Testing the Efficiency of Markov Chain Monte Carlo With People Using Facial Affect Categories

Abstract

Exploring how people represent natural categories is a key step toward developing a better understanding of how people learn, form memories, and make decisions. Much research on categorization has focused on artificial categories that are created in the laboratory, since studying natural categories defined on high-dimensional stimuli such as images is methodologically challenging. Recent work has produced methods for identifying these representations from observed behavior, such as reverse correlation (RC). We compare RC against an alternative method for inferring the structure of natural categories called Markov chain Monte Carlo with People (MCMCP). Based on an algorithm used in computer science and statistics, MCMCP provides a way to sample from the set of stimuli associated with a natural category. We apply MCMCP and RC to the problem of recovering natural categories that correspond to two kinds of facial affect (happy and sad) from realistic images of faces. Our results show that MCMCP requires fewer trials to obtain a higher quality estimate of people's mental representations of these two categories.

Identifying category representations for complex stimuli using discrete Markov chain Monte Carlo with people

“…Trials were presented on three different computers, one for each proposal type. Following previous work, each participant completed trials corresponding to different d-MCMCP chains (Martin et al, 2012; Ramlee, Sanborn, & Tang, 2017). There were four chains: two for happy faces and two for sad faces.…”

“…As a first test of d-MCMCP, we examined the categories of happy and sad faces using images of real faces. Previous work had applied the MCMCP method to estimating these categories using parameterized face stimuli, where a continuous space was derived from eigenfaces computed from a set of images (Martin et al, 2012). We used the same image database to directly compare the results of d-MCMCP and MCMCP on a matched stimulus set.…”

“…Using this insight, Sanborn, Griffiths, and Shiffrin (2010) adapted the well-known Markov chain Monte Carlo (MCMC) algorithm from computer science to sample from subjective probability distributions, a method they called Markov chain Monte Carlo with people (MCMCP). MCMCP has been used to estimate the structure of categories defined on continuous, easily parameterized stimuli, such as stick-figure animals and basic fruit shapes (Sanborn et al, 2010) or computer-generated faces (Martin, Griffiths, & Sanborn, 2012; McDuff, 2010).…”

“…The remainder of the article focuses on three experiments showing a range of applications of d-MCMP. The first experiment explores categories of happy and sad faces using real photographic images, allowing us to compare against previous results using the MCMCP algorithm (Martin et al, 2012). The second experiment applies d-MCMCP to text, exploring the words perceived to be representative of the concept of morality for people with liberal and conservative political affiliations (Graham, Haidt, & Nosek, 2009).…”

What sways people’s judgement of sleep quality? A quantitative choice-making study with good and poor sleepers

“…Also, the chains carried over from 1 participant to the next: this enhanced the power of the analysis at the cost of assuming no individual differences in how participants weighted the factors. 35 Finally, to improve the speed of data collection we set up multiple groups of chain that could be run in the same testing session: 3 good sleeper and 3 poor sleeper groups (see Supplementary Appendix 3).…”