Love More and Buy More? Behavioral Economics Analysis of Travel Preference and Travel Consumption

Abstract

This study examined the interaction between travel preference and travel consumption from the perspectives of traditional and behavioral economics and analyzed the impact of travel preference on travel consumption using survey data. The WLS regression model was adopted to examine the influence of travel preference on travel consumption and the mediating role of travel demand in the relationship between travel Preference and travel consumption. The results showed that: (1) for domestic travel, the travel preference has a positive impact on travel consumption through the mediating role of travel demand; and (2) for outbound travel, the travel preference has a positive impact on outbound travel consumption through the mediating role of outbound travel demand under the transaction utility and loss aversion effects. Thus, our findings confirm previous evidence on the relevance of tourists’ preferences as a good predictor of travel consumption in the case of Chinese travelers.

1. Introduction

Will the higher consumers’ preference for travel bring higher travel consumption? In other words, if people want to travel more, will more traveling “entities”, such as hotels, air travel, and restaurant food will be consumed? According to the explanation of traditional economics, rational consumers will make such a choice: the more substantial the degree of travel preference, the greater the travel demand, which thereby directly enhances travel consumption. Strong preference and high travel demand will undoubtedly play a positive role in travel consumption. The study of Chinese domestic travel consumption accords with the above theoretical discussion. However, domestic travel is a normal good, which is not always the case for outbound travel. We have found that traveling abroad is still a quasi-luxury for most ordinary Chinese consumers. In China, the average cost of traveling abroad exceeds that of domestic travel by a large margin [1,2,3]. Will the consumption results of Chinese travelers abroad be the same as those traveling domestically? We jump out of the traditional economic framework and study the interaction between travel preference and travel consumption from the perspective of more microscopic behavioral economics.

In behavioral economics, the perceived cost of traveling abroad usually exceeds the reference price of tourism products in tourists’ minds. The concept of reference price is based on the reference point theory, one of the most important theories in behavioral economics [4,5]. The theory says that people do not consider absolute quantities in decision-making but rather consider the quantitative differences deviating from the reference point or reference price. When people react to an object, a reference point is determined by past experiences or the average level, and decision-making is related to this reference point [4].

When the price of a commodity exceeds its reference price, the price will be regarded as a loss for consumers, resulting in the loss aversion effect. At this point, through the mediating role of travel demand, the preference for this commodity will not lead to the promotion of travel consumption but will have an inhibitory effect on travel consumption [5]. On the other hand, behavioral economics also points out that high-quality features of outbound travel implied by high prices usually prompt consumers to enjoy spending money on outbound travel. In other words, because of the existence of the transaction utility, travel demand, in turn, promotes travel preference and stimulates consumption. Therefore, whether the preference for outbound travel will increase or inhibit outbound travel consumption depends on the net effect of the loss aversion and transaction utility effects. From the perspective of econometrics, the impact of travel preference on travel consumption is uncertain. The actual effect of outbound travel preference on travel consumption abroad can be examined only by taking into account the interaction between the preference and demand, and the net effect caused by the loss aversion and the transaction utility effects. This paper finds that the transaction utility effect is larger than the loss aversion effect in the Chinese sample. Therefore, the preference for outbound travel has a positive impact on outbound travel consumption, in this case through the transaction utility effect.

In order to better analyze various factors affecting travel consumption and to measure those factors’ actual impact on travel consumption, the use of different economic models and analysis tools have been considered by researchers. It was found that travel expenditure (which in turn influences travel consumption) is crucially driven by trip-related and psychographic characteristics [6]. However, in general, the use of psychological variables in the literature is not very frequent. Additionally, their measurement is still an open question [7].

Travel preferences have been regarded as one of the most critical elements to explain travelers’ behaviors at their destination, and thus their consumption [8]. Researchers have investigated travel preference based on many criteria, such as the destination development choices [9,10,11,12], hypothetical managerial initiatives for attractions [13,14], attributes defining rural houses [15], and online travel agencies [16]. However, travel preference was not specifically addressed in early models related to travelers’ behaviors. Travel demand is composed of the three elements of motivations, perceptions, and expectations. Travel preference is regarded as one of the elements of perceptions [17]. As a result, travel decisions can be determined by travel preferences through motivations [18].

More recently, Chen et al. [19] developed the “Motivation–Preference–Intention” structural path model to show some significant effects (positive or negative) between preference and intention. Although we can see the effect of travel preference on travel consumption in the above literature, no study has evaluated tourists’ travel preferences related to their travel consumption. It seems that a higher preference for travel may not translate to higher travel consumption [8]. Overall, it is recommended that researchers should identify the problem where travelers prefer to do one thing, but in reality, they do not or do another thing [20]. However, there have been a few empirical studies showing the correlation between travel preferences and real travel consumption results.

As a theoretical and empirical concept, perceived value has received increasing attention in marketing and tourism research in the last three decades [21,22,23,24,25]. Patterson and Spreng [26] described the concept of perceived value as a cognitive-based concept which captures the benefit–sacrifice discrepancy in the same way as disconfirmation does for variations between expectations and perceived performance.

Additionally, as travel consumption makes a considerable contribution to economic growth, it is extremely important to identify factors influencing travel consumption and measure the effects on travel consumption. Therefore, much of the literature in economic analysis is used to study the influencing factors of travel consumption. The consumption theory based on the preference axiom of traditional economics reveals that the more robust the travel preference, the greater the travel demand, directly enhancing travel consumption. However, as travel consumption involves individual behavior, the traditional economic theory is far from reaching the core of the study. Therefore, it is necessary to introduce behavioral economic theory. The reference point theory in behavioral economics says that people consider the quantitative differences deviating from the reference point or reference price in decision-making. As outbound travel is a quasi-luxury for most Chinese consumers, when the price of outbound travel exceeds its reference price, the price will be regarded as a loss for travelers, resulting in the loss aversion effect. However, the reference point theory also advances the transaction utility, that the high-quality characteristics of outbound tourism implied by high prices will usually encourage consumers to enjoy the consumption of outbound tourism. Briefly, tourism demand, in turn, promotes travel preference and stimulates consumption. Travel preference and perceived value are two important variables of psychographic characteristics. Therefore, this paper studies the impact of travel preference on travel consumption from the perspective of traditional economics and behavioral economics. Specifically, the research framework of this paper is as follows: (1) to examine the effect of travel preference on domestic travel consumption consistent with traditional economics; and (2) to examine the effect of travel preference on outbound travel consumption considering the effect of loss aversion and transaction utility, which can only be explained by behavioral economics.

2. Theory and Hypotheses

In traditional economics, consumption theory based on the preference axiom has the following logic: the higher the preference for a commodity is, the higher the demand for the commodity is. Demand also leads to consumption, so the higher the demand is, the higher the consumption is. Usually, as one kind of normal service commodity, tourism complies with the consumption model of traditional economics (CMTE) shown in Figure 1 below.

The consumption model of traditional economics has a strong axiomatic status in economics. However, this position has been questioned by behavioral economics, that points out that an important prerequisite for the establishment of the model is that the market price of the commodity does not deviate from its reference price [5]. At this point, the market price will not have an impact on consumer utility, and utility is determined only by consumption [5]. The formula is expressed as U = U (C). For example, for most people, a set of travel postcards with the price of a few dollars is acceptable. Travel postcards’ market price does not deviate from its reference price, so the preference for travel postcards will certainly boost the consumption of travel postcards. Recently, domestic travel in China has become one kind of mass consumption, and is especially a necessity for urban residents [27]. It shows that domestic travel prices can be accepted by most people, which obviously do not deviate from the reference price. Therefore, the rule of domestic travel consumption should be predicted by CMTE. Thus, Hypothesis 1 is formed as follows:

Hypothesis 1 (H1).

The increase in travel preference for domestic travel can positively enhance the travel demand for domestic travel and further increase the domestic travel consumption.

Hypothesis 1 shows that domestic travel preference positively impacts on domestic travel consumption through the mediating role of domestic travel demand, as shown in Figure 2.

