Table 1. Model selection results for the effects of host characteristics (individual identity, age and sex), sample type (female fleas, male fleas or rodent blood), sampling season (spring or summer) and the interactions of these variables on (1) bacterial phylotype richness (number of distinct phylotypes per sample) and (2) bacterial phylotype diversity (Fisher's α-index) of bacterial communities.

Effect tested	Phylotype richness		Phylotype diversity
	ΔQICca	wib	ΔQICca	wib
No effect (intercept only)	0	36	8 (1)	0 (32)
Individual host identity	28	0	25 (18)	0 (0)
Host age	2	13	2 (11)	13 (0)
Host sex	2	14	2 (25)	14 (0)
Sample type	1	20	0 (37)	31 (0)
Sampling period	2	11	1 (1)	18 (22)
Host identity × host age	28	0	25 (18)	0 (0)
Host identity × host sex	28	0	25 (18)	0 (0)
Host identity × sample type	84	0	72 (48)	0 (0)
Host identity × period	59	0	54 (41)	0 (0)
Host age × period	7	1	5 (5)	2 (4)
Host sex × sample type	6	1	3 (23)	6 (0)
Sample type × period	6	2	3 (0)	8 (41)
Host identity × host age × sampling period	59	0	54 (41)	0 (0)
Host identity × host sex × sampling period	59	0	54 (41)	0 (0)
Host age × host sex × sampling period	44	0	53 (18)	0 (0)
Host age × sample type × sampling period	18	0	15 (12)	0 (0)
Host sex × sample type × sampling period	17	0	12 (7)	0 (1)

Numbers in parentheses represent results of analyses performed on bacterial communities excluding the most dominant and influential phylotypes, namely one Bartonella and two Wolbachia phylotypes. The best models are strongly supported by the data and are marked in bold and underline.

^aΔQICc: difference in QICc (corrected Akaike information criterion; Burnham and Anderson, 2002) between current and best model.

^bw_i: Akaike weight, namely, the relative likelihood of the current model, given the data and the set of models. Akaike weights are normalized across the set of candidate models to summate to one, and are interpreted as probabilities.