Leveraging Big Data for Social Good: Examining Moral Values in Immigration Discussions in News Media

Background and purpose: Scholars have long recognized the importance of leveraging large-scale data for promoting social well-being. Although a few social work scholars have begun analyzing large datasets since the late 1990s, it is still a challenge for researchers to utilize big data analytics to inform social justice issues. Immigration is one of the important and currently controversial issues in US society. Its conversations often occur in written and spoken word, and language expressed in media can play a vital role in how the US public understands immigration and immigration policies. We describe a method for analyzing large-scale text data drawn from major news resources and examined the intensity and co-occurrence of moral values and Latino Threat Narrative, a depiction of Latino immigrants as a threat to US society.

Method: A total of 440,984 news articles were collected from LexisNexis and Proquest NewsStand using “immigrant OR immigration OR Latino OR Mexican” as a search term. The news articles spanned across 13 news outlets from 1997 to 2014, including New York Times, Washington Post, USA Today, Wall Street Journal, and Los Angeles Times. First, we identified five LTN themes based on Chavez’s study (2008), which include illegal and criminality, culture threat, economic threat, reconquest, and national security. Five moral values were based on Graham et al.’s study (2009), which include harm, fairness, in-group, authority, and purity. We used the conventional approach to represent text (news articles in our case) as word feature vectors. We likewise constructed word feature vectors from keyword sets representing each moral foundation and Latino threat narrative theme. We then computed a cosine similarity of keyword vectors and news article vectors as a measure of similarity, with higher scores suggesting greater similarity. Using the median cosine similarity score as the cut-off point, we computed conditional probabilities that moral foundation themes and threat narrative themes co-occurred.

Results: From 1997-2014, in-group, the moral value used to distinguish we and they, was the most prevalent (about 26 %) among five moral foundations in Latino immigration-related news articles, while fairness was the lowest (0.3%). This pattern was consistent across all 13 news media outlets. Given the LTN “illegal and criminals” theme as the prior, the probability to observe authority (the moral values based on tradition and authority) in a same news article was 63%. Given the LTN theme “threat to national security” as the prior, the probability to observe harm (the moral values based on empathy) in the same news article was 62%.

Implication: This study demonstrated a data-driven method that could leverage large-scale data for promoting social justice issues. Our findings revealed that certain morality values, such as in-group, authority, and harm might serve as the foundation for media stigmatizing Latino immigrants in the US society. To advance the understanding of these negative stereotypes, we call for social work researchers to utilize current methods of data science to increase the capacity for analyzing big data and improving social well-being.