key: cord-267672-2ukr15se authors: Okuyama, Tadahiro title: Analysis of optimal timing of tourism demand recovery policies from natural disaster using the contingent behavior method date: 2017-08-10 journal: Tour Manag DOI: 10.1016/j.tourman.2017.07.019 sha: doc_id: 267672 cord_uid: 2ukr15se This paper examines the applicability of contingent behavior (hereafter, CB) method for analyzing dynamic processes and efficient policies in tourism demand recovery. The CB questionnaires used for this study used a hypothetical disaster situation of bird flu in Kyoto, Japan. Safety, event, visitor information, and price discounting policies were designed accordingly. Respondents were then asked about their willingness to travel time. The results showed the optimal timing for devising pertinent policies during the year. We found that the first step requires a safety information announcement, within one week, immediately after disaster site decontamination. The second step is the implementation of event information policy within 24th to 36th week after the disaster. The third step constitutes announcing visitor information within the 37th to 52nd week after the second step. The final step is the implementation of price discounting policy, until the 52nd week, immediately after the third step. Natural disasters have occasionally caused physical and economic damage to both tourist and non-tourist sites, leading to loss of tourism opportunities and the collapse of tourism industries (Murphy & Bayley, 1989; Ritchie, 2009) . Given the possibility of long-term economic deterioration due to continuing reduction in tourism demand, opportunity losses are a major concern for policymakers and the industry itself (Chew & Jahari, 2014) . The bird flu outbreak in Japan's Miyazaki prefecture in 2010 forced public officers to prohibit visitor entry to disaster areas, followed by the culling of influenza-stricken birds, which caused losses of approximately ¥8.1 billion (Miyazaki Prefecture, 2011) . The Great East Japan Earthquake, which occurred at a magnitude of 9.0, and the ensuing tsunami in Tohoku area (Eastside of Japan), in 2011, killed nearly 200,000 people. These disasters led to economic losses of ¥16.9 trillion, which included losses due to a decrement in the number of touristsdfrom 27.7 million in 2010 to 21.1 million in 2011 (Cabinet Office, Government of Japan, 2011; Kento, 2015) . The Great Kumamoto Earthquake, which occurred in Kyushu area (Westside of Japan) in 2016, caused 67 deaths and economic damages worth ¥2.4 million to ¥4.6 trillion to the Kumamoto and Oita prefectures. It further resulted in a decrease of approximately 2.3 million tourists to the Kyushu area between April and June 2016, compared to the same period in 2015 (Cabinet Office, Government of Japan, 2016; Kyushu Economic Research Center, 2016 ). E-mail addresses: okuyama@sun.ac.jp, z_okuyama@hotmail.com. Tourism Management In previous literature, tourism management studies have analyzed frameworks and methods for tourism recovery at disaster sites (Durocher, 1994; Faulkner, 2001; Huanga & Min, 2002; Mazzocchia & Montini, 2001; Wang, 2009) . For instance, Ritchie (2009, p. 262) noted that tourism crisis and disaster management models should be developed for decision-making. However, due to lack of tourism demand data with respect to disasters, few studies have examined the quantitative effects of recovery policies. Owing to insufficient research on this topic, this study examines a valuation method, while simultaneously measuring the quantitative effects and the optimal timing (order) of tourism recovery policies applying the contingent behavior (hereafter CB) method. By showing the optimal policy timing (order), we expect to contribute toward 1) helping policymakers when they may not be able to undertake rescue operations and recover disaster losses due to financial and human resources shortages simultaneously and 2) development of advance planning (the stage 1 of Faulkner, 2001) before potential disasters. The CB method design requires consideration of the realities and existence of disaster-related solutions. As it is difficulty to design and establish efficient solutions for earthquakes of large magnitudes, tsunamis, and typhoonsdwhich typically cause considerable damage across a wide areadthis study employs a bird flu scenario as a hypothetical natural disaster. The World Health Organization (2013) reported that, from 2003 to 2013, bird flu claimed 630 human lives globally. In Asia alone, 65% and 49.5% of all those infected by bird flu died in China and Vietnam, respectively. Brahmbhatt (2005) reported that bird flu decreased Vietnam's gross domestic product (GDP) by 0.4%. Moreover, the alarming possibility of a worldwide bird flu pandemic continues to exist. In such a scenario, approximately 5 million to 150 million people could die (Ministry of Foreign Affairs of Japan, 2007) . The remainder of this paper is structured as follows. Section 2 summarizes the main objectives of this study based on a review of previous studies. Section 3 describes the estimation models and survey questionnaires. Section 4 presents the estimation results. The discussion and conclusions appear in Sections 5 and 6, respectively. 2.1. Tourism demand recovery management from disasters Faulkner (2001) and Ritchie (2009) presented the frameworks of tourism demand recovery processes (strategies). Faulkner (2001) 's framework is divided into six stages: 1) the pre-event (pre-disaster) stage (stage 1) to mitigate the effects of disaster through advance planning, 2) the prodromal stage (stage 2), indicating the inevitability of a disaster, 3) the emergency stage (stage 3) to undertake rescue operations in the event of a disaster, 4) the intermediate stage (stage 4 ) that responds to the short-term needs (e.g., food, medicines) of people and companies in the disaster site, 5) the long-term recovery stage (stage 5), which includes reconstruction of infrastructure and victim counseling, and 6) the resolution stage (stage 6), which requires restoration of routine along with new and improved state establishments. The fifth and sixth stages are post-event stages, and the focus of this study. Thus, the policy effects from pre-event to the post-event stages and the feedback effects from the post-event to the pre-event stages described in Racherla and Hu (2009) are not our focus. Furthermore, the third and fourth stages would constitute the main parts of emergency policies. As mentioned in Ritchie (2009) , the quantitative valuation of recovery process is one of the most important tasks of tourism disaster management. Faulkner (2001) , thus, presented various strategies, such as restoration of business and consumer confidence, and repair of damaged infrastructures. The Ministry of Land, Infrastructure, Transport and Tourism of Japan (MLITT, 2009) states that the recovery process has to include management policies for safety information, pricing, visit campaigns, among others. Moreover, Beirman (2009) suggested the importance of media, public relations, and regional cooperation in case studies. Regardless of these suggestions, policymakers might not know which policies are effective, when they should be implemented, and which policy ordering is desirable under the provision of few quantitative valuations. The method used in this study could lead policymakers to make quick and appropriate decisions that may reduce or prevent damages related to a disaster. The Ministry of Land, Infrastructure, Transport and Tourism of Japan (2009) has published a manual (hereafter the MLITT manual) on the management of tourism demand recovery before and after the occurrence of infections, such as the bird flu. Fig. 1 shows the framework of the recovery process as per the MLITT manual in relation to the stages in Faulkner (2001) . The vertical axis shows tourism demand (tourists' choice probability) levels. The horizontal axis shows time series, where t 0 refers to the emergence time of the bird flu, t 1 denotes the time when the affected areas/sites are decontaminated, and t 2 denotes the time that the tourism demand recovers to the standard (pre-stage) demand level. Thus, the optimal policy (or policies) in this study refers to a policy or a combination of policies that can recover a tourism