Expenditure decisions of international tourists to Taiwan: application of the Heckman two-stage approach

Abstract

Based on a large sample of 30,945 international tourists to Taiwan from 2016 to 2021, this study examines how these tourists make expenditure decisions in Tawan. Simultaneously, we consider various determinants such as travel characteristics, individual characteristics, and satisfaction levels, and the COVID-19 pandemic influence tourists’ decisions regarding their expenditures on accommodation, food, transportation, shopping, and other expenses. Employing a two-stage Heckman approach, the study finds that satisfaction positively affects spending in various categories, highlighting the importance of enhancing visitor satisfaction. Tourists traveling shorter distances allocate more to accommodation, food, and shopping, while repeat visitors prioritize transportation, accommodation, and shopping. Longer stays result in higher transportation and other expenses, while shorter stays lead to increased spending on accommodation, food, and shopping. Gender, age, education, and income levels also influence spending patterns. Moreover, the COVID-19 pandemic disrupted spending, reducing accommodation and transportation expenses but increasing food and shopping spending as tourists adapted to new standards.

Keywords:

• Tourist
• expenditure
• decision
• travel characteristic
• socioeconomic characteristic
• heckman two-stage

Reviewing editor:

• Pier Luigi Sacco
• Humanities
• IULM University
• Milano
• Italy

SUBJECTS:

• Economics
• Hospitality
• Tourism

1. Introduction

Tourism is a crucial industry for many countries, contributing significantly to their economies. Understanding the purchasing habits of international tourists is critical for maximizing the economic benefits of tourism (Kim et al., 2006; Lee & Chang, 2008; Lyu & Noh, 2017). Various determinants (such as trip-related attributes, financial constraints, socio-demographic characteristics, and psychographic factors) influence the spending behavior of international tourists (Brida & Scuderi, 2013; Marrocu et al., 2015; Yang et al., 2019). Some previous studies, as indicated by studies conducted by Marcussen (2011), Massidda et al. (2020), and Yang et al. (2019), discovered that trip-related attributes (e.g., the purpose of travel, choice of accommodation, destination, duration of stay, group size, and transportation mode) significantly impact spending. These factors play a significant role in determining tourist expenditures. Additionally, financial constraints, such as income and education levels, can also influence spending decisions. Higher-income travelers tend to spend more on family well-being and to visit multiple destinations, whilse those with higher levels of education tend to spend more on travel activities (Sahoo et al., 2022; Wu et al., 2013).

Socio-demographic characteristics like gender, nationality, occupation, and family background can also influence travel choices. For instance, male-headed households may be more likely to spend on tourism than female-headed households (Sahoo et al., 2022). Age is also important, with older tourists spending more across different categories due to their higher disposable income and established travel habits. Generation Z travelers prioritize ethical and eco-friendly food choices for personal and social well-being (Orea-Giner & Fusté-Forné, 2023). Recent research by Hall et al. (2020) and Chen et al. (2022) shows that stricter border controls during the COVID-19 pandemic have led to extended stays but reduced spending on food, transportation, entertainment, and shopping. Understanding these pandemic-related shifts in spending patterns is essential for adapting to the changing dynamics of tourism.

International tourist spending behavior is influenced by various factors, making it a complex process. Spending on travel products often involves multiple stages, starting with the decision to participate in specific product spending, such as accommodation, food, transportation, shopping, and other expenses (Stynes & White, 2006). Subsequently, the tourists determine the amount they want to spend on each product. This study is based on the concept that this two-stage decision-making process can provide us with a better understanding of the factors that influence both the decision to participate and the amount spent on each travel product. We employ the Heckman sample selection approach to understand this decision-making process better (Chen & Ho, 2022; Lin et al., 2021; Lyu & Noh, 2017). The Heckman approach (Heckman, 1979) addresses potential sample selection issues and allows researchers to examine the factors influencing participation and spending on each product. By breaking down tourist spending behavior into two stages, this approach provides a more accurate understanding of what affects tourists’ spending decisions. It is essential to consider potential biases to understand tourist spending behavior better.

Our study aims to provide valuable insights into the spending habits of international tourists in Taiwan. We have conducted extensive research and examined a range of determinants that could influence their spending patterns to enable destinations to develop targeted strategies that will attract more tourists and strengthen their economies. During our research, we have considered various factors that could impact the spending behaviors of tourists. This includes analyzing satisfaction levels, travel-related factors, personal characteristics, and the effects of the COVID-19 pandemic. Our objective is to determine the most significant factors that drive spending behavior among international visitors to Taiwan by analyzing the correlations between these variables. Moreover, our findings will enable us to compare pre- and post-COVID-19 pandemic data to determine how spending patterns and relationships have changed in response to the pandemic’s impact on travel and tourism.

2. Literature review

Previous studies (Brida & Scuderi, 2013; Kieu Thi et al., 2023; Marrocu et al., 2015; Yang et al., 2019) have identified four key factors that influence the spending behavior of tourists. These include trip-related details, financial limitations, personal characteristics, and psychographic factors. Trip-related attributes encompass factors such as the purpose of the trip, choice of accommodation, destination, length of stay, group size, and mode of transportation. Financial factors also play a significant role in determining how much tourists can spend on their trips. Additionally, personal attributes (such as age, education, gender, nationality, occupation, and family background) may impact travel decisions. Finally, psychographic factors (such as tourists’ attitudes, behaviors, and motivations toward travel) are also important considerations.

Firstly, the trip-related dimension is an essential factor in determining travel expenditures, as indicated by studies conducted by Marcussen (2011), Massidda et al. (2020), and Yang et al. (2019). Marcussen (2011) surveyed 11,077 respondents and discovered that characteristics such as the type of accommodation, mode of transportation, length of stay, travel distance, the purpose of travel, age group, income level, and the traveler’s first visit all contribute to increased tourism spending. In addition, the choice of accommodation is significantly influenced by the economic impact of tourist expenditures, as highlighted in a recent study by Sahoo et al. (2022). Brida and Scuderi (2013) investigated spending patterns at the most popular Christmas markets in Northern Italy between 2008 and 2009. Their findings illustrate that trip-related factors such as the purpose of travel, place of origin, perception of the event, length of stay, and age play a critical role in determining visitor spending, as evidenced by the results of applying the double-hurdle model. According to Thrane and Farstad (2011), a tourist’s total expenses can be divided into two factors: daily spending and duration of stay. This relationship establishes a connection between a tourist’s personal characteristics, daily spending habits, and length of stay. Based on theoretical observations, it has been noted that there is a positive yet diminishing correlation between the length of a tourist’s stay and the amount of money spent on tourism.

