Memory and Probability

Imagining the Future: Memory, Simulation and Beliefs about Covid

“…To see the implications for beliefs, remember that with probability (1 − 𝜃) the DM samples statistical information and reports 𝜋, with probability 𝜃 he samples personal experiences 𝐸 and uses we prefer the current and simpler one. More broadly, it is possible that eliciting different events may elicit different beliefs, consistent with much evidence (see Bordalo et al 2022 for references). Memory can account for such effects.…”

“…This is consistent with our survey measurement, in which death is the assessed event. We could alternatively specify that the DM estimates FATALITY by separately sampling deaths, survivals, and by combining the samples, as in Bordalo et al (2022). We have checked that this formulation yields qualitatively similar implications, so 𝑆 b of lethal and non-lethal Covid experiences is the "relevant" domain specific information.…”

“…Section 3 presents a model in which people retrieve from memory either statistical information about Covid's lethality they may have seen in the media, or a subset of their own past experiences. We formalize selective memory based on well-known regularities in recall: similarity, frequency, and interference (Kahana 2012, Bordalo et al 2022.…”

“…We continue the program of unifying different belief biases based on selective memory (Bordalo et al 2016, Bordalo et al 2022. Compared to Bordalo et al (2022), we introduce simulation, which is consistent with the "analogical" reasoning of casebased decision theory (Gilboa and Schmeidler 1995, Jehiel 2005, Mullainathan et al 2008.…”

Influence of Repetition through Limited Recall

“…Our model draws on the literature of cued recall which suggests that, upon the reception of a cue, previously received messages are recalled proportionally to similarity with the cue (Kahana 2012;Bordalo, Gennaioli, and Shleifer 2017;Bordalo et al 2021). This implies that (1) repetition leads to a higher probability of recall, since there are more messages and therefore more chances to recall the sender of the message, and (2) other messages (e.g., closer in time) create interference: because of their similarity with the cue (closeness in time), they are more likely to be recalled and therefore reduce the probability of recalling any given sender (Bordalo et al 2021). Based on these two ideas, we model the probability that a source i is recalled at time t as:…”