Modeling Pedestrian Crowd Exit Choice through Combining Sources of Stated Preference Data. Issue 1 (January 2015)

• Record Type: Journal Article
• Authors: Haghani, Milad; Sarvi, Majid; Ejtemai, Omid; Burd, Martin; Sobhani, Amir
• Is Part Of: Transportation research record. Volume 2490:Issue 1(2015)
• Page Start: 84
• Page End: 93
• Publication Date: 2015-01
• Subjects: Transportation -- Periodicals; Roads; Transport -- Périodiques; Routes -- Périodiques; Routes -- Conception et construction -- Périodiques; Roads; Transportation; 388.05
• DOI: 10.3141/2490-10
• Languages: English
• ISSNs: 0361-1981

Abstract

One crucial aspect of pedestrian behavior when a facility is being evacuated is exit selection. This phenomenon, however, is difficult to capture. Recording revealed choices as the exact situations or moments in which individuals make or change their subconscious exit decisions when evacuating a place is highly ambiguous. The approach in which stated choice data are collected offers an appealing solution to tackle the problem. For the underlying factors that influence people's exit decisions to be examined, two types of stated preference (SP) data were collected and pooled: traditional stated preference data and stated preference–off–revealed preference (RP) data. The latter is from the state-of-the-art class of stated choice methods that design experiments with reference to an alternative in an individual's actual choice set. The nested logit trick model and a customized version of the generalized mixed multinomial logit model were applied to estimate the difference in variance scale of the two sectors of data and to quantify the relative contribution of the factors of distance, density, visibility, and herding behavior to exit decisions. Results showed that the SP-off-RP method, compared with the classical SP method, led to lower variance for random noise by a small margin. Compared with the nested logit trick method, the generalized mixed multinomial logit approach allowed researchers to consider more behavioral dimensions of the problem as well as accommodate the difference in scale of variance, including heterogeneity in utility weights and utility scale of individuals, correlation between alternatives, correlation of unobserved utility factors over time, and correlation between utility coefficients.