Moreover, the mode of transportation plays a critical role in personal tourism expenditures, making the connection between the starting point and the destination an essential part of a tourist’s decision-making process (Thrane & Farstad, 2011). Masiero and Zoltan (2013) found that transportation choices and travel patterns are closely correlated, indicating that tourists who opt for private modes of transportation and visit multiple destinations are positively correlated. Hough and Hassanien (2010) examined the transportation preferences of Chinese and Australian tourists visiting Scotland. Their results indicated significant differences between the two nations, highlighting the significance of origin alongside socio-demographic factors such as education, language, and previous travel experience. To further expand upon these findings, the researchers recommended examining the influence of activities and holiday expenditures on mode of transportation selection in forthcoming examinations.

Next, in terms of the individual financial constraints dimension, tourists expenditures are significantly influenced by income, as people with higher incomes tend to prioritize spending on their family’s well-being and visit multiple destinations as part of their tourism package (Gómez-Déniz et al., 2019; Park et al., 2019; Yang et al., 2019). At the same time, tourists with higher education tend to spend more money on travel activities, suggesting a positive association between visitors’ education and tourism expenditure (Sahoo et al., 2022; Wu et al., 2013). Interestingly, Sahoo et al. (2022) discovered that male-headed households demonstrated a greater propensity for tourism spending than their female-headed counterparts, suggesting a potential gender bias in resource allocation. While tourism may be perceived as an amenity for families in the lower quantile of spending, those in the higher quantiles view it as a necessity.

Regarding the socio-demographic and psychographic characteristics, food tourism is rising, with travelers increasingly valuing unique culinary experiences and destination choices (Andersson et al., 2017; Knollenberg et al., 2021; Orea-Giner & Fusté-Forné, 2023). Therefore, tourists increasingly pay more money for enjoyable food and beverage experiences (Okumus, 2021; Orea-Giner & Fusté-Forné, 2023; Rong-Da Liang et al., 2013). Orea-Giner and Fusté-Forné (2023) has demonstrated that Generation Z travelers are conscious of the impact of ethical and eco-friendly food choices on their overall personal and social well-being. However, financial and time limitations constrain their ability to consume sustainably. Prior research by Boluk et al. (2021) suggests that local food consumption can significantly preserve cultural traditions and agricultural practices. Local cuisine both attracts tourists and motivates tourism (Rong-Da Liang et al., 2013).

Besides the food and beverage dimensions, shopping expenditure emerges as a favored recreational activity among tourists, offering economic, psychological, and social benefits (Park et al., 2019; Yüksel & Yüksel, 2007). Lehto et al. (2004) examined the shopping expenses of Taiwanese tourists based on their socio-demographic details and trip features. The study found that the most significant spenders among tourists interested in shopping were women in their twenties who traveled on leisure package tours with companions. Many tourists spend a significant amount of money shopping during their travels (Bojanic, 2011; Law & Au, 2000). For instance, Mexican visitors in south-central Texas allocate over 50% of their budget to shopping, a trend also observed among Hong Kong tourists. Lyu and Noh (2017) employed the Heckman sample selection model for an investigation into the shopping patterns of global travelers. The results revealed that international tourists tend to make their buying decisions before deciding on the amount they are willing to spend on a product. Female tourists exhibited a distinct consumption pattern from males by spending less on purchasing industrial products such as electronic appliances. Previous research studies have provided evidence that the gender of tourists plays a pivotal role in explaining tourism spending (Lyu & Noh, 2017; Murphy et al., 2011). Marcussen (2011) demonstrated that women tend to spend more and engage in shopping activities during their travels. Jansen-Verbeke (1987) also identified a higher level of shopping expenditure among middle-aged women compared to younger and older women. Oh et al. (2004) discovered that gender, age, and travel objectives influence tourists’ shopping preferences. Their study revealed that female tourists exhibit a broader range of interests than males, while older travelers (aged 51–60) are more inclined towards local handicrafts and antiques. Female tourists are also more likely to explore and engage in various experiences, such as shopping and sampling local cuisine, during their trips.

In this study, we further analyze the impact of the COVID-19 pandemic on individual tourist’s spending habits. Recent research by Hall et al. (2020) and Chen et al. (2022) has examined the COVID-19 pandemic’s impact on tourist expenditures. Chen et al. (2022) notes that stricter border control policies have led to more extended stays but reduced spending on food, transportation, entertainment, and shopping. Similarly, Hall et al. (2020) found that consumer spending among foreign visitors to the Canterbury region of New Zealand experienced a significant drop during the government-mandated lockdown.

3. Materials and methods

3.1. Sampling and data collection

The data used in this study were collected from the Taiwan Tourism Bureau (TTB) (https://admin.taiwan.net.tw). This survey was created to understand the intentions, attitudes, consumption habits, and trends of international visitors to Taiwan. We employed cross-sectional data from 2016 to 2021, a final sample of 30,945 observations, to determine tourist spending. Appendix A provides the shortened questionnaire used for data collection, and Table 1 provides a detailed description of the variables. The data report represents the visitors’ spending on travel products, such as accommodation, food, transportation, shopping, and other expenses (Stynes & White, 2006). Each product’s cost is divided by the stay days and co-consumer numbers on the trip to normalize the results. This standardization provides a uniform comparison of expenditures across trips and group sizes. In our sample, short distances are Japan, China, Korea, Thailand, Vietnam, and the Philippines; medium distances are Malaysia, Singapore, and Indonesia; and long distances are the United States of America, Canada, and Europe.

Table 1. Descriptive statistics.

