Experiment: Carousel (08/18–09/04)
Setup
In this experiment, we organized the results into carousels based on aspects belonging to a given facet. Specifically, we had three arms:
- none: no carousel, the baseline
- origin: showing carousels of results for each aspect belonging to origin facet (eg. Afghanistan, India, etc.)
- type: showing carousels for each aspect belonging to type facet (eg. Kelim, Runner, etc.)
Our hypothesis was that carousels are a better presentation of content than infinite-scrolling grid. This experiment was inspired by the recent trend of websites showing carousels (most notably by Netflix, copied by many).
Mobile
Results
- carousels seem to work ok for desktop (although very few desktop users only, so no significant results)
- they definitely don’t work for mobile
- type is slightly better than origin (but no significant difference)
- about 10% of the users toggle the carousel view (more than we expected), but only doing it on desktop
- about 20% expands the carousel
