A Multilevel Model of Situational and Individual Determinants of Ethical Preferences: Multilevel and Bayesian Estimation of Attitudes to Unethical Behaviour

Abstract

This paper presents a multilevel model that examines the determinants of individual differences in ethical attitudes (as indicated in opinions of bribery) which underpin tax compliance (Alm & Torgler, 2011). The drivers of individual ethical attitudes are fundamental when examining individual taxation, but also the actions of corporate decision makers. In this way it bridges the gap between tax morale and business ethics, using elements from the ethics and criminology literature to develop a fuller model of the individual decision maker in their situation. Its use of a multilevel model provides more accurate estimates of the individual-level causal factors that lead to ethical attitudes. It uses a binomial generalized linear model and logistic regression, controlling for the multilevel nature of the data, to estimate the impact of individual characteristics on ethical attitudes. The analysis uses a range of categorical factors from the World Values Survey and augmented with national level characteristics from the IMF World Economic Outlook database. The model is re-estimated using MCMC estimation in order to control for potential biases in maximum likelihood estimation of binomial models and dependent variables with small mean values (Spiegelhalter, 2002). Factors are reviewed in terms of the estimated certainty intervals, Bayesian p-values and Estimated Sample Sizes of the MCMC model, and model evaluation uses the Deviance Information Criterion. These confirm that individual factors do have significant explanatory power, even controlling for group-effects in the multi-level model framework. The analysis produces a detailed breakdown of the impacts of explanatory factors, having controlled for variation between clusters, detailing the relevance of sub-groups in terms of age, gender, and education as well as additional factors such as religious observance, employment and income. This produces a more reliable set of estimates that produce results at odds with a number of the standard results accepted to date. It also highlights the relative weight of individual level variation in the ethical attitudes, regardless of cultural or geographic clustering. Key Points: • Examines the relationships between causal factors and ethical attitudes in the WVS/EVS Integrated dataset (N=178 country/year clusters, n=187,869) • Model incorporating causal role of several ‘demographic’ variables • Examines the role of situation in multilevel causal structure determining attitudes • Produces appropriate estimates of ethical preferences, removing misspecification bias by separating common ‘national’ influences from those varying at the individual level.