Variables	Description	Coding values	Mean	S.D.^a	%^b
Dependent variable
The first-stage probit model	Whether a tourist spends money on tourism activities or products. Dummy variables present “Yes” or “No”	Yes = 1; otherwise = 0
The second-stage OLS regression model	A natural logarithm of tourism expenditures
Exp.Accommodation	Spending on hotels	USD^c/day	60.52	154.71	–
Exp.Food	Spending on meals and beverages	USD/day	37.86	49.32	–
Exp.Transportation	Spending on transport-related charges	USD/day	18.94	30.33	–
Exp.Shopping	Spending on purchasing souvenirs, clothes, cosmetics, jewelry, and others	USD/day	41.67	114.01	–
Exp.Other expenses	Spending on entertainment (cinema, nightlife, sports) and other expenses	USD/day	12.36	67.29	–
Independent variables
Satisfaction1	Overall satisfaction level	Very poor = 1; otherwise = 0	–	–	2.16
Satisfaction2		Poor = 1; otherwise = 0	–	–	0.62
Satisfaction3		Fair = 1; otherwise = 0	–	–	5.49
Satisfaction4		Good = 1; otherwise = 0	–	–	37.99
Satisfaction5		Excellent = 1; otherwise = 0	–	–	53.74
Distance	Distance from home to a travel destination	Short distance =1; otherwise = 0	–	–	66.16
Visit Times	Number of visits to Taiwan	First time = 1, repeat time = 0	–	–	45.94
Stay days	Stay durations	Days	14.60	21.10	–
Plan days	Number of days to make a trip decision	Days	54.98	62.20	–
Gender	Gender	Male = 1, Female = 0	–	–	49.67
Generation X	Age levels	50s =1, otherwise = 0	–	–	18.8
Generation Y		30s and 40s = 1, otherwise = 0	–	–	49.71
Generation Z		10s and 20s = 1, otherwise = 0	–	–	2.23
Education1	Education levels	Under or high school = 1, otherwise = 0	–	–	16.65
Education2		College or university = 1, otherwise = 0	–	–	65.58
Education3		Graduate school = 1, otherwise = 0	–	–	17.75
Income1	Income levels	Under $14,999 = 1, otherwise = 0	–	–	14.03
Income2		$15,000–$29,999 = 1, otherwise = 0	–	–	45.08
Income3		$30,000–$39,999 = 1, otherwise = 0	–	–	12.58
Income4		$40,000–69,999 = 1, otherwise = 0	–	–	13.92
Income5		Over 70,000 = 1, otherwise = 0	–	–	14.39
COVID-19	Before COVID-19 from 2016-2019, and during COVID-19 from 2020-2021	PreCOVID-19 = 0, during COVID-19 = 1	–	–	21.67

Generation X refers to those born between 1965 and 1980. Generation Y refers to those born between 1981 and 1996. Generation Z refers to those born between 1997 and 2009.

^aS.D. is the standard deviation.

^bFrequency is determined based on values assigned with 1.

^cUSD (or $) is the currency abbreviation for the United States dollar.

3.2. Heckman sample selection model

When analyzing tourism demand data, zero expenditures can significantly impact analysis and modeling (Chen & Ho, 2022; Lin et al., 2021; Lyu & Noh, 2017). Zero observations occur when individuals do not spend money on a tourism activity or product for various reasons. However, conventional linear regression models like ordinary least squares (OLS) can lead to biased and inconsistent parameter estimates when dealing with zero observations (Lin et al., 2023; Lyu & Noh, 2017; Maddala, 1983). The Tobit model, established by Tobin, (1958), attempts to account for zero expenditures but is considered restrictive. To overcome this limitation, researchers continue to explore and develop more advanced models like the Heckman sample selection model (Heckman, 1979), which takes a two-step approach to capture the sequential decision-making process of individuals. The Heckman model uses a probit regression model in the first step to estimate the probability of participation and a second step to estimate the determinants of consumption decisions for individuals who have decided to participate.

In the first stage of the Heckman model, we estimate a probability model to determine whether tourists choose to participate in a tourism activity or purchase a product. This is done using a probit regression model as follows: (1) $$\mathit{\text{Expenditure}}{s}_i^{\ast }=X_i{\beta }_i+{\epsilon }_i,{\epsilon }_i\sim N\left(0,{\sigma }^2\right)$$(1) $$\left\{\frac{\mathit{\text{Expenditure}}{s}_i^{\ast }=1\stackrel{}{}\stackrel{}{}\mathit{\text{if}}\stackrel{}{}\mathit{\text{Expenditure}}{s}_i>0}{\mathit{\text{Expenditure}}{s}_i^{\ast }=0\stackrel{}{}\stackrel{}{}\mathit{\text{if}}\stackrel{}{}\mathit{\text{Expenditure}}{s}_i\le 0}\right.$$ where Expenditures (including accommodation, food, transportation, shopping, and other expenses) are the total amount spent by the i^th tourist for the j^th product (i, j = 1, 2, …, N), X_i represents the set of independent variables (including satisfaction level, travel and individual characteristics, and other determinants), β_i represents the vector of coefficients with regard to each explanatory factor, and ε_i represents the unobservable error term.

The typical OLS estimation with a smaller sample size inevitably leads to the possibility of sample selection bias in the second stage, which is classified as omitted variable bias (Heckman, 1979). To address this issue, the inverse Mill’s ratio (IMR) is calculated for each respondent during the first-step Probit model estimation. The IMR, or Lambda (λ) statistic, is determined by comparing the values of the standard normal density function and the standard normal cumulative distribution function as follows: (2) $$\mathit{\text{IMR}}\left(\lambda \right)=\frac{\varphi \left(X_i{\beta }_i\right)}{\Phi \left(X_i{\beta }_i\right)}$$(2) where ф denotes the standard normal density function and Φ represents the standard normal cumulative distribution function. The expression for the second stage, based on the equation used in the antecedent phase, is as follows: (3) $${\mathit{\text{Expenditures}}}_i^{*}=X_i{\beta }_i+{\epsilon }_i,{\epsilon }_i\sim N\left(0,{\sigma }^2\right)$$(3) $$\left\{\frac{\mathit{\text{Expenditure}}{s}_i^{\ast }=\mathit{\text{Expenditure}}{s}_i\stackrel{}{}\stackrel{}{}\mathit{\text{if}}\stackrel{}{}\mathit{\text{Expenditure}}{s}_{{}_i}^{\ast }>0}{\mathit{\text{Expenditure}}{s}_i^{\ast }=0\stackrel{}{}\stackrel{}{}\mathit{\text{if}}\stackrel{}{}\mathit{\text{Expenditure}}{s}_{\iota }^{\ast }\le 0}\right.$$

Incorporating the IMR as an extra independent variable in the shorted OLS model computation is a pivotal step in enhancing the accuracy and reliability of the findings (Lyu & Noh, 2017). The statistical significance of the IMR in the model clearly indicates the presence of a sample selection bias, which must be rectified. The Heckman two-stage approach is considered an appropriate procedure for addressing this issue.

