smaller budget and spend le ss time on planning than overseas travelers. Worldwide, domestic demand for tourism in term s of volume and value is more important than international tourism (Bigano et al., 2007). Domestic tour ism refers to trips made by residents of a country within the national terr itory of that country. Even t hough measures of domestic tourism are often presented for a country as a whole, it is more useful if we analyze it for specific destinations. Principal attractions bringing tourists to Florida include sunny and sandy beaches, theme parks, world class golf courses, natural parks a nd reserves, or sports, attracting about 80 million of visitors in 2004 giving touris m an important role in Florida s economy. It is important to note that different travel activities and seasons have different demands (SooCheong, 2005). Analyzing tourism demand is a useful tool for increasi ng our understanding of the importance of the different tourism demand drivers. Typically, tour ism statistics are measures of arrivals, trips, tourist nights and expenditures which often appear in total or split into categories such as business or leisure travel with estima tes usually based on sample surveys. 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% Figure 1-2. Principal tourist activities in Florida. (Source: Visitflorida.com)
16 Certainly, the choice of a travel destination is a comprehens ive process, which includes preparation for the trip and the actual behavior of the traveler s. In order to understand this, knowledge of the characteristics of potential visitors, their need s and interrelations with other competitive destinations is required. The infl uence of consumer characteristics could be examined through the effect of different vari ables like geographical origin, income, age, motivation, education, etc. Tourism Demand: The Study Tourism is a complex process that involves not only visitors and their movements, but also the destination (Latham and Edwards, 1989). Do mestic demand for tourism may be seen as competing or complementing with that by interna tional tourism. Given the difficulty of gathering information about these two demands; we analyzed only the domestic demand. U.S. domestic tourism refers to trips made by residents within the national territory of the U.S. Econometric methods have been extensively us ed to study the tourism demand in various countries and regions to determin e factors affecting demand and to forecast tourist arrivals in order to provide accurate measures for policy makers to develop appropriate policies for the sector. The purpose of econometrics models is no t purely forecasting. In addition, they attempt to explain economic or business phenomena and in crease our understandin g of the variables influencing the travel decisions (Makridakis, 1983). Past studies were useful in identifying the main economic determinants of tourism demand: income, relative prices, transport cost and specific events th at may affect tourism demand such as 9/11, sport events, etc.; adding a dummy or trend variable to represent such events, gives satisfactory result s in conformity with economic theory. The dependent variable has been widely classified as number of arriva ls/departures, length of stay, nights spent at a given tourist accommodation, share of tourist arri vals, proportion of tourist to a particular
17 destination, visit rate, etc. (L im, 1997). Tourism demand can be looked at for regions, countries, states, and cities; for different types of visito rs (international or domes tic); and for specific components such as trans portation or attractions. Research Problem For our study, having information about trips taken by a representative sample of the US population during 2001 and 2002, we were interested in determining the probability of traveling for pleasure and which were the dr iving forces for such behavior (explanatory variables). Data concerning socioeconomic characteristics such as age, sex, social group, location, purpose of traveling are clearly used in promoting dome stic tourism, since those demographics and geographic characteristics allow th e targeting of the population that is most likely to travel on long distance trips, particular ly to the state of Florida. National Household Travel Survey data were used to quantify travel behavior, relate travel behavior to the demographics of the traveler and to study the re lationship between demographics and travel preferences. Our study was intended to gain a better understandi ng of travel behavior and develop an econometric model, helping explai n the driving forces for demand of tourism in the state of Florida. The database included details about travel activities for each individual. Within this information, we had: Number of trips taken, orig in/destination, days travel ed, purpose of the trip, number of travelers, distance traveled, dates (sea son). In addition we also had information about the demographics and geographic for each indivi dual, as well as information about events happening in the time frame when they recorded their answers.
18 Research Objectives The overall objective of this study was to de velop an understanding of the factors that influence the demand for touris m. With information from th e 2001 National Household Travel Survey, effects of the different factors were esti mated as well as their influence shown with the likelihood of taking a long distance trip to any destination. Anothe r objective was to develop an econometric model that helps explain the drivin g forces for tourism demand in the U.S and particularly for the state of Florid a. We should be able at the end of this project to answer the key research questions: What was the likelihood of taking a long distance trip for pleasure? Where did they go (destinations)? What were the char acteristics of the house holds who traveled to Florida? Research Hypotheses The hypothesis tested in this research were: There is a substantial different behavior pattern for different age groups when it comes to taking long distance trips. The importance of the factors affecting the tourism demand for the average household in the US differs from that when analyzi ng only the demand of tourism for Florida. The effect of 9/11 on the demand for touris m varies among different regions in the country. Scope of the Analysis The 2001 NHTS (National Household Travel Survey) was conducted over a year period, from March 2001 through May 2002. The survey sampled approximately 45,165 households across the US. Since the survey was conducted for the United States population (50 states and the District of Columbia, but not outlying areas), this analys is was limited to the domestic tourism demand.
19 The 2001 NHTS collected data on trips to destin ations 50 miles or more away from home (long-distance travel) duri ng a four week assigned travel pe riod. Travel day da tes were assigned to all seven days of the week, including holidays. The dataset did not include information about co st of travel. It did not include information on how travel of the sampled household changed over time (sin ce this was a cross sectional survey, which means that different households were selected from the sample each time it was conducted). Therefore, we only ha d information about the trips th at the household took over the assigned period. Data about trips taken during othe r periods of time were not recorded in this survey. This study was intended to be a base model fo r the analysis of demand of tourism. When analyzing the data, information a bout a detailed description of th e trips was not used. Data about length of stay, consumption of goods and services expenses and travel related costs were not considered for our analysis. Even though the data set contained information about trips taken to different destinations within the US, our focus wa s directed to trips taken to Florida. However we also considered an analysis about the likelihood of taking a trip to the top 5 competing destinations. Methodology Step 1: Participation. In this first step we truncated the database to those people who took at least one long distance trip si nce our research interest focu sed on households who traveled. Step 2: Estimations Binary model (Probit). Among households who traveled, we estimated the likelihood of traveling for pleasure, using a Probit model. Step 3: Destination choices. In this stage we used Multinomial Logit models to compare the different destination choices, and how th e model estimates varied among them.
20 Since the response variable wa s if the person traveled or not to Florida, methods for estimating the likelihood of taking a long distance trip to Florida were implemented, using Logit (or Probit) models. Once the models were estimate d, they were used for sensitivity analysis and derive insights about the results.
21 CHAPTER 2 LITERATURE REVIEW Demand of Tourism Tourism demand is a measure of visitors use of a good or service. Such use includes the economics concept of consumption, as well as the pr esence of a visitor at a destinations, port of entry or other tourism facility, and on a transpor t vehicle, regardless of whether any exchange takes place. Therefore, visitor arrivals in a coun try or local area constitu te tourism demand since visitors benefit from the services at destination. For many regions tourism is one of the most significant economic activities in terms of economic growth and employment. World tourism demand is still growing and new or current destinations may be developed or extended in order to satisfy such growth. In this sense, tourism may be been seen as an opportunity for economic growth (Eugenio-Martin, 2005). Table 2-1. Domestic and fore ign travel in the US: 2000-2004 Year Domestic Trips (millions) Foreign Visitors (millions) 2000 1,100.8 51.2 2001 1,123.1 46.9 2002 1,127.0 43.5 2003 1,140.0 41.2 2004 1,163.9 46.1 Source: Travel Industry Asso ciation of America (2005) Tourism is a very complex and sensitive i ndustry. When analyzi ng this industry, policy makers and people involved in this industry should consider that: Tourism product is perishable Customer satisfaction depends on complementary services Leisure tourism demand is extremely sensit ive to natural and human-made disasters Tourism supply requires large, long lead-time investments in service support, equipment and infrastructure. Consumers travel for a multitude of reasons (McIntosh et al. 1995). Business people travel to make sales, attend m eetings, attend conferences and c onventions, and inspect locations.
22 In addition, the same person can plan a family vaca tion trip with different motives and resources. Other people can take a trip for recreation purposes, lik e outdoor activities, sport competitions, etc. Each one of these trip purposes may show different patterns and be affected by different factors. With these considera tions, it is valid to say that there is no consensus on a sound theoretical foundation for tourism demand, consequen tly we need to consid er conditions in a number of widely dispersed areas as they a ffect demand for a destination, place or service (Frechtling, 2001). Tourism has a range of economic impacts. Tour ists add sales, profits, jobs, tax revenues, and income to an area. The most direct effect s occur within the primar y tourism sectors -lodging, restaurants, transportation, amusements, and re tail. Through secondary e ffects, tourism affects most sectors of the economy. When conducting research on tourism demand, it is important to describe unique dimensions to a region. A short profile of tourism and economic activity in the region provides useful background fo r an economic study (Stynes, 1999). Floridas Profile The state of Florida has an area of 1 70,304 square kilometers and a population of 18,089,888 inhabitants (US Census Bureau, 2006), ra nking 4th in the U.S. behind California, Texas and New York. It is also known as the Sunshi ne State. This subtropi cal area is one of the fastest growing states. It is es timated that approximately 1,000 people move daily to Florida. In central Florida, Sea World, Universal Studios, an d Walt Disney World are big attractions that generate substantial amounts of money and bus iness. Technology industries supplement tourism as does agriculture, mainly citrus, tomatoes, landscap ing plants, and sugarcane. Florida is known around the world for its temper ate weather. The state's mild winters have made it a haven for retirees. Summers can be long and hot with showers providing much appreciated relief during the rainy season. Coasta l areas also experience gentle breezes during
23 the summer. About 40% of all U.S. exports to Latin and South America pass through Florida, being an important location for international tr ade. Tourism with 76.8 million visitors in 2004 (a record number), Florida is the t op travel destination in the world. Figure 2-1. Map of Florida Some statistics published by VI SIT FLORIDA Research depart ment that studies global consumer trends and travel patter ns to learn more about Florida' s visitors and their preferences are shown next in the following tables: Table 2-2. Calendar year visitor numbers for 2006 Visitors Percent of Total Most Recent Value(million) Change% Total Visitors 100.00 January-December 84.6 +1.2 Air Visitors 50.50 January-December 42.7 +0.2 Non-Air Visitors 49.50 January-December 41.9 +2.3 Source: Visit Florida Research Table 2-3. Total tourism spending (Tour ism/Recreation Taxable Sales) 1999-2006 Year Amount (billion) 1999 $44.60 2000 $48.50 2001 $48.60 2002 $49.50 2003 $51.50 2004 $57.10 2005 $62.00 2006 $65.00 Source: Visit Florida Research
24 Table 2-4. Top origin states by percen tage of total domestic visitors in 2005 State Percentage% New York 11.50 Georgia 10.10 New Jersey 5.6 Illinois 5.1 North Carolina 5.0 California 5.0 Pennsylvania 4.7 Ohio 4.2 Alabama 4.2 Texas 3.9 Source: Visit Florida Research Table 2-5. Seasonality of visita tion to Florida by quarter in 2005 Quarter Percentage% Jan-Mar 27.90 Apr-Jun 27.30 Jul-Sep 25.40 Oct-Dec 19.40 Source: Visit Florida Research Florida's domestic visitors said their prim ary reason for coming to the Sunshine State was for leisure (82.6%). The major type of lodging used by domestic visitors was hotel/motel/B&B (47.5%). The average length of st ay for domestic visitors to Florida was 5.2 nights. Domestic air visitors surveyed stayed an average of 5.5 nights, while auto visitors stayed an average of 5.1 nights. The top activities dom estic visitors enjoyed while in Florida were visiting the shopping, touring/sightseeing, beaches, and going to theme and/or amusement parks. Forecasting Tourism Demand An important objective of analyzing tourism de mand is to predict the most probable level of demand that is likely to occur in the pres ence of known circumstances or, when alternative policies are proposed, to show what different le vels of demand could be achieved (Archer, 1994). A tourism model is a simplified repres entation of reality, comprising a set of
25 relationships, historical information on these rela tionships, and procedures to for measuring the impact of major factors expected to in fluence tourism demand (Frechtling, 2001). Most of the past economic research about the tourism demand has been based on two approaches, one involving single equations that are specified on a prior basis, and the other comprising a system of equations based on the th eory of consumer demand. Both approaches allow us to formulate and test hypotheses concer ning the effect of particular variables on the tourism demand (Sinclair et al, 1989). On the other hand, demand of tourism and forecasting continues to be a trendy theme in the touris m literature. Some reviews by Witt and Witt (1995) show that demand forecasting, in most of the cases, is focused on economic factors. Conventional causal tourism demand forecasti ng models often provide policymakers and managers of different sectors in the touris m industry with estimates for the purpose of formulating appropriate strategies to increase their competitiveness. One of the first steps in analyzing the tour ism demand is understanding tourism behavior. The Purchase-Consumption Systems (Woodside and King, 2001; Woodside and Dubelaar, 2002) is useful for constructing a conceptual framewor k of tourism behavior. A PCS is the series of steps that a consumer undertakes to consume a product. To summarize some of these factors affecting a consumer decision when taki ng a trip, we can mention the following: Demographics and lifestyles of visitors infl uence how they frame their leisure choices. For example a household with children will cons ider leisure trips th at include activities for children, while those things are no t found among the consid erations taken by a household without children when planning a leisure trip. Unexpected events occur that influe nce the decision for leisure choices. Information, features and benefits included in selecting the destina tion may influence the choice or rejection of a given alternative. Key activities driver consolidate the d ecision to visit a se lected destination. Events learned by the visitors and activities done affect the attitude toward a destination.
