Table 2.

Final logistic regression model with random herd effect for logit-transformed seroprevalence of BLV in Japan

Variable Level	β1	SE2	z-value	P of z-value
Intercept	-0.36	0.59	-0.62	0.54
Housing conditions
Tied system	Ref.3
Loose system	0.71	0.316	2.23	0.03
Animal dehorning
No	Ref.
Yes	1.11	0.302	3.66	0.0002
Presence of horseflies in summer
Never or seldom	Ref.
Sometimes or often	-0.24	0.341	-0.70	0.49
Very high	0.82	0.321	2.56	0.01
Colostrum feeding
No	Ref.
From dam to calves	-1.11	0.52	-2.13	0.03
Pooled	-0.90	0.55	-1.65	0.10

¹estimated coefficients, ²standard error for the coefficient, ³reference category

Standard deviation in mixing distribution = 1.054, Standard error = 0.099

Starting from the full model with seven variables selected by univariate analyses, the best model was constructed on the basis of AIC. The best model with the smallest AIC included housing system, dehorning, observable presence of horseflies, and direct colostrum feeding. The coefficients (β values) indicate that loose housing, dehorning, and observation of a large number of horseflies in summer were positively associated with seroprevalence on infected farms (β values > 0, p values < 0.05). In contrast, feeding of colostrum was negatively associated with seroprevalence in the infected farms (β = -1.11, p = 0.03)