However, unlike domestic travel consumption, outbound travel is still a luxury service for most Chinese [1,2,3]. Taking the leisure survey of the National Tourism Administration of China as an example, the average monthly income of urban respondents in this survey is 7047 yuan, which is more than 47% of the national average monthly income of 4783 yuan. Even so, only less than 20% of the respondents have traveled abroad in the past year. It shows that the market price of traveling abroad in China deviates from the reference price for most people. Therefore, CMTE will not be useful to explain the consumption pattern of traveling abroad. When the market price of a commodity deviates from its reference price, preference does not necessarily lead to a consumer choice [28]. Therefore, it leads to the following Hypothesis 2:

Hypothesis 2 (H2).

The influence of travel preference on outbound travel consumption is ambiguous.

The critical difference between outbound and domestic travel consumption is that the market price of outbound travel consumption deviates from its reference price, which will lead consumers to pay special attention to the price of overseas travel. Therefore, outbound travel should not be regarded as identical products with inbound travel in consumption, especially for Chinese consumers. In this study, the price of inbound travel is anchored to the reference price as most Chinese travel domestically before they travel abroad, so the price of inbound travel will be used as the price reference for future outbound travel. As a result, in our sample pool, outbound travel seldom follows the laws of traditional consumption economics but will show instead the characteristics of consumption laws of behavioral economics. Specifically, we can say that the market price of outbound travel produces the loss aversion effect and transaction utility effect based on the relationship between reference points and loss aversion [5,29].

According to our empirical data, the mean price of outbound travel for Chinese travelers is higher than that of inbound travel. Therefore, when the inbound price is set as the reference price, the price of outbound travel will have a “loss aversion” effect on buyers. Of course, many traditional economists consider the inability to endogenously set reference points as a glaring weakness of behavioral economics. In fact, it is challenging to construct a reference point that allows participants to truly experience loss aversion without changing their motivation to participate [30]. Therefore, behavioral economists either take the previous experience as a reference point or the expected state as a reference point [30]. We use the former as a reference point in this paper.

At the same time, the purchase of outbound travel is a kind of transaction behavior. According to behavioral economics, spending money in a transaction is intrinsically pleasurable and therefore leads to positive satisfaction, namely taking a transaction produces transaction utility [31,32]. Therefore, although the loss aversion effect inhibits the purchase of outbound travel, it is also promoted by the transaction utility effect. We will explain the two effects, respectively.

The first effect is the loss aversion effect of outbound travel consumption. First, behavioral economists have found that people feel pain through spending money as well as potential satisfaction when they consume [33]. The use of magnetic resonance imaging (MRI) to study specific areas of the brain confirms the idea that preference is one kind of potential benefit and price is one kind of potential cost in the brain framework [34]. The psychological account theory of behavioral economics further reveals that when the market price of a commodity does not differ much from the reference price, the satisfaction of consumption far outweighs the pain of spending money, and the price does not affect utility [5]. However, when the difference between market and reference prices is large, people will pay special attention to prices because of the prominence effect [29]. Consumers feel the pain of spending money and then regard prices as an important loss [5]. At this time, the utility function will shift. Second, because the market price of traveling abroad deviates from its reference price, the Chinese will regard the high cost of traveling abroad as a loss. While enjoying the pleasure of traveling abroad, tourists will feel the pain of the high cost of traveling abroad. At this point, the loss aversion effect will dominate consumption [5]. Finally, in view of Chinese traveling consumption abroad, when the preference for traveling abroad rises resulting in the increase of the demand for traveling abroad, the market price of traveling abroad is higher than the reference price in people’s hearts. People will feel the pain brought by the high price and high consumption, which will limit the actual consumption of travel. This partly explains why only a small percentage of people in high-income groups have traveled abroad.

The second effect is the transaction utility effect of outbound travel. Behavioral economists recognize that high consumption behaviors such as traveling abroad do not only cause loss aversion, but also have a positive impact on consumption. Price represents the acquisition of value in a commodity or service, so the price is both a loss and a gain [35]. Furthermore, high prices that deviate from the reference price also indicate high quality and future enjoyment. Spending money can bring happiness, so the price can produce transaction utility [5]. Finally, in the case of outbound travel, the transaction utility can cause the demand for outbound travel, which in turn strengthens the preference for outbound travel. Specifically, the pleasure consumers feel from spending money on traveling abroad makes them feel more worthwhile spending money on traveling abroad, which leads to a higher demand for traveling abroad, sending a positive signal to consumers themselves and promoting their preference for traveling abroad. These processes are called self-signing by consumers [36]. Therefore, the transaction utility effect makes the interaction between travel preference and travel demand, and has a positive effect on travel consumption.

Based on the above two discussions, the rule of outbound travel consumption is illustrated in Figure 3.

The meaning of Figure 3 is expressed in hypothesis 3.

Hypothesis 3 (H3).

The net effect of the loss aversion effect and the transaction utility effect determines the practical role of the preference of traveling abroad to outbound travel consumption.

Hypothesis 3 shows that outbound travel preference negatively affects outbound travel consumption through the mediating role of outbound travel demand under the loss aversion effect. However, the interaction between outbound travel preference and outbound travel demand has a positive impact on outbound travel consumption under the transaction utility effect. The net role of the two effects determines the direction of preference for consumption.

3. Data and Sample Selection

3.1. Data Collection

We used data from the 2016 Chinese Tourism and Leisure Survey (CTLS 2016) conducted by the China National Tourism Administration. CTLS 2016 targeted citizens who were 18 years of age or above living and in China for one year or more. CTLS 2016 was designed to gather detailed information on Chinese travel consumption, travel preference, demographics, and other leisure-related data. The method of multi-stage stratified random sampling combined with population proportion sampling was adopted in the CTLS 2016. The survey samples covered the 16 cities of Beijing, Shanghai, Guangzhou, Chengdu, Wuhan, Shenyang, Xian, Nanjing, Jilin, Zhuhai, Lanzhou, Yangzhou, Qiqihaer, Lianyungang, Kaifeng, and Zunyi. Turning to the investigation method, the survey was mainly based on the online survey, supplemented by intercept surveys. Thus, the data in the CTLS 2016 were comprised of 8149 observations. As the survey covered detailed information on travel preference and travel consumption, the CTLS 2016 was well suited to the purpose of this study. However, since the survey was done in 2016, it failed to report the tourism situation in China after the COVID-19 outbreak, which is a weakness of the research.

The survey data in the CTLS 2016 contains responses from 8149 respondents, from which 4982 respondents had domestic travel experience and 40 of these were excluded from the total due to missing values or their refusal to answer questions. Four of 1308 respondents who had both domestic travel and outbound travel experience were excluded because of missing values or their refusal to answer questions. As a result, samples of domestic travel are 4942, and samples from outbound travel are 1304.

3.2. Variables

In this paper, travel consumption is a dependent variable, which refers to domestic travel consumption and outbound travel. In CTLS 2016, respondents replied to the question “Please review your domestic Travel Consumption in 2016; how much have you spent on it?” (Questionnaire D2b) and “Please review your outbound Travel Consumption in 2016, how much have you spent on it?” (Questionnaire D4b). They also responded to the detailed information about their travel consumption on transportation, accommodation, food and beverage, tickets, shopping, and other activities. We also used the value of travel consumption to measure travel consumption.

The two central independent variables in this study are travel preference and travel demand (perceived value), in which travel demand is a mediator. In this paper, we examine whether travel preference mediated by travel demand can significantly influence travel consumption. A customer evaluates what is fair, right, or deserved for the perceived cost of the offering, including monetary payments and non-monetary sacrifices to build up her/his preference [37,38]. Similarly, travel preference is defined as the relative attitude of travelers toward a trip for a destination [16,39]. The standard measure of travel preference is to obtain the corresponding problem data through the questionnaire survey and then analyze the travel preferences using conjoint analysis [40,41,42,43,44], choice experiment [45,46,47], and other methods. The corresponding problem mainly refers to asking tourists about the specific tendency of food, housing, transportation, travel, shopping, and entertainment at different levels. Therefore, to measure travel preference, respondents in this study answer the question, “How many days do you want to spend on travel each time?” (Questionnaires C102/C202).

