Can Hedonic Ratings Predict Pairwise Comparisons?

Back in 1950, 𝗣𝗲𝗿𝘆𝗮𝗺 introduced the 9-𝘱𝘰𝘪𝘯𝘵 𝘩𝘦𝘥𝘰𝘯𝘪𝘤 𝘴𝘤𝘢𝘭𝘦 to replace the cumbersome multiple pairwise comparison approach.

Today, most consumer studies rely on 𝘀𝗲𝗾𝘂𝗲𝗻𝘁𝗶𝗮𝗹 𝗺𝗼𝗻𝗮𝗱𝗶𝗰 𝗲𝘃𝗮𝗹𝘂𝗮𝘁𝗶𝗼𝗻𝘀, making direct preference comparisons rare — yet the interest in both 𝗮𝗯𝘀𝗼𝗹𝘂𝘁𝗲 𝗹𝗶𝗸𝗶𝗻𝗴 and 𝗿𝗲𝗹𝗮𝘁𝗶𝘃𝗲 𝗰𝗵𝗼𝗶𝗰𝗲 remains strong.

What if we could get both — accurate 𝗽𝗿𝗲𝗳𝗲𝗿𝗲𝗻𝗰𝗲 𝗽𝗿𝗲𝗱𝗶𝗰𝘁𝗶𝗼𝗻𝘀 from 𝗺𝗼𝗻𝗮𝗱𝗶𝗰 𝗵𝗲𝗱𝗼𝗻𝗶𝗰 𝗿𝗮𝘁𝗶𝗻𝗴𝘀?

In collaboration with The Institute for Perception, we explored this question by collecting both 𝗵𝗲𝗱𝗼𝗻𝗶𝗰 𝗿𝗮𝘁𝗶𝗻𝗴𝘀 and 𝗽𝗮𝗶𝗿𝗲𝗱 𝗽𝗿𝗲𝗳𝗲𝗿𝗲𝗻𝗰𝗲 data from the 𝘴𝘢𝘮𝘦 𝘤𝘰𝘯𝘴𝘶𝘮𝘦𝘳𝘴.

Using 𝗧𝗵𝘂𝗿𝘀𝘁𝗼𝗻𝗶𝗮𝗻 𝗺𝗼𝗱𝗲𝗹𝗶𝗻𝗴, we:

• Estimated underlying hedonic d′ values

• Converted these into predicted pairwise proportions

• Compared them with actual preference outcomes

𝗞𝗲𝘆 𝗳𝗶𝗻𝗱𝗶𝗻𝗴𝘀:

1. Predicted and observed preferences were aligned in direction.

2. A 𝟮-𝘁𝗼-𝟭 𝗿𝗲𝗹𝗮𝘁𝗶𝗼𝗻𝘀𝗵𝗶𝗽 emerged: to match real pairwise outcomes, the preference ratio from hedonic ratings must be roughly 𝗱𝗼𝘂𝗯𝗹𝗲𝗱.

○ For example, a 60/40 predicted split corresponds to an actual 70/30 preference.

○ This scaling reflects consumers’ variability in the use of the hedonic scale, increasing the underlying variance of the data.

🔍 Another demonstration of the value of 𝗧𝗵𝘂𝗿𝘀𝘁𝗼𝗻𝗶𝗮𝗻 𝗺𝗼𝗱𝗲𝗹𝘀 — and a useful tool for researchers looking to link how people like to how they choose.

📄 Research overview: https://lnkd.in/gMfS48mU