Abstract: Moral Foundations Theory (MFT) is a framework with enormous influence across the social sciences, but it has encountered both theoretical and empirical critiques, largely centered around its claims of pluralism and modularity. Functional neuroimaging, together with multivariate analytic techniques, may begin to shed some light on these formerly intractable issues. Participants (N=27) made moral wrongness judgments about vignettes representing each of the moral foundations while undergoing an fMRI scan, and were tested on their memory for the depicted transgressions. A data-driven spatiotemporal partial least squares (PLS) analysis revealed two latent variables (LVs) related to emotional empathic arousal and binding violations, respectively. Importantly, these LVs did not differentiate the foundations from each other, nor did they reflect the standard superordinate structure posited by MFT. Further, the LV patterns served as better fixed effects than the traditional superordinate categories in mixed effects model comparisons of moral judgment and memory accuracy data. A second round of model comparisons revealed that models incorporating information about trait-empathy levels and emotional reactions to the vignettes led to better model performance for moral judgments and memory accuracy, despite increasing the number of parameters. We take these results to suggest that morality is not as modular and domain-specific as argued by MFT. Rather, the neural data support constructivist theories of morality which argue that it arises from interactions among domain-general processes, such as emotion, empathy, and conceptualization.
mft_memory.Rmd
contains code for all of the behavioral analyses, as well as code for generating all of the figures in the paper. All behavioral data can be found in data/behavior
. data/qualtrics
contains data from all of the post-task questionnaires, and data/qualtrics/scoring
contains the scoring keys for surveys that are not specific to this experiment.
PLS analyses were conducted using a PLS toolbox in MATLAB developed at the Rotman Research Institute. Datamats and text files used for the PLS analyses can be found in data/PLS/analysis files
. Preprocessed nfiti files used in the PLS analysis will soon be available on OpenNeuro. Result .mat
files for the mean-centered and non-rotated analyses can be found in data/PLS/mean-centered
and data/PLS/non-rotated
, respectively. Each of these directories also contains .csv
files with values relevant for plotting extracted from the .mat
results file.
Email maria.khoudary [at] duke.edu.