26 In their work March and Woodside (2005) also de scribe seven trip components that are to be considered when collecting data on tourism behavior: Destinations; Route/mode to and in the principal destination; Accommodation during the stay in the destination Activities engaged in the destination Regions visited at the destination Attractions visited; and Gifts and other purchases made in the destination area The first decision any person has to make con cerns the choice between traveling or not within a period of time. The researcher, depending on the purpos e of the analysis, must set a time interval. Usually, this may range from a period of a year to a period that contains extended periods. Available literature suggests that vari ables such as age, education, income, labor conditions, characteristics of the place of residence and size and composition of the household or family may be significant when deciding to travel or not. Among the forecasting models using multivariate techniques, multiple regression analysis is the most used and relevant technique for estimating tourism demand. Although multiple regression models and approaches may assume different forms (e .g., Logit, Probit models), the basic multiple regression model is represented as Y = 0 + x + 2x nxn + (2-1) 0 = intercept x nxn = linear effect of vari ous independent variables = error term In a paper reviewing forecasting techniques for international tourism demand, Witt and Witt (1995) provide a good perspective on the different techniques used for forecasting international tourism demand, including a comparis on of the accuracy of the techniques. With reference to multiple regression, they note the need for improved model specification while recognizing that the more recent models have improved in this respect.
27 In order to model participati on decisions, we consider it as a binary choice, denoted by Y such that, Y if household or individual decides to travel and Y not to travel. We assume Pr Y is linked to a set of exoge nous variables as suggested ea rlier, which might be those already shown above. Traditional linear probability models are not recommended to be used to estimate the probability function because it would present nonnormal errors, heteroskedasticity and logical inconsistency, since prediction of pr obabilities may lie out of range ( 0,1). It is recognized that the suggested model for binary choice estimations is the latent variable model. This model considers the existence of a latent variable Y composed by two parts; one observed by the researcher, which includes all the socioeconomic variables, and another part that it is unobserved by the researcher and corresponds to heterogeneity reasons among tourists. Thus the model can be represented as: Y x 2-2) where denotes unobserved part or error term. It is necessary to choose a distribution and a value for the variance of The most common approaches assumes that is independently and identically distributed, either following a normal distribution with zero mean and variance of one, or following a logistic distribution. If we assume that follows the former distribution we ar e employing the well-known Probit model, and if we assume the latter distribution we ar e employing the also well -known logit model. Any of these distributions can be employed for th e participation decision and both present similar results. Finally, maximum likelihood estimation is applied to the model in order to estimate parameters of interest.
28 Introduction to the Probit For our research, we considered as a first step the application of a Probit model to determine the likelihood of taking a long distance pleasure trip. Prob it regression is an alternative log-linear approach to handling categorical depe ndent variables. Its assumptions are consistent with having a categorical dependent variable assumed to be a proxy for a continuous normal distribution. Like logit or logist ic regression, the researcher fo cuses on a transformation of the probability that Y, the dependent, equals 1. Where the logit transformation is the natural log of the odds ratio, the function used in Probit is the inverse of th e standard normal cumulative distribution function. Probit regression assumes th e categorical dependent reflects an underlying quantitative variable and it uses th e cumulative normal distribution. In practical terms, Probit models usually co me to the same conclusions as logistic regression and have the drawback th at Probit coefficients are more difficult to interpret (there is no equivalent to logistic regression's odds ratios as effect sizes in Pr obit), though the choice is largely one of personal preference. Both the cu mulative standard normal curve used by Probit as a transform and the logistic (l og odds) curve used in logistic regression display an S-shaped curve. Though the Probit curve is slightly steeper differences are small. Because of its basis on the standard normal curve, Probit is not recomme nded when there are many cases in one tail or the other of a distributio n. Logit and Probit analysis generally arrive at the same conclusions for the same data, but the logit and Probit coeffici ents differ in magnitude. The Probit model is defined as Pr (Y=1|X=x) = ( x) (2-3) Where is the standard cumulative nor mal probability distribution and x is called the Probit score or index. Since the Probit model is based on a normal distribution, interpreting
29 Probit coefficients requires thinking in the Z (normal quantile) metric. The interpretation of a Probit coefficient, is that a one-unit increase in the predictor leads to increasing the Probit score by standard deviations. Variable Figure 2-2. Classic Probit model Introduction to the Multinomial Logit Model Another approach used for our research wa s to compare the proba bility of taking long distance trips to Florida or to th e other top five destinations in the US. For this purpose, we applied a multinomial logit model. When categor ies are unordered, multinomial logit regression is one often-used strategy. It is widely us ed to predict consumer preferences. The common situation to apply this mode l is: One response variable Y with J levels and; one or more explanatory or predictor variable s (the predictor variables may be quantitative, qualitative or both).
30 Suppose a Y (response variable) has J categories, one value of the Y is designated as the reference category. The probability success in other categories is compared to the probability of success in the reference category. For a Y with J ca tegories, this requires the calculation of J-1 equations, one for each category relative to the re ference category (Agresti, 2002). Hence, if the first category is the reference, then, for j = 2, J. The predicted probabilities are given by (2-4) (2-5) When applying this model, the observations need to be mutually independent. If a person took more than one trip within a definable peri od, then the combinati on of destinations are numerous. Hence, in order to measure the tr adeoff among alternative de stinations, we limited the logit model to only those individual that took just one trip in the reporting period. If a model such as the logit is based on data that was first selected from another broader database, selective bias is a possibility unless one accounts for t hose factors contributing to the initial selection. For example, in the subsequent models th e models for traveling to six destinations were estimated after selecting only those traveling for pleasur e and only one trip. Clearly, there are factors influenc ing the decision to travel for pl easure. Using probit models one can first estimate the likelihood of traveling for pleasure and then in the second stage estimated the likelihooh of different destination among t hose that traveled. Ignoring those factors influencing the likelihood of tr aveling for pleasure must be considered in the second stage estimation. A useful approach is to first es timate the probit model and then identify the probabilities for traveling for pleasure using a well estimated statistica l measure known as the Inverse Mills Ratio (IMR). Intitutively, IMR s imply provides a measure of all of those factors J j j i j i ij i ix x P x j y2exp 1 exp ) | Pr( J j j i i i ix P x y2 1exp 1 1 ) | 1 Pr(
31 leading to the likelihood for trav eling for pleasure. The inverse Mills' ratio (sometimes also called 'selection hazard') is used in regression analysis to take account of a possible selection bias. If a dependent variable is censored, i.e. not for all observations a positive outcome is observed, it causes a concentration of observations at zero values. In a first step, a regression for observing a positive outcome of the dependent variab le is modeled with a probit or logit model. The estimated parameters are used to calculate th e inverse Mills' ratio, wh ich is then included as an additional explanatory variable in the second level estimation. If the coefficient for the IMR is not statistically significant then there is no se lective bias and one can simply work with those households who just traveled for pleasure. Wher eas, if the IMR is statistically different from zero, then selective bias occurs if the IMR is not included in the logic model (second stage model).
32 CHAPTER 3 DATA DESCRIPTION The database used for this research was the 2001 National Household Travel Survey (NHTS). It provides information to assist tran sportation planners and policy makers who need comprehensive data on travel and transportation pa tterns in the United States. In this research, the NHTS data were used primarily for gain ing a better understanding of travel behavior. The 2001 NHTS is a based on a cross sectional database that was conducted over a period from March 2001 through May 2002. Travel days fo r daily-travel trip reporting were assigned for all seven days of the week, including all ho lidays. The first travel day assigned was March 29, 2001. The last travel day assigned was Ma y 4, 2002. The assigned travel period for longdistance trip reporting was the four -week period ending with the travel day. So, the last travel period assigned was April 7 through May 4, 2002. Th e survey was conducted over at least a 12month period so that seasonal variations in trav el are represented. A fact or affecting seasonality variation the climate which is an important consid eration for tourists choi ce of destination (Lise and Tol, 2002) The 2001 NHTS took 14 months, rath er than 12 to complete. This was because interviewers were trained in wa ves and it took a few months to tr ain all the interviewers needed for the study. The intent was to repr esent travel across an entire year. The target population for the 2001 National H ousehold Travel Survey (NHTS) was all non-institutional persons liv ing in households, excluding group qua rters, in the 50 states and the District of Columbia, during the data co llection period, March 19, 2001 through May 9, 2002. To obtain information from this target population, the NHTS relied on a sample design that used a list-assisted random digit dialing (RDD) methodology. Understanding the NHTS data collection was esse ntial to properly using and interpreting the data. The NHTS was conducted as a telephon e survey, using Computer-Assisted Telephone
33 Interviewing (CATI) technology. The sample desi gn was a list-assisted random digit dialing (RDD) telephone number sample. The list-assist ed telephone frame was constructed using a commercially compiled database. The data base contained approximately 65,000,000 listed residential telephone numbers nationwide, and was updated continuously as new White Page Directories were published. The lis ted telephone numbers were coll apsed to the 100-bank level, providing a count of listed households for each bank. (A 100-bank is a set of 100 telephone numbers with the same first eight digits, that is the same area code, exchange, and the next two digits.) For the national sample, all telephone numb ers, including unlisted numbers, in the frame of 100-banks had an equal probability of sel ection. Each quarter, a new sampling frame was constructed and a sample was selected for us e until a new sample was drawn. Sampling frames were constructed as of December 2000, Ma rch 2001, June 2001, September 2001 and December 2001. Each quarterly sample was pre-screened for business numbers and non-working numbers. Two methods were used to reduc e the cost associated with identifying nonr esidential telephone numbers. One was to match all the sampled tele phone numbers against a file of Yellow Pages Directory listings of business numbers. This proc edure was done automatically after selecting the sample of telephone numbers. The second method was an automated procedure developed by the commercial vendor to detect non-working numbers. The NHTS sample fell short of the target population for two groups of households: (1) households without telephones and (2) households w ith telephone numbers in 100-banks that had no listed residential numbers. Because data were collected by telephone interview, households without telephones were not included by design. Th is design decision excluded about three to four percent of the households in the U.S. th at do not have telephones, and who tend to have
34 lower incomes and more children than households with telephones. As a result, the 2001 NHTS could potentially underestimate certa in types of travel behavior such as use of public transport. Once a sample telephone number was selecte d, an advance letter was mailed to the household if a mailing address for that telephone number was available from vendors that specialize in providing a ddresses for both listed and unlisted telephone numbers. The letter was signed by the Secretary of Transportation, No rman Mineta. The pre-household interview transmittal package included the letter, a five-d ollar cash incentive, and a brochure introducing the survey. About a week afte r the advance letter mailing, an interviewer made the first telephone call to the household a nd attempted to speak with an adult household member. This household member was administered the Household Interview. The first portion of the interview included screening questions to determine if th e telephone number was residential. Eligible residential households were administered the complete household questionnaire. Each household that completed a household interview was sent a diary package. The package was sent via Priority Mail soon after the hous ehold interview was completed. The mailing was timed to reach the household a few days prior to its assigned travel day. Each diary package contained: A letter from the U.S. DOT, A brochure describing the survey, An envelope with a diary and a two-dolla r cash incentive for e ach household member, A reminder card showing th e assigned travel day, A map demarcating places over 50 miles from the household, and An odometer mileage form listing the household's vehicles. The next contact with the household was on th e day before the household's travel day. An NHTS interviewer called to find out if the househol d had received the diary package and had any questions about the survey. Th e person answering th e telephone was asked to remind household members to complete their travel diaries on the following day.
35 From the interviewed households, we were in terested in those who took at least one long distance trip) during the assigne d travel period. From this popul ation we analyzed what the likelihood was of taking long distance trip s to Florida, general and by purpose. The number of responses that we analyzed in this survey is 45,165. We only took the 45,000 respondents who traveled less than 30 trip s. From these 45,000 respondents, 98.1 percent (44,156) took at least one long distance trip and 1.9 percent (844) did not travel more than 50 miles during the four weeks assigned period. These observations were used to analyze and derive insights about tourism demand. More than 50 Miles 98.1% Less than 50 miles 1.9% Figure 3-1. Respondents that tr aveled more than 50 miles From the population who took at least one l ong distance trip, the di stribution of distance traveled shows that 50 percent of the samp le (21,508 respondents) tr aveled 100 miles and 85 percent traveled 300 miles or less.
36 0 100 200 300 400 500 600 700 800 900 1000 1100 1200 1300 1400 1500 Milles 0510152025 # Respondents (1,000) Figure 3-2. Distribution of distance traveled The distribution of the origin region of respondents is shown in Figure 3-3. 1 = New England 5.8% 2 = Middle Atlantic 12.7% 3 = East North Central 17.3% 4 = West North Central 10.0% 5 = South Atlantic 17.2% 6 = East South Central 5.7% 7 = West South Central 9.7% 8 = Mountain 7.1% 9 = Pacific 14.5% Figure 3-3. Distribution of origin region of respondents
37 The distribution of destinati ons of the trips among interv iewed households is shown in Figure 3-4. It gives an idea about the principal destinations th at households chose when they decided to take a long distance trip. The 6 main de stinations chosen are: California, Florida, New York, Pennsylvania, Ohio and Texas. From th ese, our research interest focused on the households whose destination was Florid a, when making long distance trips. 7 = R e f u s e d 8 = D o n t K n o w A K = A l a s k a A L = A l a b a m a A R = A r k a n s a s A Z = A r i z o n a C A = C a l i f o r n i a C O = C o l o r ad o C T = C o n n e c t i c u t D C = D i s t r i c t o f C o l u m b i a D E = D e l a w a r e F L = F l o r i d a G A = G e o r g i a H I = H a w a i i I A = I o w a ID = I d a h o I L = I l l i n o i s I N = I n d i a n a K S = K a n s a s K Y = K e n t u c k y L A = L o u i s i a n a M A = M a s s a c h u s e t t s M D = M a r y l an d M E = M a i n e M I = M i c h i g a n M N = M i n n e s o t a M O = M i s s o u r i M S = M i s s i s s i p p i M T = M o n t a n a N C = N o r t h C a r o l i n a N D = N o r t h D a k o t a N E = N e b r a s k a N H = N e w H a m p s h i r e N J = N e w J e r s e y N M = N e w M e x i c o N V = N e v a d a N Y = N e w Y o r k OH = O h i o O K = O k l a h o m a O R = O r e g o n P A = P e n n s y l v a n i a R I = R h o d e I s l a n d S C = S o u t h C a r o l i n a S D = S o u t h D a k o ta T N = T e n n e s s e e T X = T e x a s U T = U t a h V A = V i r g i n i a V T = V e r m o n t W A = W a s h i n g t o n W I = W i s c o n s i n W V = W e s t V i rg i n i a W Y = W y o m i n g 0.00 1.00 2.00 3.00 4.00 5.00 Respondents (1,000) Figure 3-4. Distribution of destin ations among interviewed households From the sampled population, 32.6 % took only one long distance trip during the assigned period, and 80% took five or less trips. From the population whose destination was Florida, 59.8% took only one long distance trip and 90% took five or less trips.
