*2.9. Statistical Analysis*

All statistical analyses were performed using the SAS statistical program [39]. First, it was performed the normality test on all variables using the UNIVARIATE procedure. BW, DMI, DWG and FCR data were analysed for each period with a completely randomized design with repeated measures over time, using the MIXED procedure. Initially, initial BW was included as a covariate to adjust the variables DWG, DMI and final BW. However, this covariate was removed from the model because it was not significant (*p* > 0.05). Different variance–covariance structures were verified to fit the statistical model, and the compound symmetry structure showed the best fit according to the criteria of the lowest values of BIC and AIC [40]. The full statistical model used was:

$$\mathbf{Y}\_{\rm ijk} = \mu + \mathbf{T}\_{\rm i} + \mathbf{P}\_{\rm j} + (\mathbf{T} \times \mathbf{P})\_{\rm ij} + \mathbf{A}\_{\rm k} + \mathbf{e}\_{\rm ijk} \tag{1}$$

where Yijk represents the value measured at period j and treatment i for the lamb k, μ represents the overall mean, Ti represents the fixed effect of HM treatments (i = 1, 2, 3, 4), Pj represents the fixed effect of the period within four feeding periods (j = period 1: 1–14, period 2: 15–28, period 3: 29–42 and period 4: 43–56 d), (T × P)ij represents the fixed effect of interaction between treatment and period, Ak represents the random effect of lambs provided different diets (k = 1, 2, 3, . . . 36), and eijk represents the random residual error.

On the other hand, data on carcass characteristics, animal organs and meat quality were analysed using the GLM procedure. Each lamb was considered an experimental unit. Initially, final BW was included as a covariate to adjust all variables (carcass characteristics, organs and meat quality). However, this covariate was removed from the model because it was not significant (*p* > 0.05). The statistical model used was: Yijk = μ + Ti + eij, in which μ is the mean value, Ti is the treatment effect (fixed), and eij is the error term.

Linear and quadratic orthogonal polynomials were used to evaluate the effects of HM level on all variables evaluated. Means of treatments were compared using the Tukey test, and significant differences were considered when *p* ≤ 0.05. In addition, a trend was considered when *p* > 0.05 and ≤0.10.
