It’s the Great Eigen-pumpkin, Charlie Brown!

Each Halloween, households in the US and Canada (and a few other places too) create a wide range of intriguing face stimuli: Jack-O’-Lanterns, which are traditionally schematic faces carved into pumpkins or other decorative gourds. Though typically regarded as a mere seasonal decoration, I suggest that these also offer an opportunity to examine features of face perception and recognition via a production task. To that end, I present an analysis of a collection of Jack-O’-Lantern drawings created by children between the ages of 2 and 14 years old in response to cartoon pumpkins with varying external contours. Using Principal Components Analysis, I investigate a simple question: Does the external contour influence the mean and covariance structure of the internal features children choose to draw? I find that it does. Specifically, wider pumpkins have average image structure that is less coherent, and pumpkin top-heaviness vs. bottom-heaviness decreases the proportion of variance captured by low-order principal components.