Laboratory experimentation and simulation of discrete direction choices: Investigating hypothetical bias, decision-rule effect and external validity based on aggregate prediction measures

Introduction: Laboratory experiments have recently become popular methods for understanding decision-making behaviour of humans in evacuations. When designed for individual-level data extraction, these experiments can be analysed by econometric models. These models can subsequently be implemented as...
Ausführliche Beschreibung

Introduction: Laboratory experiments have recently become popular methods for understanding decision-making behaviour of humans in evacuations. When designed for individual-level data extraction, these experiments can be analysed by econometric models. These models can subsequently be implemented as part of broader computational tools that simulate evacuation processes. In taking this approach, a modeller will face several major questions at both experiment design and estimation/implementation phases: (I) Can the behaviour be instead inferred, with adequate accuracy, from stated choice surveys (i.e. hypothetical bias)? (II) Given that these laboratory experiments are performed in specific geometric layouts, are their modelling outcomes transferable to geometric layouts other than that of their origin (i.e. external validity)? (III) At the modelling phase, how critical is to determine whether to set the decision rule as random regret minimisation as opposed to random utility maximisation (i.e. decision rule)? This study investigates how each of these three problems impact on prediction outcomes when these models are employed to simulate an evacuation system.Methods: Using three sets of experimental observations of discrete direction choice (one stated-choice dataset and two revealed-choice datasets) and by integrating their associated choice models with the crowd motion simulation tool that we have developed, we examined these questions mainly based on aggregate simulation outputs, including evacuation times, exit utilisations and movement patterns.Findings: Our findings showed that the effect of decision rule specification on the prediction of aggregate measures was less noticeable and of little practical importance compared to the effect of hypothetical bias. Counterpart regret and utility models resulted in very similar simulated movement patterns, evacuation times and exit utilisations. Changing the source of the model estimation, however, made relatively bigger differences in those predictions, although not to drastic levels. In general, however, models obtained from independent experiments showed great degrees of parameter similarity and prediction consistency, while the most noticable difference between them was related to their scales. Despite the scale difference, models estimated from one experiment well replicated independent observations of the other experiment. In other words, models dervied from all three data sources proved almost equally valid for accurately replicating macro-scale observations.Applications: These questions are of practical importance in the design and analysis of laboratory evacuation experiments and in establishing their external validity/transferability and also in prioritising modelling issues.Future directions: The question of decision rule could be revisited using more recent versions of the random regret model (than the 2010 variant, applied here). The question can also potentially be extended to comparisons between the econometric and machine-learning methods. The question of experimental validity needs further investigation in relation to the aspects of evacuation decision-making other than the direction choice. Ausführliche Beschreibung

Autor*in:	Haghani, Milad [verfasserIn] Sarvi, Majid [verfasserIn]

Format:	E-Artikel
Sprache:	Englisch

Erschienen:	2019

Schlagwörter:	Evacuation experiments Evacuation simulation Laboratory crowd experiments Pedestrian dynamics Discrete choice models Model transferability Contextual bias External and internal validity

Übergeordnetes Werk:	Enthalten in: Transportation research / A - Amsterdam [u.a.] : Elsevier Science, 1979, 130, Seite 134-157
Übergeordnetes Werk:	volume:130 ; pages:134-157

DOI / URN:	10.1016/j.tra.2019.09.040

Katalog-ID:	ELV003137295