demand level immediately after t 1 is closest to or over the standard demand level at t 0 (t 2 ). The tourism demand process was categorized into Periods 1 to 4. Period 1 almost corresponds to stages 1 and 2 of Faulkner (2001) ; Period 2, to stages 3 and 4; and Periods 3 and 4, to stages 5 and 6, respectively. One of the aims of this study is to examine the recovery process by estimating the demand function after t 1 in Period 3. Theoretically, tourism demand is determined by travel prices to tourism sites, individual, or household income, and site attributes data, such as nature, safety levels, and leisure amenities (Dann, 1981; Dwyer, Forsyth, & Dwyer, 2010) . Tourism policy evaluations, which are based on demand function approaches, measure policy effects from these factor (policy variable) changes (e.g., discounting the prices and improving attributes). While micro (consumer behavior) data are frequently used for the demand analyses (Fleming & Cook, 2008; Phaneuf, Kling, & Herriges, 2000) , the difficulty of researching such data from the time series of independent consumer behaviors sometimes dissuades researchers from analyzing the dynamic processes of demand functions, especially in disaster. Macro data following the time series are also used for tourism demand analysis (Song & Witt, 2000; Wang, 2009) . Using data from the World Health Organization (2013), Kuo, Chen, Tseng, Ju, and Huang (2008, 2009) showed the negative impacts of the Severe Acute Respiratory Syndrome (better known as SARS) and bird flu on tourism activities. Page, Song, and Wu (2012) also showed that the outbreak of bird flu decreased the number of tourists to England. Kuo et al. (2008) and Chang et al.'s (2012) results indicated that analyses based on social statistics may not always be able to estimate the impacts of bird flu on tourism, given the influences of external macro-impact factors, such as economic trends, terrorism, and temperature. The CB method could overcome these micro and macro data issues, as it analyzes individual behaviors. This method enables researchers to analyze individual behaviors under (researcher -designed) hypothetical situations, and it is used in cases where observable data are limited. For instance, Whitehead, Johnson, Mason, and Walker (2008) used observable data by asking respondents the number of times they visited hockey games depending on game intervals and seat prices. Phaneuf and Earnhart (2011) measured recreational benefits of lakes using trip data under hypothetical trip time and prices. Whitehead, Dumas, Herstine, Hill, and Buerger (2010) valued the benefits of improving the widths of beaches using actual and contingent trip data collected under different accessibility conditions to beaches of various widths. Using the CB method, Whitehead (2005) analyzed hurricane evacuation behaviors by asking respondents about their order of fleeing from their homes depending on the strength of the hurricane. Overall, few previous studies have incorporated the time series factor into the CB method. This study, on the other hand, examined CB questionnaires from previous studies, such as Phaneuf and Earnhart (2011) , and developed a new CB questionnaire with time series factors for analyzing the policy timings and orderings. This study mainly examined the effects of information and pricing policies. Three information policies are included in the CB questionnaire. The first is the provision of safety information, by the Japanese government and Kyoto prefectural governments, to ensure safety in traveling to Kyoto prefecture, from the discussions of stage 5 of Faulkner (2001) and the MLITT manual. Note that tourists sometimes may not visit a disaster site without safety information. The second is the provision of event information of the respondents' preferred events that have been performed and/or new tourism facilities that have been established in Kyoto. The third is the provision of visitor information regarding the number of tourists who have already visited Kyoto. The second and third information policies are designed for improving (respondents') destination images referred by Beirman (2003) , Chew and Jahari (2014) , and Ritchie (2009) . Previous studies suggested that pricing policies, such as decrements of travel, hotel, and food costs, have a positive effect on tourism demand (Garrod & Fyall, 2000; Laarman & Gregersen, 1996) . H.I.S. (2016), a major Japanese tourism company, also implemented a tourism campaign (price discounting) in collaboration with the Japanese government to recover tourism demand from Japan to France, following the 2015 terrorist attacks in France. Thus, this study employed price discounting. Thus, tourism demand recovery will be delayed if its effects are not revealed in the recovery process. Thus, this study also examined the pricing policy effect through comparisons with information policies. Previous studies employed the time series analysis (Eilat & Einav, 2004; Gurudeo, 2012; Song, Li, Witt, & Fei, 2010) and the random utility model (Baltas, 2007) in analyzing tourism demands. Here, the CB questionnaires collect "yes/no" response data on individual decisions on travel, and thus, this study used the logit model for the estimations. Let X be a vector of variables and b 0 be a transported vector of parameters. pr, defined by equation (1), is the probability of obtaining a negative response ("no") from the ith respondent. Equation (2) shows the log-likelihood function. The estimations were performed using the glm function in R ver 3.01. 3.2. Survey questionnaires The following hypothetical site conditions were considered for the CB questionnaires: i) short distance from all respondents' homes to reduce the number of rejected responses typical with long distance travel, ii) actual bird flu experiences to add reality to the hypothetical situation described in the CB questionnaires, and iii) use of a famous site to avoid wrong answers resulting from respondents' unawareness. The questionnaires were in Japanese. This study selected Kyoto prefecture, one of Japan's most famous tourism sites, as the site for the hypothetical case. Fig. 2 shows the location of Kyoto prefecture with the hypothetical bird flu outbreak. Kyoto prefecture is located in the central part of Japan (E135 45 0 , N35 01 0 ), satisfying condition (i). Its area is 4613.21 km 2 , and the Japan Sea lies towards its north, while Nara Prefecture lies to its south and Mie prefecture is located towards its east. The population in 2014 was 2.6 million (Kyoto Prefectural Government, 2016) , and, in 2014, approximately 55,636 thousand Japanese and foreign tourists visited Kyoto (Kyoto City Government, 2014). Kyoto was the capital of Japan nearly 1100 years from the eighth century onwards. Many historical temples and shrines that were built during the period still exist. Seventeen historic sites, such as Kiyomizu-dera Temple and Nijo Castle, are recognized as World Heritage Sites. Kyoto has three major festivals, which attract tourists: the Aoi-matsuri Festival in early summer, the Gion-matsuri Festival in mid-summer, and the Jidai-matsuri Festival in fall. These events are hugely popular with both Japanese and foreign travelers. The bird flu outbreaks (Fig. 2) in this study are categorized as domestic bird flu, wild bird flu, and other bird flu (Ministry of Agriculture, Forestry, and Fisheries, 2011) . Kyoto prefecture is hypothesized to have a case of wild bird flu, satisfying condition (ii). Condition (iii) was confirmed by the survey questionnairesdrespondents were asked whether they had heard of and ever been to Kyoto. All respondents answered "yes" to the former question, while 92.6% of them answered "yes" to the latter question, thus satisfying condition (iii). The CB questionnaire was implemented in four steps, as shown in Appendix 1. The first step (Appendix 1-A) was to research respondents' credible information source to avoid respondents' distrust in an information source, described in Appendix 1-D. Specifying respondents' credible information sources reduces the rate of rejection responses in the CB questionnaires. The second step was to provide explanations on bird flu (Appendix 1-B); the third, to present a hypothetical bird flu outbreak (Appendix 1-C); and the fourth, to implement the CB questionnaire (Appendix 1-D). Alternatives and results of respondents' credible information sources are shown in Table 1 and Appendix 1-A. The most credible information source is public information (35.48%). The contents of explanations on bird flu are almost same as those in the introduction. Fig. A1 in Appendix 1-B shows the actual and hypothetical areas with bird flu outbreak to the left and right, respectively. In the explanation, respondents were also explained that the bird flu outbreak in Kyoto had not resulted in any damages to humans and food items. As shown in the right panel of the figure, designating all of Kyoto prefecture as a hypothetical area with bird flu outbreak would avoid misunderstandings among the respondents (and hence, any wrong answers). Otherwise, some respondents might respond that they visited areas affected by the bird flu outbreak against their will. In addition, the fact that the flu outbreak in Kyoto prefecture was attributed to wild birds might have made the hypothetical situation more realistic. Announcements made after disasters play an important role in tourism demand recovery. Safety information would particularly alleviate tourists' anxiety and assure them that any danger from the situation has passed. This study designed three hypothetical information policies (see Appendix 1-D): safety information (Information 1), event information (Information 2), and visitor information (Information 3), as mentioned in section 2.3. Combinations of the policies were presented to the respondents via the CB questionnaires. The combinations were mainly categorized as Type A and B, based on the inclusion or exclusion of safety information, respectively. IP denotes the dummy variables for both Types, respectively: 1 for yes, 0 for others. The subscript safe (e.g., IP safe ) means safety information (Type A) is provided, while nosafe (e.g., IP nosafe ) indicates none is provided. Thus, the IP nosafe variable is used to explore the natural recovery process of tourism demand over time. The superscripts event and visitor (e.g., IP event and IP visitor ) mean event and visitor information are provided, respectively. Thus, IP event safe and IP visitor safe indicate mixed (simultaneous) policies of event and visitor information with safety information, respectively. IP event nosafe and IP visitor nosafe indicate single policies of event and visitor information without safety information, respectively. For example, IP event safe ¼ 1 means the respondents selected a WTT value to a CB questionnaire when event and safety information are simultaneously provided; IP event nosafe ¼ 1 means the respondents selected a WTT value when only event information is provided. Estimating the travel cost variables (hereafter TC) allowed the analysis of the effects of price discounting policies on increasing tourism demand. Firstly, a hypothetical tourism design is considered. Two-day and one-night travel times were assumed as the hypothetical travel times. The Japan Tourism Agency (2013) reported that the average total overnight stay days in trips per year and average number of trips per year by the Japanese in 2011 were 2.08 and 1.3, respectively. Thus, the average overnight trip days per trip is 2.08/1.3 ¼ 1.6 in 2011, 2.09/1.32 ¼ 1.58 in 2010, and 2.38/ 1.46 ¼ 1.63 in 2009. The average values of overnight trip days per trip indicate that almost half of the Japanese do not (or would not like to) travel for over 1.6 days. Two-day and one-night travel times were also assumed to avoid reject responses due to longer travel times and expensive cost. The website of JTB (a Japanese major travel company, http:// www.jtb.co.jp/) showed that the maximum travel cost in January 2012 was ¥40,000 per trip. A sum of ¥20,000 (half the maximum value) was employed to help respondents understand the price difference. As a hypothetical campaign for price discounting, a sum of ¥1000 was presented as the lowest price level (such as a positive follow up campaign in Ritchie, 2009, p. 187) . In examining the survey, it is expected that some respondents might consider ¥1000 too expensive for a travel trip to Kyoto. For example, respondents living in Takatsuki city in Osaka prefecture could travel to and from Kyoto within ¥780 in March 2017 (West Japan railway company, http://www.westjr.co.jp/global/en/; in English). The influence of low price design on tourism demand was confirmed through simulationda larger price discounting would lead to larger tourism demand recovery compared to information policies. Respondents were asked about their willingness-to-travel time (hereafter WTT) after the bird flu outbreak was resolved under the hypothetical situation, given information policies and travel costs. The minimum WTT was designed as "a (1) week" to ensure sufficient planning time for respondents. Depending on respondents' annual income, the maximum WTT was "a year (52 weeks)." For periods exceeding a year, respondents' different budget constraints would have resulted in varied consumption schemes. Thus, the other alternatives were "a month (4 weeks)," "three months (12 weeks)," "six months (24 weeks)," "nine months (36 weeks)," and "Never (0 weeks)." The values within the parentheses denote the value of the WTT variable. Appendix 1-D shows an example of the CB questionnaire. The matrix-type answer format was used in the questions (Broberg & Br€ annlund, 2008; Evans, Flores, & Boyle, 2003; Wang & Heb, 2011) . Each respondent was assigned to either group A or B (with or without safety information), after which they answered three questionnaires (one for each of the price levels). To avoid order effects, respondents were randomly assigned to either Type A or B, and then asked to answer another Type questionnaire (Bateman & Langford, 1997; Halvorsen, 1996) . A dummy variable, ODR, was used to examine the order effect, where ODR took 1 for the first questionnaire (Type A or B), and 0 for the other. Respondents who selected "Never" in Type A (including safety information) with TC of ¥1000 (the lowest travel cost) were asked about their reasons for rejection using a free response format. The questionnaire showed four reasons for rejection: "Anxiety about being infected by the bird flu (RANXIETY)," "Distrust safety information provided by central and local governments (RDTRUST)," "Not willing to travel to Kyoto (RNWILL)," and "Others (ROTHER)." Finally, "yes/no" response data for the logit model were constructed from the WTT data (Bishop & Heberlein, 1979; Habb & McConell, 2002) . Let WTT ji (WTT j 2f1; 4; 12; 24; 36; 52g) denote the time that respondent i is willing to travel, given the hypothetical situation. Respondent i's positive ("yes") response to WTT ji conduces positive ("yes") responses for the periods WTT ki ! WTT ji (ksj); otherwise, it conduces negative ("no") if responses or the periods WTT hi < WTT ji (hsj). The "yes/no" response data for WTT were constructed accordingly, and the pooling data were employed for the estimations. Table 2 shows the questionnaires for the individual characteristics. As observed in previous studies, respondents' gender, age, income, jobs, and educational status were analyzed. Respondents' travel experience (EXKYOTO), interest to tour Kyoto (INTKYOTO), and anxiety about the bird flu (AXBF) were also studied as impact factors for the tourism demand recovery process. Note that questions on EXKYOTO, INTKYOTO, and AXBF are asked to respondents before the questionnaire on the credible information in Appendix 1-A; questions on other variables are asked after the CB questionnaire. This study estimated the following three models. Model 1 was used to estimate the parameters of the main variables (WTT, information policies from IP safe to IP visitor nosafe , TC, and ICM) and reasons for rejection. Model 2 included the main reasons for rejection and individual characteristic variables to conduct statistical evaluations for all parameters. Here, Model 2 also confirmed the signs and statistical evaluations of parameters by WTT periods; the pooling data of a period were used. Thus, the WTT variable was not included. A part of the result is examined in the result and discussion section, and Appendix 2 shows all the results. Model 3 was formulated by eliminating the statistically insignificant variables from Model 