38 1 32.6% 2 21.2% 3 11.6% 4 8.0% 5 5.0% 6 2.7% 7 1.9% 8 1.7% 9 1.1% 10 0.8% >10 13.4% 1 59.8% 2 17.4% 3 7.3% 4 4.4% 5 2.8% 6 1.6% 7 1.2% 8 0.9% 9 0.6% 10 0.6% >10 3.4% Figure 3-5. Number of trips ta ken during the assigned period. A) U.S. population. B) Florida. To account for the seasonality effect, a variable that specified the month when the trip was taken was included in the model. A distribution of the months is shown in Figure 3-6. February and March were the months when the majority of trips were taken, when traveling to any destination and for Florida. A B
39 y 0 1 = J a n u a r y 0 2 = F e b r u a r y 0 3 = M a r c h 0 4 = A p r i l 0 5 = M a y 0 6 = J u n e 0 7 = J u l y 0 8 = A u g u s t 0 9 = S e p t e m b e r 1 0 = O c t o b e r 1 1 = N o v e m b e r 1 2 = D e c e m b e r0.00 1.00 2.00 3.00 4.00 5.00 6.00 7.00 Respondents (1,000) 2001 2002 A 0 1 = J a n u a r y 0 2 = F e b r u a r y 0 3 = M a r c h 0 4 = A p r i l 0 5 = M a y 0 6 = J u n e 0 7 = J u l y 0 8 = A u g u s t 0 9 = S e p t e m b e r 1 0 = O c t o b e r 1 1 = N o v e m b e r 1 2 = D e c e m b e r0.00 0.10 0.20 0.30 0.40 Respondents (1,000) 2001 2002 B Figure 3-6. Seasonality distributi on. A) U.S. populat ion. B) Florida. A variable to account for the effect of an unexpected event was also included in the analysis. The graph shows that most of the trips were taken after th e 9/11 events, but we have to take into account seasonality influences and the number of months when the survey was recorded.
40 After 9/11 61.7% Before 9/11 38.3% After 9/11 (FL) 65.3% Before 9/11 (FL) 34.7%U.S.FL Figure 3-7. Pre and Post 9/11 distribution The income, race, age, education, purpose, employment and urban distributions for the sample population and for the respondents who tr aveled to Florida are shown in the following figures. $24,999 or less 16.0% $25,000 to $49,999 28.4% $50,000 to $74,999 22.3% $75,000 to $99,999 14.8% $100,000 or more 18.4% $24,999 or les 17.9% $25,000 to $49,999 24.6% $50,000 to $74,999 23.3% $75,000 to $99,999 12.9% $100,000 or more 21.4%U.S. FL Figure 3-8. Income distribution
41 W h i t e ( = 1 ) B l a c k ( = 2 ) H i s p a n i c ( = 3 ) W h i t e / H i s p a n i c ( = 4 ) A l l o t h e r s ( = 5 )0 10 20 30 40 Respondents (1,000) Race US W h i t e ( = 1 ) B l a c k ( = 2 ) H i s p a n i c ( = 3 ) W h i t e / H i s p a n i c ( = 4 ) A l l o t h e r s ( = 5 )0.00 0.50 1.00 1.50 2.00 Race FL Figure 3-9. Race distribution. A) U.S. population. B) Florida. For the respondents who took l ong distance trips, the age gr oup between 45 and 65 years old was the lasgest proportion, closely followed by the (25-45 years old) age group, for US population and for those traveling to Florida. When accounting for education, the majority of the sampled population (63% for US population and 65% for those whose destination was FL) had at least some college education. A B
42 Under 25 year 21.3% 25 years to 44 years 31.7% 45 year to 65 years 35.8% 65 years and over 11.2% Under 25 year 22.8% 25 years to 44 years 27.4% 45 year to 65 years 34.4% 65 year s and ov e 15.4% U.S.FL Figure 3-10. Age distribution College Education 63.3% No College Education 36.7% College Education 65.3% No College Education 34.7% U.S.FL Figure 3-11. Education distribution. The two principal purposes when making the decision of taking a long distance trip are business and pleasure. For the sampled population, 55 percent of them reported taking pleasure trips. From the population who traveled to Florida, 60 percent of them repo rted pleasure as the purpose for the trip. Also, the majority of the population taking long distance trips was reported as employed and urban.
43 1=Busine 28.2 2=Pleasu 55.1 3=Combin 13.2 4=Othe 3.5 1=Busine 24.3 2=Pleasu 60.9 3=Combin 11.8 4=Othe 3.0 Figure 3-12. Purpose distribution. U.S. Fl Employed 65.8% No Employed 34.2% Employed 61.8% No Employed 38.2% U.S. Fl URBAN 69.5% RURAL 30.5% URBAN 80.5% RURAL 19.5% Figure 3-13. Distributions. A) Employment B) Urban/rural A B
44 A list and description of the va riables included for our model is shown in Table 3-1. A list of the states included in each region is shown in Table 3-2. Table 3-1. Variables description Description Variable DCENSUS: Household origin region DCENSUS1 1 = New England DCENSUS2 2 = Middle Atlantic DCENSUS3 3 = East North Central DCENSUS4 4 = West North Central DCENSUS5 5 = South Atlantic DCENSUS6 6 = East South Central DCENSUS7 7 = West South Central DCENSUS8 8 = Mountain DCENSUS9 9 = Pacific DMTH: month when the trips was taken DMTH1 01 = January DMTH2 02 = February DMTH3 03 = March DMTH4 04 = April DMTH5 05 = May DMTH6 06 = June DMTH7 07 = July DMTH8 08 = August DMTH9 09 = September DMTH10 10 = October DMTH11 11 = November DMTH12 12 = December DPOST: Trip taken before or after 9/11 DPOST1 (1) Before 9/11 DPOST1 (-1) After 9/11
45 Table 3-1. Continued Description Variable DINC: Household income in $ DINC1 <= $24,999 DINC2 (>$24,999 &<=$49,999) DINC3 (>$49,999 &<=$74,999) DINC4 (>$79,999 &<= $99,999) DINC5 $100,000 or more DRACE: Household race DRACE1 1=White, DRACE2 2=Black DRACE3 3=Hispanic DRACE4 4=White/Hispanic DRACE5 5=All others DAGE: Household age DAGE1 <25 DAGE2 (>=25 &<45) DAGE3 (>=45 & <65) DAGE4 >=65 DEDUC: Respondent education level DEDUC (1) College Education DEDUC (-1) No College Education DPURP: Purpose of the tr ip DPURP1 1=Business DPURP2 2=Pleasure DPURP3 3=Combined DPURP3 4=Other EMPL: Employment status of the respondent EMPL(1) Worker EMPL(-1) No Worker URBAN: Urban or rural origin URBAN(1) Urban URBAN(-1) Rural TNUM: # of trips during the assign ed period TNUM # of Trips
46 Table 3-2. Distributions of states by regions States for each Region 1. New England 2. Mid-Atlantic 3. East North Central 4.West North Central Maine New York Ohio Minnesota New Hampshire New Jersey Indiana Iowa Vermont Pennsylvania Illinois Missouri Massachusetts Michigan Nebraska Rhode Island Wisconsin Kansas Connecticut North Dakota South Dakota 5. South Atlantic 6. East South Cent ral 7. West South Central 8. Mountain Maryland Kentucky Arkansas Montana Delaware Tennessee Louisiana Wyoming Washington D C Alabama Oklahoma Colorado Virginia Mississippi Texas Idaho West Virginia New Mexico North Carolina Nevada South Carolina Arizona Florida Utah 9. Pacific Washington Oregon California
47 CHAPTER 4 MODEL SETTING AND ESTIMATIONS The main focus of this research was to de velop a base model to study tourism demand and then analyze the likelihood of taking a long di stance trip for pleasure among the sampled population. The data set included 45,165 households reporting for an assigned four weeks period over the reporting months from November 2000 through May 2002. For each entry a household indicated if they took a pleasure trip or not, he nce giving us a binary de pendent variable model pointing to the appropriatene ss of using Probit modeling. Probit Model Estimates The implementation of a Probit model allows us to calculate the probabi lities of traveling for pleasure and showing how these probabilitie s change across each independent variable influencing taking a pleasure tr ip. Letting Pleasure represent the binary response, the Probit model can be represented as: ) ( ) | 1 Pr( iX X Pleasure (4-1) 2 1 ) | 1 Pr(exp2 /2dz X Pleasurez XB (4-2) The probability that a respondent will take a pl easure trip is given by the values of the explanatory variables Xi, which represent attributes and char acteristics expected to impact the likelihood of taking a pleasure trip. is the standard cumulative normal distribution function (CDF) with zero mean and variance equals 1. Th e Probit model assumes that there is a basic response variable y defined by: i i iX y (4-3)
48 In equation (4-3) i represent the respondent, is a vector of coefficients that represents the effect of the independent variables on the dependent variable, and i a stochastic random error for each respondent. If the latent variable (Y ) is greater than zero, th e observed variable is 1 (the person took a pleasure trip) and 0 otherwise. In the Probit model the likelihood of taking a pl easure trip is estimated using the following independent variables denoted as Xi : income, age, race, education level, employment status, urban status, pre or post 9/11, mont h, region and number of trips. Each independent variable is binary with up to J categories within the variable group, hence ( J -1) dummy variable are needed for each group. Clearly, J can differ with each group. The actual Probit model is denoted as ) ) 4 ( ) 5 ( ) 5 ( 1 ) 12 ( ) 9 ( ( ) | 1 Pr(35 34 33 3 1 29 29 4 1 24 4 1 20 20 11 1 8 8 1 0TNUM URBAN EMPL DAGE DAGE DEDUC DRACE DRACE DINC DINC DPOST DMTH DMTH DCENSUS DCENSUS X Pleasurej j j j j j j j j j j j j j j (4-4) Since each dummy group is mutually exclusive, methods for dealing with the well-known dummy variable trap must be adopted. One appr oach is to set the sum of the coefficients for each dummy to zero so as to estimate the effect of each qualitative variable on the probability of taking a pleasure trip as shown in equation (4-5 ). For the Probit and Logit estimations, the last category coefficient within a dummy set of variables is not shown since it is the negative sum of the categories estimated. The intercept value o represents the unweighted mean value across all discrete categories. Hence, using this method to restrict the sum of th e dummy coefficients to zero the intercept ( o) represents the mean value of the base category or in other words the
49 unweighted average household. This method is adopted for its convenience when there are many dummy variables in a model. 01 k i i or 1 1 k i i k (4-5) The variable estimates for the Probit model for taking a pleasure trip with the corresponding statistics are shown in Table 4.1. The inverse mills ratio wa s calculated from this model in order to check for sele ction bias discussed earlier. Th e estimates for taking a pleasure trip, limiting this selection to t hose respondents who took only one trip is shown in Table 4.2. When looking at the supporting statistics for each model, ev en though it is desirable to have a high goodness-of-fit, the Rsquare has limited interpreta tion and would generally be expected to be low in large cross-sectional data se ts such as used in this analysis. The t-statistic shown in the tables are compared to a t-valu e at 95% confidence (value=1.96). The estimators considered statistically significantly different fr om the average are thos e whose t-statistic is exceed |1.96| or the absolute value of +/-1. 96. Another measure would be observing the P-value (< 0.05) for those estimators statistically different from the average. The last category for each set of dummy group is not shown since it is the category being expressed in terms of the other category coeffici ents (see equation 4-5). Table 4-1 reports the estimated coefficients and accompanying distri bution properties of each estimate. When accounting for regional effects, the sign for the es timates of regions four (West North Central), six (East South Central) and se ven (West South Central) were negative, which means that the likelihood of taking a pleasure trip was lower for the respondents fr om these regions than for the average. Associated t-values show regions one (New England), two (Mid-Atlantic), three (East North Central) and six (East Sout h Central) were statistically significantly different from the average. Seasonality January to May and Se ptember had negative signs indicating that the
