MEASURING BRAND SIMILARITY
9/13/18

Before I start, I'd like to thank Grailed for letting me use their data for this post!

If you’re interested in clothes at all, it should be pretty intuitive that a brand like Supreme is more similar to Bape than A.P.C., and more similar to Palace than Rick Owens. To prove this, you could point to the brands’ similar ethos, their heavy use of printed graphics, or their roots in modern youth culture. However, these reasons are all relatively unscientific, and get harder as the brands you think about become less obviously separable - is Off-White more similar to Gucci or Adidas?

Instead, I’d argue that one of the easiest ways to gauge two brands’ similarity is by the fraction of people who own both brands, essentially allowing for peoples’ taste to answer the question of similarity. “More people own both Bape and Supreme than Bape and A.P.C., so therefore Bape is more similar to Supreme than A.P.C.” This statistic encompasses things like brand ethos, “aesthetic”, and price, all things that consumers, consciously or not, factor in every time they buy a piece of clothing. This allows us to make quantitative comparisons between brands based on a generally very qualitative set of criteria, making the question of similarity easily answered regardless of brand.

For this statistic to actually mean anything significant, it wouldn’t be enough for me to just take inventory of my friends’ wardrobes - a much larger sample size is needed. Instead, I turned to Grailed, an online marketplace for secondhand luxury menswear (note: this post is not sponsored/endorsed by Grailed in any way). On Grailed, people list the clothing they wish to sell, and it is possible to see all of the items that a particular user has for sale. In essence, each of these pages is a small snapshot of a user’s wardrobe.

An example of a user’s items for sale.

By compiling these snapshots from nearly 2,000 Grailed users, I was able to build up a dataset of wardrobes that could be analyzed to determine the similarity score between any two brands (calculated as the proportion of people who were selling one brand given that they were also selling the other). Many of the more intuitive relations were confirmed - streetwear fans of Supreme tend to also like Palace, clean and minimal Acne Studios fans tend to like other understated yet refined brands like A.P.C. and Common Projects. However, dig in a little further and some more unexpected results appear - who knew that vintage clothing and Issey Miyake had so much in common?

Take a look for yourself below and check out which brands are most similar to your favorite brand. Maybe you’ll discover some new designers in the process! (Some of the more popular brands take a few seconds to populate)

# Brand Match
search for a brand to find similar brands!



Lastly, I also want to address some potential drawbacks of my general approach. First, my approach makes the assumption that the items users are selling are an accurate reflection of their own taste. While some Grailed users do act as online stores for a very diverse range of brands and thus would violate this assumption, I believe that this assumption holds for the majority of users on Grailed. My approach also becomes less applicable for less common brands, as the smaller sample size for those brands increases the likelihood of an outlier negatively impacting the results. This could be remedied by scraping more data from Grailed in the future. Overall, while my approach has its flaws, I believe that it still produces meaningful results.

If you found this post interesting, check up the follow-up post, CATEGORIZING PERSONAL STYLE! In that post, I use this same data to automatically cluster users into different styles.