**4. Methods**

To understand how cultural tourism could act as people's livelihood, brewing social resilience in the relocated community, this study applied in-depth semi-structured interviews and participant observations during fieldwork from December 2017 to March 2019. Interviews allowed us to understand people's perception, thoughts, and experiences on culture from an internal angle. At the same time, participant observations provided an external viewpoint to understand what kind of cultural messages were passed from the Zhulu to outsiders via tourism activities.

Interviews were conducted in person, following an interview guide (see Appendix A) for consistency, with flexibility across interviews. The interview guide was developed based on concepts excerpt from the literature on social resilience, cultural tourism, and post-disaster recovery. The interviewed residents were asked detailed questions on the community status quo and development, local livelihood structure and the tourism situation, cultural transformation and merging with recovery work, Morakot typhoon's impact, and the reconstruction process. We also asked interviewees from the public sector questions on post-disaster recovery strategy, tourism promotion objectives in Zhulu, and the relation between cultural revival, post-disaster recovery, and local industrial strategies. Questions to tourists were mainly about their tourist behaviors. The researchers conducted interviews according to each circumstance and could change the order of questions, adding questions within the research objectives that we found important or were in the interviewee's interest.

All interviews were audio-recorded with the participants' oral permission. We used purposive sampling, by which several residents, who sell cultural goods in Zhulu, were interviewed first. These persons were then asked to recommend others who might be suitable and willing to be interviewed. Then, snowball sampling was used to identify others until the participants offered no new information relating to the study's purpose. The interviewed residents are all currently living in Zhulu, including community organizations' leaders and members, and residents who have promoted or are promoting tourism. The interviewees' representation was not based on the sampled size, but rather because they were appropriate and informative in responding to the research topic. All participants were informed of the study's research aims and its ethical considerations before they agreed to be interviewed. The interviews lasted, on average, from 45 to 120 minutes, were conducted in Chinese Mandarin, and were carried out in places where interviewees were comfortable to talk, such as their homes. Table 2 provides basic descriptions of the 25 interviewees with coded identification numbers to protect their privacy. The interviews were transcribed for narrative analyses and discourse analyses.

Participant observation sessions were conducted to observe Zhulu residents' daily lives and their tourism activities. Through the observation, we were able to understand how they weave Tsou culture into tourism to revive their post-disaster livelihood and how they use culture to attract tourists. Although the researchers are Han, we had no difficulties interacting with the Tsou people because there has been strong interaction between Tsou and Han over hundreds of years [45] (the Alishan National Scenic Area, with an administrative boundary that covers the entire Tsou living area, was formally established in 2001).

Participant observation took place in the tribal market, tribal artistic village, permanent houses, and activity center on occasions such as guided tours, Tsou traditional dance, the culture sharing festival, culture and health station courses for the tribal elders, Zhulu tribe culture co-learning courses for children, and lunar new year tourist activities (Figure 4). Informal interviews were held during the observations. When the researchers participated in activities that were not open to the public, such as courses for residents, they introduced themselves and requested permission before participating

and asking questions (informal interviews). While participating in activities open to the public, such as festivals, the researchers did not reveal their identity unless informal interviews were conducted (when researchers verbally revealed the research topic to obtain people's consent). Observational data were recorded in many forms, including written field notes, photos, and videos. Field notes were made extensively during fieldwork to record what was seen and heard.


**Table 2.** Classification and characteristics of interviewees.

**Figure 4.** Participant observation: (**left**) aboriginal dance activity; (**right**) culture and health station courses.

This research involved 13 participant observations with coded identification numbers from P01 to P13. Although both researchers were not indigenous Tsou, most Tsou people have extensive interaction with the Han Chinese, use Chinese Mandarin as their main language of communication, and interact with tourists who are mainly Han Chinese. Thus, the researchers' Han ethnicity did not have any impact on the collection of data.

All interviews were subsequently transcribed. Field notes, interview transcriptions, and secondary data (like published reports, statistics, and maps) were cross-referenced during analysis. Editing analysis

was used to develop categories and then crystallize concepts for interpretation. Text, narrative, and discourse analyses were used during the process [46].

The research limitations were mainly in the data. First, post-disaster recovery is a continuous process. The researchers conducted fieldwork from 2017 to 2019 and thus could not get first-hand data on events before 2017. Second, both the disaster and the relocation happened more than five years ago; some interviewees might have vague memories about past events. To limit the bias caused by these limitations, we used mixed methods to collect data, including interviews and observation, as well as collecting second-hand data from multiple sources to triangulate the obtained first-hand data.