50 likelihood of taking trips during those months was lower than the average, April, June, August, September and October were not statistically sign ificant different from the seasonal average. The coefficient estimate for DPOST1 was very small and not statistically significant. Across income levels, the response was as expected with the negative sign for the lowest income and increasingly positive as income rises. Blacks had a negative sign for the estimator, which indicates that this category was somehow less lik ely to take long distance trips for pleasure. When analyzing the significance, category of ra ce one and two were statistically significant different for the average. For level of educati on the likelihood of taking a pleasure trip showed little relationship with education. For age ca tegories, ages between 25 and 45 years old and between 45 and 65 years old had a negative coeffi cient estimate and all age categories were statistically significant. Those empl oyed were are less likely to take pleasure trips than those who were not employed and this estimate was statisti cally significant. Note that not employed also included retirees. And urban respondents were mo re likely to take pleasure trips and the coefficient estimate was also statistically significant. Table 4-2 shows the Probit estimates for the likelihood of taking pleas ure trips for those respondents who only took one trip during the 4 weeks recorded period. When accounting for regions, the estimates for regions four (West No rth Central), six (East South Central), seven (West South Central) and eight (Mountain) had negative sign, wh ich means that the likelihood of taking a pleasure trip was lower for these regions than for the average. Also regions one (New England) and six (East South Centra l) were statistically significantly different from the average. For seasonality effect, the coefficients for es timates of the months from January to May and the month October had a negative sign, whic h indicates that the lik elihood of taking trips during those months was lower than the average. Also only the months of January, February,
51 April, May, July and November were statistical ly significant different fr om the base category. The coefficient estimate for DPOST1 was negative and not statistically significant. For income levels, only for category one there was a negative sign for the coefficient estimate and category one (< $25000), three (between 50 and 75 thousand dollars) a nd four (between 75 and 100 thousand dollars) were statistically significant. Fo r race categories black had a negative sign for the estimator, which indicates that this categor y was somehow less likely to take long distance trips for pleasure. All other responses ar e similar to that discussed for Table 4-1. Multinomial Logit Model Estimates As a second approach for this research, we we re interested in anal yzing the likelihood of traveling to competing locations among those respondents who took a pleasure trip. Th e analysis was limited to the top 6 destinations among the trav elers from the sample. The destinations were: California, Florida, New York, Ohio, Pennsylvani a and Texas. To address this analysis, the recommended model for multiple unordered nominal choices was the well-known Multinomial Logit Model (MNLM). Consider a nominal dependent variable Y that has J categories. One value (typically the first, the last, or the value with the highest frequency) of the outcome is designated as the reference category. The probability of occurrenc e in other categories is compared to the probability of the occurrence in the reference category. For a dependent variable with J categories, this requires the ca lculation of J-1 equations, one fo r each category relative to the reference category, to describe the relations hip between the dependent variable and the independent variables. The choi ces for a destination represent the categories, with no implied order or ranking. Also, to ensure that the choi ces were mutually exclusive, we selected the respondents who took only one pleasure trip duri ng the assigned 4 weeks period, avoiding any
52 problems of having several joint travel events dur ing the period. Equations 4-6 and 4-7 show the specific multinomial logit model. The probability of choosing different destina tions depended on 35 explanatory variables where corresponds to the function of equation 4-5. Table 4-3 shows the coefficient estimates for the multinomial logit model. (4-6) ) ) 4 ( ) 5 ( ) 5 ( 1 ) 12 ( ) 9 ( ( ) | Pr(35 34 33 3 1 29 29 4 1 24 4 1 20 20 11 1 8 8 1 0IMR URBAN EMPL DAGE DAGE DEDUC DRACE DRACE DINC DINC DPOST DMTH DMTH DCENSUS DCENSUS X j Yj j j j j j j j j j j j j j j (4-7) It was expected that the same variable (X i) presented a different impact across the destination. Table 4-4 provides the estimated parameters and supporting statistics. These estimates correspond to the values in the equation 4-7. Tables 4-4, 4-5 and 4-6 show the Multinomial Logit estimates for the top six destinations in the country. Since in the logit model as show n in equation (4-6) the estimated coefficients are in both the numerator and denominator and, as su ch, it is difficult to l ook at the signs of the estimates and directly draw inferences about the effects of each variable. The directional impact of each variable is best seen using the simulatio ns for showing the direction effects in the next chapter. 6 1 exp 1 exp ) | Pr(2 j for x x P x j yJ j j i j i ij i i J j j i i i ix P x y2 1exp 1 1 ) | 0 Pr(
53 Consider for example the effect of region acros s the six states. For California, estimates for respondents located in the Mid-At lantic, West North Central a nd East South Central regions were not statistically significant, while estimates fo r the other five states were significant. In the case of Florida, estimates for households located in New England, East North Central and West South Central region were not stat istically significant compared to the average. Also coefficient estimates for regions East North Central and West North Central had negative sign. For New York, regions East North Central and East S outh Central were not st atistically significant different from the average. For Pennsylvania only the region Mountain was not statistically significant and for Ohio, regions West North Central and West South Central were not statistically significant. The months of the year (seasona lity), showed estimates that were of particular importance when selecting a destination for pleasure trips. For California, the mont hs of May, September and November were statistically significant diffe rent from the average. For Florida, January, February, March (winter) were st atistically significant. Also J une and August were significant. For New York, only the month of October was sta tistically significant different for the average. For Pennsylvania all of the estimates were not statistically significant. For Texas, May and October were statistically significant and for Ohio only April was stat istically significant. For DPOST (if the respondent traveled before the 9/11 event) effect only the estimates for Florida and Texas had a positive effect. Also onl y the coefficient estimates for California, Florida and Ohio were statistically si gnificant different from the average. When comparing income estimates to an av erage household, they were statistically significant when looking at income levels between $ 50,000 and $74,999 for New York and Pennsylvania, and $25,000 to $49,999 for California and Texas estimates. Income category
54 number 1 (less than $25,000) was statistically si gnificant for Texas, and category 4 ($75,000 to $99,999) was statistically significant for Pennsylvania. On the other hand, all income coefficients for Pennsylvania were positive, and with exception of income category one (<$25,000), the rest of the coefficient estimates for Texas and Oh io were positive. For New York, only income category 4 ($75,000 to 99,999) had a negative sign For Florida only income category one had a positive sign. For California, only income cat egory 3 (between $ 50,000 and $74,999) had a positive effect. Race estimates were not statistically significant different from the average for New York and Ohio. For California only categor y 4 (White/Hispanic) was not statistically significant different from the average. For the ca se of Florida only category 1 (White) was not significantly different from the average. In the case of Pennsylvania only category 1 (White) was statistically significant from the average. For Texas categories 2 and 3 (Black and Hispanic) were statistically significant different from the average. Estimates for college education were statistic ally significant for California and New York; and when looking at sign of the coefficient es timates, only the ones for Pennsylvania and Texas had a negative effect. Again recall that direc tly interpreting the signs can be misleading. Age estimates were not statistically significant different from the average for all of the age categories across the six states with exception of category 2 (between 25 and 45 years old) for the state of Texas. In the case of employment, the coefficien ts for any of the states were statistically significant and the coefficients for Pennsylvania and Texas had negative sign. For Urban, only the coefficients of the estimates for New York and Pennsylvania were not statistically significant and the sign for the coefficient estimates for Ne w York and Ohio were negative. The Inverse Mills Ratio was not statistically significant for any of the six destina tions thus indicating no selection bias when just usi ng those traveling for pleasure.
55 Table 4-1. Probit estimates for taking a pleasure trip Parameter Estimatet-statisticP-value C 0.6413334.96160[.000] DCENSUS1 0.098093.94268[.000] DCENSUS2 0.036682.07893[.038] DCENSUS3 0.071094.61002[.000] DCENSUS4 -0.03272-1.69695[.090] DCENSUS5 0.004780.30848[.758] DCENSUS6 -0.17429-6.96998[.000] DCENSUS7 -0.03853-1.94153[.052] DCENSUS8 0.034531.54005[.124] DMTH1 -0.10617-4.42681[.000] DMTH2 -0.08860-4.06868[.000] DMTH3 -0.05442-3.05827[.002] DMTH4 -0.03237-1.69310[.090] DMTH5 -0.13967-4.71785[.000] DMTH6 0.007650.25888[.796] DMTH7 0.079182.83548[.005] DMTH8 0.027911.03056[.303] DMTH9 -0.00900-0.30774[.758] DMTH10 0.014210.50594[.613] DMTH11 0.128644.82844[.000] DPOST1 0.003320.23758[.812] DINC1 -0.03086-2.10049[.036] DINC2 0.011861.00596[.314] DINC3 0.016401.29748[.194] DINC4 0.047053.16969[.002] DRACE1 0.050933.13667[.002] DRACE2 -0.11122-3.76951[.000] DRACE3 0.036941.01023[.312] DRACE4 0.022800.65089[.515] DEDUC -0.00168-0.23491[.814] DAGE1 0.2439518.07820[.000] DAGE2 -0.08503-6.93692[.000] DAGE3 -0.09798-8.65458[.000] EMPL -0.19478-23.00820[.000] URBAN 0.050426.93609[.000] TNUM -0.19282-58.78850[.000] Number of observations = 44156 Number of positive obs. = 24290 Mean of dep. var. = .550095 R-squared = .199723
56 Table 4-2. Probit taking a pleasure tr ip for those who took only one trip Parameter Estimate t-statisticP-value C 0.41867 18.31710[.000] DCENSUS1 0.09014 2.58319[.010] DCENSUS2 0.04618 1.89776[.058] DCENSUS3 0.03717 1.76314[.078] DCENSUS4 -0.04190 -1.55392[.120] DCENSUS5 0.02495 1.16908[.242] DCENSUS6 -0.16333 -4.65077[.000] DCENSUS7 -0.01045 -0.38139[.703] DCENSUS8 -0.02140 -0.70427[.481] DMTH1 -0.19617 -5.90943[.000] DMTH2 -0.08176 -2.75337[.006] DMTH3 -0.01780 -0.75108[.453] DMTH4 -0.08769 -3.21026[.001] DMTH5 -0.12036 -2.96909[.003] DMTH6 0.03015 0.73978[.459] DMTH7 0.13317 3.47671[.001] DMTH8 0.02945 0.79062[.429] DMTH9 0.07256 1.74185[.082] DMTH10 -0.06613 -1.69250[.091] DMTH11 0.11969 3.25745[.001] DPOST1 -0.02775 -1.45939[.144] DINC1 -0.07244 -3.65510[.000] DINC2 0.00760 0.46959[.639] DINC3 0.03981 2.25371[.024] DINC4 0.04569 2.23748[.025] DRACE1 0.08757 3.99934[.000] DRACE2 -0.19646 -4.97695[.000] DRACE3 0.04836 0.99955[.318] DRACE4 0.07374 1.54308[.123] DEDUC -0.00648 -0.65190[.514] DAGE1 0.21316 11.99310[.000] DAGE2 -0.03989 -2.35013[.019] DAGE3 -0.11301 -7.18042[.000] EMPL -0.11848 -10.30920[.000] URBAN 0.06391 6.25900[.000] Number of observations = 22278 Number of positive obs. = 15357 Mean of dep. var. = .689335 R-squared = .035491
57 Table 4-3. Supporting statistics for Logit Mode l taking a pleasure trip to any of the 6 destinations, TNUM=1, including IMR Choice Frequency Fraction De pendent variable: ZFARST 0 10226 0.6659 Number of observations = 15357 1 1456 0.0948 Scaled R-squared = .687563 2 953 0.0621 LR (zero slopes) = 14200.8 [.000] 3 738 0.0481 Number of Choices = 107499 4 664 0.0432 5 815 0.0531 6 505 0.0329