As shown in Table 1, the dependent variable for the first-stage probit model was whether a tourist spent money on a tourism activity or product, coded as 1 if yes and 0 if otherwise. The second-stage model’s dependent variable was the tourist’s daily expenditure, including accommodation, food, transport, shopping, and other expenses in Taiwan. Due to its truncated nature, the second-stage model could be dominated by a non-normal error distribution, as highlighted by Mauldin et al. (2001) and Jang and Ham (2009). Therefore, the dependent variables for the second stage are travel expenses, which were standardised as a natural logarithm of travel expenditures. Based on the literature review, the independent variables in this study were categorized into three groups: travel characteristics (e.g., travel distance, visit times, stay durations, plan durations), individual characteristics (e.g., gender, generation, education level, income level), and other determinants (e.g., overall satisfaction level, COVID-19 pandemic).

4. Results

4.1. Descriptive statistics of the sample

Table 1 shows that on average, each tourist spends 171 USD per day. The highest expenses are for accommodation, with an average of 61 USD per day, followed by shopping at 42 USD per day, food at 38 USD per day, and transportation at 19 USD per day. The lowest spending is on other items, with an average of 12 USD per day. Satisfaction levels among tourists were generally high, with 53.74% of respondents rating their experience as excellent and 37.99% rating it as good. Only a tiny percentage of tourists rated their experience as poor or very poor. The majority of tourists were short-haul travelers, with 66% of visitors coming from countries like Japan, China, Korea, Thailand, Vietnam, and the Philippines. Meanwhile, 44% of visitors were medium- and long-haul travelers, with Malaysia, Singapore, Indonesia, the United States, Canada, and Europe among the top source markets. Almost half of all visitors were first-time tourists, comprising 45.94% of the total.

The average length of stay for tourists was 14.6 d, while the average time it took to decide to take a trip was 55 d. There was an almost equal distribution of male and female tourists, with males accounting for 49.67% of visitors and females making up 50.33%. Generation Y was the largest demographic group, with nearly half of all tourists belonging to this age group. Those from Generation X and Z groups tended to travel less, with only 18.8% and 2.23% of tourists coming from these age groups, respectively. In terms of education level, the majority of tourists were college or university graduates, comprising 65.58% of the total. Graduate school alums accounted for 17.75%, while 16.65% of visitors had either an undergraduate or high school degree. Finally, tourists with an average income of $15,000–$29,999 had the highest proportion, comprising 45.08% of the total. Other income groups showed no significant difference, with an approximate range of 12.58%–14.39%.

4.2. Decisions on expenditure participation

The first-stage probit models aim to understand the heterogeneous expenditure decisions for five travel products. The predictors of expenditure participation decisions differed from those in the second-stage expenditure models. It is evident that whether or not to purchase a product relies on the main habits of international travelers (Lyu & Noh, 2017; Murphy et al., 2011). According to Table 2, the results indicate a good fit with the data. Most tested independent variables showed a significant relationship (p < .05) with travel expenditures.

Table 2. Results of the first-stage probit model estimations.

Variable	Outcome D(Exp.Accommodation)		Outcome D(Exp.Food)		Outcome D(Exp.Transport)		Outcome D(Exp.Shopping)		Outcome D(Exp.Other expenses)
	Coeff	t-stat	Coeff	t-stat	Coeff	t-stat	Coeff	t-stat	Coeff	t-stat
Satisfaction2	0.155	(1.34)	0.0962	(0.59)	0.922**	(2.05)	0.229*	(1.94)	0.554**	(2.63)
Satisfaction3	0.323***	(5.01)	0.402***	(4.23)	0.861***	(13.13)	0.353***	(5.48)	0.422***	(6.63)
Satisfaction4	0.367***	(6.66)	0.403***	(5.33)	0.852***	(15.88)	0.444***	(8.15)	0.322***	(5.82)
Satisfaction5	0.322***	(5.89)	0.504***	(6.66)	0.793***	(14.92)	0.419***	(7.75)	0.0677**	(2.24)
Distance	0.115***	(5.88)	0.131***	(3.92)	0.0360*	(1.66)	0.324***	(16.65)	0.016	(0.88)
Visit Times	0.394***	(21.79)	−0.086***	(−2.67)	0.207***	(10.16)	0.148***	(8.02)	0.0513***	(3.24)
Stay days	−0.440***	(−43.72)	−0.319***	(−18.37)	−0.568***	(−49.77)	−0.199***	(−20.07)	0.427***	(38.31)
Plan days	0.0390***	(5.16)	0.0257*	(1.95)	0.0552***	(6.64)	0.177***	(23.53)	0.124***	(18.09)
Gender	0.0645***	(3.63)	−0.0649**	(−2.07)	0.0393**	(1.99)	−0.350***	(−19.16)	−0.0408**	(−2.54)
Generation X	0.0415**	(2.78)	−0.0957**	(−2.37)	0.0461*	(1.74)	0.0925***	(4.04)	−0.0385*	(−1.84)
Generation Y	−0.121***	(−6.61)	−0.0349	(−1.05)	−0.0311	(−1.49)	−0.127***	(−6.87)	0.0458***	(2.76)
Generation Z	−0.145**	(−2.42)	−0.191**	(−2.01)	−0.215***	(−3.46)	−0.114*	(−1.74)	0.0316*	(1.59)
Education2	0.126***	(5.31)	0.0473	(1.13)	0.0860***	(3.24)	−0.0873***	(−3.4)	−0.00201	(−0.09)
Education3	0.157***	(5.14)	0.136**	(2.57)	0.103***	(3.07)	0.129***	(4.12)	0.0610**	(2.15)
Income2	0.0522**	(1.96)	−0.0889*	(−1.73)	−0.0789**	(−2.56)	−0.110***	(−3.71)	−0.0923***	(−3.97)
Income3	0.0884***	(2.59)	0.0244	(0.35)	−0.119***	(−3.03)	−0.124***	(−3.37)	−0.0693**	(−2.35)
Income4	0.102***	(3.04)	−0.136**	(−2.19)	−0.0964**	(−2.52)	0.237***	(6.73)	−0.014	(−0.48)
Income5	0.301***	(8.58)	−0.0891	(−1.42)	0.0969**	(2.45)	0.223***	(6.25)	0.0223	(0.7)
Constant	1.102***	(15.1)	2.152***	(18.64)	1.199***	(15.94)	0.842***	(11.49)	−1.434***	(−19.97)
Observations	30,945		30,945		30,945		30,945		30,945
LR χ^2	6,939		672		4,659		6,109		5,875
Pseudo R^2	0.200		0.081		0.177		0.186		0.142