Travel demand is defined as the amount of money that tourists are willing and able to afford to purchase certain travel services and goods. It can be measured in multiple ways. Song and Li [48] reviewed the literature on tourism demand prediction published in the period 2000–2007, and it was found that 20 measures of tourism demand were used in these studies, including tourist arrivals (departures), and tourist expenditure (receipts), length of stay, nights spent at tourist accommodation, etc. Tourist arrivals (departures), tourist expenditure (receipts) are the most adopted among them. In this study, the logarithmic value of the highest acceptable single travel consumption (questionnaire C8) was used to measure travel demand. Although this measure was not common in previous studies, on the one hand, there is no other index closer to travel demand in CTLS 2016. On the other hand, it is reasonable to use this index because the higher the demand, the higher the acceptable single travel consumption is.

We control many variables according to the literature. In economics, the consumption function of Keynesian models is given by C = a + b * Y, in which C is consumption, Y is disposable income, and a and b are parameters [49]. As a result, individual income level could be the essential control variable for travel consumption [50]. Moreover, socio-demographic variables could be controlled for as well when studying the decision framework of travel consumption. It is necessary to control gender [6], age [7], educational [51], and occupational factors. Occupation is divided into five categories: employed, unemployed, retired, student, and others, which used the means of dummy variables. The employed are considered as the reference group (questionnaire B3a). Furthermore, trip-related characteristics also need to be controlled [51]. First, transportation is often introduced into the framework when considering the influencing factors of travel consumption [52,53,54]. Specifically, we controlled transportation by airplane, cruise ship, bus, car, high-speed train, or another way (questionnaires D901). Second, Marrocu, Paci, and Zara [6] revealed that accommodation is a relevant determinant to travel consumption. Therefore, we controlled for accommodation levels in questionnaire D1201. Third, stay time is found to be a significant factor related to travel consumption in most studies [50,55,56]. In conclusion, we controlled trip-related variables such as transportation, accommodation, and stay time.

3.3. Descriptive Statistics

Table 1. Descriptive Statistics for Domestic Travelers.

Core Variables	Mean	SD	Range	Med
Domestic Travel Consumption	8.604	0.814	4.605–11.408	8.666
Travel Preferences	5.687	3.754	1–90	5
Travel Demand	8.332	0.808	4.585–11.513	8.517
Control variables
Income	8.671	0.646	4.595–11.513	8.666
Gender
Female	0.518	0.500	0–1	1
Male	0.482	0.500	0–1	0
Age	44.006	13.801	18–78	43
Education:
Bachelor degree or above	0.434	0.496	0–1	0
Other	0.566	0.496	0–1	1
Occupation
Employed	0.762	0.426	0–1	1
Retired	0.168	0.374	0–1	0
Student	0.025	0.156	0–1	0
Unemployed	0.034	0.182	0–1	0
Other	0.011	0.103	0–1	0
Transportation:
Airplane	0.497	0.500	0–1	0
Cruise	0.010	0.098	0–1	0
Bus	0.070	0.254	0–1	0
Car	0.066	0.248	0–1	0
High-speed train	0.274	0.446	0–1	0
Regular Train	0.080	0.271	0–1	0
Other	0.004	0.063	0–1	0
Accommodation:
Two-star or below	0.485	0.500	0–1	0
Three-star	0.303	0.460	0–1	0
Four-star or above	0.206	0.405	0–1	0
Other	0.005	0.072	0–1	0
Stay Time	4.532	3.136	0–99	4

The number of observations is 4942.

reports descriptive statistics of the respondents who only took domestic travel in 2016, including domestic travel consumption, travel preference, travel demand, and control variables. It indicates that the respondents’ average travel consumption in this study is 8.604. In addition, the level of travel preference (mean value = 5.687) and travel demand (mean value = 8.332) are all above the average. In addition, the respondents’ average income per month is 8.671. The gender distribution of the respondents is average by the male (48.2%) and female (51.8%). The mean age of the respondents is 44 years and 43.4% of them have a bachelor’s degree or above. Most of the respondents are employed (76.2%). In addition, half of the respondents take an airplane to travel (49.7%). Two stars or below hotels (48.5%) are the most frequent kind of accommodation.

Compared to domestic travel consumption, it is shown that the respondents’ average consumption for outbound travel is 9.450, as shown in

Table 2. Descriptive Statistics for Outbound travelers.

Core Variables	Mean	SD	Range	Med
Outbound Travel Consumption	9.450	0.762	6.957–11.695	9.473
Travel Preference	7.782	3.879	1–60	7
Travel Demand	9.444	0.900	4.595–13.122	9.616
Control variables
Income	8.884	0.639	6.215–11.513	8.854
Gender
Female	0.517	0.500	0–1	1
Male	0.483	0.500	0–1	0
Age	43.966	13.867	18–70	42
Education:
Bachelor’s degree or above	0.520	0.500	0–1	1
Other	0.480	0.500	0–1	0
Occupation:
Employed	0.785	0.411	0–1	1
Retired	0.168	0.374	0–1	0
Student	0.015	0.123	0–1	0
Unemployed	0.028	0.164	0–1	0
Other	0.005	0.068	0–1	0
Transportation:
Airplane	0.755	0.430	0–1	1
Cruise	0.018	0.132	0–1	0
Bus	0.018	0.134	0–1	0
Car	0.021	0.145	0–1	0
High-speed Train	0.140	0.347	0–1	0
Regular Train	0.048	0.213	0–1	0
Other	0	0	——	0
Accommodation:
Two-star or below	0.359	0.480	0–1	0
Three-star	0.343	0.475	0–1	0
Four-star or above	0.297	0.457	0–1	0
Other	0	0.028	0–1	0
Stay Time	7.227	10.334	1–300	6

The number of observations is 1304.

. The respondents’ average income is 8.884, which is quite higher than those who only took domestic travel in 2016. The gender distribution of the respondents is average by the male (48.3%) and female (51.7%). More than half of the respondents (52.0%) have a bachelor’s degree or above. The two-star or below the hotel (35.9%) is the most frequent kind of accommodation for the outbound travel respondents. Furthermore, the average stay time is longer than that of the domestic group.

4. Empirical Strategy

To examine the influence of travel preference on travel consumption, we constructed three classical linear regression models [57]. The classical linear regressions were specified according to the following linear formulation:

$$Y_d=\alpha +\beta \left(TP\right)+{\sum }_{i=1}^{18}{r}_i{X}_i+\epsilon$$

(1)

$$TD=\alpha +\beta \left(TP\right)+{\sum }_{i=1}^{18}{r}_i{X}_i+\epsilon$$

(2)

$$Y_d=\alpha +\beta \left(TP\right)+g\left(TD\right)+{\sum }_{i=1}^{18}{r}_i{X}_i+\epsilon$$

(3)

where Y_d is domestic travel consumption per person per year for the sample of 4942 tourists surveyed, α is the constant term, TP is travel preference, β is the coefficient of travel preference, X_i is the control variable, r_i is the coefficient of X_i, ε is stochastic disturbance term capturing all influences on tourist consumption other than those associated with the measured regressions. In the second and third models, we added travel demand. TD is travel demand; g is the coefficient of TD. Models (1)–(3) allowed us to evaluate Hypotheses 1 that travel preference influences tourism consumption with the mediating role of travel demand in the relationship between travel preference and domestic travel consumption.