58 Table 4-4. Multinomial Logit estimation (Logit taki ng a pleasure trip to California and Florida, TNUM=1, including IMR) Parameter Estimate t-statistic P-value Parameter Estimate t-statisticP-value California Florida C1 -0.07335 -0.03775 [.970]C2 -3.19683 -1.88892[.059] DCENSUS11 -1.10786 -2.80340 [.005]DCENSUS120.31680 1.43079[.152] DCENSUS21 -0.45840 -1.68126 [.093]DCENSUS221.27580 9.84349[.000] DCENSUS31 -0.61264 -2.62376 [.009]DCENSUS32-0.26313 -1.50766[.132] DCENSUS41 -0.07745 -0.33790 [.735]DCENSUS42-1.20924 -4.91779[.000] DCENSUS51 -0.93475 -3.89447 [.000]DCENSUS522.25083 26.24390[.000] DCENSUS61 -1.25464 -1.76211 [.078]DCENSUS620.74004 2.28856[.022] DCENSUS71 -0.71568 -1.98059 [.048]DCENSUS720.12658 0.64659[.518] DCENSUS81 1.14723 6.88515 [.000]DCENSUS82-1.98180 -5.31023[.000] DMTH11 0.45622 1.90026 [.057]DMTH12 0.52534 2.44022[.015] DMTH21 -0.32023 -1.56656 [.117]DMTH22 0.45373 2.48935[.013] DMTH31 0.20004 1.47241 [.141]DMTH32 0.46619 3.73914[.000] DMTH41 0.20708 1.52600 [.127]DMTH42 0.14467 1.16799[.243] DMTH51 0.91639 2.89434 [.004]DMTH52 -0.55489 -1.90401[.057] DMTH61 0.15188 0.90111 [.368]DMTH62 -0.39269 -2.24041[.025] DMTH71 -0.27663 -1.28727 [.198]DMTH72 -0.31878 -1.59790[.110] DMTH81 -0.19877 -1.16851 [.243]DMTH82 -0.53653 -3.13874[.002] DMTH91 0.43209 2.58656 [.010]DMTH92 -0.27853 -1.36333[.173] DMTH101 -0.19032 -1.04328 [.297]DMTH102 0.24408 1.41737[.156] DMTH111 -0.54890 -2.04329 [.041]DMTH112 -0.00760 -0.03016[.976] DPOST11 -0.28810 -3.51155 [.000]DPOST12 0.19496 2.48313[.013] DINC11 -0.06527 -0.64575 [.518]DINC12 0.05293 0.58286[.560] DINC21 -0.36415 -4.75118 [.000]DINC22 -0.06260 -0.88386[.377] DINC31 0.06115 0.75091 [.453]DINC32 -0.02390 -0.30813[.758] DINC41 -0.13381 -1.15008 [.250]DINC42 -0.07912 -0.72585[.468] DRACE11 -0.98065 -7.60561 [.000]DRACE12 -0.15702 -1.29544[.195] DRACE21 0.87558 3.02841 [.002]DRACE22 -1.09680 -4.31314[.000] DRACE31 0.42983 2.60987 [.009]DRACE32 0.67211 3.20435[.001] DRACE41 -0.13011 -0.75020 [.453]DRACE42 0.65312 3.97561[.000] DEDUC1 0.14320 3.11347 [.002]DEDUC2 0.01475 0.35173[.725] DAGE11 -0.71689 -1.76255 [.078]DAGE12 0.01304 0.03657[.971] DAGE21 0.22260 1.35565 [.175]DAGE22 -0.14776 -1.00065[.317] DAGE31 0.26161 1.41450 [.157]DAGE32 -0.03939 -0.24567[.806] EMPL1 0.59918 1.72314 [.085]EMPL2 0.04939 0.16211[.871] URBAN1 0.27244 2.53731 [.011]URBAN2 0.28661 3.09484[.002] IMR1 -6.29088 -1.78013 [.075] IMR2 -0.24363 -0.07931[.937]
59 Table 4-5. Multinomial Logit es timation (Logit taking a pleasure trip to New York and Pennsylvania, TNUM=1, including IMR) Parameter Estimate t-statistic P-value Parameter Estimate t-statisticP-value New York Pennsylvania C3 -3.96012 -1.96360 [.050]C4 -7.19675 -3.47488[.001] DCENSUS13 2.06803 8.58589 [.000]DCENSUS140.81496 2.50482[.012] DCENSUS23 3.77028 24.27060 [.000]DCENSUS243.90919 20.66180[.000] DCENSUS33 -0.47908 -1.83527 [.066]DCENSUS340.86855 3.56431[.000] DCENSUS43 -1.76331 -3.80063 [.000]DCENSUS44-1.87390 -3.47323[.001] DCENSUS53 0.63102 3.56545 [.000]DCENSUS541.55983 8.26626[.000] DCENSUS63 -0.64796 -1.20935 [.227]DCENSUS64-2.49161 -2.58553[.010] DCENSUS73 -1.65358 -2.57666 [.010]DCENSUS74-1.48868 -2.28879[.022] DCENSUS83 -1.10915 -2.84646 [.004]DCENSUS84-0.58178 -1.52950[.126] DMTH13 -0.15655 -0.58363 [.559]DMTH14 -0.33695 -1.22855[.219] DMTH23 -0.26848 -1.21785 [.223]DMTH24 -0.38059 -1.66587[.096] DMTH33 0.07413 0.47320 [.636]DMTH34 -0.18973 -1.16078[.246] DMTH43 0.26541 1.85192 [.064]DMTH44 0.20528 1.38839[.165] DMTH53 -0.05132 -0.15081 [.880]DMTH54 -0.22125 -0.64622[.518] DMTH63 0.07765 0.37868 [.705]DMTH64 0.09595 0.45443[.650] DMTH73 -0.00481 -0.02062 [.984]DMTH74 0.36922 1.56718[.117] DMTH83 -0.36532 -1.78124 [.075]DMTH84 0.14071 0.72766[.467] DMTH93 0.22064 1.09613 [.273]DMTH94 -0.26304 -1.12032[.263] DMTH103 0.37973 2.05678 [.040]DMTH104 -0.37014 -1.51349[.130] DMTH113 -0.05214 -0.18335 [.855]DMTH114 0.43124 1.47321[.141] DPOST13 -0.06448 -0.65026 [.516]DPOST14 -0.02313 -0.22509[.822] DINC13 0.08723 0.76845 [.442]DINC14 0.00829 0.07372[.941] DINC23 0.00295 0.03459 [.972]DINC24 0.04067 0.47347[.636] DINC33 0.26666 3.08310 [.002]DINC34 0.21484 2.38530[.017] DINC43 -0.16811 -1.31655 [.188]DINC44 0.27393 2.18336[.029] DRACE13 0.08200 0.50658 [.612]DRACE14 0.43157 2.49099[.013] DRACE23 0.02493 0.08238 [.934]DRACE24 0.01137 0.03758[.970] DRACE33 0.53068 1.66482 [.096]DRACE34 -0.16911 -0.41620[.677] DRACE43 -0.35375 -1.12626 [.260]DRACE44 0.08473 0.28456[.776] DEDUC3 0.19679 3.82827 [.000]DEDUC4 -0.02513 -0.50606[.613] DAGE13 0.26690 0.64049 [.522]DAGE14 0.77426 1.82208[.068] DAGE23 0.10953 0.63147 [.528]DAGE24 -0.23446 -1.32200[.186] DAGE33 0.08062 0.42121 [.674]DAGE34 -0.22588 -1.16376[.245] EMPL3 0.14121 0.40008 [.689]EMPL4 -0.30326 -0.83893[.402] URBAN3 -0.03439 -0.31991 [.749]URBAN4 0.21423 1.93127[.053] IMR3 -0.82190 -0.22457 [.822] IMR4 4.39344 1.17318[.241]
60 Table 4-6. Multinomial Logit es timation (Logit taking a pleasure trip to Texas and Ohio, TNUM=1, including IMR) Parameter Estimate t-statistic P-value Parameter Estimate t-statisticP-value Texas Ohio C5 -7.75322 -3.23416 [.001] C6 -6.40747 -0.01163[.991] DCENSUS15 -1.11901 -1.65883 [.097] DCENSUS16-2.03758 -2.20486[.027] DCENSUS25 -0.01063 -0.02995 [.976] DCENSUS261.81472 7.31683[.000] DCENSUS35 -0.76053 -2.10155 [.036] DCENSUS362.91354 11.95110[.000] DCENSUS45 0.43690 1.85099 [.064] DCENSUS46-0.54545 -1.39474[.163] DCENSUS55 -0.50769 -1.67646 [.094] DCENSUS560.52557 2.06172[.039] DCENSUS65 -1.30397 -1.95967 [.050] DCENSUS661.02893 2.12708[.033] DCENSUS75 4.59912 26.87170 [.000] DCENSUS76-0.09509 -0.21255[.832] DCENSUS85 0.04803 0.15775 [.875] DCENSUS86-1.55253 -2.36110[.018] DMTH15 0.29054 0.95401 [.340] DMTH16 0.05225 0.18772[.851] DMTH25 0.27338 1.06922 [.285] DMTH26 -0.33469 -1.32799[.184] DMTH35 -0.11964 -0.67882 [.497] DMTH36 -0.10879 -0.62586[.531] DMTH45 0.00223 0.01401 [.989] DMTH46 0.51292 3.41870[.001] DMTH55 -1.01086 -2.55001 [.011] DMTH56 0.65553 1.84232[.065] DMTH65 -0.12417 -0.52150 [.602] DMTH66 0.26965 1.15931[.246] DMTH75 -0.38442 -1.35234 [.176] DMTH76 0.21952 0.87824[.380] DMTH85 -0.40625 -1.79797 [.072] DMTH86 -0.03607 -0.16637[.868] DMTH95 -0.19774 -0.85178 [.394] DMTH96 -0.08322 -0.41540[.678] DMTH105 0.68303 3.08582 [.002] DMTH106 -0.23542 -1.10486[.269] DMTH115 0.18623 0.54614 [.585] DMTH116 -0.46596 -1.47585[.140] DPOST15 0.15687 1.56179 [.118] DPOST16 -0.29223 -2.70637[.007] DINC15 -0.36222 -2.84563 [.004] DINC16 -0.08720 -0.68762[.492] DINC25 0.22756 2.40785 [.016] DINC26 0.12577 1.40429[.160] DINC35 0.19302 1.83148 [.067] DINC36 0.09455 0.96853[.333] DINC45 0.19650 1.22898 [.219] DINC46 0.01133 0.08368[.933] DRACE15 -0.06148 -0.38525 [.700] DRACE16 3.52212 0.00639[.995] DRACE25 -1.60386 -4.65645 [.000] DRACE26 4.30560 0.00781[.994] DRACE35 1.15341 5.14640 [.000] DRACE36 -14.37520 -0.00652[.995] DRACE45 0.35515 1.78189 [.075] DRACE46 3.30572 0.00600[.995] DEDUC5 -0.05258 -0.93506 [.350] DEDUC6 0.05846 1.11479[.265] DAGE15 0.73720 1.43862 [.150] DAGE16 -0.27619 -0.61571[.538] DAGE25 -0.54440 -2.59322 [.010] DAGE26 0.19997 1.07088[.284] DAGE35 -0.31237 -1.34672 [.178] DAGE36 0.01978 0.09593[.924] EMPL5 -0.53637 -1.22322 [.221] EMPL6 0.38536 1.02112[.307] URBAN5 0.39037 3.06794 [.002] URBAN6 -0.24967 -2.23606[.025] IMR5 5.91433 1.36203 [.173] IMR6 -3.63216 -0.92316[.356]
61 61 CHAPTER 5 SIMULATION ANALYSIS Drawing from the Probit estimates, the next fundamental step was to calculate the probabilities of traveling for pleasure and show ing how these probabilities differed across those variables included in the Probit model (Table 4-2). Since the Probit is based on a normal distribution, the likelihood of tr aveling from pleasure can be estimated by simply expressing the predicted values (latent variable) for any comb ination of the explanat ory variables and then insert that the predicted values into a cu mulative normal distribution. The corresponding cumulative probability is the likelihood of traveli ng based on the variables used. As a starting point, the combination can be the average values for each variable in equation (4-4), thus giving the likelihood of traveling based on the average household. Next, any of the groups such as employment can be simulated to give the probabi lity while leaving the other variables at their mean values. The minimum and maximum probabil ities for a particular group provides a direct measure of the potential influence of that variab le (i.e., dummy group) on traveling for pleasure. The extent of the impact is simply the range from the max to min probabilities. Clearly, these ranges can be compared across variables in or der to rank the variables in terms of their importance to changing the likelihood of travelin g or the probability of taking a long distance pleasure trip. For the first analysis, the e ffects were ranked over the diffe rent groups of explanatory variables. Summarizing, the method of ranking us ed in this section c onsist of the following steps: first the impact on the probability from changing each value of th e explanatory variables was calculated while holding all othe r variables fixed at their aver age value. Then the difference between the minimum and maximu m probabilities was calculated a nd the effects were ranked by the difference. Similar procedures were al so used for the logit destination analyses.
62 62 Ranking the Probabilities of Traveling for Pleasure Figure 5.1 shows the ranking of the effect of each set of variables for those respondents who took one pleasure trip during the assigned period. 0.13 0.11 0.10 0.09 0.06 0.03 0.03 0.00 0.00 Probability range ranked 0.000.050.100.15 Probability range Employment Age Season Region Race Urban Income Sept 11 Educ Ranking the effects of variables influence the probability of traveling for pleasure 0.600.650.700.750.80 Probality of traveling for pleasure Figure 5-1. Ranking of the effect s of variables on the probabil ity of taking a pleasure trip From the results shown in Figure 5-1 the vari able Employment was ranked on the top bar, which means that this variable had the largest effect when deciding for a pleasure trip (long distance) during the assigned period. Differences in Employment st atus can cause the change of probability by 13.2 percent points as shown in the right portion of Figure 5-1. The second important factor influencing th e likelihood of taking a pleasure trip was Age, which among age levels can cause a change of 11.3 percentage units on the estimated probability, establishing that different ages exhibit important effects on trav eling behavior when an alyzing tourism demand. Third, Season was also an important factor si nce it can change the estimated probability by 10.4 percentage points among the different categories for months. This is understandable since travel
63 63 patterns are expected to change during the different seasons of th e year. Fourth, Region of origin of the respondent was also an important fact or to consider when examining tourism demand since it can change the value by 6.5 percentage leve ls. The last three variab les that also have a relative importance for analyzing tourism demand were Race of the respondent which could change the likelihood of taking a pleasure trip by 5.3 percentage points, Urban (3.4 percentage change) and Income (3 percentage change). The 9-11 effect and Education did not have significant impacts in the change of the like lihood of taking a long distance pleasure trip compared to the other explanatory variables. Probabilities of Traveling Across Each Variable Analyzing the change in the probabilities acro ss the same type of variables (each set of explanatory variables) showing the direction of the effect of each variable. The average probability of taking a long distan ce pleasure trip was calculated with the Probit model using the average value of each variable. Holding the other variables constant at their average values, the variable of interest was simulated among the di fferent values of the category. Following, an analysis of these changes are shown. On aver age, the probability of taking a long distance pleasure trip during a month period is 0.689, thus indicating that for a majority of households traveling was the norm. For the Income category, si mulated effects of the different income levels are presented in Figure 5-2. The vertical axis shows the different levels of income, while the horizontal axis gives the corresponding change in probability of the different income levels. For each case the probability changes were predicted, while holding other variables to the average levels. The likelihood of taking a long distance pl easure trip decreased fo r income category one (less than $25,000) and to level five (more than $100,000) from an average of 68.96% to 67.88% and 67.41% respectively; while its bigger increm ent was for income category four from an a average of 68.96% to 70.5%.
64 64 0.690 0.679 0.693 0.695 0.705 0.674 Average $24,999 or less $25,000 to $49,999 $50,000 to $74,999 $75,000 to $99,999 $100,000 or more Income levels 0.0000.2500.5000.7501.000 Probability of traveling for pleasure Figure 5-2. Proportion of variation in relation to the average by income 0.690 0.768 0.660 0.655 0.668 Average Under 25 year 25 years to 44 years 45 year to 65 years 65 years and over Age groups 0.0000.2500.5000.7501.000 Probability of traveling for pleasure Figure 5-3. Proportion of variation in relation to the average by Age
65 65 For the Age categories, Figure 5-3 shows that there was an increment in the likelihood of taking a long distance pleasure trip compared to the average for age group number one (< 25 years old) changing the probability from an average of 68.96% to 76.82%. The other 3 age categories had a negative variation from 68.96% on average to 65.99%, 65.52% and 66.85% for age group 2, 3 and 4 respectively. 0.690 0.693 0.637 0.688 0.683 0.683 Average White (=1) Black (=2) Hispanic (=3) White/Hispanic (=4) All others (=5) Race 0.0000.2500.5000.7501.000 Probability of traveling for pleasure Figure 5-4. Proportion of variation in relation to the average by Race For Race category, similar to the above simu lation, in each position the rest of the independent variables were set to average level. The figure shows that only for race category two (black) there was a significant decrement of the probability of taking a pleasure trip from an average of 68.96% to 63.65 percent. For the othe r four categories the probability remained very close to the average (69.28% for White, 68.80% for Hispanics, 68.33% for White/Hispanics, and 68.35% for all other races).
66 66 0.690 0.689 0.690 0.632 0.764 Average College Education No College Education Employed No Employed Education and Employment 0.0000.2500.5000.7501.000 Probability of traveling for pleasure Figure 5-5. Proportion of variation in relati on to the average by Education and Employment When analyzing the change of probability for the variable College E ducation, variation for both responses remained very close to the average, not changing in a significant percentage the probability of taking a pleasure trip. For the variable Employ ment, the likelihood of taking a pleasure trip decreased from an average of 68.96% to 63.20% for respondents who were employed at the time of taking the survey, a nd increased from 68.96% to 76.42% for those respondents who were un-employed. For Urban variable, the likelihood of taking a pl easure trip decreased from the average of 68.96% to 66.43% for those respondents who lived in rural areas and sligh tly increased for those respondents living in urban area s (from 68.96% to 69.85%). When considering the effect of September 11 (i.e., 9/11) on the likelihood of taki ng a pleasure trip, Figur e 5-6 shows that the variation from the average probability to be extremely small.
67 67 0.690 0.699 0.664 0.691 0.689 Average Urban Rural Before 9/11 After 9/11 Location and Pre-Post 9/11 0.0000.2500.5000.7501.000 Probability of traveling for pleasure Figure 5-6. Proportion of variation in relati on to the average by Location and 9-11 event 0.690 0.654 0.660 0.672 0.679 0.642 0.693 0.717 0.700 0.687 0.695 0.732 0.746 Average January February March April May June July August September October November December Seasonality 0.0000.2500.5000.7501.000 Probability of traveling for pleasure Figure 5-7. Proportion of variation in relation to the average by Season
68 68 For seasonality effect on the variation from the average pr obability of taking a pleasure trip, Figure 5-7 shows the vari ation among different months. The largest increments from the average occurred for July, November and Decembe r when the probabilities changed from an average of 68.96% to 71.66%, 73.25% and 74.65% respectively. On the other hand, the months of January, February and May showed the most si gnificant decreases with respect to the average going from 68.96% to 65.38%, 66% and 64,19% respectively. The last set of variables an alyzed was region of the respondent. Figure 5-8 shows the proportion of variation from the average for the di fferent region of origin of the respondents. Region six (East South Central) was the region that showed the largest decrease on the probability of taking a long distan ce pleasure trip going from an average of 68.96% to 62.46%. In contrast, regions one and th ree (New England and East Nort h Central) were the ones that showed the larger positive increment from the average (to 71.81% and 70.94% respectively). 0.690 0.718 0.698 0.709 0.675 0.687 0.625 0.673 0.697 0.686 Average New England (1) Middle Atlantic (2) East North Central (3) West North Central (4) South Atlantic (5) East South Central (6) West South Central (7) Mountain (8) Pacific (9) Regions 0.0000.2500.5000.7501.000 Probability of traveling for pleasure Figure 5-8. Proportion of variation in relation to the average by Region
69 69 Destination Probabilities As a second part of our analysis, the estimates calculated for the Multinomial Logit Model (Tables 4-4, 4-5 and 4-6) were in corporated into a simulation pro cedure to analyze the effect of each set of variables on the probability of taki ng a long distance pleasure trip for the top six domestic destinations. 0.095 0.062 0.048 0.043 0.053 0.033 CaliforniaFloridaNew YorkPennsylvaniaTexasOhio U.S. states 0.000 0.040 0.080 0.120 Figure 5-9. Average probability of ta king a pleasure trip by destination Comparing the effect of the di fferent Income categories among the top six destinations, we can see in Figure 5-10 that for the states of California and Florida the trend was similar, increasing the likelihood of taki ng a pleasure trip to these des tinations among the higher income categories. For the state of New York, the inco me category that was most likely to take a pleasure trip to this destination was the thir d category ($50,000 to $75,000). For the states of Ohio and Texas the probability of taking a pleasu re trip did not change substantially among the different income categories. For the state of Pennsylvania, the probability of taking a pleasure trip to this destination was almost the same across the different income categories, but it decreased notably for the highe r income category (> $100,000).
70 70 $ 2 4 9 9 9 o r l e s s ( C A ) $ 2 5 0 0 0 t o $ 4 9 9 9 9 ( C A ) $ 5 0 0 0 0 t o $ 7 4 9 9 9 ( C A ) $ 7 5 0 0 0 t o $ 9 9 9 9 9 ( C A ) $ 1 0 0 0 0 0 o r m o r e ( C A ) $ 2 4 9 9 9 o r l e s s ( F L ) $ 2 5 0 0 0 t o $ 4 9 9 9 9 ( F L ) $ 5 0 0 0 0 t o $ 7 4 9 9 9 ( F L ) $ 7 5 0 0 0 t o $ 9 9 9 9 9 ( FL ) $ 1 0 0 0 0 0 o r m o r e ( F L ) $ 2 4 9 9 9 o r l e s s ( N Y ) $ 2 5 0 0 0 t o $ 4 9 9 9 9 ( N Y ) $ 5 0 0 0 0 t o $ 7 4 9 9 9 ( N Y ) $ 7 5 0 0 0 to $ 9 9 9 9 9 ( N Y ) $ 1 0 0 0 0 0 o r m o r e ( N Y ) $ 2 4 9 9 9 o r l e s s ( P A ) $ 2 5 0 0 0 t o $ 4 9 9 9 9 ( P A ) $ 5 0 0 0 0 t o $ 7 4 9 9 9 (P A ) $ 7 5 0 0 0 t o $ 9 9 9 9 9 ( P A ) $ 1 0 0 0 0 0 o r m o r e ( P A ) $ 2 4 9 9 9 o r l e s s ( T X ) $ 2 5 0 0 0 t o $ 4 9 9 9 9 ( T X ) $ 5 0 0 0 0 to $ 7 4 9 9 9 ( T X ) $ 7 5 0 0 0 t o $ 9 9 9 9 9 ( T X ) $ 1 0 0 0 0 0 o r m o r e ( T X ) $ 2 4 9 9 9 o r l e s s ( O H ) $ 2 5 0 0 0 t o $ 4 9 9 9 9 (O H ) $ 5 0 0 0 0 t o $ 7 4 9 9 9 ( O H ) $ 7 5 0 0 0 t o $ 9 9 9 9 9 ( O H ) $ 1 0 0 0 0 0 o r m o r e ( O H Incomes by states 0.000 0.040 0.080 0.120 Figure 5-10. Effect of Income Level on the probab ility of taking a pleasure trip to any of the six top destinations When analyzing the effect of age on the likeli hood of taking a pleasure trip to any of the 6 destinations, we can see different trends acr oss the states and deri ve very interesting observations. When selecting California as the destination for a pleasure trip, the change in probability among the different age categories was al most irrelevant. When selecting Florida as the destination for a pleasure trip, we can s ee that the younger population (< 25 years old) was slightly more likely to take a trip than age category two (between 25 and 45 years old) and 3 (between 45 and 65 years old), but the probability went up for the older population (> 65 years old). For Texas, the change in probabilities was similar to Florida. When selecting New York as the destination for a pleasure tr ip, the younger population was more lik ely to take a trip to this state, and the probability decreased when age increased. For Pennsylvania and Ohio the probability practically did not change among the different age categories.
71 71 U n d e r 2 5 y e a r ( C A ) 2 5 y e a r s t o 4 4 y e a r s ( C A ) 4 5 y e a r t o 6 5 y e a r s ( C A ) 6 5 y e a r s a n d o v e r ( C A ) U n d e r 2 5 y e ar ( F L ) 2 5 y e a r s t o 4 4 y e a r s ( F L ) 4 5 y e a r t o 6 5 y e a r s ( F L ) 6 5 y e a r s a n d o v e r ( F L ) U n d e r 2 5 y e a r ( N Y ) 2 5 y e a rs t o 4 4 y e a r s ( N Y ) 4 5 y e a r t o 6 5 y e a r s ( N Y ) 6 5 y e a r s a n d o v e r ( N Y ) U n d e r 2 5 y e a r ( P A ) 2 5 y e a r s t o 4 4 y e a rs ( P A ) 4 5 y e a r t o 6 5 y e a r s ( P A ) 6 5 y e a r s a n d o v e r ( P A ) U n d e r 2 5 y e a r ( T X ) 2 5 y e a r s t o 4 4 y e a r s ( T X ) 4 5 y e ar t o 6 5 y e a r s ( T X ) 6 5 y e a r s a n d o v e r ( T X ) U n d e r 2 5 y e a r ( O H ) 2 5 y e a r s t o 4 4 y e a r s ( O H ) 4 5 y e a r t o 6 5 y e a rs ( O H ) 6 5 y e a r s a n d o v e r ( O H )Age groups by states 0.000 0.040 0.080 0.120 Figure 5-11. Effect of Age on the probability of taking a pleasure trip to any of the six top destinations For Race, when comparing the variation on the pr obability of taking a pleasure trip for the different categories across the si x destinations, it is shown in Figure 5-12 that white respondents trend was the same as the average probability for each state, California being the preferred destination, closely followed by Florid a, New York and Texas, then Pennsylvania and Ohio. For black respondents the likelihood of visiting California was three times higher that visiting any other state, then the preferred dest inations were Ohio, New York and Pennsylvania, with Florida and Texas as the last two prefer ences. For Hispanic respondents, the preferred destinations when taking a pleasure trip was Ca lifornia followed by Florida, and its remarkable to note that the probability of ta king a pleasure trip to Ohio is al most not significant for this race. For Hispanic/White respondents, the probabilities of taking a pl easure trip to California or Florida were at the top and were very close valu es. For all other races, th e probabilities of taking a pleasure trip to California and Florida were al most the same, New York and Pennsylvania were less important and Ohio ha d the higher preference.