2. Thus, Model 3 provided the final estimation results, and was used for the simulations. The robustness of the signs and statistical evaluations of the parameters were checked using the estimation models. Impacts of individual characteristics on decisions were revealed by the results of Model 3. Model 1 To do so, this study employed the criterion as less than 1% of the p-value. CONT denotes a constant. In all models, the expected parameters Xb The applicability of the CB method was examined using the expected signs of the parameters. b visitor and b event were used to check whether preference ordering on information policies would be preserved. For example, b visitor Estimation results were used to simulate policy effects into the tourism demand recovery process. Determining a pre-disaster demand level could help to understand the policy effects through simulation analyses. This study designed the standard level as 92.6% following INTKYOYO, in Table 2 , because actual behavior data (EXKYOTO) would be inadequate for the stated preference-based simulations due to differences between revealed and stated behaviors (Whitehead, 2005) . Moreover, it was assumed that ODR ¼ 0 eliminated the order effects in all simulations. The probabilities for a "yes" response from a week to a year were simulated by applying the estimated parameters ( b b) in Model 3 and the mean values (X) in Tables 2, 5 and 6 to equation (1), that is Tourism demand recovery processes by Model 3 were simulated under the following simulation conditions (SCs). Here, the interpretations for the superscripts and subscripts of SCs are same as IPA and IPB variables, as mentioned earlier. A policy variable that equals 1 (e.g., IP safe ¼ 1) refers to an implementation of the policy, and 0 (e.g., IP safe ¼ 0) refers to no implementation. For simplicity, the policy effect was assumed to sustain from the starting point (1st week) to the terminal point (52nd week). For example, the effect of safety information beginning at the 1st week continues until the 52nd week. Ten SCs are examined. The SC nosafe showed a natural recovery process, indicating an increase in the number of tourists without policies (all information variables equal to zero). Next, the single effect of safety information (IP safe ) is observed for the process under SC safe donly IP safe ¼ 1. The demand recovery processes by the mixed effects of the safety and event information (IP event safe ), and of safety and visitor information (IP visitor safe ), appear under SC event safe and SC visitor safe , respectively. The conditions under SC nosafe to SC full might be insufficient to allow tourism demand recovery to the pre-disaster demand level. Hypothetical policies that overcome the reasons for rejection were designed from SC ranx full to SC allrr full , based on SC full . SC ranx full is a policy to overcome anxiety about getting infected with bird flu (the superscript notation, ranx) under SC full , SC ranx&rdtrust full is a policy to overcome both anxiety and distrust toward government information (the superscript notation, ranx&rdtrust) under SC full , and SC allrr full is a policy to overcome all reasons for rejection (the superscript notation, allrr) under SC full . Simulating the travel cost (price) discount due to the simulated tourism demand from SC safe to SC allrr full would help compare effects of price discounting under different information policies. Simulated tourism demands were calculated based on the 52nd week probability levels from SC safe to SC allrr full . Let STD 52 be a simulated "yes"response (tourism demand) probability for 52 weeks calculated from SC safe to SC allrr full , and let TC be the mean value presented to respondents in Table 5 below. Then, the simulated discounting costs (SDC) were calculated by equation (4). The simulated "pure" price discounts for overcoming RANXIETY, RANXIETY, and RDTRUST, and all reasons for rejection, were calculated as the SDC values from SC ranx full , SC ranx&rdtrust full , and SC ranx full , minus the SDC value from SC full . Here is an inner product of the vectors of the mean values and the parameters of the other variables. The survey was conducted by an internet research company for 20-to 69-year-old residents living in 18 major cities of Japan, in February 2012. The respondents' average age and rates of numbers in the cities were designed to be as consistent as possible with the national survey data for 2010 (Statistics Bureau, 2011) . The company paid online reward points (available for shopping in registered stores) to respondents for motivating them to participate. The questionnaires were sent by e-mail to 17,277 respondents who had registered with the company. Overall, 2128 of 17,277 respondents satisfied the above conditions. However, the exact response rate was unknown due to the lack of information collected by the company on the number of non-participants. Data were collected from 2128 respondents. Table 3 presents the individual characteristics of the respondents. The proportion of male respondents was 63.53%, with an average age of 42.45 years. The corresponding national survey of Japan indicated that these values were 50.17% and 41.98 years, respectively, in 2011. The average household income was approximately ¥6.57 million, while the national survey recorded a value of ¥5.482 million in 2011 (Ministry of Health, Labour and Welfare, 2013). The data indicated that the values corresponding to the male respondents and households in this survey would be slightly higher. However, the average age was almost the same. The highest "yes" response rates for jobs and educational statuses were about 0.11% for homemakers (JBHM) and 55.87% for undergraduate university students (EDUV). Tourism experience at Kyoto (EXKYOTO) was 17.62%, but 92.6% respondents were interested in visiting Kyoto (INTKYOTO). Finally, 23.07% respondents felt anxious about the bird flu outbreak (AXBF). Table 4 shows the results of the CB questionnaires, including IP safe and IP event nosafe . The second row shows the results for combinations of policies. The third, fourth, and fifth rows show the means, standard deviations, and number of respondents who answered "Never," respectively, for ¥1000 to ¥40,000. IP nosafe for ¥40,000, respectively. The minimum and maximum numbers for the respondents who replied "Never" were 296 (the third row for IP event safe ) and 932 (the fifth row for the IP nosafe ), respectively. Finally, Table 5 shows the pooling data for the logistic simulations. Table 6 shows the reasons for rejection for 463 respondents. One hundred and sixty-three respondents answered "Anxiety about being infected by the bird flu." The second and third reasons for rejection were "Distrust safety information provided by central and local governments" and "Other," respectively. The estimation results in Table 7 showed that the parameters of all the models had the same signs, indicating the low probability of multi-collinearity. The obtained parameter signs satisfied expec- Thus, the CB questionnaires in this study proved to be appropriate for studying individual preferences. The results of Model 2 showed that income (ICM), part-time job (JBPTJ), high school education (EDHS), vocational college education (EDVC), and anxiety about being infected by the bird flu (AXBF) were not statistically significant. The results of Model 3 showed that the parameter signs of GND, AGE, JBTS, JBFP, JBST, EDJC, EDUV, EXKYOTO, and INTKYOTO were positive, while those of MAR, JBSOM, JBHM, and EDTC were negative. Finally, the statistically significant and positive sign of b ODR indicated that the probability of attaining a "yes" response was influenced by the order effect. The results of main variables by WTT periods from Model 2 are shown in Table 8 . The complete details of all results are shown in Note: N ¼ 2128. Standard errors are in parentheses. Number of respondents who answered "Never" appears in brackets. Appendix 2. The estimated parameters of main variables (from b safe to b TC ) without income parameters (b ICM ) are statistically significant in all periods, while the b ICM are statistically significant in first week and 52 week, though not in other periods. The parameter signs from b safe to b TC are same in Models 1 to 3. Furthermore, the signs of b