*4.5. Statistical Analysis*

Agreement between the final reports (i.e., Oxoid and Xebios) and the two media (i.e., MWY and BCYEα) was assessed by two-by-two contingency tables, through Cohen's κ coefficient. Comparison between *Legionella* spp. counts obtained by Xebios and Oxoid media was performed by a Wilcoxon signed-rank test.

To quantify the difference between counts on different media, our analyses took into account the decimal logarithm of counts observed. The median difference was chosen as the estimator of the actual difference for each comparison, as indicated by Wilcoxon's nonparametric analysis.

The two 2 × 2 tables were compared using the Deming–Stephan method [22,23]. Specifically, the first table (Oxoid counts, see Table 3a) was transformed through the algorithm developed by Mosteller et al. [24] in order to obtain another table (Table 3c). These could be considered as the expected values for the Xebios table according to the results of the Oxoid count. Therefore, values in Table 4c were compared to the actual observed Xebios counts (Table 3b), with Pearson's chi-squared test.

The sensitivity of *Legionella* spp. detection was calculated by comparing the number of positive plates for a given medium with the cumulative yield of *Legionella* spp. from all four media. Selectivity for each method was defined as the number of plates that suppressed the growth of organisms that were not *Legionella* spp. Sensitivity and selectivity were compared using the exact McNemar test.

All statistical analyses were performed using the statistical software R ("stats" package, version 3.6.3) [25]. For all analyses, the level of significance was set at α = 0.05.