In models (4)–(6), we used outbound travel consumption as a dependent variable and repeated the regression process of domestic travel consumption in models (4)–(6). Models 4–6 allowed us to evaluate Hypotheses 2 and 3 that travel preference influences travel consumption through the mediating role of travel demand in the relationship between travel preference and outbound travel consumption.

$$Y_o=\alpha +\beta \left(TP\right)+{\sum }_{i=1}^{18}{r}_i{X}_i+\epsilon$$

(4)

$$TD=\alpha +\beta \left(TP\right)+{\sum }_{i=1}^{18}{r}_i{X}_i+\epsilon$$

(5)

$$Y_o=\alpha +\beta \left(TP\right)+g\left(TD\right)+{\sum }_{i=1}^{18}{r}_i{X}_i+\epsilon$$

(6)

where $Y_o$ is outbound travel consumption per person per year for the sample of 1304 tourists surveyed, the rest is the same as domestic travel.

Heteroscedasticity generally affects the efficiency of the estimator and therefore the individual significance of the estimates. Heteroscedasticity can be treated by the weighted least square (WLS). Therefore, at this time, a WLS estimation was chosen.

In addition, travel consumption will affect people’s travel experience, which will have a negative impact on people’s travel preferences. Therefore, there may be endogenous problems caused by reverse causality in the model. Because the cross-sectional data is used in this paper, variables of the lag period cannot be used as a tool variable to solve the endogenous problem, so they must be strictly exogenous tool variables. However, there are few studies on the explanatory variable of travel preference and there is no research on the method of using tool variables, and no suitable tool variables are found in CTLS 2016. Therefore, this article does not discuss endogenous issues for the time being.

5. Results and Discussion

5.1. The Effects of Travel Preference and Travel Demand on Travel Consumption

The three hypotheses were tested using mediated multiple regression [57], where domestic and outbound travel consumption regressed onto the predictors and control variables separately. Use OLS to test hypothesis 1–3 before using WLS regression. The results are shown in the table. White’s test results show that the p value is equal to 0.000, strongly rejecting the original assumption of homoscedasticity and believing that there is heteroscedasticity. So WLS should be used to test Hypothesis 1–3.

First, hypothesis 1 was tested as seen in

Table 3. WLS Regression Model for Domestic Travel Consumption.

Variables	Model 1	Model 2	Model 3
Travel Preference	0.019 ***	0.043 ***	0.016 ***
	(4.910)	(10.325)	(4.511)
Travel Demand			0.287 ***
			(25.846)
Control Variables
Income	0.293 ***	0.271 ***	0.227 ***
	(15.208)	(13.419)	(11.797)
Gender
Male
Female	0.045 **	0.042 *	0.030
	(2.108)	(1.884)	(1.424)
Age	−0.001	−0.001	−0.000
	(−0.633)	(−0.706)	(−0.411)
Education:
Bachelor degree or above Other
Other	−0.039 *	−0.065 ***	−0.022
	(−1.655)	(−2.598)	(−0.936)
Occupation status:
Employed
Other	−0.156	0.131	−0.174 *
	(−1.525)	(1.223)	(−1.731)
Retired	0.054	−0.018	0.064 *
	(1.479)	(−0.460)	(1.800)
Student	0.148 *	0.270 ***	0.077
	(1.954)	(3.402)	(1.053)
Unemployed	−0.105 *	−0.020	−0.114 **
	(−1.763)	(−0.327)	(−2.011)
Main transportation mode:
Airplane
Boat	−0.208 *	0.095	−0.250 **
	(−1.944)	(0.846)	(−2.254)
Bus	−0.492 ***	−0.225 ***	−0.493 ***
	(−11.302)	(−4.942)	(−11.900)
Car	−0.261 ***	−0.112 **	−0.248 ***
	(−5.864)	(−2.405)	(−5.723)
High-speed rail	−0.197 ***	−0.097 ***	−0.179 ***
	(−7.707)	(−3.607)	(−7.031)
Train	−0.464 ***	−0.330 ***	−0.343 ***
	(−11.350)	(−7.707)	(−8.914)
other	−0.580 ***	−0.491 ***	−0.477 ***
	(−3.395)	(−2.752)	(−3.206)
Accommodation:
Two stars or below
Three stars	0.040	0.041	0.017
	(1.611)	(1.579)	(0.685)
Four stars or above	0.173 ***	0.113 ***	0.165 ***
	(5.936)	(3.726)	(5.627)
Other	−0.873 ***	0.244	−0.930 ***
	(−5.594)	(1.491)	(−6.226)
Length of stay	0.038 ***	−0.003	0.022 ***
	(10.481)	(−0.857)	(9.854)
Constant	5.909 ***	5.840 ***	4.157 ***
	(33.095)	(31.241)	(22.204)
Observations	4942	4942	4942
R-squared	0.197	0.110	0.290
Ajust_R2	0.194	0.107	0.287
F-value	63.53	32.18	100.49
Indirect effect	0.010 ***
Direct effect	0.005
Total effect	0.014 ***

Model 1 and Model 3’s Dependent variable are Domestic Travel Consumption per person in 2016. Model 2’s Dependent variable is Travel Demand. Estimation method: WLS Regression. t-statistics are in parentheses. * p < 0.1; ** p < 0.05; *** p < 0.01.

. The result shows that hypothesis 1 is supported. The travel preference, as hypothesized, has a significant positive effect on travel consumption at 1% level (p < 0.01) in Model (1) after controlling the effect of control variables. Our findings confirm previous evidence on the relevance of travel preference as a good predictor of tourists’ consumption. Model 2 shows that travel preference has a positive and significant effect on travel demand. When we put travel demand and travel preference into the model as shown in Model (3), travel demand is significant, and travel preference is still positively significant, which indicates that travel demand has a mediating effect between travel preference and domestic travel consumption. The results of the BG test show that there is no autocorrelation.

When it comes to the outbound travel consumption, travel preference and travel demand were entered in the first and third steps. Like domestic travel, travel preference has a positive and significant effect on outbound travel consumption in Model 4 after controlling the effect of control variables. Travel preference has a significant positive effect on travel demand at 1% level (p < 0.01) in Model 5 after controlling the effect of control variables. Model 6 regresses travel consumption to travel demand and travel preference at the same time. As shown in Model 6, travel demand is significant (p < 0.05). The results of BG test show that there is no autocorrelation. The result shows that Travel Preference is positively related to outbound travel consumption due to the transaction utility effect. On the other hand, the positive mediation effect on outbound consumption shows that the net effect of the loss aversion and the transaction utility effects are positive because the transaction utility effect outweighs the loss aversion effect.

5.2. Other Effects on Travel Consumptions

The regression results reported in Table 3 and

Table 4. WLS Regression Model for Outbound Travel Consumption.

Variables	Model 4	Model 5	Model 6
Travel Preference	0.025 ***	0.048 ***	0.006
	(4.290)	(6.636)	(1.299)
Travel Demand			0.316 ***
			(16.068)
Control Variables
Income	0.296 ***	0.295 ***	0.208 ***
	(8.679)	(7.171)	(6.373)
Gender
Male
Female	0.007	−0.031	0.011
	(0.180)	(−0.663)	(0.308)
Age	0.005 **	0.002	0.006 ***
	(2.416)	(0.914)	(3.399)
Education:
Bachelor degree or above Other
Other	−0.120 ***	−0.145 ***	−0.098 **
	(−2.725)	(−2.742)	(−2.387)
Occupation status:
Employed
Other	0.028	0.288	−0.176
	(0.099)	(0.825)	(−0.649)
Retired	0.012	0.010	0.010
	(0.184)	(0.118)	(0.156)
Student	0.422 **	0.768 ***	0.183
	(2.421)	(3.623)	(1.071)
Unemployed	−0.356 ***	0.005	−0.322 ***
	(−3.016)	(0.038)	(−3.014)
Main transportation mode:
Airplane
Boat	0.016	−0.155	0.027
	(0.112)	(−0.883)	(0.198)
Bus	−0.437 ***	−0.323 *	−0.282 **
	(−3.128)	(−1.942)	(−2.286)
Car	−0.379 ***	−0.388 **	−0.138
	(−2.886)	(−2.462)	(−1.182)
High-speed rail	−0.195 ***	−0.294 ***	−0.114 **
	(−3.480)	(−4.372)	(−2.227)
Train	−0.443 ***	−0.617 ***	−0.216 ***
	(−4.896)	(−5.675)	(−2.809)
Accommodation:
Two stars or below
Three stars	−0.005	−0.013	0.028
	(−0.102)	(−0.231)	(0.645)
Four stars or above	0.274 ***	0.119 **	0.264 ***
	(5.624)	(2.024)	(5.687)
Other	−0.113	1.511	−0.578
	(−0.149)	(1.579)	(−0.534)
Length of stay	0.005 ***	−0.011 ***	0.009 ***
	(2.959)	(−5.155)	(10.425)
Constant	6.424 ***	6.557 ***	4.214 ***
	(20.280)	(17.145)	(12.993)
Observations	1304	1304	1304
R-squared	0.200	0.171	0.335
Ajust_R2	0.189	0.159	0.325
F-value	17.88	14.69	33.99
Indirect effect	0.013 ***
Direct effect	0.006
Total effect	0.020 ***