ICM are negative from the 1st week to the 12th week, and positive from the 24th week to the 52nd week. Table 9 shows the simulation results for SC nosafe to SC full . The minimum and maximum values were 0.0934 and 0.2431 for the 1st week, 0.1095 and 0.2770 for the 4th week, 0.1646 and 0.3804 for the 12th week, 0.2855 and 0.5546 for the 24th week, 0.4476 and 0.7164 for the 36th week, and 0.6753 and 0.8664 for the 52nd week, respectively. Table 11 shows the results of the price discounting (prices after the discount) policy simulations. The minimum and maximum values were ¥9984 for SC ranx full and ¥56,867 for SC full , respectively. The applicability of the CB method and information policy effects were evident from the results of Model 3 in Table 7 . First, the signs of the parameters were as expected. Second, the results b visitor safe > b event safe and b visitor nosafe > b event nosafe indicated that the preference orderings were preserved. Thus, the CB method can be used to analyze tourism demand recovery from disasters. The orderings indicated that providing visitor information could be more effective than event information. Next, the ordering b safe > b visitor nosafe > b event nosafe indicated that safety information could have the highest effect among all information policies. Most tourists were not willing to travel to disaster sites without safety information. The ordering b visitor safe > b event safe > b safe indicated that mixed policies could have a greater recovery effect over single policies. These results also supported the applicability of the CB method for estimating preferences. The findings confirmed the mixed effects of safety and other information. A pure effect of event information was that b event safe À b safe ¼ 0.3020 >b event nosafe . Furthermore, according to the visitor information, b visitor safe À b safe ¼ 0.3836 >b visitor nosafe . These results indicated that mixed policies could generate synergetic effects. Moreover, the fact that the value of b safe was larger than the pure effect values indicated that providing safety information could have the highest effect. Table 8 indicates that the parameters of information policies totally tend to increase in spending periods (although there are cases of increment and decrement by periods). For example, the estimated parameters of IP safe (b safe ) are 0.3972 in the 1st week and 0.7098 in the 52nd week, respectively. The estimated parameters of IP visitor safe (b visitor safe ) are 0.7604 in the 1st week and 1.1833 in the 52nd week, respectively. The reason could be that the respondents' anxiety toward the bird flu outbreak decreased as time passed; respondents might come to think the bird flu outbreak would not occur. Thus, the travel cost parameters (b TC ) decrease from À0.0166 in the 1st week to À0.0311 in the 52nd week. This indicates that the price effects are enhanced over time due to an increase in the number of willing-to-travel respondents by reducing (overcoming) their anxiety. Table 8 also shows that the signs of ICM parameters (b ICM ) turned negative from the 1st week to the 12th week to positive from the 24th week to the 52nd week b ICM was statistically significant for the 1st week and 52nd week, and insignificant from the 4th week to the 36th week. The results indicated that tourism at disaster sites was an inferior good (that decreases corresponding to increments of ICM) in the period just after its occurrence, but changed to a normal good (that increases corresponding to increments of ICM) as time passed. This can be attributed to the fact that a tourism site in disaster would be considered a low-quality good that was not be preferred to other non-disaster tourism sites (Loomis & Walsh, 1997, p. 91) . Finally, the results for the individual characteristics indicated that the following persons were willing to travel provided information announcements were made: male respondents (because of their tolerance level, they typically suffer less anxiety about being infected by the bird flu compared to women); the elderly, temporary staff, freelance professionals, and university students (because they have ample free time to plan and travel); and those interested in traveling to Kyoto. Information policies could be effective at attracting these persons to travel. Otherwise, persons who were married, operating self-owned businesses, homemakers, and those educated at technical colleges were not swayed by information policies, possibly because of the fear of catching the infection themselves and/or infecting their children (the negative influence of tourists' perceived risks for tourism sites described in Law, 2006 and Rittichainuwat & Chakraborty, 2009 ). Furthermore, they could also possibly have little time to travel because of their jobs or study commitments (the substitution between work and leisure time described in Weiermair, 2006). Table 9 shows that information policies cannot help postdisaster tourism demand levels (the maximum was approximately 86.6% for SC full ) recovery to reach the pre-disaster demand level (92.6%). Overcoming the reasons for the rejection in Table 10 makes it possible for the post-disaster demand level to come close to the pre-disaster demand level (91.2% for SC allrr full ). Thus, an issue in recovering tourism demand is how to compensate for demand losses by the reasons for rejection. The results of price discounting policies (prices after the discount) in Table 11 showed that the SDT values for SC safe , SC event safe , SC visitor safe , SC full , and SC allrr full exceed the average travel cost per trip (¥20,333 in Table 5 ). This indicated that the tourism demand recovery, to levels of information policy by price discounting, would not be feasible due to the high discounting rates. Here, it might be feasible to discount ¥16,094 (a discount rate of about 80%) of SC ranx&rdtrust full that corresponds to overcome RANXIETY (the anxiety for the bird flu) and RDTRUST (the distrust toward the safety information from governments). For example, The Japanese government has implemented a maximum rate of 70% as a price discount for travel to Kyushu area in order to encourage economic recovery after the Kumamoto Earthquake of 2016 (Kumamoto Prefecture Tourist Federation, 2016) . Thus, the 80% price discounting could be feasible as a policy for tourism demand recovery. Early implementation of tourism demand recovery policies after decontaminating the disaster site could shorten recovery periods. However, sometimes, lack of information in planning may act as an obstacle to implement the policies before the disaster. This section reexamines the estimation and simulation models based on the above findings in order to consider an optimal timing of tourism demand recovery policies. The estimation and simulation results were used for determining the optimal policy orderings. First, the estimation model is reexamined. The estimation results of Table 8 indicate that the main variables would be influenced by WTT periods. Thus, Model 4 is designed based on Model 3 as equations (5) and (6). Here, including the ICM variable with information of the 1st week and the 52nd week could help confirm different parameter signs by one week and 52 weeks, as shown in Appendix 2, and to improve estimation accuracy of the estimation model. In equation (5), D 1 and D 52 are dummy variables; D 1 ¼ 1 for WTT ¼ 1 (0 if others) and D 52 ¼ 1 for WTT ¼ 52 (0 if others); the g ICM1 and g ICM52 are parameters of the interaction term variables of ICM with D 1 and D 52 , respectively. Similarly, different policy effects can be observed by taking the interaction term with WTT variable. The notations from g safe to g TC in equation (6) ) are same as in Model 3. The expected signs are positive for g safe , g event safe , g visitor safe g event nosafe , g visitor nosafe , and g ICM52 , and negative for g TC and g ICM1 from results in Tables 7 and 8 Next, the simulation procedure was examined. First, it is assumed that only a single policy can be implemented in a period; policymakers cannot implement multiple policies simultaneously. For example, event information could not be announced in one week if safety information was announced at that time. In this case, the safety information (IP safe ) must necessarily be the first policy