Moral Foundations in Partisan News Sources

Daniel Preotiuc-Pietro, Dean Fulgoni and Jordan Carpenter
November 5, 2015

Although almost everyone agrees that some things are morally good and some things are morally bad, the specific form of these beliefs can differ throughout the population. What is egregious to one person: harming marginalized communities, banning sugary soft drinks, refusing to go to church, etc.; can be considered completely trivial or even be endorsed by someone else.

The Moral Foundations Theory [1,2,3] was developed to model and explain these differences. Under this theory, there are a finite number of basic, moral values that people can intuitively support, but not necessarily to the same extent across the population. The five moral foundations are:

  1. Care/Harm:
    The valuation of compassion, kindness, and warmth, and the derogation of malice, deliberate injury, and inflicting suffering.
  2. Fairness/Cheating:
    The endorsement of equality, reciprocity, egalitarianism, and universal rights.
  3. Ingroup loyalty/Betrayal:
    Valuing patriotism and special treatment for ones own ingroup.
  4. Authority/Subversion:
    The valuation of extant or traditional hierarchies or social structures and leadership roles.
  5. Purity/Degradation:
    Disapproval of dirtiness, unholiness, and impurity.

Under this theory, a person who strongly endorses the value of ‘Care/Harm’ will be appalled at an action that causes suffering, while someone who endorses ‘Authority’ will support an action that supports the social hierarchy. These responses would be immediate, emotional, and intuitive.

Moral Foundations Theory explains some political disagreements between people on opposite sides of the political spectrum. Liberals tend to strongly endorse the values of ‘Care’ and ‘Fairness’, but they consider ‘Ingroup loyalty’, ‘Authority’, and ‘Sanctity’ to be irrelevant to morality. On the other hand, conservatives tend to endorse all five moral foundations, though they do not support ‘Care’ and ‘Fairness’ as strongly as liberals do [3].

Therefore, some of the most contentious political conflicts center on issues where each side focuses on a value not equally shared by the other side. For instance, the issue of police violence against minorities can be thought of as a conflict of values: Liberals abhor the unfairness of violence instigated against a marginalized social group, while conservatives hate the anti-authoritarian, public disrespect of legitimate authority figures.


In their study of Moral Foundations Theory, Graham and Haidt developed a dictionary which is meant to capture people’s spontaneous moral reactions. For each of the five moral foundations, this dictionary contains two categories of words: recognition of the value being demonstrated (virtue, denoted with +) and recognition of the value being violated (vice, denoted with ).

For instance, the ‘Harm+’ subdictionary contains words describing compassion and care: ‘safety’, ‘protection’, ‘shelter’, etc. Likewise, the ‘Authority-‘ subdictionary contains words that describe rebellion: ‘heretic’, ‘riot’, ‘nonconformist’, etc. Therefore, in theory, the extent to which a person spontaneously uses the words in each subdictionary captures their concern for each moral foundation.


As part of the Media Cloud Data Challenge, we thought this would be a perfect way to explore the extent to which the moral foundations are invoked in news discourse.

We retrieved word counts using the MIT Media Cloud API, using a query that samples word counts from articles that match specific tags. In this case, we were interested in articles filtered by partisan bias and political issue. We searched for articles belonging to either ‘liberal’, ‘conservative’ and ‘libertarian’ news sources, and using the controversy tag, we retrieved articles labelled with of issue and partisan side, the top 1,000 unique words were identified and mapped to their respective counts with a sample rate of 10,000. These word counts were later smoothed for consistency: using the set union of all words in the dataset, if any seen word did not appear in one of the ten word count maps, that word was mapped to the map’s minimum value. Additionally, each moral foundation was mapped to the aggregate number of its associated words across all issues. This was done using the moral foundations lexicon, which maps a few hundred words and stems to their associated moral foundations.


As a first exploratory step, we analysed the similarity in word usage across issues, disregarding partisanship. We computed similarity as the cosine similarity between the word frequency distributions:

Similarity in word usage across all issues, disregarding partisanship. Similarity is represented by cosine similarity between the word frequency distributions.

Probably the most notable feature of these findings is a cluster of sex-related issues (`teen pregnancy’, `abortion’, `contraception’, `sex education’) which seem to all be discussed in similar ways. Surprisingly, three issues that seem related, the violence against Eric Garner, Freddie Grey, and Trayvon Martin, do not contain especially similar patterns of words.


Focusing on a few issues, we wanted to see what are the different words each partisan side (we only limited to liberal vs. conservative for clarity) uses when mentioning a story. We illustrate this via word clouds. These were generated using the word count maps for each partisan bias and controversy combination. For each issue, its respective conservative and liberal word counts were compared using the log odds ratio to determine the relative use of the most polarized words. Log odds ratio is a widespread technique that performs well in identifying which features are most associated with a particular label – in this case, which words are dominantly used by conservatives or liberals.

