When making decisions under risk, people must evaluate the consequences of their decisions and the associated probabilities. Past work argues that these evaluations depend on people’s cognitive abilities, which affect risk-taking behavior. We test this using two hypotheses. The first posits that risk preferences are linked to cognitive abilities such that people with higher (or lower) cognitive abilities are less (or more) likely to take risks. The alternative assumes that people with low cognitive abilities are simply more likely to make unsystematic mistakes and therefore show more random choice behavior. In the majority of studies, we find the architecture of the risk preference elicitation tasks is such that random choice behavior leads to a classification of risk aversion. This means that the observed correlation between cognitive abilities and risk preferences might be spurious and mediated by the bias for random choice behavior in the task architecture. In the current work, we conduct a systematic literature search and run a meta-analysis of 30 studies. We examine different aspects of choice architecture and identify possible causes of previous, divergent findings. We find no link between risk-taking and cognitive abilities. More importantly, we show that the correlation between risk aversion and cognitive abilities is heavily affected whenever the choice set’s architecture is biased by random choice behavior. We introduce a generalized scheme to measure how elicitation methods are affected by random choice (i.e. Random-Choice Risk-Taking (RCRT) bias), discuss its implications, and make recommendations for future work utilizing risk elicitation measures. Keywords: risk preferences, cognitive ability, Raven’s cognitive reflection test, multiple price list, meta-analysis