More general panel data models for hospitality and tourism research

Purpose - This paper introduces more advanced panel data specifications that would exploit heterogeneity and allow for arbitrary forms of autocorrelation and heteroskedasticity in the error terms. Design/methodology/approach - In line with Assaf and Tsionas (2019a, 2019b), this paper builds on the Mundlak device to propose panel data models to allow for random slope coefficients, as well as time slope coefficients. This paper allows for arbitrary heteroskedasticity and autocorrelation, thus mitigating possible model misspecification. This paper develops and estimates the model in a Bayesian framework. This paper's methods can be generalized to many nonlinear models including limited dependent variable models. Findings - This paper compares several competing models such as a classical panel data model, which has only firm effects. This paper also examines the role of standard deviations in the formation of firm effects and time effects in the Mundlak device. This paper clearly shows that our framework introduces the best flexibility and model fit. Research limitations/implications - This paper illustrates the importance of using more flexible models (i.e. unit-specific and time-varying coefficients) for future estimation of panel data in the field. Originality/value - This paper discusses techniques that will improve panel data estimation in the hospitality and tourism literature.

Automated summary

This paper introduces more advanced panel data specifications that exploit heterogeneity and allow for arbitrary forms of autocorrelation and heteroskedasticity in the error terms. It builds on the Mundlak device to propose panel data models that allow for random slope coefficients, as well as time slope coefficients. The paper develops and estimates the model in a Bayesian framework, and the methods can be generalized to many nonlinear models.