Detecting “True" Waste

In the summer of 2021, I conducted a real-world, randomised controlled trial at a COVID-19-restricted festival site with approximately 1,000 participants. The results suggest that adding exemplars to waste sorting signage improves sorting accuracy. Since then, I have been investigating the mechanisms underlying this effect. Drawing on theories of categorisation, I approach waste items as quantifiable along three dimensions: frequency, prototypicality, and ease of sorting (“sortability”). I collected 1,016 images of waste items, of which 792 were selected and rated on the three dimensions by around 200 Swedish participants. Using hierarchical Thurstone–Mosteller modelling, I have ranked the items by frequency and prototypicality, while hierarchical logit-based models have helped identify which items are easier to sort.

Based on these item-level characteristics, I have selected 287 items that serve as “idealised” exemplars spanning high and low values across the three dimensions. In my talk, I will present results from a two-alternative forced-choice experiment in which approximately 400 participants were shown a waste item and a category sign, and asked whether the item belonged in that fraction. Depending on condition, the sign displayed either zero, one, or five exemplars. The data are still being analysed, but preliminary results suggest that providing more exemplars improves sorting performance—particularly for difficult categories, for identifying true positives, and among individuals who generally struggle with sorting.

I may not yet have the answer to how best to support accurate waste sorting, but I will show you some of my many plots, share how signal detection theory applies to waste sorting, and welcome your input on where to take the project next.

__
Open for external guests.

Please request access to Zoom link in advance if participating virtually.