Mixed model procedure

The data were evaluated for normality of residual distribution before analysis (PROC UNIVARIATE; SAS Institute, 2003 ) and all blood variables that were not normally distributed were logarithmically transformed. Data on production variables, plasma metabolites, BCS, and rectal temperature were analyzed using the MIXED MODEL procedure (SAS Institute, 2003 ) for repeated measures according to the following model:
in which, Y_ijklm is the dependent variable, µ is the average experimental value, Cow_i is the random effect of cow, Treatment_j is the fixed effect of treatment j (j = CL or HS), Time_k is the fixed effect of time k (k = number of day or week), (Treatment × Time)_jk represents the effect of the interaction between treatment and time, E_ijkl is the sampling error and e_ijklm is the error term.
Time (day or week) was modeled as a repeated measurement by using a first-order autoregressive covariance structure which was determined by the lowest Bayesian information criterion. When the interaction between treatment and time was significant (P ≤ 0.05), pair-wise comparisons of the individual means were performed using the Tukey–Kramer test. Differences between treatments were declared significant at P ≤ 0.05 and differences from P > 0.05 to P ≤ 0.10 were considered as trends.

Pregnancy rate to AI was defined as the number of animals diagnosed pregnant at the first pregnancy diagnosis following AI divided by the total number of animals submitted to AI. The MIXED Model procedure (SAS Institute Inc., Cary, NC) was used to analyze all binominal data (AI pregnancy rate and onset of puberty [n = 211], luteolysis by the time of insemination [n = 117]). The model included year, location, animal BW, BCS, AI technician, and treatment, with treatment as the fixed effects. Year, location, animal BW, BCS, RTS, and AI technician were considered random for all data analyses. Effects of BW, BCS, AI technician, treatment, and the appropriate interactions on AI pregnancy rate were initially evaluated within location and by the respective interactions with treatment. Terms with a significance value of P > 0.20 were removed from the complete model in a stepwise manner to derive the final reduced model for each variable. A statistical significance was reported at a P < 0.05. A tendency was reported at a 0.05 < P < 0.1.