Given the global acceptability of the model, we used it to test the next two hypotheses. Hypothesis 2 stated that, in new venture TMTs, the level of cohesion would be positively related to the level of cognitive conflict experienced during decision making. Similarly, Hypothesis 3 stated that, in new venture TMTs, the level of cohesion would be negatively related to the level of affective conflict experienced during decision making. An examination of the structural path coefficients, depicted in Fig. 2, shows that both dimensions of perceived cohesion were significantly related to cognitive conflict. However, contrary to our expectations, the relationship between cognitive conflict and sense of belonging was negative. Thus, Hypothesis 2 received only partial support. As predicted, feelings of morale was negatively related to the level of affective conflict. However, sense of belonging was unrelated to affective conflict. Thus, Hypothesis 3 also received partial support.
Our final hypothesis stated that cohesion would be positively related to new venture performance. We tested this hypothesis with hierarchical regression, the results of which are provided in Table 3. In the first step of the analysis, we developed models predicting sales growth and profitability from the control variables. In the second step, the conflict variables were entered into the model to test the direct effects of cognitive and affective conflict on the performance variables. While we did not specify hypotheses regarding these relationships, the model shown in Fig. 1 indicated implied relationships. We then developed a third model, including the two cohesion variables. This was done to assess the effects of cohesion on performance while controlling for these other influences. We tested both dimensions of cohesion together inasmuch as they occur together. This also allows us to test each, while controlling the effects of the other. While the correlation among our predictors may introduce some multicollinearity, the effect of that multicollinearity is to inflate the standard error of our predictors. As such, the presence of any multicollinearity would not bias our significance tests (Belsey et al., 1980). In addition, a scan of the correlation matrix provides little evidence of multicollinearity (Hair et al., 1995). Given that the direction of the relationship is consistent with our expectations, the actual test of the hypothesis is the significance of the increase in R2 between the two models, which is the proportion of variation in performance attributable to cohesion.