Note(s): 1. Classification of Cohen (1988): small effect (f² = 0.02), medium effect (f² = 0.15), and large effect (f² = 0.35)

2. ESize, Industry, and Hostility are the control variables in our model

3. For the Q² Predict results, we report the outcomes for the model without control variables, using 3 folds, a fixed seed, and 10 repetitions. The Q² predict values are above the threshold of zero (e.g. Shmueli et al., 2019)

4. The discussion focuses on the model without control variables for three reasons. First, none of the control variables were statistically significantly associated with MC effectiveness or with the forecast macro-functions, indicating that they add little explanatory information to the model. Second, the structural coefficients remain virtually unchanged across the two specifications (e.g., the responsiveness–planning coefficient is 0.352 without controls and 0.327 with controls), suggesting that the estimates are robust and not driven by omitted-variable bias. Third, parsimony is advisable given our sample size (n = 86), as adding controls consumes degrees of freedom

5. As a robustness check for the model with control variables, we recomputed the standard errors of the structural coefficients using the stable exponential adjustment method (STBL3) proposed by Kock (2018), applied to the estimated path coefficients. Examination of the bootstrap path-coefficient distributions indicated unimodal distributions, and the inner-model variance inflation factors were all below conventional thresholds (maximum = 2.861, in the linear model with control variables); in our view, the attenuation of significance when controls are added is therefore attributed to the loss of statistical power in a sample of n = 86 rather than to bimodality or multicollinearity. Under stable standard errors (STBL3), the hypothesized structural relationships in the model with controls remain statistically significant: Responsiveness → ForrolePlan (β = 0.327, SE = 0.098, p = 0.001), Responsiveness → ForroleCon (β = 0.303, SE = 0.099, p = 0.002), Responsiveness → ForroleEv (β = 0.353, SE = 0.097, p < 0.001), ForrolePlan → MC Effectiveness (β = 0.352, SE = 0.097, p < 0.001), and ForroleEv → MC Effectiveness (β = 0.251, SE = 0.100, p = 0.012), whereas the direct effect Responsiveness → MC Effectiveness (β = 0.105, SE = 0.105, p = 0.315) and ForroleCon → MC Effectiveness (β = −0.002, SE = 0.108, p = 0.985) remain non-significant. For more information about STBL3, see Kock (2018)