Individual Submission Summary
Share...

Direct link:

Estimating the Effects of Digital Foreign Interferences

Sat, August 31, 2:00 to 3:30pm, Hilton, Tenleytown East

Abstract

The 2016 U.S. elections witnessed an outrageously appalling and widespread disturbance in US history: foreign interference in an election via data-driven digital platforms. Facebook, Twitter, and Google found thousands of paid ads by Kremlin-linked Russian operatives designed to sow division within the public by targeting specific segments of US voters with divisive issues (Congressional Hearings, November, 2017). Congressional studies commissioned by the Senate Intelligence Committee (DiResta et al., Howard et al., December 2018) revealed extensive, widespread Russian disinformation campaigns on social media platforms including Facebook/Instagram and Twitter, reaching tens of millions of the US voting age population. A study (Kim et al., 2018) discovered that “suspicious” groups that did not have any public footprints ran divisive issue campaigns targeting battleground states including Pennsylvania and Wisconsin. Many of them later turned out to be Kremlin-linked Russian groups. Many of them later turned out to be Kremlin-linked Russian groups, implying that more foreign groups might have operated but been unidentified or untraceable. The list of foreign groups identified by intelligence community’s investigations keeps growing ever since its first revelation.

Despite its normative implications, however, little empirical research has been conducted to systematically assess the impact foreign interference had on elections. This is partly due to the limited availability of empirical data in estimating the effects of digital foreign inference, but also to the lack of appropriate conceptual and methodological approaches. The proposed study strives to overcome such limitations by multi-methods approaches and ultimately attempts to estimate the effects of digital foreign inference.

To this end, we will utilize multiple data sets: the “population data” of paid ads run by the Internet Research Agency, a Kremlin-linked Russian operation (over 3,500 unique ad with impression and engagement indicators) and the surveys of the users exposed to such Russian ads including four-wave panel data. The surveys of the users (n=~11,000 for the baseline survey; n=1,200 for the following waves) exposed to ads were collected by our own research team with a user-based, real-time ad tracking tool.
We will first use the “population data” of paid ads on Facebook provided by the House Permanent Select Committee on Intelligence (HPSCI) to assess targeting algorithms features (e.g., targeting keywords, descriptions of behavioral targeting) and then predict full user-level information by matrix completion methods (e.g., random forest) by utilizing our survey data.

A preliminary analysis on the matrix completion has been completed. Unlike tech platforms’ claims, the results indicated that even though they might have not widely utilized tech platforms’ geographic targeting features, Russians indeed targeted battleground states including Pennsylvania and Wisconsin, where Trump won with razor thin vote margins in the 2016 presidential election. The processes by which digital foreign interference may have affected election outcomes will be discussed.

Author