Authors:
(1) Howard Zhong, MIT CSAIL ([email protected]);
(2) Mark Hamilton, MIT CSAIL, Microsoft ([email protected]).
Table of Links
3 Results
5. Conclusion/ Acknowledgements/ References
A Appendix
A.2 Detailed NFT Information & A.3 Google NFT Searches Map
3.2 Aggregate NFT Market
We conduct similar analysis for gender and race bias across the NFT market. We found 44 collections with gender labels but only 3 with race labels.
3.2.1 Gender
For the 44 NFT collections chosen as detailed in Section 2.2, we calculate p-values to determine whether male NFT prices are greater than female NFT prices. P-values are calculated from a paired log t-test marked to weekly mean with a outlier trim percentile of 2.5%. To satisfy preconditions of a t-test, we apply a log transformation to make prices follow a more normal distribution and remove outliers. We mark to weekly mean price, which is common for low transaction volume markets such as NFT markets [7]. From the analysis, we find no clear relationship between market cap and gender bias. However, we do demonstrate that male prices are not statistically significantly higher than female prices for the broader NFT market.
Figure 5 displays collections with male price statistically significantly (α = 0.05) higher than female price (right red dots), female price statistically significantly higher than male price (left red dots), and no statistically significant difference (black dots). There is no clear relationship between market cap and mean price difference across gender, but the figure does answer whether male NFTs are priced higher than female NFTs across the aggregate market.
First, the mean across all collections of the price difference of male and female NFTs is only 0.63%. Furthermore, at a significance level of α = 0.05, only 11 out of 44 collections have male NFTs valued more than female NFTs, whereas 7 collections have female NFTs valued more than male NFTs. Therefore, there is not enough evidence to conclude there is a statistically significant difference between male and female prices for the broader NFT market.
3.2.2 Race
In addition to CryptoPunks, we obtained race labels for the Avastar and Dynamic Duelers collections. From mean and median statistics in Table 4, we find that light-skinned NFTs on average tend to be sold for higher prices that darker-skinned counterpart. The result is less significant for Dynamic Duelers. For statistics in Table 4 and 5, we use outlier trim level of 2.5%.
We also conduct hypothesis tests in Table 5 to determine the statistical significance of whether lighter-skinned NFTs are worth more than darker-skinned NFTs. With a majority of the p-values less than 0.2 and our conclusion from CryptoPunks, the results suggest lighter-skinned NFTs are worth more than darker-skinned NFTs, though not necessarily at a statistically significant level. Because this analysis is on a small subset of the entire NFT market, more data is required to generalize our conclusion to the entire NFT market.
This paper is available on arxiv under CC BY 4.0 DEED license.