*3.2. Data Analysis*

To analyze the text-based communication records, qualitative analysis is considered appropriate [12,14,41,49,50]. Specifically, using netnography and coding skills from the qualitative analysis [51], the qualitative data analysis proceeded in the following four steps.

First, first-level coding. This is also referred to as open coding in classic qualitative analysis, where topics are generated from words or sentences of the original material [51]. Researchers of this study coded each line of communication records as well as user' comments after the consultation experience using the language of doctors or users. To ensure the validity and reliability of qualitative coding, three researchers read and coded the original communication records independently. After each of their initial coding was completed, they go through all the coding results and discuss di fferent opinions through in-depth discussion until they reached consensus.

Second, second-level coding. This is also referred to as axial coding in classic qualitative analysis, where topics are consolidated and abstracted to categories and sub-categories based on comparison and contrast [51]. Usually, the categories and sub-categories may appropriate the terms and phrases from the literature. As a result, first-level codes in our study were further classified into informational and emotional dimensions. Through this step, doctors' information-related behaviors and emotion-related behaviors that cause users' dissatisfaction, as well as users' information-related behaviors and emotion-related behaviors that cause doctors' dissatisfaction, are obtained.

Third, Third-level coding. This is also referred to as selected coding in classic qualitative analysis, where categories are connected to tell a logical story of the intended phenomenon [51]. We counted the frequencies of each identified category and selected categories with high frequencies to form the complete model that explains the antecedents of poor DPR in the mobile context. Based on these selected categories, challenges of mobile technologies identified using the CMC literature and interaction behaviors of doctors and users identified in the communication records are connected. Specifically, each of the researchers tried to understand the underlying reasons behind doctors' and users' mobile misbehaviors by referring to the CMC features identified by the CMC literature. To ensure the validity and reliability of the classification, three researchers conduct this step independently and converge opinions through in-depth discussion.

Forth, developing coding schemes. Based on the above three steps, we developed a coding schemes, and use this coding scheme to code subsequent consultation records. To ensure the reliability and validity of the codes, di fferent researchers repeat the above coding steps and compare the codes and data to reach a converged opinion. The above coding steps repeat until there are no new themes, categories, or sub-categories that are generated to explain the original data.
