Statistical Analysis

All statistical analysis was done using STATA version 15 (StataCorp LP, College Station, TX). To determine whether to retain a variable in the final multivariate model, univariate methods were used and the factor retained if P < 0.35 (Niranjan et al., 2005). The sow was the experimental unit for statistical analysis. In this 2 × 2 factorial arrangement both factors, MOI and DOF, and their interaction served as fixed effects. Score on day 0 was included as a linear covariate for lesion quantity, lesion severity, lameness, and BCS models. Sow served as the random effect unless otherwise noted. A P-value of <0.05 was treated as significant.

Farrowing success was analyzed as a binary outcome using a mixed effect logistic regression model with degree of familiarity and method of introduction as fixed effects and breeding group as the random effect. Total born was defined as the sum of live born, still born, and mummies. Birth weight was calculated for those piglets that were born alive. Total born and birthweight were analyzed by mixed effect generalized linear model where the family was Gaussian and the link function identity. The MOI, DOF, farrowing room (pens versus stalls), and parity served as fixed effects. Sow served as the random effect for total born, and replicates the random effect for birthweight. Live born was analyzed by generalized linear model with a binomial distribution, grouped by total born, and the link function was logit. The model included both factors and their interaction, farrowing room, and parity as fixed effects. Sow served as the random effect. A Kruskal–Wallis test was used to test the effect of DOF on whether a sow was bred by day 5 postweaning.

A mixed effect ordinal regression model was used to model lesion quantity, and lesion severity. The lesion quantity model included DOF, MOI, and their interaction, the interaction of DOF and day postweaning, day postweaning, region, and parity as fixed effects. The lesion severity model included DOF and MOI as well as their interaction, the interaction of both DOF and MOI with day postweaning, the day postweaning, region, and parity as fixed effects. A Bonferroni correction was used to account for multiple comparisons between parities, and DOF and MOI on different days. The correlation between lesion severity and lesion quantity was done using a spearman rank sum correlation.

Due to the low number of animals scored as a 2 (n = 18), and no animal scored as a 3, lameness was condensed to lame (score ≥ 1.5) or not lame (score <1.5). Lameness was assessed with a mixed effect binomial regression model including DOF, MOI, the interaction of DOF and MOI with the day postweaning, the day postweaning, and parity as fixed effects and the sow as random effects. A Bonferroni correction was used to account for multiple comparisons between multiple days and parities.

BCS was analyzed with a mixed effect ordinal regression model including DOF, MOI, the interaction of DOF and MOI, the interaction of DOF and MOI with the day postweaning, parity, and day postweaning as fixed effects. A Bonferroni correction was used to account for multiple comparisons between days, DOF and MOI on different days, and parities. Throughout the study only scores of 2, 3, and 4 were recorded.