Data processed, 2023.

* p < .1, ** p < .05, *** p < .01.

To be specific, regarding accommodation spending outcome, several independent variables exhibited significant positive relationships with lodging expenditure. Visitors who reported higher satisfaction, traveled shorter distances, visited for the first time, had longer plan durations, were males in their 50s, had higher education and income levels, were more likely to participate in spending on accommodation. However, a few variables, such as stay durations and generations Y and Z, had significant negative relationships with lodging expenditure. This suggests that visitors who stayed longer and were older tended to participate in spending on accommodation.

Regarding the outcome of food spending, there are some opposite results with accommodation spending or insignificant results. Some independent variables, such as satisfaction, plan durations, and education level, were positively correlated with spending on food. This indicated that visitors who reported higher satisfaction levels, longer plan durations, and graduate degrees were more likely to participate in spending on food. On the other hand, variables like visit times, stay durations, gender, and generation X and Z had negative relationships with food expenditure. This indicated that visitors who frequent visits had shorter stays, and younger female visitors were more likely to participate in spending on food. Income levels did not clearly influence food spending habits among tourists.

In term of transportation spending outcomes, visitors who are more satisfied and travel shorter distances are likelier to participate in transportation spending. Additionally, those who are visiting for the first time, have longer plan durations, are males of older ages, and have higher education and income levels also tend to contribute more to transportation spending. However, a few variables have a negative relationship with transport expenditure, such as stay durations and generation Z. This means that visitors who stayed for a shorter period of time and were younger were less likely to participant in spending money on transportation.

Regarding purchase spending outcomes, tourists who are satisfied with their vacations and travel short distances are more inclined to spend money shopping. Furthermore, first-time travelers, older females, longer trips anticipated, and higher levels of education and money are more likely to contribute to shopping spending. Those who stayed shorter were more inclined to spend money on shopping. Regarding other expenses, tourists who are satisfied with their trips and begin on their first trip, have longer extended stays planned, are younger females, and have better levels of education are more likely to contribute to other expenses.

Based on our analysis of decisions on the expenditure participation of tourists, we found that several independent variables had a positive relationship with lodging, food, transportation, shopping, and other expenditures. These variables included higher satisfaction, short travel distances, first-time visits, longer plan days, short stay durations, older tourists, and higher education and income levels.

4.3. Decisions on expenditure consumption second-stage probit analyses

For the second stage, this study used OLS models to explore the effects of various independent variables on tourists’ decisions to consume five travel products: accommodation, food, transportation, shopping, and other expenses. Unlike the first-stage probit model, which only included some independent variables, this stage incorporated several travel characteristics (such as satisfaction, distance, visit times, and stay days) and individual characteristics (such as gender, generation, education level, and income level) that influence travel spending decisions. The results of our study indicate that plan duration was more relevant in determining the decision to engage than spending behavior in travel-related spending activities. Therefore, we excluded the plan duration variable from the second-stage model, which had already been incorporated into the first-stage model. The exclusion procedure in the Heckman model was used based on previous research to obtain a more reliable parameterization (Lyu & Noh, 2017; Sartori, 2003). We then estimated the second-stage models, considering the IMR data for each respondent group. As shown in Table 3, our findings reveal that four models had statistically significant IMR data. This indicates a sample selection issue when the decision to participate in travel spending activities is excluded, as suggested by Heckman, (1979). The Heckman two-stage models were an appropriate method for our study, and Lambda values were a crucial factor in determining respondents’ spending behavior. The results of the second-stage OLS model estimations are presented in Table 3.

Table 3. Results of the second-stage truncated OLS model estimations.