Model 4 and Model 6’s Dependent variable are Outbound Travel Consumption per person in 2016. Model 5’s Dependent variable is Travel Demand. Estimation method: WLS Regression. t-statistics are in parentheses. * p < 0.1; ** p < 0.05; *** p < 0.01.

also include several other independent variables. In terms of travel consumption, most of the control variables are significant, especially the groups of economic constraints and trip-related characteristics. First, our findings confirm previous evidence on the relevance of income as one of the most important drivers of travel consumption as a case of economic constraint variables. Second, among the socio-economic characteristics, the occupation status was mostly found to significantly influence travel consumption. The result of occupation status is in line with previous evidence. The more interesting finding is that students tend to spend more than the employed. As for age, this outcome may be because, as argued by Brida and Scuderi [7], the age effect is very sensitive to how the variable is measured. Travel consumption is also marked by significant differences in the level of education. An individual with a bachelor’s degree or above tends to spend more money than the lower-educated group. Finally, we examine the trip-related characteristics. According to our findings, air travel, as expected, is associated with the highest consumption. With traveling by airplane being the reference group, other variables of the travel transportation mode are significantly negative to the consumption. Consequently, the present results support previous research by showing that mode of transportation is an important predictor of travel consumption [52,53,54]. The differences in travel consumption between the five categories of accommodations are comparable to those between the various modes of transportation. Compared to respondents choosing two stars or below hotels, travel consumption tends to increase for respondents choosing hotels of four stars or above. This is probably due to the fact that the hotel price is proportional to the star levels of hotels. In our study, the length of stay is found to be positively related to travel consumption.

Overall, our findings confirm the main results of the previous empirical literature on travel consumption. Moreover, we show how travel consumption is characterized by a great degree of heterogeneity on an individual’s travel preference. Based on behavioral economics theory, if the price of outbound travel is above the reference price, then travelers face the effects of loss aversion and transaction utility. Finally, the higher degree of travel preference still causes more outbound travel consumption. Although we have accounted for factors related to heterogeneity by adding a wide set of control variables and the WLS regression, the results may not be robust. In other words, the main findings of this study may be methodological artifacts. This issue is investigated in the next section by different means of specification tests, as discussed in Section 5.5, which is specifically designed to analyze robustness.

5.3. Sub-Regional Test

Next, we analyzed the influence of travel preference on tourism consumption. We wanted to check if the relationship between travel preference and consumption was the same as different levels of economic development and different social environments of the cities in which the samples are located. First,

Table 5. WLS Regression Model for Travel Consumption for cities in East China.

Variables	Domestic			Outbound
	Model 1	Model 2	Model 3	Model 4	Model 5	Model 6
Travel Preference	0.013 *	0.020 ***	0.010	0.043 ***	0.066 ***	0.026 *
	(1.688)	(2.701)	(1.530)	(3.050)	(3.226)	(1.963)
Travel Demand			0.248 ***			0.282 ***
			(10.078)			(6.190)
Other variables	——	——	——	——	——	——
Constant	5.843 ***	5.709 ***	4.513 ***	8.198 ***	6.575 ***	6.498 ***
	(14.947)	(13.630)	(11.330)	(11.333)	(6.383)	(9.052)
Observations	953	953	953	218	218	218
R-squared	0.201	0.106	0.293	0.238	0.207	0.376
Ajust_R2	0.185	0.088	0.277	0.173	0.140	0.319
F-value	12.38	5.82	19.28	3.67	3.08	6.66
Indirect effect	0.004 **			0.017 ***
Direct effect	0.009 *			0.021 *
Total effect	0.013 **			0.039 ***

Model 1 and Model 3’s Dependent variable are Domestic Travel Consumption per person in 2016. Model 2’s Dependent variable is Travel Demand. Model 4 and Model 6’s Dependent variable are Outbound Travel Consumption per person in 2016. Model 5’s Dependent variable is Travel Demand. Estimation method: WLS Regression. t-statistics are in parentheses. * p < 0.1; ** p < 0.05; *** p < 0.01.

shows that the influence of domestic travel preference on travel consumption is not significant. However, these results are consistent with Table 3 and Table 4 overall. Therefore, in the eastern cities, the transaction utility of tourists’ outbound travel is higher than the loss aversion effect, so the outbound travel preferences have a positive impact on travel consumption, and the travel demand plays a positive mediating role in the process.

Second,

Table 6. WLS Regression Model for Travel Consumption for cities in Northeast and Central China.

Variables	Domestic			Outbound
	Model 1	Model 2	Model 3	Model 4	Model 5	Model 6
Travel Preference	0.013 *	0.020 ***	0.010	0.043 ***	0.066 ***	0.026 *
	(1.688)	(2.701)	(1.530)	(3.050)	(3.226)	(1.963)
Travel Demand			0.248 ***			0.282 ***
			(10.078)			(6.190)
Other variables	——	——	——	——	——	——
Constant	5.843 ***	5.709 ***	4.513 ***	8.198 ***	6.575 ***	6.498 ***
	(14.947)	(13.630)	(11.330)	(11.333)	(6.383)	(9.052)
Observations	953	953	953	218	218	218
R-squared	0.201	0.106	0.293	0.238	0.207	0.376
Ajust_R2	0.185	0.088	0.277	0.173	0.140	0.319
F-value	12.38	5.82	19.28	3.67	3.08	6.66
Indirect effect	0.004 **			0.017 ***
Direct effect	0.009 *			0.021 *
Total effect	0.013 **			0.039 ***

Model 1 and Model 3’s Dependent variable are Domestic Travel Consumption per person in 2016. Model 2’s Dependent variable is Travel Demand. Model 4 and Model 6’s Dependent variable are Outbound Travel Consumption per person in 2016. Model 5’s Dependent variable is Travel Demand. Estimation method: WLS Regression. t-statistics are in parentheses. * p < 0.1; ** p < 0.05; *** p < 0.01.

shows the results of cities in northeast and central China. The result in Table 5 is consistent with the results of Table 3 and Table 4 and says that in Northeast and Central China cities, tourists’ domestic travel preferences and outbound travel preferences both have a positive impact on travel consumption, and the travel demand plays a positive mediating role in the process.

Finally,

Table 7. WLS Regression Model for Travel Consumption for cities in West China.

