Graham, J., Haidt, J., & Nosek, B. A. (2009). Liberals and conservatives rely on different sets of moral foundations. Journal of personality and social psychology,96(5), 1029.
In a previous post I reviewed the results and implications of this paper. This post will focus on the paper’s use of linguistic analysis to identify distinct moral foundations in the texts of liberal and conservative church sermons. Church sermons were used because sermons are generally written by individuals rather than a collection of writers used for political speeches.
The authors relied mostly on the Linguistic Inquiry and Word Count (LIWC) software program for their text analysis, plus some supplemental word count analysis. Additional resources are available at their website moralfoundations.org .
The LIWC product website provides the following description of the program:
“LIWC2007 is designed to accept written or transcribed verbal text which has been stored as a text or ASCII file using any of the popular word processing software packages… LIWC2007 accesses a single file or group of files and analyses each sequentially, … As each target word is processed, the dictionary file is searched, looking for a dictionary match with the current target word. If the target word matches the dictionary word, the appropriate word category scale (or scales) for that word is incremented. As the target text file is being processed, counts for various structural composition elements (e.g., word count and sentence punctuation) are also incremented. With each text file, approximately 80 output variables are written as one line of data to a designated output file. This data record includes the file name, 4 general descriptor categories (total word count, words per sentence, percentage of words captured by the dictionary, and percent of words longer than six letters), 22 standard linguistic dimensions (e.g., percentage of words in the text that are pronouns, articles, auxiliary verbs, etc.), 32 word categories tapping psychological constructs (e.g., affect, cognition, biological processes), 7 personal concern categories (e.g., work, home, leisure activities), 3 paralinguistic dimensions (assents, fillers, nonfluencies), and 12 punctuation categories (periods, commas, etc).”
Before analysing 69 liberal sermons (178k words) and 34 conservative sermons (137k words), the authors built a custom dictionary to categorise words reflecting the 5 moral foundations harm/care, fairness/reciprocity, ingroup/loyalty, authority/respect and purity/sanctity. Words were lemmatized into foundation confirming categories and foundation violating categories. Similar to what I have done for my LSA python program custom dictionary but more aggressively into 5 moral virtues and 5 vices plus a morally neutral category. Calculating the word frequency for each word grouped into one of the 11 moral categories, the top 23 words (or their stems) were selected where they contributed to a greater than 0.02% difference in overall usage between the liberal and conservative sermons. Each word selected was blind reviewed in context to address the chance that some words being alphabetically the same may have different meanings or inflections depending whether the user is a liberal or a conservative. An weighted adjustment was then applied to these words to compensate.
Their results are summarised in this table:
Interestingly the text analysis didn’t perfectly match the outcomes from the self-assessment surveys. While words linked with the moral foundations of Harm and Fairness were more frequently used on liberal sermons matching the surveys, the Ingroup moral foundation also rated more highly in the liberal sermons rather than the conservative sermons. This unexpected outcome seems to be due to words like community and nation being used by liberal preachers as being rooted in the idea of individual autonomy rather than group loyalty per se. My view is that this discrepancy arises due to ingroup/loyalty being evolutionarily based on kin-altruism rather than the more libertarian preferences for community. Words like obligation, familiar and sameness need to be added and community type words removed.
While some of the results were unexpected the excellent p values suggest that different political groups do have distinct differences in moral language, both different words used, their inflection and in their frequency. Their words reflect different moral foundations leading to distinctly different decision architecture types.
This research suggests that if we are not careful and fail to learn the moral language of others we could end up with the political equivalent of the Tower of Babel. Linguistic analysis may help us bridge the gap.