Causal Drivers of Sustainable Social Media Engagement in the Textile Industry: A Double Machine Learning Approach

Green digital marketing is critical for advancing sustainability in the textile sector. As brands aim to reduce their environmental impact and engage ethically conscious consumers, identifying effective social media formats is essential. This study proposes a causal inference framework that integrates Double Machine Learning (DML), a method for estimating treatment effects in high dimensional observational data, with theDoWhyplatform for treatment effect estimation and refutation testing. The framework controls for sentiment polarity, posting time, weekday/weekend status, and sustainability keywords, ensuring robust Average Treatment Effect (ATE) estimates. Empirical analysis reveals that Instagram Reels produce the strongest positive impact on engagement, measured as the combined number of likes and comments for each post. In contrast, Videos and Carousel Albums reduce interaction. Among all estimation methods tested, the DML model produced comparatively precise and stable estimates, yielding narrower confidence intervals and stronger refutation performance than the baseline approaches. The study provides strong causal evidence; practical generalisation should consider platform dynamics and potential unobserved influences. Across 20,768 posts in 2024, DML yields tighter confidence intervals (CI) and smaller placebo errors than OLS/PSM/PSS/NDIM. Robustness is demonstrated via bootstrap CIs, and placebo effects are examined through permuted treatments, random and hidden common-cause refuters, and subset analyses. Effects generalise within the window and context, and temporal or platform limits were noted.Managerial Relevance Statement: This paper provides causal evidence, robust to the placebo effect and includes hidden confounder checks, that short form Reels lift engagement by$\sim$19 interactions per post on average, while carousels and long videos reduce interaction. Managers can translate this into action by Reallocating 15-25% of content volume from long videos or carousels to Reels for awareness phases; Structuring campaigns as two-stage funnels using Reels for attention and carousels or long videos for education and brand values; utilising weekly A/B “what-if” pilots seeded with covariates, including weekday/time, sentiment, and sustainability keywords to localise effects; and finally, tracking KPI deltas such as incremental interactions, post engagement rate, and not just raw counts alone. For policymakers, the results inform evidence based public campaigns like textile-waste reduction, prioritising attention-efficient, low-cost formats to amplify sustainability messages. This paper also contributes to the following SDG: SDG 9, SDG 12, and SDG 13.