Variable	Outcome ln(Exp.Accommodation)		Outcome ln(Exp.Food)		Outcome ln(Exp.Transport)		Outcome ln(Exp.Shopping)		Outcome ln(Exp.Other expenses)
	Coeff	t-stat	Coeff	t-stat	Coeff	t-stat	Coeff	t-stat	Coeff	t-stat
Satisfaction2	0.252***	(6.49)	0.00505	(0.13)	0.819***	(15.07)	0.0842	(1.42)	0.989***	(4.00)
Satisfaction3	0.0847	(1.49)	0.106*	(1.88)	0.159***	(2.7)	0.0597	(0.67)	1.831***	(6.22)
Satisfaction4	0.0672***	(3.5)	0.00322	(0.17)	0.0610***	(2.96)	−0.0222	(−0.73)	1.755***	(5.52)
Satisfaction5	0.0515***	(5.21)	0.0482***	(5.03)	0.0488***	(4.91)	0.0542***	(3.81)	1.337***	(4.36)
Distance	0.0213*	(1.96)	0.0768***	(7.22)	−0.0149	(−1.37)	0.127***	(5.73)	0.117	(1.52)
Visit Times	−0.126***	(−6.03)	0.0347***	(3.91)	−0.0364***	(−3.07)	−0.0341**	(−2.25)	0.364***	(4.99)
Stay days	−0.191***	(−13.84)	−0.246***	(−15.76)	0.134***	(5.53)	−0.345***	(−21.74)	0.704***	(6.32)
Gender	0.0328***	(3.26)	0.0181**	(2.06)	0.0159*	(1.66)	−0.170***	(−8.39)	−0.504***	(−7.27)
Generation X	0.0471***	(3.85)	0.0418***	(3.63)	0.0571***	(4.61)	0.0929***	(4.96)	−0.470***	(−5.2)
Generation Y	−0.0772***	(−7.78)	−0.0340***	(−3.86)	−0.0300***	(−3.05)	0.00338	(0.22)	−0.373***	(−5.19)
Generation Z	−0.0392	(−1.32)	−0.132***	(−4.45)	0.0329	(0.98)	−0.0599	(−1.33)	−0.469**	(−1.98)
Education2	−0.0577***	(−3.65)	−0.118***	(−10.15)	−0.166***	(−12.55)	−0.129***	(−6.98)	−0.350***	(−3.79)
Education3	−0.0202	(−1.14)	−0.157***	(−9.94)	−0.142***	(−8.22)	−0.139***	(−5.55)	0.162	(1.36)
Income2	0.0638***	(4.94)	0.0744***	(5.7)	0.0176	(1.23)	−0.0825***	(−4.12)	−0.313***	(−2.86)
Income3	0.148***	(8.34)	0.0845***	(5.31)	−0.00938	(−0.51)	−0.0483*	(−1.87)	−0.405***	(−2.97)
Income4	0.244***	(13.78)	0.125***	(7.43)	0.000206*	(2.01)	0.0736***	(2.67)	0.244*	(1.87)
Income5	0.412***	(20.78)	0.218***	(12.76)	0.0113**	(2.61)	0.0845***	(2.76)	0.414***	(3.1)
COVID − 19	−0.0224**	(−2.88)	0.0316***	(3.5)	−0.0942***	(−9.44)	0.129***	(3.22)	−0.834***	(−3.89)
Constant	4.520***	(94.19)	3.846***	(140.38)	3.115***	(104.06)	4.197***	(87.84)	−5.219***	(−5.75)
IMR (λ)	−0.786***	(−4.68)	−1.474***	(−4.83)	−1.920***	(−17.35)	−0.459***	(−3.74)	4.675***	(7.8)
Selected (N)	23,221		30,021		26,234		24,017		18,782
R^2	0.136		0.168		0.110		0.124		0.085
σ	0.641		0.705		0.741		1.027		4.490

Data processed, 2023.

* p < .1, ** p < .05, *** p < .01.

A few of the coefficients of the travel characteristics variables in the five models in Table 3 are statistically significant compared to the results of the first-stage probit model estimations. This implies that the second-stage travel expenditure decisions differed from the initial decision-making process for spending participation. To be specific, the satisfaction levels were generally significant in influencing travel expenditures, with satisfaction5 being especially significant for all models (p < .05). This indicates that tourists who reported higher satisfaction levels tended to spend more money on various travel products compared to those with lower satisfaction (Satisfaction2–5). However, some satisfaction levels (e.g., satisfaction2 and satisfaction4) were insignificant in the food and shopping models, whereas the accommodation model was insignificant for satisfaction3. The distance variable had a significant positive coefficient sign in several models (accommodation, food, and shopping). This suggests that international tourists who traveled shorter distances tended to spend more money on these travel commodities. This finding is consistent with the results of the first-stage shopping participation models. The exceptions for the transport and other expenses models were insignificant. The visit times variable had significant negative coefficient signs in some models (accommodation, transport, and shopping). This means that international tourists who visited Taiwan for repeat trips tended to spend more on accommodation, transportation, and shopping. However, in the food model, visit times had a significant positive coefficient sign, indicating that tourists visiting Taiwan for the first time tended to spend more money on food. The predictor of stay days reveals different coefficient signals. Specifically, some coefficients of the stay days variable showed significant negative signals in the three models (accommodation, food, and shopping), whilse other expenses showed significantly positive signs, which aligns with the first-stage participation models. This suggests that visitors with shorter stay durations tend to spend more on accommodation, food, and shopping while spending less on other expenses. On the other hand, the variable of stay days showed a significant positive coefficient sign, which is inconsistent with the first-stage transportation participation models. This indicates that international tourists with longer stays tend to spend more on transportation.

Like travel characteristics, a few of the coefficients of the individual characteristics variables in the five models in Table 3 were statistically significant compared to the results of the first-stage Probit model estimations. This implies that the second-stage spending participation decision-making process differed from the initial one for spending participation. For instance, the predictor of gender had a significantly positive coefficient signal in the accommodation and transportation models but negative coefficients in the shopping and other expenses models. This is consistent with the first-stage shopping participation models, indicating that male tourists spent more money on accommodation and transportation but less on shopping and other expenses. Moreover, there was a difference in spending on food compared to the first-stage shopping participation models, as male tourists spent more money on food than females. Interestingly, the significant positive signs of Generation X reflect that older tourists in their 50s spent more money on accommodation, food, transportation, and shopping (with the exception of other expenses). On the other hand, younger tourists (aged in their 10s–40s) belonging to Generations Y and Z had significant negative coefficients or were insignificant in five models, indicating that they spent less on travel products. Another interesting predictor is the education variable, which showed significant negative coefficients. This is inconsistent with the first-stage shopping participation models and indicates that tourists with higher education (college, university, and graduate school) spent less on travel expenditures than tourists who had under/high school. Visitors with a higher education are more careful with their travel expenses than those with a lower education. We have found clear evidence for differences in travel spending patterns based on the annual income level of visitors. This result is inconsistent with the first-stage spending participation models. Specifically, we noticed that all coefficients of income variables showed significant positive signs in five models, indicating that tourists of any income level are likely to spend money on accommodation and food. However, we also observed that higher-income tourists tend to spend more on transportation, shopping, and other expenses.

