Shiyu Dou – Data Analytics

Risky Choice Framing Effect is Partially a Result of Mismatched Options

Theories of risk preference are fundamental to decision making and behavioral economics. In risky-choice framing studies, participants typically choose between a sure option and a risky option in either the gain frame (e.g., money will be saved) or the loss frame (e.g., money will be lost). The risky-choice framing effect is general tendency for people to be risk-averse in the gain frame and risk-seeking in the loss frame (Tversky & Kahneman, 1981). Descriptions of the sure options can include the good outcome only (e.g., “$5,000 will be saved” in the gain frame), the bad outcome only (e.g., “$10,000 will not be saved” in the gain frame), or both outcomes. The same is true of the risky option. There are 9 combinations of sure and risky options in the gain frame and 9 corresponding combinations in the loss frame, yielding 18 possible combinations. Interestingly, most published studies (including the original) have relied on mismatched comparisons that involve one combination in the gain frame and a different combination in the loss frame. Some comparisons yield the usual framing effect, but other comparisons can amplify, eliminate, or reverse the effect.

We propose a simpler gist-based model that makes the same predictions as Fuzzy Trace Theory (Broniatowski & Reyna, 2018; henceforth B&R) for the choices in question, though for slightly different reasons. In our model, options are coded +1 if they include only the good outcome, –1 if they include only the bad outcome, and 0 if they include both. For any particular choice, the appeal of the risky option is a function of Gist(Risky) – Gist(Sure). This Gist Difference variable can have integer values ranging from –2 to 2, with more positive scores indicating a stronger gist-based preference for the risky option. In our preliminary analysis, we tested this model on the studies reported by B&R, using mixed-effects logistic regression to predict the proportion of participants choosing the risky option in each cell of each study. The results indicated a strong effect of Gist Difference, with no residual effect of Frame. The implication is that risky-choice framing effects are entirely due to mismatched comparisons of gain and loss cells that differ in the gist of the options.

In a large online experiment, we assigned each of 949 MTurk workers to one of the 18 combinations of option descriptions described above. Each participant made choices in 4 domains (disease, investment, wildfire, and drought). As anticipated, mixed-effects logistic regression indicated a strong effect of Gist Difference. However, in contrast to our results for B&R’s collection of studies, there was a substantial residual framing effect in our experiment. These results indicate that risky-choice framing effects may not be completely explained by the gist of the options, as FTT predicts. Instead, participants’ choices appear to be multiply determined: partly by gist (as in FTT) and partly by frame (as in PT).