**2.3.3 Hypotheses testing**

The cluster analysis evaluates the similarities of perceptions among the stakeholders based on all pairs at once, but it does not allow drawing conclusions about the analogy of the perceptions between stakeholders. In fact, a cluster can indicate a high similarity within its subjects, and still display significant differences among them when tested pair wise. Finally two hypotheses were tested. The first hypothesis was the similarity of perceptions between the villages and the experts. This test is relevant because experts play a decisional role in law and policy making although the results are implemented at local level. The second test concerned the similarity of perceptions between villages managing different forest use types. The data indicating the absence or presence of an indicator against a consolidated reference list of indicators is of binary character. Thus, the Phi coefficient [mean square contingency coefficient] was calculated (Janssen and Laatz, 2010). To test the significance of this Phi coefficient, the Pearson Chi-square was applied if the expected cell frequencies were all - 5, otherwise the Fisher exact probability test was used (Janssen and Laatz, 2010; Sachs, 2002), both with a significance level of = 0,05. The hypotheses were:

*H0: Perceptions of X and Y are not associated H1: Perceptions of X and Y are associated* 

If there is no association between variables, the answers of stakeholders are independent, meaning that the C&I sets are NOT similar: Perceptions of X Perceptions of Y.

Accordingly, if there is an association between variables, the answers are dependent, meaning that the C&I sets ARE similar: Perceptions of X = Perceptions of Y.