The COVID-19 pandemic (COVID-19) had varying effects on different types of travel expenditures. According to some expenditure models, the coefficients associated with the pandemic had negative signs in categories like accommodation, transportation, and other expenses. This indicates that the pandemic had a detrimental impact on spending in these areas, which could be due to factors like travel restrictions and safety concerns. On the other hand, the coefficients associated with the COVID-19 pandemic had positive signs in models related to food and shopping. This could mean that some tourists have shifted their spending priorities and are allocating more budget to dining and shopping while reducing spending on other aspects of their travel experience. Changes in travel behavior during the pandemic, such as a preference for outdoor activities and dining, have resulted in increased spending in these categories. These findings likely reflect travel behavior’s complex and evolving nature during the pandemic.

This study found some fascinating results about tourist’s spending habits. Apparently, those who reported higher satisfaction with their travel experiences tended to spend more money on travel products overall. Travelers who traveled shorter distances spent more money than those who traveled farther. Additionally, repeat visitors tended to spend more on accommodation, transportation, and shopping. The study also revealed that gender and education level played a role in spending decisions among travelers. Unfortunately, the COVID-19 pandemic negatively impacted spending in categories like accommodation and transportation. However, some tourists have shifted their priorities and are now spending more on food and shopping instead.

5. Discussion and implication

Higher overall satisfaction has been found to positively correlate with increased spending in various categories, which implies that creating a positive overall experience for tourists can directly impact their willingness to spend. As such, destination managers and businesses should prioritize improving visitor satisfaction to encourage higher spending. Nevertheless, lower satisfaction levels were deemed insignificant in the food, accommodation, and shopping models.

Based on the data analysis, tourists traveling shorter distances tend to spend more on accommodation, food, and shopping. This could be because they save on transportation expenses and have more discretionary income to allocate to these categories. However, distance plays a smaller role in determining spending on transportation and other expenses, suggesting that other factors, such as the nature of the trip or personal preferences, have a more significant impact. Interestingly, repeat tourists spend more on accommodation, transportation, and shopping, possibly because they are more familiar with the destination and comfortable exploring various options. Meanwhile, first-time visitors tend to allocate more of their budget to trying local cuisine, an expected behavior when exploring a new destination.

Based on the data analysed, the spending habits of tourists in Taiwan vary depending on the length of their stay (Massidda et al., 2020; Thrane & Farstad, 2011). The transportation model showed an opposite signal direction in the second-step truncated OLS models. The stay-day variable showed that international tourists with longer stays tend to spend more on transportation and other expenses (Thrane & Farstad, 2011). This could be because tourists with extended stays have more time to explore a broader range of attractions within the destination and neighboring regions. Therefore, they may invest in various transportation options to facilitate their exploration. Longer-staying tourists are also more likely to engage in other spending activities such as entertainment activities, cultural experiences, and other leisure expenses. On the other hand, visitors with shorter stays tended to spend more on accommodation, food, and shopping. Tourists with shorter stays prefer to make the most of their limited time and choose more comfortable or convenient lodging options. They may also be more inclined to dine out frequently and explore local cuisine during their brief visit. Additionally, tourists with shorter stays allocated a more significant portion of their budget to shopping, engaging in more souvenir shopping or spontaneous purchases during their stay.

Male tourists spent more on accommodations and transportation, which could be related to traditional gender roles and spending patterns (Sahoo et al., 2022). They may value comfort and convenience more than female tourists. On the other hand, female tourists allocate higher resources to shopping and incidental expenses, which could reflect differences in interests or priorities (Lyu & Noh, 2017; Marcussen, 2011). However, males spent more on food, suggesting that dining experiences are important to them. As for age, older tourists in their 50s tend to spend more across different categories, possibly due to their greater disposable income and established travel habits. They prioritize comfort and quality, which leads to increased spending. Younger tourists, on the other hand, tend to spend less on travel products overall. This could be due to budget constraints or a preference for budget-friendly options. They may also prioritize experiences over material purchases.

Tourists with higher education levels (college, university, and graduate school) spend less on travel expenses than those with lower education levels (under or high school), as stated by Lyu and Noh (2017). However, other studies (e.g., Sahoo et al. (2022), Wu et al. (2013) have shown inconsistent findings, suggesting that tourists with higher education may spend more money on travel activities. This finding challenges the conventional assumption that higher education correlates with higher spending during travel. The negative coefficients suggest that tourists with higher education may be more financially conscious and careful with their travel expenses. They prioritize budgeting and value-conscious decisions while planning their trips. On the other hand, tourists across all income levels spend money on accommodation and food during their travels, as evidenced by the positive coefficients in the five models (Gómez-Déniz et al., 2019; Park et al., 2019; Yang et al., 2019). This suggests that these categories are relatively consistent priorities for all travelers, regardless of their income. However, the data also shows that higher-income tourists spend more on transportation, shopping, and other expenses, indicating differences in spending patterns between income groups. Perhaps these higher-income travelers are more likely to allocate their budget to activities, shopping for souvenirs, and other experiences.

The COVID-19 pandemic impacted negatively spending in some categories (accommodation, transportation, and other expenses). This can be attributed to travel restrictions, safety concerns, and a reduced willingness to engage in traditional travel activities (e.g., entertainment). However, tourists adjusted their spending priorities during the pandemic, as evidenced by the positive coefficients for the food and shopping model (Chen et al., 2022; Chen & Ho, 2022; Hall et al., 2020). This shift may be linked to a desire for safer and more localized experiences, such as dining outdoors or exploring local markets. The pandemic accelerated trends of "staycations" and "eat local," which shows that travelers are resilient and adaptable, able to pivot and find alternative ways to enjoy their trips. As the pandemic persisted, tourists and businesses continued adapting to changing conditions, which is crucial for the tourism industry’s survival and recovery. Constant innovation and flexibility are necessary in these unprecedented times.

6. Conclusion, limitations, and further research areas

Most tourism destinations are interested in encouraging tourists to spend more money on various travel products because this spending contributes to the local economy. The study’s primary objectives are to identify the factors related to travel and individual characteristics that influence the spending behaviors of international tourists in Taiwan. We notably considered other determinants, such as satisfaction levels and the COVID-19 pandemic. Our research employed a unique approach by utilizing a sequential decision-making process as our framework instead of traditional economic methods. This process involves international tourists first deciding if they want to purchase travel products and then determining how much they are willing to spend on those products. We employed the Heckman sample selection approach to analyze their spending habits, which involves two stages: deciding whether to spend on specific products and determining how much to spend. The statistically significant IMR statistics confirmed the appropriateness of the sequential decision-making procedure for expenditure tendencies. This two-stage technique was consistent with previous research on tourist’s general consumption behavioral patterns. The study also found significant heterogeneous spending behaviors for travel products among international tourists. These differences are more pronounced in the second stage (how much to spend) than in the first stage (whether to spend).

