Mental Sampling as Algorithm of Anchoring and Repulsion Effects

In general, people’s reasoning process is not independent. Both the anchoring effect and the repulsion effect are produced by prior irrelevant information. Based on three basic models, this technical report aims to develop a computational model to simulate anchoring and repulsing simultaneously. Metropolis-Hastings algorithm serves as a metaphor of anchoring and adjustment heuristic. Bayesian amortized sequential sampling and self-consistency produce repulsion effect. After a comparison with empirical data, the results show that MCMC-BASS and MCMC-BASS-SC are the two potential candidates. In general, MCMC-BASS-SC produces more significant repulsion effect than MCMC-BASS. Moreover, with the optimal stopping rule, MCMC-BASS and MCMC-BASS-SC provide a rational basis for the trade-off in the cognitive process. Biases might not be evidence of human irrationality but the results of rationality with limited information and computational resources.

Instruction for the Code: (1) "Decision Matrix" produces the decision matrix for the optimal stopping, so it needs to be run first (2) All files named by models can run independently (3) "heatmap" needs simulated data from other files