Variables	Domestic			Outbound
	Model 1	Model 2	Model 3	Model 4	Model 5	Model 6
Travel Preference	0.042 ***	0.041 ***	0.031 ***	−0.009	−0.003	−0.011
	(4.105)	(3.669)	(3.243)	(−0.686)	(−0.182)	(−1.157)
Travel Demand			0.303 ***			0.420 ***
			(10.916)			(9.066)
Other variables	——	——	——	——	——	——
Constant	5.985 ***	6.040 ***	4.178 ***	7.160 ***	6.405 ***	3.748 ***
	(12.602)	(11.358)	(8.743)	(7.928)	(5.723)	(4.552)
Observations	672	672	672	190	190	190
R-squared	0.252	0.162	0.383	0.235	0.311	0.566
Ajust_R2	0.232	0.140	0.366	0.159	0.243	0.520
F-value	12.93	7.43	22.50	3.10	4.56	12.37
Indirect effect	0.013 ***			−0.003
Direct effect	0.018 ***			−0.006
Total effect	0.031 ***			−0.009

Model 1 and Model 3’s Dependent variable are Domestic Travel Consumption per person in 2016. Model 2’s Dependent variable is Travel Demand. Model 4 and Model 6’s Dependent variable are Outbound Travel Consumption per person in 2016. Model 5’s Dependent variable is Travel Demand. Estimation method: WLS Regression. t-statistics are in parentheses. *** p < 0.01.

shows the results of cities in west China. The results of models 1–3 in Table 7 are consistent with Table 3 results. However, the travel preferences in Model 4 and Model 5 are not significant with a negative sign. Therefore, we can conclude that in the western region, outbound travel preferences do not have a positive role in promoting travel consumption, and travel demand has not played an active mediating role in this process. It may be that the level of economic development in the western region is relatively backward and the consumption concept of local residents is conservative, which leads to the transaction utility effect to be lower than the loss aversion effect.

5.4. Sub-Gender Test

To compare whether male and female travel preferences have the same impact on travel consumption, the total sample was divided into two groups according to gender.

Table 8. WLS Regression Model for Travel Consumption for Male.

Variables	Domestic			Outbound
	Model 1	Model 2	Model 3	Model 4	Model 5	Model 6
Travel Preference	0.035 ***	0.058 ***	0.029 ***	0.019 **	0.033 ***	0.007
	(5.955)	(9.222)	(5.283)	(2.295)	(3.227)	(0.948)
Travel Demand			0.284 ***			0.293 ***
			(17.423)			(9.915)
Other variables	——	——	——	——	——
Constant	6.152 ***	6.066 ***	4.273 ***	6.077 ***	5.958 ***	4.136 ***
	(24.302)	(22.729)	(15.945)	(14.204)	(11.581)	(9.310)
Observations	2385	2385	2385	630	630	630
R-squared	0.186	0.118	0.274	0.228	0.234	0.346
Ajust_R2	0.180	0.111	0.268	0.206	0.212	0.327
F-value	30.01	17.58	46.97	10.61	10.97	17.98
Indirect effect	0.014 ***			0.011 ***
Direct effect	0.015 ***			0.005
Total effect	0.029 ***			0.016 **

Model 1 and Model 3’s Dependent variable are Domestic Travel Consumption per person in 2016. Model 2’s Dependent variable is Travel Demand. Model 4 and Model 6’s Dependent variable are Outbound Travel Consumption per person in 2016. Model 5’s Dependent variable is Travel Demand. Estimation method: WLS Regression. t-statistics are in parentheses. ** p < 0.05; *** p < 0.01.

shows the results for men. Travel preference in Models 1, 2, 4, and 5 are significantly positive and travel demand in Models 3 and 6 are also significantly positive, thus the results of men were consistent with those of Table 3 and Table 4.

Table 9. WLS Regression Model for Travel Consumption for Female.

Variables	Domestic			Outbound
	Model 1	Model 2	Model 3	Model 4	Model 5	Model 6
Travel Preference	0.000	0.024 ***	−0.008	0.027 ***	0.053 ***	0.004
	(0.041)	(4.292)	(−1.607)	(3.173)	(5.024)	(0.642)
Travel Demand			0.269 ***			0.340 ***
			(17.796)			(12.515)
Other variables	——	——	——	——	——	——
Constant	5.603 ***	5.601 ***	3.936 ***	6.883 ***	6.884 ***	4.473 ***
	(22.937)	(22.009)	(15.602)	(14.944)	(12.438)	(9.649)
Observations	2557	2557	2557	674	674	674
R-squared	0.226	0.124	0.331	0.190	0.143	0.337
Ajust_R2	0.221	0.118	0.326	0.170	0.121	0.320
F-value	41.18	19.93	66.13	9.60	6.84	19.60
Indirect effect	0.006 ***			0.013 ***
Direct effect	−0.003			0.007
Total effect	0.003			0.021 ***

Model 1 and Model 3’s Dependent variable are Domestic Travel Consumption per person in 2016. Model 2’s Dependent variable is Travel Demand. Model 4 and Model 6’s Dependent variable are Outbound Travel Consumption per person in 2016. Model 5’s Dependent variable is Travel Demand. Estimation method: WLS Regression. t-statistics are in parentheses. *** p < 0.01.

shows the results of women. Travel preference in Model 1 is not significant, while travel preferences in Models 2, 4, and 5 are significantly positive, and travel demand in Models 3 and 6 are also significantly positive, thus the results for women were consistent with those of Table 3 and Table 4 overall. Therefore, we conclude that there was no significant difference in the results between men and women. Thus, we can say that the original results are robust.

5.5. Robustness Test

In this section, we test the stability of our baseline estimates to alternative regression measures: robust regression and quantile regression results are reported in

Table 10. Huber-White Robust Standard Errors HC1.

Variables	Domestic				Outbound
	Model 1	Model 2	Model 3	Model 4	Model 5	Model 6
Travel Preference	0.014 ***	0.029 ***	0.005	0.020 ***	0.037 ***	0.006
	(3.331)	(6.256)	(1.209)	(3.363)	(4.279)	(1.418)
Travel Demand			0.340 ***			0.367 ***
			(20.044)			(13.952)
Other variables	——	——	——	——	——	——
Constant	5.959 ***	5.901 ***	3.953 ***	6.492 ***	6.693 ***	4.039 ***
	(30.594)	(27.624)	(19.275)	(18.693)	(16.332)	(11.581)
Observations	4942	4942	4942	1304	1304	1304
R-squared	0.195	0.107	0.297	0.196	0.155	0.354
Root MSE	0.731	0.766	0.684	0.688	0.833	0.617
F-value	55.51	24.80	87.83	——	——	——

Model 1 and Model 3’s Dependent variable are Domestic Travel Consumption per person in 2016. Model 2’s Dependent variable is Travel Demand. Model 4 and Model 6’s Dependent variable are Outbound Travel Consumption per person in 2016. Model 5’s Dependent variable is Travel Demand. Estimation method: Robustness Regression. t-statistics are in parentheses. *** p < 0.01.

,

Table 11. MacKinnon-White SE HC2.

Variables	Domestic				Outbound
	Model 1	Model 2	Model 3	Model 4	Model 5	Model 6
Travel Preference	0.014 ***	0.029 ***	0.005	0.020 ***	0.037 ***	0.006
	(3.154)	(5.941)	(1.162)	(3.301)	(4.004)	(1.405)
Travel Demand			0.340 ***			0.367 ***
			(19.862)			(13.880)
Other variables	——	——	——	——	——	——
Constant	5.959 ***	5.901 ***	3.953 ***	6.492 ***	6.693 ***	4.039 ***
	(30.397)	(27.486)	(19.245)	(18.628)	(16.240)	(11.539)
Observations	4942	4942	4942	1304	1304	1304
R-squared	0.195	0.107	0.297	0.196	0.155	0.354
Root MSE	0.731	0.766	0.684	0.688	0.833	0.617
F-value	55.14	24.62	87.34	——	——	——

Model 1 and Model 3’s Dependent variable are Domestic Travel Consumption per person in 2016. Model 2’s Dependent variable is Travel Demand. Model 4 and Model 6’s Dependent variable are Outbound Travel Consumption per person in 2016. Model 5’s Dependent variable is Travel Demand. Estimation method: Robustness Regression. t-statistics are in parentheses. *** p < 0.01.

,

Table 12. Long-Ervin SE HC3.

