Shape Matters

Shape Matters: Evidence from Machine Learning on Body Shape-Income Relationship

Suyong Song 1 and Stephen Baek 2
1 Department of Economics, The University of Iowa
2 Department of Industrial and Systems Engineering, The University of Iowa
{suyong-song, stephen-baek} (at) uiowa (dot) edu


 

Abstract

We study the association between physical appearance and family income. In most previous studies, physical appearance was measured by imperfect proxies from subjective opinion based on surveys. Instead, we use the CAESAR data which have 3-dimensional whole body scans to mitigate the issue of possible reporting errors and measurement errors. We show there are significant nonclassical reporting errors in the reported height and weight by comparing them with measured counterparts and show that these discrete measurements are too sparse to provide a complete description of the body shape. We use the graphical autoencoder built on deep machine learning to obtain intrinsic features consisting of human body shapes and estimate the relation between body shapes and family income. We also take into account a possible issue of endogenous body shapes by utilizing proxy variables and control functions approaches. The estimation results show that there is a statistically significant relationship between physical appearance and family income and the associations are different across the gender. This supports the hypothesis on the physical attractiveness premium in the labor market outcomes and its heterogeneity across the gender. Our findings also highlight the importance of correctly measuring body shapes to provide adequate public policies for the healthcare.

Keywords: Physical attractiveness premium, non-Euclidean data, deep machine learning, graphical autoencoder


 

Figure 1. Summary of the estimation results for family income equation. Estimated coefficients and bootstrapped 90% confidence bands are reported. The left panel presents results from the conventional body measures and the right panel reports results from the deep-learned body parameters through the graphical autoencoder.


 

Figure 2. Conditional expectation of the reporting errors in height conditional on the true height (top) and weight conditional on the true weight (bottom).


 

Figure 3. Body shape parameters derived from the graphical autoencoder for male (left) and female (right). 3D body shape models are arranged in accordance with their body shape parameters, with increments of -3 std., -1.5 std., 0, 1.5 std., and 3 std. with respect to the mean in each direction.


 

Publications

Journals & Proceedings

  1. Shape matters: Evidence from machine learning on body shape-income relationship
    Song, S. & Baek, S.
    Working Paper

  2. Economic models with non-Euclidean data
    Baek, S. & Song, S.
    In Proceedings of the 2018 Joint Statistical Meetings (JSM2018), Vancouver, Canada, July, 2018.

Conference Presentations

  1. Shape matters: Evidence from machine learning on body shape-income relationship
    Baek, S. & Song, S.
    In 88th Southern Economic Association Annual Meetings (SEA2018), Washington, D.C., November, 2018.

  2. Shape matters: Evidence from machine learning on body shape-income relationship
    Baek, S. & Song, S.
    In 28th Annual Meeting of Midwest Econometrics Group (MEG2018), Madison, WI, October, 2018.

  3. Estimation of economic models with non-Euclidean data
    Baek, S. & Song, S.
    In New Frontiers in Econometrics 2018, Stamford, CT, June, 2018.

Invited Talks/Seminars

  1. Shape matters: Evidence from machine learning on body shape-income relationship
    New York University, Stern School of Business, New York City, NY, February, 2019.

  2. Shape matters: Evidence from machine learning on body shape-income relationship
    Federal Reserve Bank of Kansas City, Kansas City, MO, January, 2019.

  3. Shape matters: Evidence from machine learning on body shape-income relationship
    Georgia Institute of Technology, Department of Economics, Atlanta, GA, January, 2019.

  4. Shape matters: Evidence from machine learning on body shape-income relationship
    Microsoft Research, Redmond, WA, December, 2018.


 

Download

Will be available soon.