“…First, we note that the Friedman index, which is suggested to assess the likelihood of tacit collusion, predicts poorly if correlated with φ Nash (ρ = 0.213, p = 0.330), but is positively and significantly correlated with φ Walras (ρ = 0.593, p = 0.003). Second, in order to control for potential dependencies between treatments from the same study, i.e., different base levels of tacit collusion between experimental settings, the following three-level linear random-intercept model is estimated: 8 The following treatments reported in Table II are not considered in this step of the inter-study analysis: Incomplete information (Fouraker and Siegel [1963]), Random matching (Orzen [2008]), Hard and Hardest (Bosch-Domènech and Vriend [2003]), and DMNC/TMNC in which participants are managers instead of students (Waichman et al [2014]). Fouraker and Siegel [1963]), and ξ s is the error component shared between observations from the same study.…”