72 72 W h i t e ( C A ) B l a c k ( C A ) H i s p a n i c ( C A ) W h i t e / H i s p a n i c ( C A ) A l l o t h e r s ( C A ) W h i t e ( F L ) B l a c k ( F L ) H i s p a n i c ( F L ) Wh i t e / H i s p a n i c ( F L ) A l l o t h e r s ( F L ) W h i t e ( N Y ) B l a c k ( N Y ) H i s p a n i c ( N Y ) W h i t e / H i s p a n i c ( N Y ) A l l o t h e r s ( N Y ) Wh i t e ( P A ) B l a c k ( P A ) H i s p a n i c ( P A ) W h i t e / H i s p a n i c ( P A ) A l l o t h e r s ( P A ) W h i t e ( T X ) B l a c k ( T X ) H i s p a n i c ( T X ) W hi t e / H i s p a n i c ( T X ) A l l o t h e r s ( T X ) W h i t e ( O H ) B l a c k ( O H ) H i s p a n i c ( O H ) W h i t e / H i s p a n i c ( O H ) A l l o t h e r s ( O H )Race groups by states 0.000 0.040 0.080 0.120 0.160 0.200 0.240 Figure 5-12. Effect of Race on the probability of taking a pleasure trip to any of the six top destinations When accounting for College Education, the variation in the probability of taking a pleasure trip among states is shown in Figure 5-13 Even though the values were slightly lower for those respondents who did not have college edu cation, the trend in de stinations preferences was almost the same, only varying for the stat e of New York. For Employment status of the respondent, the variation of th e probabilities between the two categories (employed or not employed) for the 6 destinations followed th e same trend and had similar values for both categories. For Urban or Rural condition of the respondent, the variation of the probability among the six states is shown in Figure 514. As shown in the figure, the likelihood of taking a pleasure trip to any of the top destinations was lower for ru ral respondents (except for those whose selected destination was New York or Ohio). Also the th ree preferred destinatio ns for urban respondents were California, Florida and Texas (in that orde r), while for rural respondents, the order of the preferred destinations was California, New York and Texas. Another important observation for
73 73 this graph is that the vari ation of the probability of ta king a pleasure trip among urban respondents was bigger (about 7%) than the va riation among rural respondents (about 2%). Considering the effect of the 9/11 event on the va riation of the probability of taking a pleasure trip across the 6 destinations, we could not totally charge the variation of this probability to this event since this variable was also affected by seasonality. However, Figure 5-14 shows that the probability of taking a pleasure trip to the 2 prefe rred destinations (California and Florida) went up for California after the event and went down for Florida. For New York and Ohio, it also went up after the 9/11 event. For Pennsylvania it rema ined almost constant, and for Texas it went slightly down. C o l l e g e E d u c a t i o n ( C A ) N o C o l l e g e E d u c a t i o n ( C A ) C o l l e g e E d u c a t i o n ( F L ) N o C o l l e g e E d u c a t i o n ( F L ) C o l l eg e E d u c a t i o n ( N Y ) N o C o l l e g e E d u c a t i o n ( N Y ) C o l l e g e E d u c a t i o n ( P A ) N o C o l l e g e E d u c a t i o n ( P A ) C o l l e g e E d uc a t i o n ( T X ) N o C o l l e g e E d u c a t i o n ( T X ) C o l l e g e E d u c a t i o n ( O H ) N o C o l l e g e E d u c a t i o n ( O H ) E m p l o y e d ( C A ) N o E m p l o y e d ( C A ) E m p l o y e d ( F L ) N o E m p l o y e d ( F L ) E m p l o y e d ( N Y ) N o E m p l o y e d ( N Y ) E m p l o y e d ( P A ) N o E m p l o y e d ( PA ) E m p l o y e d ( T X ) N o E m p l o y e d ( T X ) E m p l o y e d ( O H ) N o E m p l o y e d ( O H Education and employment 0.000 0.040 0.080 0.120 Figure 5-13. Effect of Educati on and Employment on the probabil ity of taking a pleasure trip to any of the six top destinations
74 74 U r b a n ( C A ) R u r a l ( C A ) U r b a n ( F L ) R u r a l ( F L ) U r b a n ( N Y ) R u r a l ( N Y ) U r b a n ( P A ) R u r a l ( P A ) U r b a n ( TX ) R u r a l ( T X ) U r b a n ( O H ) R u r a l ( O H ) B e f o r e 9 / 1 1 ( C A ) A f t e r 9 / 1 1 ( C A ) B e f o r e 9 / 1 1 ( F L ) A f t e r 9 / 1 1 ( F L ) B e f o r e 9 / 1 1 ( N Y ) A f t e r 9 / 1 1 ( N Y ) B e f o r e 9 / 1 1 ( P A ) A f t e r 9 / 1 1 ( P A ) B e f o r e 9 / 1 1 ( T X ) A f t e r 9 / 1 1 ( T X ) B e f o r e 9 / 1 1 ( O H ) A f t e r 9 / 1 1 ( O H )Location and Pre-Post 9/11 0.000 0.040 0.080 0.120 Figure 5-14. Effect of Urban/Rural origin a nd 9-11 Event on the probability of taking a pleasure trip to any of the six top destinations Seasonality is one of the sets of variable s that had a significa nt variation among the different destinations during the different seas ons. For the state of California, which was the destination that ranked as the first choice when taking a pleasure trip, the peak months where the likelihood went up were May and September, a nd the lower demand was for the months of October, November, December and February. For the state of Florida, the tourism demand for this destination rose precisely for the winter se ason (months of December, January, February and March) and went down for the summer season (May June, July and August). For the state of New York, the likelihood of taking a pleasure trip to this destination stayed close to the mean for the majority of the months, having the higher de mand for the months of September and October. For Pennsylvania, the demand also stayed close to the mean during the major part of the year, having its lowest points in the m onths of September and October. The state of Texas also had the higher demand for tourism during the winter se ason (December, January, and February) and the lowest points for the summer (May to August). The demand for tourism for the state of Ohio also found its lowest points for winter (December to March).
75 75 JanFebMarAprMayJunJulAugSepOctNovDec 0.000 0.040 0.080 0.120 0.160 Average probability by state and season CA JanFebMarAprMayJunJulAugSepOctNovDec 0.000 0.040 0.080 0.120 0.160 Average probability by state and season FL JanFebMarAprMayJunJulAugSepOctNovDec 0.000 0.040 0.080 0.120 0.160 Average probability by state and season NY JanFebMarAprMayJunJulAugSepOctNovDec 0.000 0.040 0.080 0.120 0.160 Average probability by state and season PA JanFebMarAprMayJunJulAugSepOctNovDec 0.000 0.040 0.080 0.120 0.160 Average probability by state and season TX JanFebMarAprMayJunJulAugSepOctNovDec 0.000 0.040 0.080 0.120 0.160 Average probability by state and season OH Figure 5-15. Effect of Seasonality on the probability of taking a pl easure trip to any of the six top destinations The last set of variables that we analyzed was the Region categor y. First, it is important to note that each one of the six top destinations ranked as the firs t option for the region where they are in when selecting a destination for a pleasure trip (with the exception of Pennsylvania, given that New York and Pennsylvania are located in the same geographic region), which showed the preference of the respondents for traveling closer to their home town. It was of particular importance the case of region 9 (Pacific), wh ich showed California as the only important destination for a pleasure trip, the other impor tant case was region seven (West South Central), which only showed Texas as the important dest ination when making a pl easure trip and also
76 76 showed a slight importance for Florida as a dest ination selection. New York ranked as the first destination choice for the respondents from re gions one (New England) and two (Middle Atlantic). Florida ranked as the first destination choi ce for respondents from regions five (South Atlantic) and six (East South Ce ntral) and ranked second for regions one (New England) and three (East North Central). N e w E n g l a n d M i d d l e A t l a n t i c E a s t N o r t h C e n t r a l W e s t N o r t h C e n t r a l S o u t h A t l a n t i c E a s t S o u t h C e n t r a l W e s t S ou t h C e n t r a l M o u n t a i n P a c i f i c0.000 0.200 0.400 0.600 Probability of traveling for pleasure for orgin to: CA N e w E n g l a n d M i d d l e A t l a n t i c E a s t N o r t h C e n t r a l W e s t N o r t h C e n t r a l S o u t h A t l a n t i c E a s t S o u t h C e n t r a l W e s t S ou t h C e n t r a l M o u n t a i n P a c i f i c0.000 0.200 0.400 0.600 Probability of traveling for pleasure for orgin to:FL N e w E n g l a n d M i d d l e A t l a n t i c E a s t N o r t h C e n t r a l W e s t N o r t h C e n t r a l S o u t h A t l a n t i c E a s t S o u t h C e n t r a l W e s t S ou t h C e n t r a l M o u n t a i n P a c i f i c0.000 0.200 0.400 0.600 Probability of traveling for pleasure for orgin to:NY N e w E n g l a n d M i d d l e A t l a n t i c E a s t N o r t h C e n t r a l W e s t N o r t h C e n t r a l S o u t h A t l a n t i c E a s t S o u t h C e n t r a l W e s t S ou t h C e n t r a l M o u n t a i n P a c i f i c0.000 0.200 0.400 0.600 Probability of traveling for pleasure for orgin to:PA N e w E n g l a n d M i d d l e A t l a n t i c E a s t N o r t h C e n t r a l W e s t N o r t h C e n t r a l S o u t h A t l a n t i c E a s t S o u t h C e n t r a l W e s t S ou t h C e n t r a l M o u n t a i n P a c i f i c0.000 0.200 0.400 0.600 Probability of traveling for pleasure for orgin to:TX N e w E n g l a n d M i d d l e A t l a n t i c E a s t N o r t h C e n t r a l W e s t N o r t h C e n t r a l S o u t h A t l a n t i c E a s t S o u t h C e n t r a l W e s t S ou t h C e n t r a l M o u n t a i n P a c i f i c0.000 0.200 0.400 0.600 Probability of traveling for pleasure for orgin to:OH Figure 5-16. Effect of Region on the probability of taking a pleasure trip to any of the six top destinations
77 77 Since this research study focused on Florida tourism demand, the last approach that we performed was to analyze the change in the probability of taking a pleasure trip for this destination, using the estimates from the Mul tinomial Logit Model (selecting only Florida). These estimates were incorporated into a simulation procedure (similar to the analysis performed for the simulation of the Probit model in the first s ection of this chapter) to analyze the effect of each set of variables on the probability of taki ng a long distance pleasure trip. Then, we ranked the effects of the different groups of explanator y variables. The method of ranking used on this section was the same used on the first section of this chapter. Figure 5.17 shows the ranking of the effect of each set of variables for those re spondents who took one pleasure trip to Florida during the assigned period. 0.24 0.09 0.05 0.03 0.02 0.02 0.01 0.00 0.00 Probability range ranked 0.000.100.200.30 Probability range Region Race Season Urban Sept 11 Age Income Employment Educ Ranking the effects of variables influence the probability of traveling to Florida 0.000.040.080.184.108.40.206.28 Probality of traveling for pleasure Figure 5-17. Ranking of the effects of variables on the probability of taking a pleasure trip (FL)
78 78 From the results shown in Figure 5-17 we obs erve that the variable Region was ranked on the top bar, which means that this variable had th e largest effect when deciding for a long distance pleasure trip to Flor ida during the assigned period. The change among regions caused the variation of the probability by 23.81 percent points. The next im portant factor that influenced the likelihood of taking a pleasure trip to Flor ida was race, which among categories can cause a change of 8.9 percent points in the estimated probability, showing that different race categories had different behaviors when sele cting Florida as a destination fo r a pleasure trip. Third, Season (which also ranked third when ranking the vari ables that affected the likelihood of taking a pleasure trip to any destination in the country) wa s also an important fact or since it changed the estimated probability by 5.2 percen t points. Fourth, urban or rural origin of the respondent was also an important factor to consider when examining tourism demand for Florida since it changed the value by 2.6% percent points. The la st three variables that also had a relative importance for analyzing tourism demand for the st ate of Florida were the 9/11 event that can change the estimated probability by 2.17% points, Age of the respondent that could change the likelihood of taking a pleasure trip by 1.8 per cent points, and Income (1.17 percent points change). Employment and Education did not have a substabtive impact in the change of the likelihood of taking a long distan ce pleasure trip to Florida co mpared to other explanatory variables. Also, an analysis of the change in the proba bilities of taking a pleasure trip to Florida across the same type of variable s (each set of explanatory variab les) was conducted. The average probability of taking a long distan ce pleasure trip was calculated with the MNLM estimates for Florida, this value was calculated by using th e average value of each variable. Then, while holding the others variables constant at the average value, the vari able of interest was simulated
79 79 among the different values of the category. On av erage, the probability of taking a long distance pleasure trip to Florida during a month period was 0.06205. For income category, the simulated effects of the different income levels are presented in Figure 5-18. The likelihood of ta king a pleasure trip to Florida decreased for income category two (between $25,000 and $50,000), three (betw een $50,000 and $75,000) and four (between $75,000 and $100,000) from an average of 6.21% to 5.90%, 5.92% and 5.82%respectively; while its bigger increment was for income category five from an average of 6.21% to 7%. 0.062 0.065 0.059 0.059 0.058 0.070 Average $24,999 or less $25,000 to $49,999 $50,000 to $74,999 $75,000 to $99,999 $100,000 or more Income levels 0.0000.0250.0500.0750.100 Probability of traveling to Florida Figure 5-18. Proportion of variation in relati on to the average by income (for Florida) For Age categories, Figure 5-19 shows that there was an increment in the likelihood of taking a long distance pleasure tr ip to Florida compared to th e average for age group one (< 25 years old) changing the probability from an average of 6.21% to 6.33% and for age group number five going up from 6.21% to 7.43%, which shows that the younger population and the older population were the most lik ely to travel to Florida fo r pleasure. The other two age categories had a negative variation.
80 80 0.062 0.063 0.056 0.061 0.074 Average Under 25 year 25 years to 44 years 45 year to 65 years 65 years and over Age groups 0.0000.0250.0500.0750.100 Probability of traveling to Florida Figure 5-19. Proportion of variation in rela tion to the average by Age (for Florida) For Race category, Figure 5-20 shows that only for black respondents, there was a significant decrease in the probabi lity of taking a pleas ure trip to Florida from an average of 6.21% to 2.66%. For white respondents, the probab ility stayed almost the same as the average. For the other three categories, the probability went up (11.13% for Hispanics, 11.56% for White/Hispanics, and 9% for all other races). 0.062 0.062 0.027 0.111 0.116 0.090 Average White (=1) Black (=2) Hispanic (=3) White/Hispanic (=4) All others (=5) Race 0.0000.0500.1000.150 Probability of traveling to Florida Figure 5-20. Proportion of variation in rela tion to the average by Race (for Florida)
81 81 When analyzing the change in probability for the variable College E ducation, the variation for both responses remained very close to average, not changing in a considerable percentage the probability of taking a pleasure trip to Florida. For the variable employment status, the likelihood of taking a pleasure trip to Florida remained almost constant for both categories (which contrasted with the analysis of the genera l population for which employment status was a variable that caused a significan t variation in the probabilities). 0.062 0.062 0.062 0.063 0.061 Average College Education No College Education Employed No Employed Education and Employment 0.0000.0250.0500.0750.100 Probability of traveling to Florida Figure 5-21. Proportion of variation in relati on to the average by Education and Employment (for Florida) For Urban variable, the likelihood of taking a pl easure trip to Florida decreased from an average of 6.21% to 4.35% for those respondents living in rural areas and increased for those respondents living in urban areas (from 6.21 to 6.92%). For the e ffect on the likelihood of taking a pleasure trip to Florida after the 9/11 even ts, Figure 5-22 shows that the variation on the average probability for those respondents who trav eled before the 9/11event was higher than for those who traveled after the event.
82 82 0.062 0.069 0.044 0.077 0.056 Average Urban Rural Before 9/11 After 9/11 Location and Pre-Post 9/11 0.0000.0250.0500.0750.10 0 Probability of traveling to Florida Figure 5-22. Proportion of variation in relation to the average by Urban/Rural origin and 9-11 event (for Florida) For the Seasonality effect in the variation from the average probability of taking a pleasure trip to Florida, Figure 5-23 shows the vari ation among the different months. The bigger increments from the average occurred for the months of December, January, February and March, which enforced our belief that Florida wa s a preferred destination for the winter season, given its warmer temperatures compared to othe r regions og the country. For the month of April it started to decrease. On the other hand, fo r the months of May, June, July, August and September the probability of taking a pleasure tr ip to Florida decrease d substantially, and it started to increase again for the months of October and November.
