A Novel Method to Determine a Custom Sample Size for Image-Based Instagram Content Analysis

Abstract

Abstract Objectives Most Instagram users are young people, and social media is often used to search nutrition information. Health interventions aimed at young people should consider such information sources. Content analyses of Instagram images offer insights into types of content that may influence nutrition-related decision making and health behaviors. However, the number of analyzed images in existing studies has varied, and methods to determine data-specific sample sizes to reach saturation have not been explored. We aimed to develop a method to determine sample sizes for image-based content analyses on Instagram. We piloted the method and determined the reliability by identifying the saturation point for content categorized under two separate nutrition-related hashtags. Methods Instagram ‘top posts’ for a 21-year-old user were searched using hashtags ‘mindfuleating’ and ‘intuitiveeating’. 1200 images from each were extracted. Hashtag-specific coding frameworks were constructed inductively by two authors, initially coding the image- and text-based elements of the first 90 images collaboratively. Next, increments of 45 images were coded independently, then compared, solving disagreements by discussion. The process was repeated until saturation occurred when no new codes emerged. This was repeated seven weeks later to determine reliability. Results The coding frameworks constructed for #mindfuleating at first and second capture comprised 63 and 74 distinct codes, with saturation occurring at 360 and 405 images, respectively. The #intuitiveeating frameworks comprised 83 and 86 codes, with saturation at 450 and 495 images, respectively. The codes captured detailed pictorial content (e.g., ‘female’, ‘White’, ‘young adult’) and text (e.g., ‘nutrition information’, ‘relationship with food’). For both hashtags, the number of image-based codes decreased while text-based codes increased between coding. Conclusions Variations in coding frameworks and sample sizes over a short timeframe reflect the dynamic nature of Instagram content. Assessment of diet trends on social media requires accurate sampling to ensure nuances of a specific topic are captured, while research efficiency benefits from reduced data redundancy. Funding Sources NHMRC Peter Doherty Early Career Fellowship; Sydney Medical School Foundation, The University of Sydney.