I only glanced at the paper, but I wonder how much of this is explained by just random chance?
It looks like they used multiple choice quizzes to determine both knowledge in science and a propensity to respond "don't know" indicating confidence. Any "don't know" response was counted as an incorrect response, while a correct guess increased the participants "science knowledge".
Thus, a willingness to guess something at random in the multiple choice test would both increase "science knowledge" as well as make the participant appear overconfident.
I mean, the data modeling assumed people guessing were doing so completely at random without eliminating any options (In the section "Simulation").
If I'm looking at the right document, one question was about which city out of Chicago, New York, and LA have the greatest annual temperature range (accompanied with a plot).
Almost all respondents said New York or Chicago, rather than LA or "All equal".
It looks like they used multiple choice quizzes to determine both knowledge in science and a propensity to respond "don't know" indicating confidence. Any "don't know" response was counted as an incorrect response, while a correct guess increased the participants "science knowledge".
Thus, a willingness to guess something at random in the multiple choice test would both increase "science knowledge" as well as make the participant appear overconfident.