A panel data regression model in which it is assumed that the unobserved heterogeneity of the entities (typically, firms or consumers) in the panel is a random variable. This unobserved heterogeneity is treated as a component of the error term and can be group-specific (a single draw for each cross-sectional entity that enters the regression identically in each period) or time-specific (a single draw for each time period that enters the regression identically for every entity). See also fixed effects.