in order to permit visiting the site after solving the disaster. Thus, an efficient ordering of event information, visitor information, and price discounting were examined. The IP event safe and IP visitor safe variables are used as event and visitor information due to the synergetic effects with the safety information, as discussed in section 5.1. Thus, the IP event nosafe and IP visitor nosafe variables are designed as zero in all simulations. The price discounting policy is designed as an 80% discount of the mean value of TC (TC  0:2) from the result in section 5.3. The 40% discount cases (TC  0:6; arbitrarily decided) are also simulated for comparisons. Note that ODR ¼ 0 for all simulations. The terminal point is also the 52nd week. A policy effect is assumed to sustain from the starting point at j weeks (j 2{1, 4, 12, 24} due to four recovery policies) to the terminal point. For Table 12 The starting points and adjusted willingness-to-travel time. 1st week 4th week 12th week 24th week 36th week 52nd week AWTT 1 1 4 12 24 36 52 AWTT 4 0 1 9 2 1 3 3 4 9 AWTT 12 0 0 1 1 3 2 5 4 1 AWTT 24 0 0 0 1 1 3 2 9 Note: the super script a means the starting points of ATT j s. Simulations for analyzing effective policy ordering. Xb simplicity, the 36th and 52nd weeks are not selected. Even if the 36th and 52nd weeks are considered as a starting point, the orderings of recovery levels of tourism demand and the optimal policy ordering would be same due to the linearity of WTT variable in the interaction terms. Thus, the recovery levels of tourism demand would differ. j ¼ 1 is determined to implement the safety information policy as mentioned above. Thus, j'2{4, 12, 24} are the main points for analyzing the optimal ordering. An issue in simulation is how to treat the time delay. For example, the effect of event information policy with WTT variable (WTT  g event safe  IP event safe in equation (6)) beginning at the 4th week (WTT ¼ 4) is calculated as 4  b g event safe  1, even though the beginning time is j ¼ 4; the exact calculation is Thus, values of the adjusted WTT at jth week (AWTT j ) for a policy shown in Table 12 are used by replacing the WTT value in the interaction terms (e.g., AWTT event 4  g event safe  IP event safe ; the superscripts of AWTT i indicate the policy name). For example, AWTT event 4 means the event information policy begins from the 4th week, and the AWTT 4 value at that time is 1 (thus, 1  b g event safe  1 is calculated). In the next (12th) week, 9 is assigned as the AWTT 4 value (i.e., 9  b g event safe  1). (5) and (6) were adjusted as b b 0 event g safe in order to eliminate the safety information effect. Thus, the simulation model was redesigned as equation (5) 0 and (6) 0 . Here, k, l, and m 2{4, 12, 24} and kslsm. The mean values in Table 3 Finally, the 11 patterns of policy orderings listed in Table 13 are examined. Case 1 corresponds to SC nosafe (natural recovery) in Table 9 . Cases 2 to 4 show the single effects of information policies as time passes by. Case 2 corresponds to SC safe in Table 9 , whereas Cases 3 and 4 do not correspond to SC event safe and SC visitor safe due to the synergetic effects that overestimate tourism demand recovery levels than those calculated by a single policy. Case 5 corresponds to the price discounting policy. Here, the 80% price discounting policy (that compensates for the demand losses from the reasons for rejections, RANXIETY and RDTRUST) was assumed from discussions in section 5.3. Cases 6e11 were designed to simulate the optimal policy order. For example, Case 6 shows that the safety information policy begins from the 1st week (ATT 1 ), the event information policy from the 4th week (ATT 4 ), the visitor information policy from the 12th week (ATT 12 ), and the price discounting policy from the 24th week (ATT 24 ). The estimation results in Table 14 showed that only g visitor nosafe is not statistically significant. The signs of estimated parameters are same as Model 3. The ordering of degree of information policy parameters was also same as that of Model 3db visitor Otherwise, the results of interaction term variables showed that g event safe > g visitor safe > g safe >g event nosafe >g visitor nosafe ; the visitor information parameters were smaller than the event information parameters. Similarly, the ordering of pure effects of event and The estimation results indicated that safety information would have the highest effect on demand recovery. The ordering of event and visitor information are reversed with and without the time series factor. The reason could be that the visitor information based on the actual behavior might give respondents a sense of trust regarding the safety of a disaster site regardless of the periods. Deutsch and Gerard (1955) showed that an individual's behavior is sometimes influenced by the information obtained from another as evidence about reality. Furthermore, McFerran, Dahl, Fitzsimons, & Morales. (2010) showed that a part of the purchase behavior was determined by other consumers' purchase quantities. The parameters of visitor information show little changes as time passes by (also see Table 11 ). Second, combinations of these information policies would generate synergetic effects on tourism demand recovery with and without the time series factor. Third, the parameters of ICM  D 1 and ICM  D 52 in Table 14 showed that tourism at disaster sites would be an inferior good in the period immediately after the disaster, and thereafter, changing to a normal good with time. The third result and the negative sign of price parameter might also indicate that researchers would be allowed to analyze the cost-benefit ratio using the consumer surplus for reconstructing infrastructures in tourism sites in a year after the disaster (Johansson, 1987) . Finally, the AIC and R2 values of Model 4 indicate that it is more suitable for simulations due to the lower and higher values than the ones of Models 1 to 3, respectively. Table 15 shows the simulation results from Cases 1e5. As references, values in brackets show SC nosafe and SC safe values in Table 9 for Cases 1 and 2, respectively. The values of Cases 2e5 in the "1st week" column showed the recovery amounts by policies immediately after disaster site decontaminationdthe first is safety information, followed by visitor information, price discounting, and event information (Policy effect order 1; PEO1). The "52nd week" column showed that the order changes to safety information followed by visitor information, event information, and finally price discounting (PEO2). Thus, implementing safety information as the first step is valid. The result indicates the following: First, planning requires dynamic policy effect analyses because of the possibility of changing policies with time, in contrast to static analyses, due to lack of information. Moreover, it would be necessary to consider applicable tourism management frameworks, in disaster, corresponding to policy effect changes with time. Second, the pricing policy after solving a disaster will not have a significant effect, in contrast to previous studies, possibly due to respondents' anxiety matching income effects. Table 8 also shows that the b TC decreases from À0.0203 in week 36 to À0.0311 in week 52. This result indicates that rapid and broad announcements of information polices could be more important. The simulation results of price discounting also indicated that the low price design in the CB questionnaire might not cause bias due to the small differ-enced0.7238 and 0.7057 in 52 weeks for 80% and 40% price discounting in Table 15 , respectively. The simulation results from Cases 6e11 are shown in Table 16 . In the 80% price discounting case, the minimum and maximum values in 52 weeks are 90.83% in Case 11 and 91.50% in Case 6. Thus, the ordering of Case 6 could recover tourism demand in 52 weeks to near the standard demand level (92.6%). The result showed that the optimal policy ordering is Case 6, which implements safety information as the first step, followed by event information, visitor information, and price discounting, in contrast to the results of policy effect orderings in Table 15dCase 9 for PEO 1 and Case 8 for PEO 2. The results indicated that the policy orderings based on the effects would not be