Donald Trump’s fact-checked tweets were recently used by researchers to develop a model that can detect his lies. New findings published in Psychological Science show that Trump’s use of language changed in predictable ways when he tweeted out falsehoods.
Lies have been linked to a shift in the way people speak in previous studies. A common pattern among those who lie is the reduction in the use of sensory-perceptual words and pronouns. Ecological validity, on the other hand, is often lacking in that research.
Because deception research takes place in a controlled environment, it is critical for researchers to conduct real-world testing to ensure that their methods work in the real world. Although the ground truth must be established in order to conduct real-world deception detection research, this was explained by the study’s author Sophie van der Zee, an assistant professor at Erasmus School of Economics.
“This can be extremely difficult, or even impossible, to accomplish. Even if a person is found guilty, it is difficult to determine exactly which parts of a suspect’s statement were fabricated from court transcripts. It’s also possible to get data from politicians’ statements that have been fact-checked. Deception detection models didn’t exist because no politician’s communications were consistently fact-checked. To finish it all off, there was Trump.
Between November 2017 and January 2018, as well as February 2018 and April 2018, the researchers collected tweets from @realDonaldTrump. It was necessary to remove tweets with long quotes and those that contained only web links from the data. Cross-referencing these two datasets with fact-checking articles from the Washington Post helped the researchers figure out whether or not any of Trump’s tweets were accurate.
A total of 142 tweets, or 30.28 percent of the total, were found to be factually incorrect in the first dataset. According to the second dataset, 111 tweets (22.93 percent) were deemed incorrect.
To determine which of Trump’s tweets were true and which were false, Van der Zee and her colleagues used a text analysis program called Linguistic Inquiry and Word Count.
Trump’s tweets were found to contain fewer emotional words, more tentative words, more negations, more cognitive processing words, fewer first-person pronouns, and more third-person pronouns in line with previous research on deception. True tweets contained fewer six-letter words than his false tweets, but they had a higher word count overall.
Ex-President Trump made “the majority of fact-checked incorrect statements,” Van der Zee said in a PsyPost interview.
Three-quarters of the time, Van der Zee and her colleagues could correctly predict whether Trump’s tweets were factually correct or incorrect based on their linguistic data. Afterwards, the scientists put their brand-new deception model to the test in comparison to other models. According to Van der Zee, “personalized deception detection outperformed the existing deception detection models in the literature.
Added Van der Zee: “Our paper also constitutes a warning for all people who share information online.” Information people post online can be used against them, that much was already known. A person’s trustworthiness can be inferred based on the words they use when sending information over the internet, which we have demonstrated using only publicly available data.
However, as with any research, there are some limitations to the study.
According to Van der Zee, “We analyzed language in tweets sent by @realDonaldTrump.” It’s possible that the ex-President did not write every tweet himself, which would introduce noise into the dataset. Possibly, a cleaner dataset would lead to higher prediction rates than the current dataset’s 74 percent.”