Building composite indices in the age of big data - Application to honey bee exposure to infectious and parasitic agents.

Pollinator insects play a crucial role in maintaining biodiversity and agricultural production worldwide. Yet they are subject to various infectious and parasitic agents (IPAs). To better assess their exposure to IPAs, discriminative and quantitative molecular methods have been developed. These tools produce large datasets that need to be summarised so as to be interpreted. In this paper, we described the calculation of three types of composite indices (numerical, ordinal, nominal) to characterize the honey bee exposure to IPAs in 128 European sites. Our summarizing methods are based on component-based factorial analyses. The indices summarised the dataset of eight IPAs quantified at two sampling times, into synthetic values providing different yet complementary information. Because our dataset included two sampling times, we used Multiple Factor Analysis (MFA) to synthetize the information. More precisely, the numerical and ordinal indices were generated from the first component of MFA, whereas the nominal index used the first main components of MFA combined with a clustering analysis (Hierarchical Clustering on components). The numerical index was easy to calculate and to be used in further statistical analyses. However, it contained only about 20% of the original information. Containing the same amount of original information, the ordinal index was much easier to interpret. These two indices summarised information in a unidimensional manner. Instead, the nominal index summarised information in a multidimensional manner, which retained much more information (94%). In the practical example, the three indices showed an antagonistic relationship between