Variables	Domestic				Outbound
	Model 1	Model 2	Model 3	Model 4	Model 5	Model 6
Travel Preference	0.014 ***	0.029 ***	0.005	0.020 ***	0.037 ***	0.006
	(2.964)	(5.605)	(1.110)	(3.182)	(3.500)	(1.374)
Travel Demand			0.340 ***			0.367 ***
			(19.603)			(13.681)
Other variables	——	——	——	——	——	——
Constant	5.959 ***	5.901 ***	3.953 ***	6.492 ***	6.693 ***	4.039 ***
	(30.102)	(27.266)	(19.173)	(18.415)	(15.944)	(11.403)
Observations	4942	4942	4942	1304	1304	1304
R-squared	0.195	0.107	0.297	0.196	0.155	0.354
Root MSE	0.731	0.766	0.684	0.688	0.833	0.617
F-value	54.57	24.35	86.50	16.20	11.96	29.20

Model 1 and Model 3’s Dependent variable are Domestic Travel Consumption per person in 2016. Model 2’s Dependent variable is Travel Demand. Model 4 and Model 6’s Dependent variable are Outbound Travel Consumption per person in 2016. Model 5’s Dependent variable is Travel Demand. Estimation method: Robustness Regression. t-statistics are in parentheses. *** p < 0.01.

,

Table 13. Quantile Regression (0.3).

Variables	Domestic				Outbound
	Model 1	Model 2	Model 3	Model 4	Model 5	Model 6
Travel Preference	0.010 ***	0.027 ***	−0.001	0.016 ***	0.043 ***	0.005
	(2.596)	(8.277)	(−0.382)	(2.689)	(5.070)	(1.038)
Travel Demand			0.438 ***			0.435 ***
			(27.787)			(18.601)
Other variables	——	——		——	——	——
Constant	5.378 ***	5.403 ***	3.123 ***	6.323 ***	6.294 ***	3.630 ***
	(23.636)	(27.599)	(13.841)	(16.869)	(11.817)	(10.161)
Observations	4942	4942	4942	1304	1304	1304
Pseudo R2	0.131	0.054	0.215	0.117	0.083	0.231

Model 1 and Model 3’s Dependent variable are Domestic Travel Consumption per person in 2016. Model 2’s Dependent variable is Travel Demand. Model 4 and Model 6’s Dependent variable are Outbound Travel Consumption per person in 2016. Model 5’s Dependent variable is Travel Demand. Estimation method: Quantile Regression. t-statistics are in parentheses. *** p < 0.01.

,

Table 14. Quantile Regression (0.5).

Variables	Domestic				Outbound
	Model 1	Model 2	Model 3	Model 4	Model 5	Model 6
Travel Preference	0.010 **	0.023 ***	0.002	0.017 ***	0.038 ***	0.002
	(2.555)	(8.097)	(0.546)	(2.688)	(5.477)	(0.338)
Travel Demand			0.392 ***			0.414 ***
			(26.735)			(17.771)
Other variables	——	——	——	——	——	——
Constant	5.661 ***	5.909 ***	3.517 ***	6.449 ***	5.549 ***	3.561 ***
	(24.740)	(33.582)	(16.742)	(16.228)	(12.706)	(10.007)
Observations	4942	4942	4942	1304	1304	1304
Pseudo R2	0.117	0.046	0.185	0.100	0.114	0.201

Model 1 and Model 3’s Dependent variable are Domestic Travel Consumption per person in 2016. Model 2’s Dependent variable is Travel Demand. Model 4 and Model 6’s Dependent variable are Outbound Travel Consumption per person in 2016. Model 5’s Dependent variable is Travel Demand. Estimation method: Quantile Regression. t-statistics are in parentheses. ** p < 0.05; *** p < 0.01.

,

Table 15. Quantile Regression (0.7).

Variables	Domestic				Outbound
	Model 1	Model 2	Model 3	Model 4	Model 5	Model 6
Travel Preference	0.012 ***	0.027 ***	0.004	0.019 ***	0.035 ***	0.007
	(3.100)	(9.281)	(1.064)	(3.199)	(5.537)	(1.058)
Travel Demand			0.310 ***			0.349 ***
			(18.463)			(12.378)
Other variables	——	——	——	——	——	——
Constant	6.198 ***	6.077 ***	4.365 ***	6.492 ***	6.363 ***	4.402 ***
	(26.847)	(34.510)	(18.156)	(17.027)	(15.843)	(10.232)
Observations	4942	4942	4942	1304	1304	1304
Pseudo R2	0.095	0.070	0.137	0.102	0.081	0.179

Model 1 and Model 3’s Dependent variable are Domestic Travel Consumption per person in 2016. Model 2’s Dependent variable is Travel Demand. Model 4 and Model 6’s Dependent variable are Outbound Travel Consumption per person in 2016. Model 5’s Dependent variable is Travel Demand. Estimation method: Quantile Regression. t-statistics are in parentheses. *** p < 0.01.

and

Table 16. Quantile Regression (0.9).

Variables	Domestic				Outbound
	Model 1	Model 2	Model 3	Model 4	Model 5	Model 6
Travel Preference	0.007	0.022 ***	0.008 *	0.032 ***	0.036 ***	0.013
	(1.556)	(5.955)	(1.764)	(3.163)	(3.294)	(1.304)
Travel Demand			0.183 ***			0.276 ***
			(8.774)			(6.050)
Other variables	——	——	——	——	——	——
Constant	7.124 ***	7.166 ***	6.069 ***	6.639 ***	6.978 ***	4.540 ***
	(24.694)	(31.328)	(20.293)	(10.345)	(10.035)	(6.517)
Observations	4942	4942	4942	1304	1304	1304
Pseudo R2	0.077	0.019	0.095	0.119	0.103	0.182

Model 1 and Model 3’s Dependent variable are Domestic Travel Consumption per person in 2016. Model 2’s Dependent variable is Travel Demand. Model 4 and Model 6’s Dependent variable are Outbound Travel Consumption per person in 2016. Model 5’s Dependent variable is Travel Demand. Estimation method: Quantile Regression. t-statistics are in parentheses. * p < 0.1; *** p < 0.01.

respectively.

This paper uses the Huber–White robust standard errors HC1 to test the relationship between preference and consumption. It shows that the travel preference and travel demand statistical significance and signs are similar to Table 3 and Table 4 results.

This paper used MacKinnon–White SE HC2 to test the relationship between preference and consumption. It shows that the results in Table 11 are similar to Table 10.

We also used the Long–Ervin SE HC3 to test the relationship between preference and consumption. It shows that the results in Table 12 are similar to Table 10 and Table 11 results.

Here, we utilized quantile regression to test the relationship between preference and consumption. Table 13 reports the results of the regressions in the tri-sectional quantiles. It shows that travel preference in Models 1, 2, 4, and 5 is significantly positive, and travel demand in Models 3 and 6 is significantly positive. These are consistent with Table 3 and Table 4 results as well.

Table 14 shows the results of the regressions in the quintile, which are also consistent with Table 3 and Table 4.

Table 15 reports the results of the regressions in the seventh percentiles, which are still consistent with Table 3 and Table 4.

Table 16 shows the results of the regressions in the ninth percentiles. It shows that travel preference in Model 1 is not significant, while travel preference in Models 2, 4, and 5 are significantly positive, and travel demand in Models 3 and 6 are also significantly positive, thus the results of Table 16 are consistent with those of Table 3 and Table 4 overall.

Table 10, Table 11, Table 12, Table 13, Table 14, Table 15 and Table 16 clearly support the findings of the more parsimonious WLS regression analysis reported in Table 3 and Table 4. We conclude that our estimated travel preferences and perceived values are remarkably robust to the usual specification tests.

There may be ineffective travel demand in residents with unattended children or elderly people in need of care, thus affecting the impact of travel preference on travel consumption. Therefore, we removed the residents with unattended children aged 3 years and below and elderly people with poor health and used the WLS regression method. It was found in

Table 17. Robustness Test.

