# Directly Observing and Characterizing Adolescents' Self-Generated Social Media Posts: Protocol for Creation and Implementation of a Cyberethnography Informed Codebook

## TL;DR

This study outlines a method to observe and analyze adolescents' social media posts over time to better understand their well-being.

## Contribution

The paper introduces a replicable cyberethnography-based codebook for analyzing adolescent social media content with high interrater reliability.

## Key findings

- Interrater reliability scores were consistently high across platforms and years (AC1 scores above 0.87).
- The codebook allows for longitudinal contextual analysis of self-generated social media content.
- The method supports cross-analysis with well-being data for identifying patterns.

## Abstract

Adolescent social media research has primarily focused on frequency of platform use and self-report measures. There has been limited focus on the self-generated content posted by adolescents and how this might relate to their well-being.

This study describes a researcher-observed codebook for characterizing adolescents’ self-generated content in a longitudinal sample.

Participants in the study provided informed assent (and parental informed consent) for researchers to follow them and passively observe their self-generated content on Instagram (Meta Platforms), TikTok (ByteDance Ltd), Facebook (Meta Platforms), and X (formerly known as Twitter; X Corp). Guided by Bronfenbrenner’s social ecological biopsychosocial model, the research team created a codebook incorporating prior cyberethnographic observation of self-generated social media content. After codebook refinement, coders (research staff and student research assistants) were trained through multiple rounds of test coding, and the codebook was applied to participant data with periodic quality control measures to ensure interrater reliability.

This study was funded in early 2023 and began data collection in March 2023 and will conclude in 2027. So far, the interrater reliability agreement scores (AC1) between coders have shown strong interrater reliability. For Year 1, scores were Facebook 0.89, Instagram 0.89, TikTok 0.88, X 0.87, and combined 0.88; for Year 2, Facebook 0.95, Instagram 0.96, TikTok 0.96, X 0.96, and combined 0.96. This project provides replicable guidance to categorize social media data from adolescent participants using human coders who can contextualize content through longitudinal observation. The method that our team chose and followed paved the way for many strengths to be recognized as well as lessons learned by our team that allowed for adaptation and growth to occur while this study has been ongoing.

Cyberethnography, the chosen method for this research protocol, has allowed this research project to collect self-generated content for adolescent social media in a comprehensive manner. Thus, allowing our team to be able to cross-analyze this data with the well-being data that are being collected under the grander project for patterns. Sharing our protocol through this paper will also allow other researchers to draw from our methodology for future projects to aid in social media and adolescent understanding.