Can photography influence politics? At last Wednesday’s Data Science Lunch Seminar Series, Assistant Professor L. Jason Anastasopoulos from the University of Georgia explained how he is using photography-as-data to measure photography’s role in building trust between politicians and their constituents.
Take former president Lyndon B. Johnson, for example. Although many regard Johnson as a civil rights supporter, he actually voted against every civil rights bill before he took office. Yet, this snag is rarely discussed. On one hand, Johnson’s previous voting record may have been overlooked simply because of the major civil rights bills he passed during his presidency.
But, Anastaspolous argues that the rapid transformation of Johnson into a civil rights leader was also facilitated by photography. When Johnson took office in 1964, he hired Yoichi Okamoto, the White House’s first photographer, to snap pictures of him alongside civil rights leaders of the day, thus bolstering his reputation as a civil rights supporter.
Political scientists have since developed a theory to define the behaviors that politicians perform to maintain trust between themselves and their constituents: ‘Homestyle’. This involves three components: (1) resource allocation, (2) Washington activities, and (3) self-presentation (eg. either by themselves or with other individuals).
Anastaspolous’ research focuses on how photography that exhibits the last two ‘Homestyle’ components can affect political support. His short case study of racial influence in the pictures of US Congressman Lou Barletta has already yielded intriguing results. Anastaspolous selected Barletta in particular because he typically snaps pictures of himself with other individuals in the same room, meaning that contextual bias is reduced as the environment remains constant.
Interestingly, Anastaspolous found that when Barletta was photographed alone, 58% of the participants in the experiment guessed that he was a Republican. But, when photographed next to an African-American, 62% guessed that was a Democrat. Barletta also appeared trustworthier when photographed next to a woman.
The next stage of Anastaspolous’ project involves using machine learning to build a neural network to classify race so that he can repeat his case study on Barletta, but on a larger scale. After gathering over 50,000 annotated images of African-American, Asian, Hispanic, and White individuals, Anastaspolous is training his race classifier to learn from the annotated images so that it can independently tag pictures in an un-annotated data-set of 192,000 images of US House and Senate members.
While analyzing ‘photographic homestyle’ is a new approach, Anastasplous’ preliminary research already suggests that it is a valuable lens for those who are interested in the factors that influence voter psychology.
by Cherrie Kwok