Variables	Domestic				Outbound
	Model 1	Model 2	Model 3	Model 4	Model 5	Model 6
Travel Preference	0.020 ***	0.047 ***	0.015 ***	0.027 ***	0.049 ***	0.007
	(4.908)	(10.494)	(3.930)	(4.359)	(6.245)	(1.359)
Travel Demand			0.290 ***			0.316 ***
			(24.252)			(14.894)
Other variables	——	——	——	——	——	——
Constant	5.795 ***	5.868 ***	3.910 ***	6.445 ***	6.556 ***	4.376 ***
	(30.468)	(29.593)	(19.454)	(18.811)	(15.776)	(12.398)
Observations	4279	4279	4279	1109	1109	1109
R-squared	0.199	0.109	0.293	0.193	0.167	0.330
Ajust_R2	0.195	0.105	0.290	0.179	0.153	0.318
F-value	55.52	27.38	88.15	14.45	12.13	28.22
Indirect effect	0.011 ***			0.013 ***
Direct effect	0.005			0.009 *
Total effect	0.016 ***			0.022 ***

Model 1 and Model 3’s Dependent variable are Domestic Travel Consumption per person in 2016. Model 2’s Dependent variable is Travel Demand. Model 4 and Model 6’s Dependent variable are Outbound Travel Consumption per person in 2016. Model 5’s Dependent variable is Travel Demand. Estimation method: WLS Regression. t-statistics are in parentheses. * p < 0.1; *** p < 0.01.

that both domestic and outbound travel has the same positively significant characteristics as the original results, and the coefficient is greater than the original results. After removing the possibly invalid samples, we concluded that the impact of domestic travel preference on domestic travel consumption becomes greater, and the interaction between outbound travel preference and demand has a more obvious impact on outbound travel consumption.

The above findings show that the three hypotheses are supported. Domestic travel preference positively impacts domestic travel consumption through the mediating role of domestic travel demand. Outbound travel preference can significantly affect outbound travel consumption. Specifically, outbound travel preference is positively related to outbound travel consumption due to the transaction utility effect. Meanwhile, travel preference has a negative impact on travel consumption through the mediating role of travel demand under the loss aversion effect.

6. Conclusions, Contributions and Limitations

6.1. Research Conclusions and Theoretical Contributions

This study examined the interaction between travel preference and travel consumption from the perspectives of traditional economics and behavioral economics and analyzed the impact of travel preference on travel consumption. Based on the theory of the reference point, this paper adjusted the consumption model of traditional economics, constructed the influence mechanism model of preference-demand-consumption for travel, and attempted to explore the influencing factors of different rules between domestic travel and outbound travel, drawing the following conclusions.

First, for domestic travel, travel preference has a positive impact on travel consumption through the mediating role of travel demand. Considering preference as one of the elements of attitude in the travel behavior model [17], the existence of an attitude-behavior gap, does not necessarily translate into behaviors in some contexts [58,59]. In this sense, travel preference may not translate into behavior, thus it is necessary to identify the relationship between travel preference and travel consumption. In the analysis of travel consumption in traditional economics, this study leads into more micro-behavioral economics, introduces the analysis of perceived value, such as travel demand, and attempts to construct a logically closed loop between travel preference and travel consumption. Patterson and Spreng [26] described the concept of perceived value as a cognitive-based concept, which captures the benefit–sacrifice discrepancy in much the same way disconfirmation does for variations between expectations and perceived performance. In this study, the empirical analysis of domestic travel, given by Models 1–3 prove the positive impact of travel preference on travel consumption and the significant mediating role that travel demand plays between travel preference and travel consumption.

Second, for outbound travel, travel preference can significantly affect travel consumption. The results of this study show that the increase in travel preference significantly stimulates travel consumption, and the results of Model 4 verify that there is a significant correlation between travel preference and travel consumption. Models 4–6 verify that travel demand plays an active mediating role between travel consumption and preference. This result indicates that the market price of traveling abroad in China deviates from the reference price for most people. This will lead consumers to pay special attention to the price of traveling abroad, resulting in the loss aversion and transaction utility effects [5,29]. The former restrains travel consumption and the latter promotes travel consumption. Therefore, the consumption pattern of traveling abroad is not supported by the traditional consumption model CMTE. When the market price of a commodity deviates from its reference price, preference does not necessarily lead to consumers’ choice [28]. Therefore, we constructed another logically closed loop after considering the net effects of the loss aversion and transaction utility effects to simulate the relationship between travel preference and travel consumption.

Third, the interaction between outbound travel preference and outbound travel demand has a positive impact on outbound travel consumption under the transaction utility effect. The preference for outbound travel has a negative impact on outbound travel consumption through the mediating role of outbound travel demand under the loss aversion effect. The net role of the two effects determines the direction of the impact of travel preference on travel consumption. The mediating effect of travel demand can finally verify that travel preference and travel demand can have a significant impact on travel consumption. The loss aversion effect will make travel preference and demand inverse correlation and thus reduce travel consumption. The transaction utility effect will lead travel demand to positively affect travel preference, thus stimulating travel consumption. At this point, the mediating effect of travel demand largely determines the final trend of travel consumption.

The theoretical contribution of this conclusion lies in discussing the influence of travel preference on travel consumption, and distinguishing the different laws of domestic and outbound travel consumption, especially using the method of behavioral economics, and in attempting to explain the factors that determine the real travel consumption behind the different laws. The model confirmed that travel preference has a positive mediating effect on travel consumption in domestic travel. When traveling abroad, travel preference cannot directly determine consumption, but we should take into account the net effects of two effects. We summarize that the decrease in travel consumption is caused by the loss aversion and the increase of travel comes from the transaction utility effect.

6.2. Research Inspiration and Management Significance

The conclusions of this study provide ideas and a basis to precisely stimulate travel consumption, grasp the focus of travel product publicity and marketing, and build a travel advantage development strategy. It has enlightening significance for the management practice of travel enterprises, which is embodied in the following two aspects. Hypothesis 1 and Hypothesis 3 show that the consumption laws of outbound travel and domestic travel should be separated, and that marketing strategies at the national level and travel enterprise level should have their own emphasis.

First, for domestic travel enterprise marketing, Hypothesis 1 proves that travel preference has a positive impact on domestic travel consumption. Therefore, tourism enterprises can first subdivide the market according to different travel preferences of tourists based on preferences for marketing promotion.

Second, for outbound travel marketing organizations, our results show that the influence of outbound travel preference on travel consumption is affected by the loss aversion and the transaction utility effects, and the net effect of outbound travel consumption is positive. These results show that travel preference can also positively affect outbound travel consumption for Chinese consumers. Therefore, if foreign destinations want to attract more Chinese tourists, they can use the same strategy as for domestic travel. It was found that the net effects of the loss aversion and transaction utility effects of residents in the cities of west China have no positive effect on travel consumption. Therefore, if foreign destinations want to attract tourists from west China, they can use the price strategy in marketing to reduce the impact of the loss aversion effect.

Third, we conducted a further regression analysis based on the gender of tourists. The regression results are basically consistent with the overall results, and there is no difference between men and women. Therefore, from the perspective of travel preference, there is no need to implement different marketing strategies for men and women and increase meaningless marketing costs.

6.3. Research Limitations and Future Expectations

This study also has some limitations. First, the survey samples include only big cities, not small cities and rural areas in China, so the samples cannot be studied on the impact of domestic travel preferences on tourism consumption in small cities and rural areas. Domestic tourism is not an entirely normal good but a quasi-luxury for the residents in small cities and rural areas in China due to the relatively backward economy and conservative consumption concept. Therefore, just like outbound tourism, it is necessary to analyze the influence of domestic tourism preference on tourism consumption through the reference point theory of behavioral economics in a future study. Second, this study only examined the basic market of domestic travel and outbound travel in the data collection and data analysis and does not involve the analysis of individual characteristics or group discussion for more typical travel consumers. In the future, researchers can study different characteristics of travel groups, such as gender, age, and occupation. The validity of this study could be further improved by comparing the impact of different types of tourists’ preferences on travel consumption. Third, the psychological analysis is still facing many theoretical gaps, which need more discussion and testing to find possible influencing factors. Finally, the construction of the model still needs repeated cross-comparison with the actual situation, in order to constantly adjust the adaptability of the model. For example, we can select data in countries and regions with different scales, expand survey years, infer whether the impact of psychological factors on travel consumption mentioned in this study has historical evolution laws and universality, and then constantly explore its increased social significance.