The word clouds consist of 80 words each, which are the top 40 most conservative and top 40 most liberal words by log odds ratio. Large text size indicates a high total frequency for that word. Because some words are used much more than others, the relative frequencies were scaled by taking the logarithm (base 2) of each, such that the size differential between words is more reasonable.


For color, red identifies the word as polarized conservative, while blue identifies the word as polarized liberal. Deeper color indicates a stronger correlation with that partisan bias, such that deep red words are strongly conservative while deep blues ones are strongly liberal.

Word usage in partisan news stories about ‘Abortion’. Darker red indicates more conservative, darker blue indicates more liberal. Larger size indicates higher word frequency (log-scaled).

Word usage in partisan news stories about ‘Climate Change’. Darker red indicates more conservative, darker blue indicates more liberal.Larger size indicates higher word frequency (log-scaled).


Word usage in partisan news stories about ‘Contraception’. Darker red indicates more conservative, darker blue indicates more liberal. Larger size indicates higher word frequency (log-scaled).


Word usage in partisan news stories about ‘Ferguson’. Darker red indicates more conservative, darker blue indicates more liberal. Larger size indicates higher word frequency (log-scaled).

Word usage in partisan news stories about ‘Freddie Gray’. Darker red indicates more conservative, darker blue indicates more liberal. Larger size indicates higher word frequency (log-scaled).

The patterns of word usage are largely face valid, containing specific aspects of issues of special concern to liberals and conservatives respectively. In general, most partisan sides appear to be focusing on the opposing side and combating their views. For instance, several word clouds feature each side mentioning contentious and famous opposing voices: conservatives mention ‘Gore’, ‘Pelosi’, and ‘Sharpton’, while liberals mention ‘Koch’, ‘Huckabee’, and ‘Paul’. In some topics, there are clearly different frames in use. For example, liberals mention concrete issues about climate change (e.g. ‘fracking’, ‘renewable’, ‘pollution’, ‘fossil’), while conservatives seem to discuss more about their political opposition (e.g. ‘Obama’, ‘democrats’).


Finally, we look at the usage of moral foundations for each of the three side across all issues, and further at the moral foundations invoked by liberals and democrats for each issue.

Relative moral foundation usage for each partisan side and across all issues.

Moral foundation usage by issue for liberal news sources.

Moral foundation usage by issue for conservative news sources.

Though these results are somewhat murky, several clear patterns emerge.

In conformance to theory, liberal sources are far more likely to mention harm and suffering across issues, while conservative sources are more likely to address ‘loyalty’. The most striking is the use of ‘fairness+’ and ‘loyalty-‘ for libertarians. Libertarians’ patterns of moral values have been found to differ from those of both liberals and conservatives[4]. From our analysis, libertarians sources put high emphasis on ‘fairness’, but have the lowest scores on the
harm’ dimension.

Of particular interest here is the cell for ‘loyalty-‘, which, for the issue of Freddie Grey, is strong for liberals and even stronger for conservatives. This difference probably stems from a particular word within the ‘loyalty-‘ subdictionary: ‘riot’. Liberal sources were more likely to refrain from using this word, preferring the more sympathetic ‘uprising’, which is not included in the Moral Foundations dictionary.

Another difference regarding usage of different foundations is the high incidence of ‘loyalty-‘ for conservatives in the ‘NSA Snowden’ issue. Indeed, conservatives sources frame Snowden as being a ‘traitor’ or ‘disloyal’ to his country, while liberals frame the story in terms of harm caused to the country. On the issue of ‘climate change’, liberals ‘harm-‘ is the most prevalent of all issues, while conservative sources use ‘harm-‘ relatively very little and frame the issue more in terms of textit{loyalty}.

We plan to continue quantitatively studying the moral foundations by examining statements and news at a large scale to see the effects of framing a controversial issues in terms of certain moral foundations. In very recent news, research has shown that framing an issue to appeal to other’s moral foundations is more effective than by framing it in yours.


  1. Haidt J. & Joseph (2004). Intuitive Ethics: How Innately Prepared Intuitions Generate Culturally Variable Virtues. In Daedalus, pp. 55-66.
  2. Haidt, J., & Graham, J. (2007). When morality opposes justice: Conservatives have moral intuitions that liberals may not recognize. In Social Justice Research, pp. 98-116.
  3. Graham, J., Haidt, J., & Nosek, B. A. (2009). Liberals and conservatives rely on different sets of moral foundations. In Journal of Personality and Social Psychology
  4. Iyer, R., Koleva, S., Graham, J., Ditto, P., Haidt, J. (2012). Understanding Libertarian Morality: The Psychological Dispositions of Self-Identified Libertarians. In PLoS ONE

This blogpost was written by Daniel Preotiuc-Pietro, Dean Fulgoni and Jordan Carpenter. We would like to express our gratitude to the Media Cloud team and especially Hal Roberts for the support in using the platform over the summer.