2.3. Coding and Analysis

The final codebook was applied by LL and MW to the posts from both periods. Google translate was used as needed to translate captions and hashtags into English. Since the study focuses on e-liquid marketing, posts made by users with no ties to the e-cigarette industry (e.g. individual social media users who did not indicate any brand/store sponsorship or affiliation) and posts that did not promote e-liquid were excluded from further analysis. Given the possibility of purchasing e-liquids from abroad and the fact that exposure to social media marketing does not confine itself to national borders, promotional posts were coded regardless of country of origin. All final coding was performed using MAXQDA 2018 software. Coders held weekly meetings to discuss coding questions, refine definitions, and identify any new themes emerging during the coding process. A random sample of 10% of posts (n=100) were double coded by LL and MW across the two coding periods in order to ensure reliability. The estimated intercoder reliability coefficient (Cohen’s Kappa) was on average 0.805, with 96.5% agreement between the two coders. In addition to qualitative analysis, we also conducted descriptive quantitative analysis using a robust F-test (using Stata “fstar” module) for continuous variables and Pearson Chi-square Test for categorical variables to explore differences in posting characteristics between different types of users. All quantitative analyses were done with STATA 14 (StataCorp. 2015).