83 83 0.062 0.087 0.085 0.084 0.062 0.035 0.041 0.044 0.038 0.046 0.070 0.058 0.072 Average January February March April May June July August September October November December Seasonality 0.0000.0250.0500.0750.100 Probability of traveling to Florida Figure 5-23. Proportion of variation in relati on to the average by Season (for Florida) 0.062 0.045 0.058 0.025 0.011 0.243 0.072 0.019 0.005 0.005 Average New England (1) Middle Atlantic (2) East North Central (3) West North Central (4) South Atlantic (5) East South Central (6) West South Central (7) Mountain (8) Pacific (9) Regions 0.0000.1000.2000.300 Probability of traveling to Florida Figure 5-24. Proportion of variation in relati on to the average by Re gion (for Florida)
84 84 Finally, Figure 5-24 shows the variation from the average on the likelihood of traveling to Florida for pleasure for the different regions of origin of the respondents. Region five (South Atlantic) was the region that showed the biggest increment on the probability, going from an average of 6.21% to 24.30%, which was reasonable since the state of Florida is located in this region. In contrast, regions 3( East North Central), 4(West North Central), 7 (West South Central), 8 (Mountain) and 9 (P acific) showed an important decrement from the average probability, which was also reasonable since those were distant regions from Florida.
85 CHAPTER 6 SUMMARY AND CONCLUSIONS Tourism is one of the fastest growing industrie s in the world that accounts for nearly 11% of the worlds gross domestic product, therefor e explaining the importance of studying this industrys behavior. This resear ch project focused on analyzi ng and understanding the factors that affect the tourism demand. The main object ive was to identify the significant factors that influenced the probability of taking long distan ce pleasure trips among th e general population in the country as a general approach and also for Florida as a part icular case. Policy makers and tourism organizations could use the obtained ba se model and information to develop market strategies for further growth in the sector. Results from this research provide information necessary to analyze factors influe ncing a tourists decision to take a pleasure trip, to establish a rank of the variables driving this response, and to compare the va riation of the effect of the variables among different destination choices. Data used in this resear ch were provided by the 2001 National Household Travel Survey (NHTS), which recorded about 45,000 usable responses. Households completing the survey provided responses about very specific demographic, geographic and other general queries including among those if they took or not a long distance pleasure trip during the assigned recording period (4 weeks). As a first step, using Probit model specification, selected variables were incorporated to estimate the impact of each set of variables on th e likelihood of taking a pleasure trip. From this analysis, with exception of the 9/11 effects and e ducation level, the majority of the variables included in the model were statistically signifi cant. Simulation analyses were performed to evaluate the impact of the different variables on the likelihood of taking a pleasure trip relative to the average household. When ranking the variati on in the probability caused by each set of variables, the 9 /11 effects and e ducation level also ranked at the bottom. In contrast employment
86 status, age, season and region of respondents were the variables th at caused the biggest variation in probability of taking a long distance pleasure trip. Second, an analysis among the top six competi ng destinations when taking a pleasure trip was conducted; thus, a Multinomial Logit Model was estimated. Then, simulations analyses were performed to evaluate the impact across states of the different variables. For the average consumer, the likelihood of taking a pleasure trip to California was 9.48 percent; to Florida, 6.21 percent; to Texas, 5.31 percent; to New York, 4. 81 percent; to Pennsylvania, 4.30 percent; and to Ohio, 3.30 percent. For each set of variables included in the model, a comparison of the variation on trends and probabilities of traveling for pleasure among the six states was conducted. Finally (since Florida was the state of main interest for our re search) selecting the estimates of the Multinomial Logit Model for this destination, an analysis of the variation from the average of the probability of taking a pleasure trip wa s performed by each category of variables included in the model. A ranking of the effect of each set of variables was presented, showing that the importance of the variables for th is specific destination varies from the general population. In this case region, race and season we re the variables that caused major variations from the average of the probability, while employment and education level were not significant and ranked at the bottom. From the results found in this research, a cl earer understanding of th e important variables that drive tourism demand for the country and fo r Florida was obtained. This can be used for different purposes like targeti ng population when conducting advert ising programs or developing appropriate policies for the sector. This study found that region of origin of the respondent highly infl uenced the choice of destination when deciding to take a pleasure trip, being they most likely to prefer a destination
87 that is closer to their point of departure. Al so, the race of the respondent had an important role when selecting a travel destination. Another fi nding was that choice of de stination for a pleasure trip widely varied with seasonality, for example for Florida and Texas winter was the season that increases the demand for tourism, while for Calif ornia and Ohio this season presented a lower level of demand. The 9/11 event did not seem to have had a relevant impact on the variation in demand for domestic tourism in the country; in contrast when analyzing the effect of this variable for specific destinations like California, Florida and Ohio it was significant, though the effect of this variable could not be clearly studi ed since only data from one year was available and this variable could be affected by seasona lity. Income category had only a small impact on the likelihood of taking a pleasure trip and educat ion level of the respondent was not significant.
APPENDIX A SIMULATED PROBABILITIES (A)
89 SIMNUM VAR# PURPOSEPROBIT PL EASUREP(OTHERS)Ca Fl NY Penn Tx Oh 1 1 1 20.6896071570.6658852640.0948 1 0.0620560.0480560.0432380.053070.032884 2 Income 2 1 20.6788008890.6787585460. 087394 0.0648430.0501280.0452070.0456070.028062 3 2 2 20.6932397620.6749785210.0802 42 0.0589940.0481590.043530.0577050.036392 4 2 3 20.6947585530.6467210740.0994 17 0.0592130.0568010.0468510.0560690.034928 5 2 4 20.7049179820.6657904530.0952 52 0.0582350.0412720.0508110.0539750.034663 6 2 5 20.6741515040.6686228520.1118 28 0.0699370.0432240.0306090.0492460.026533 7 Age 3 1 20.768181580.6444310710.0972 02 0.0632990.0566270.0480730.0536270.03674 8 3 2 20.6598978720.6789634660.0948 51 0.0561430.0489020.0402370.0460210.034883 9 3 3 20.6552625590.6735135430.0945 32 0.0611320.0471550.0416890.0529790.028999 10 3 4 20.6684616740.6641778380.0987 19 0.0743030.0311360.0383340.0611850.032146 11 Race 4 1 20.692794160.6741176260.0859 23 0.0617020.0487780.044790.0513710.033319 12 4 2 20.6365614650.6537761440.1524 11 0.0265570.0450120.0446310.0270480.050566 13 4 3 20.688093440.563978220.150801 0.1113570.0665330.0225540.0847765.84E-10 14 4 4 20.6833103420.6070260810.1208 09 0.11560.0328450.0346330.0636160.025472 15 4 5 20.6834948390.4652062180.1077 69 0.0902670.0193250.0179130.0642770.235242 16 Educ 5 1 20.6892026380.6591742110.0983 91 0.0621360.0529030.0414240.0519920.03398 17 5 2 20.6903333250.6802297510.0864 15 0.0619360.0388910.0464970.0547540.031277 18 Employment 6 1 20.6320831720.659166716 0.09518 0.0626770.0494220.0457110.0537630.034079 19 6 2 20.7641750460.6725316210.0967 62 0.0614970.0468280.0389640.05110.032316 20 Urban 7 1 20.698582890.6524627330.1007 77 0.0691680.0470580.0446170.0563080.02961 21 7 2 20.6642832560.7167335760.0637 5 0.043520.0516060.0390380.0449040.040448 22 11-Sep 8 1 20.6909812060.6726698690. 080175 0.077350.0450.0421180.0580310.024657 23 8 2 20.6887519540.6552031290.1048 09 0.0556490.0495830.0436580.0504270.040672
90 APPENDIX B SIMULATED PROBABILITIES (B)
91 SIMNUM VAR# PURPOSEPROBIT PL EASUREP(OTHERS)Ca Fl NY Penn Tx Oh 24 Season 9 1 20.6538349590.6352062960. 101173 0.0871280.040890.038370.0678550.029378 25 9 2 20.6600112770.6801740040.0705 99 0.0847620.0389440.0373150.0664250.02178 26 9 3 20.6719018580.6506459290.0974 94 0.0837880.0501250.0379370.0522540.027756 27 9 4 20.6794858040.6308266010.1007 29 0.0623630.0553650.0486720.0537870.048256 28 9 5 20.6419468760.6685788030.1186 81 0.0354360.0447880.0472230.0375060.047786 29 9 6 20.6930572130.6658866640.1049 07 0.0411530.0520640.0447840.0479510.043254 30 9 7 20.7166648840.6790182550.0954 96 0.0443990.0487790.0497630.0364730.046071 31 9 8 20.6998277930.707759610.0924 44 0.0377160.0375680.0495140.0398070.035191 32 9 9 20.6874393920.6660098330.1158 15 0.0458940.0605730.033210.0472790.031218 33 9 10 20.6952549330.6492430170.0897 12 0.0698450.0677440.0267820.0688770.027797 34 9 11 20.7324644680.6811832560.0898 51 0.058480.0479780.0484060.0467750.027327 35 9 12 20.7461404690.6632501010.0828 55 0.0719930.0451870.0476310.0597170.029368 36 Region 10 1 20.7181822450.8147177360.0086 69 0.0448260.1093890.0190170.0022810.001101 37 10 2 20.6980577090.4028121010.0068 01 0.0576470.2778580.2283490.0040810.022453 38 10 3 20.7094143480.7893768140.0128 42 0.0246950.0082890.0205240.0034780.140795 39 10 4 20.6745802910.9427635030.0186 5 0.0111390.0026490.0020.0183080.004491 40 10 5 20.6873602710.6743421140.0063 0.2430650.0210020.0406850.0043870.010219 41 10 6 20.6245875570.8951963120.0034 2 0.0723030.0072140.0014260.004740.015701 42 10 7 20.6725808750.4615155730.0046 86 0.0190190.0014410.0014680.5083280.003543 43 10 8 20.6973426160.8999225970.0726 87 0.0049330.0049850.0060040.0096950.001774 44 10 9 20.6858587420.4850829240.5013 08 0.0051670.0034720.0030840.0013290.000557
92 LIST OF REFERENCES Agresti A. 2002. Categorical Data Analysis, 2nd edn. John Wiley & Sons, Inc: Hoboken, NJ. Archer B.1994. Demand Forecasting and Estimation. In Travel, Tourism, and Hospitality Research: A Handbook for Managers and Researchers, 2nd edn, Edited by Ritchie JRB, Goeldner CR (eds). Wiley and Sons: New York; 105-114. Bigano A, Hamilton J, Lau M, Tol R, Zhou Y. 2007. A Global Database of Domestic and International Tourist Numbers at National and Subnational Level. International Journal of Tourism Research 9:147-174. Eugenio-Martin J. 2003. Modeling Determinants of Tourism Demand As A Five-Stage Process: A Discrete Choice Methodological Approach. Tourism and Hospitality Research. 4(4): 341-354. Frechtling D. 2001. Forecasting Tourism Demand: Methods and Strategies. ButterworthHeinemann: Jordan Hill, Oxford. Latham J, Edwards C. 1989. The Sta tistical Measurement of Tourism. Progress in Tourism, Recreation and Hospitality Management. 1: 55-76. Lim C. 1997. Review of Intern ational Tourism Demand Models. Annals of Tourism Research. 24(4):835-849. Lise W, Tol R. 2002. Impact of Climate on Tourist Demand. Climatic Change 55:429-449. Makridakis S, Winkler R. 1983. Averag es of Forecasts: Empirical Results. Management Science 29: 987. March R, Woodside A. 2005. Tourism Behaviour: Travellers Decisions and Actions. Cabi Publishing: Cambridge, Ma. McIntosh R, Goeldner C, Ritchie J. 1995. Tourism Principles, Practices, Philosophies, 7th edn John Wiley and Sons, Inc: New York, NY. Sinclair T, Blake A, Sugiyarto G. 1989. The Economics of Tourism. Progress In Tourism, Recreational and Hospitality Management 1: 1-27. Soocheong J, Cai L, Morrison A, OLeary J. 2005. The Effects of Travel Activities and Seasons on Expenditure. International Journal of Tourism Research 7:335-346. Stynes D. 1999. Economic Impacts of Tourism. Department of Park, Recreation & Tourism Resources, Michigan State University. East Lansing, MI. Taylor P. 2006. Getting Them to Forgive and Fo rget: Cognitive Based Marketing Responses to Terrorist Acts. International Journal of Tourism Research 8: 171-183.
93 Witt S, Witt C. 1995. Forecasting Tourism De mand: A Review of Empirical Research. International Journal of Forecasting 11: 447-475. Woodside A, Dubelaar C. 2002. A General Th eory of Tourism Consumption Systems: A Conceptual Framework and an Empirical Exploration. Journal of Travel Research 41: 120-132. Woodside A, King R. 2001. Tourism Consumption Systems: Theory and Empirical Research. Journal of Travel and Tourism Research 10(1):3-27.
94 BIOGRAPHICAL SKETCH Mayra Gissella Villacis was born in 1982 in the city of Guayaquil, Ecuador. The second of four children, she grew up in Guayaquil too. Sh e earned her Bachelor of Science degree in food processing engineering from the Escuela Superi or Politecnica del Litoral in Ecuador in 2005, minoring in quality assurance. After graduati on, she gained some experience working as a laboratory analyst for the department of Quality Control of UNICOL S.A. In 2006, she entered the Master of Scien ce program at the Department of Food and Resource Economics at the University of Florid a. Her research interests focused on applied economics for agricultural markets.