optimal, that is, they may not achieve the highest recovery of tourism demand due to the time delay (in Table 13 ) and the policy effect changes by period. The optimal policy ordering would be valid. After announcing safety information, policymakers and/or companies could encourage tourists to visit the site by events. Then, informing the situation, the price discounting could enhance the increment of the number of tourists. The validity of designing the price discounting in the final step is also supported from another viewpoint. Canina, Enz, and Lomanno (2006) stated that price discounting helps increase tourism supply (e.g., hotel rooms), but this comes at a cost to revenues. To elaborate, (extreme) price discounting might have a positive effect on tourism demand recovery, whereas it might deteriorate the finances of the government and/or the tourism company. The measurement of price discount rate without deterioration of finances of other stakeholders should be considered in future research. Finally, the optimal timing of implementing the policies is discussed based on Table 17 and drawn from Table 8 . Table 17 shows the changes of synergetic effects of mixed policies (b event safe À b safe and b visitor safe À b safe ) and b TC by WTT periods. Here, we assume two conditions: i) the policy ordering of Case 6 is employed and ii) the policies were implemented one by one. Announcing safety information in the first step as soon as possibledwithin week 1, immediately after decontaminating the disasterdwould result in faster tourism demand recovery at such sites. Since the b event safe À b safe values decrease from the 36th to the 52nd week, it is preferable to time the second step (event information) within 24th to 36th weeks after the disaster. The increment of b visitor safe À b safe values from the 36th to the 52nd week and condition (ii) indicated that it would be timed until the 52nd week after the 37th week for the third step timing (visitor information). Thus, it would be appropriate to implement the fourth step (price discounting) immediately after the third step due to the decrement of b TC values from the 36th to the 52nd week. In summary, it is preferable to implement announcing the safety information (the first step) within one week, the event information (the second step) within 24th to 36th week after the first step, visitor information (the third step) within 37th to 52nd week after the second step (e.g., 40th week), and the price discounting until the 52nd week immediately after the third step (e.g., 41th week). Note that the timing relies on the framework of this study. More proper policy timing analyses are needed for the (optimal) recovery process of restructuring infrastructures of tourism sites as described in Faulkner (2001) . Disasters in tourism sites urge policymakers to implement effective tourism demand recovery policies through optimal timing. However, the difficulty of micro data collection in disasters and external factors in macro data make it difficult for policymakers and researchers to analyze policy effects. This study examined the CB method with the time series factor as an appreciable method for analyzing the policy effects and timing (orderings) after solving the disaster. A bird flu emergency in Kyoto prefecture, Japan, acts as the hypothetical disaster in the CB questionnaire. The respondents were asked about their WTT under this hypothetical situation, assuming combinations of three information policies (safety, event, and visitor information) and three travel costs (¥1,000, ¥20,000, and ¥40,000). The alternatives for WTT were designed to range from week 1 to week 52. The estimation results indicated the following. First, safety information has the highest effect on demand recovery. The second is event information, if it relates to the time series factor (the willingness to travel time variable), and visitor information, if not. Event information would be more important in considering the optimal timing due to its relation with the time series factor. Second, combinations of these information policies could generate synergetic effects on tourism demand recovery. Third, tourism at disaster sites would be an inferior good in the period immediately after the disaster, changing to a normal good with time. The simulation results indicate the following. First, the necessity to analyze the dynamic policy effect due to changes with time would require researchers to examine applicable management frameworks corresponding to the effect changes (e.g., announcing music festivals in a tourism site until the 4th week or informing the visitors' looks of enjoying the festivals to other non-visited tourists through the media until the 12th week). Second, the pricing policy, after solving a disaster, would not have a significant effect due to tourists' anxiety. Third, the optimal policy ordering and timings are determined as follows: the provision of safety information within a week, event information within 24th to 36th week after the disaster, visitor information within the 37th week to the 52nd week, and price discounting until the 52nd week immediately after the third step (e.g., 41th week). Here, the optimal policy ordering results indicated that the policy orderings based on the effects would not be optimal due to the time delay and the policy effect changes. The results conclude that the method of this study could be useful for analyzing dynamic tourism demand processes by recovery policies. The method aims to show policymakers the policy effects and the optimal policy ordering for recovery from disaster damages, especially through advance planning. Moreover, the method of this study could be applicable to other disasters, such as earthquakes, hurricane, crimes, and terrorism, by modifying the contents of policies in the CB questionnaire (the solutions for the issues are necessary as well). For example, the policy contents for recovering demands from terrorism would include the announcement of arresting terrorists, establishment of security cameras to monitor terrorism, and police deployment. Finally, this study has certain limitations. The first is the difficulty in reducing the number of CB questionnaires when the number of policies or periods increasedthe order effect might occur by the answer format of this study. For example, the choice experiments, which could possibly implement the same analysis in this study, were not employed owing to its difficulty. The second limitation is how to design and estimate the policy effects from the pre-event to the post-event stages, and the feedback effects from the post-event to the pre-event stages, for improving disaster planning, training and education, among others. The third is to research more realistic trip information (e.g., the type and quality of accommodation, food and beverage, and distances from respondents' homes to the tourist site) in order to estimate more realistic price effects, especially in low-price level situations (¥1000 in hypothetical prices). The simulation results could be improved Note: the values of 40% price discounting cases are in parentheses; (À) means the 40% price discounting case was not simulated due to nontarget of simulations; the standard demand level is assumed as 92.6% in 52 weeks; the value of SC ranx&rdtrust full in 52 weeks is 0.8994 in Table 10 . by collecting these details, and thus further studies could take up these challenges. Additionally, new research could also involve adjusting the difference between real and hypothetical behaviors, and applying the proposed method to varied disaster events, such as earthquakes and tsunamis. No other relationships, conditions, and circumstances present a potential conflict of interest. In general, the bird flu influenza virus infects many birds living in naturedmainly water birds, such as ducks (Anas). Bird flu does not frequently infect humans, but it rarely does so when a person touches or is in close contact with an infected bird. In recent years, cases of H5N1 type bird flu influenza virus infection on humans has been reported and developed. The observed bird flu influenza type in Japan is H5N1. In Asian countries, it has also been reported that symptoms of seriously ill patients infected by the H5N1 type are pneumonia, multiple organ dysfunction, among others, while the main symptom of mildly ill patients infected by H7N7 type observed in the Kingdom of the Netherlands is conjunctivitis. The WHO reported that 568 persons were infected, of whom 334 persons were dead from 2003 to November 2011. Another reason that public organizations pay serious attention to the bird flu is the influenza pandemic caused by various bird flu influenza viruses through infections from birds to humans. Hence, the pandemic might rapidly expand across the world. Currently, the Japanese government prevents the bird flu outbreak by euthanizing infected (or probably infected) birds. Note that the bird flu shot (vaccination) has not been implemented in Japan. Please read the explanation below and answer the next questionnaire. In Kyoto prefecture, a wild bird infected by the bird flu was observed and decontaminated (see the left panel in Fig. A1 ). However, the spread of infections to humans and food items was not confirmed. It is expected to be difficult in future to undertake protective measures for whole infection cases of the bird flu due to wide action ranges of wild birds. Here, the bird flu is assumed to be observed in the entire area of Kyoto prefecture (see the right panel in Fig. A1 ) in the next questionnaire. This bird flu outbreak is called an influenza pandemic, as described above. Here, the outbreak is assumed to expand only to birds, and not to humans and food items, among others. Note that the area of the outbreak is assumed as only the Kyoto prefecture. This questionnaire requests you to provide your assumed travel behavior under the hypothetical situations described below. Please read the information carefully. Note that the travel costs (¥1,000, ¥20,000, and ¥40,000) differ by the questions presented to you. [Hypothetical situation: Travel to Kyoto prefecture]. Imagine that you plan to travel to Kyoto prefecture alone. During your planning, you realize that a bird flu emergency has occurred throughout Kyoto prefecture (see the Table below) . You are also aware that the Japanese local governments could decontaminate the bird flu-affected areas without causing human and physical damages. Then, you are provided each of following three pieces of information by a credible information source. Please answer what would be the earliest period you would be willing to travel to Kyoto prefecture, depending on the information provided to you. (Please provide this answer even if you live in Kyoto in the present time). Your travel period is assumed to be two days and one night, with the assumed travel cost being ¥X b (including accommodation and the travel fee). Assume that there are no factors (job, homework, etc.) impeding your travel in each period. Q. Please answer your main reason for selecting "Never" from Information 1 to 3. Please write box here [ ]. Appendix 2. Estimation results by periods. Information 1 Information 2 Information 3 After a year (52 weeks) After nine months (36 weeks) After six months (24 weeks) After three months (12 weeks) After a month (4 weeks) After a week Never a: Only the descriptions provided in the brackets were shown for the Type B group. b: X, yen ¼ {1000 (special price for campaign), 20,000, 40,000}. Econometric models for discrete choice analysis of travel and tourism demand Budget-constraint, temporal, and questionordering effects in contingent valuation studies Restoring tourism destinations in crisis Crisis and post-crisis tourism destination recovery marketing strategies Measuring values of extra-market goods: Are indirect measures biased Avian influenza: Economic and social impacts An alternative interpretation of multiple bounded WTP data: Certainty dependent payment card intervals Estimations of economic losses from the Great East Japan Earthquake (Higashi-nihon dai-shinsai niokeru higaigaku no suikei nitsuite) On the effects of the Great Kumamoto Earthquake in 2016 (Heisei 28 nen Kumamoto-jisin no eikyou-sisan nitsuite) Why discounting doesn't work: A hotel pricing update Estimating price effects in an almost ideal demand model of outbound Thai tourism to East Asia Destination image as a mediator between perceived risks and revisit intention: A case of post-disaster Japan Tourist motivation: An appraisal A study of normative and informational social influences upon individual judgment Recovery marketing: What to do after a natural disaster. The Cornell Hotel and Restaurant Administration Quarterly Tourism economics and policy (Aspects of tourism texts) Determinants of international tourism: A three dimensional panel data analysis Multiple bounded uncertainty choice data as probabilistic intentions Towards a framework for tourism disaster management The recreational value of Lake McKenzie, Fraser island: An application of the travel cost method Managing heritage tourism Modeling tourist arrivals using time series analysis: Evidence from Australia We love paris campaign Valuing environmental and natural resources: The econometrics of non-market valuation Ordering effects in contingent valuation surveys. Environmental and Resource Economics Earthquake devastation and recovery in tourism: The Taiwan case Chapter 1 section 2 tourism trends in Japan The economic theory and measurement of environmental benefits Current status and issues on tourism in Tohoku area: Toward one trillion tourism industry market (Tohoku-tiiki niokeru kanko no genjou to kadai: Kankousangyou no icchouenka wo mezashite). Bank of Japan Reports and Research Papers Kyushu-hukkou-wari Estimating the impact of avian flu on international tourism demand using panel data Assessing impacts of SARS and avian flu on international tourism demand to Asia Total research on tourism activity in Kyoto (Kyotokanko-sougou-chosa) Annual statistics of Kyoto prefecture in 2014 (heisei 26 Kyoto-hu-toukei-hakusho) (Chapter 2) Population and Household Influences of the great Kumamoto earthquake for Kyushu's economy Pricing policy in nature-based tourism The perceived impact of risks on travel decisions Recreation economic decisions: Comparing benefits and costs Earthquake effects on tourism in central Italy Status on the highly pathogenic bird flu in 2010 (heisei 22 nen niokeru koubyougensei-tori-infuruenza no kakuninjoukyou) Critical events in Miyazaki prefecture (Miyazaki-ken no ugoki 2011 kikijishou). Miyazaki no ugoki Tourism and disaster planning. Geographical Review Assessing the impacts of the global economic crisis and swine flu on inbound tourism demand in the United Kingdom Combining multiple revealed and stated preference data sources: A recreation demand application Estimation and welfare calculations in a generalized corner solution model with an application to recreation demand A framework for knowledge-based crisis management in the hospitality and tourism industry Crisis and disaster management for tourism (Aspects of tourism) Perceived travel risks regarding terrorism and disease: The case of Thailand Tourism demand modelling and forecasting: How should demand be measured? Tourism demand modelling and forecasting: Modern econometric approaches National survey at 2010. Website of the Ministry of Internal Affairs and Communications The impact of crisis events and macroeconomic activity on Taiwan's international inbound tourism demand. Tourism Management Estimating individual valuation distributions with multiple bounded discrete choice data Environmental risk and averting behavior: Predictive validity of jointly estimated revealed and stated behavior data Valuing beach access and width with revealed and stated preference data Consumption benefits of national hockey league game trips estimated from revealed and stated preference demand data Dr Tadahiro Okuyama completed his PhD in Economics at Tohoku University, and is an associate professor at Regional Design and Development This work was supported by JSPS KAKENHI Grant Number JP15K03522. The author thanks the participants at the Workshop of Economics and Regional Studies in 2016 at Ryukoku University and the referees for their advice on improving this study. Appendix 1. Explanations and Questionnaires on contingent behavior (original sentences were written in Japanese) 1. No problem 2. I feel anxiety* 3. I cannot imagine. *respondents who select alternative 2 were assigned for 1; 0 for others.