Based on our findings, international tourists’ spending behavior in Taiwan is influenced by various factors. The level of satisfaction experienced by tourists positively affects their spending across different categories, highlighting the significance of enhancing overall satisfaction. Additionally, the distance tourists travel plays a role in their spending habits, with shorter-distance travelers allocating extra resources to accommodation, food, and shopping. Repeat tourists spend more on transportation, accommodation, and shopping, while first-time visitors prioritize local cuisine. Longer stays lead to increased spending on transportation and other expenses, while shorter stays result in higher expenditures on accommodation, food, and shopping. Gender differences also play a role, with males spending more on accommodations and transportation, whilse females allocate more resources to shopping and incidental expenses. Age is another factor, with older tourists spending more across categories and valuing comfort and quality. Income levels also play a role, with higher-income tourists allocating more of their budget to transportation, shopping, and other expenses. However, tourists with higher education levels spend less, challenging assumptions about education and spending. Finally, the COVID-19 pandemic has negatively impacted accommodation and transportation spending but has led to positive shifts, increasing spending on food and shopping as tourists adapt to new travel behavior patterns.

Limitations and further research areas

Our analysis provides valuable insights into tourist spending behavior but also has drawbacks. Specifically, there is limited reflection of recent developments or changes in tourism trends in the post-COVID-19 pandemic era. Incomplete consideration of external factors such as economic conditions and travel regulations. Additionally, the data used in the analysis may rely on self-reported tourist information, which can be biased and inaccurate (e.g., recall and social desirability biases). The analysis may potentially omit emerging trends or recent shifts in spending patterns. Lastly, non-exhaustive coverage of all variables influences tourist spending behavior.

Several areas still require further research in terms of tourist spending behavior and its driving forces. Further studies can employ advanced data analytics to uncover hidden patterns and predictive factors that impact spending. Predictive modeling can be instrumental in forecasting spending based on various variables. Furthermore, further research can focus on post-pandemic travel behavior. To be specific, research the lasting effects of the pandemic on tourist spending as the world emerges from the COVID-19 pandemic. In addition, further research can analyze whether new spending patterns and preferences continue in the post-pandemic era. Lastly, consider the adoption of digital payment methods among tourists and how they impact spending behavior. With the growing popularity of mobile wallets and cryptocurrencies, cashless transactions have become more accessible than ever before, and we must examine how this trend affects tourist spending.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Funding statement

The author(s) received no financial support for the research, authorship, and/or publication of this article.

Additional information

Notes on contributors

Kieu-Thi Phan

Kieu-Thi Phan is a doctoral student currently pursuing her degree at the Department of International Business at the National Kaohsiung University of Science and Technology in Kaohsiung, Taiwan. She is also an officer at the Tra Vinh Statistical Office, which the General Statistics Office in Vietnam manages. She has a keen interest in finance research and statistics. E-mail: [email protected].

Sheng-Hung Chen

Sheng-Hung Chen is currently a full professor at the Department of International Business at the National Kaohsiung University of Science and Technology in Kaohsiung (Taiwan). He earned Ph.D. in Applied Economics from National Chung Hsing University in 1997 in Taichung (Taiwan). He teaches Financial Management and International Financial Management in the full-time MBA program and at the undergraduate level. His main research interests are in empirical study on global banking. Currently, He is doing research in the areas of bank management, risk management, energy finance, and ESG. He has published research in the Journal of Banking and Finance, International Review of Economics and Finance, and Quarterly Review of Economics and Finance. Email: [email protected].

Jie-Min Lee

Jie-Min Lee is a chairman and Professor, Department of Shipping and Transportation Management, National Kaohsiung University of Science and Technology, Kaohsiung 811213, Taiwan. Email: [email protected].

Ca-Van Pham

Ca-Van Pham is a Ph.D. student at the Department of International Business, National Kaohsiung University of Science and Technology Kaohsiung, Taiwan. He is also a Lecturer at the School of Economics and law, Tra Vinh University, Vietnam. His primary research mainly focuses on Enterprise Innovation, ESG, accounting information quality, and Knowledge Management. Email: [email protected]

Appendix A.

Questionnaire annual survey of visitors expenditure and trends in Taiwan

Table

A. The total number of trips made to Taiwan in the previous three years, except the current one. B. The amount of time the visitor had to organize this vacation before departing for Taiwan C. The number of days to stay in Taiwan throughout this trip D. The modes of transportation (including personal trips without travel agencies and group or individual tours planned by a travel company) E. How much did this trip to Taiwan cost in total? (Money type, excluding payments in advance) F. Travel spending, including lodging costs G. Food costs, including any out-of-home meals H. Costs associated with local transport (high-speed rail, railway, MRT, aircraft, coach, bus, taxi, ferry, shared bicycles, rental cars, and others) I. Spending for leisure (movies, clubs, playing golf, and others) J. Purchases related to shopping (such as those for apparel, jewelry, trinkets, skin care products, and specialty items from the area). K. Additional expenditures L. Details on the individual (gender, nationality, age, annual earnings, and others) L1. Nationality L2. The visitor’s home country or region L3. Age ranges: (12) 12 − 19; (20) 20 − 29; (34) 30 − 39; (55) 50 − 59; (6) 60 − 65; and (7) 66 and older L4. Yearly earnings (in US dollars) (1) Under 9,999, (2) 10,000–14,999, (3) 15,000–29,999, (4) 30,000–39,999, (5) 40,000–69,999, (6) 70,000–99,999, and (7) Over L5. Gender: (1) Male; (2) Female L6. Education includes the following: (1) Elementary, junior, senior, or vocational schools; (2) College or university; (3) Graduate school (master’s or doctoral); (4) Other

Monetary value = US dollar. Only inbound visitors to Taiwan were asked about their expenditure questions.