Optimization and Validation of a Linear Appraisal Scoring System for Milk Production-Linked Zoometric Traits in Murciano-Granadina Dairy Goats and Bucks

Round 1

Reviewer 1 Report

Dear authors,

I read this paper with high interest, but found some mistakes.

Line 51 - abbreviation CAPRIGRAN is previously described in line 45

Table 1 - Nipple diameter has wrong measure - in degrees

Line 186 - how are you used compass?

Line 275 - comma lost

And another one. Most research using PCA has graphical illustrations - QQ-plot and other.

Best regards

Author Response

Reviewer 1

Comments and Suggestions for Authors

Dear authors,

I read this paper with high interest, but found some mistakes.

Response: We thank the reviewer for his/her careful appreciation of our paper, as it makes it improve its quality.

Line 51 - abbreviation CAPRIGRAN is previously described in line 45

Response: Corrected.

Table 1 - Nipple diameter has wrong measure - in degrees

Response: Corrected.

Line 186 - how are you used compass?

Response: Drill compass is a tool used in zoometry that consists of two curved branches finished in the shape of a button and articulated by a screw. One of the branches articulates a fixed arch, graduated in cm, that slides through an opening attached to the arch. Pairs of compasses vary in regards the part of the body that we aim to measure (splitter, interior or exterior measurements). As it is a regularly used tool in zoometrics we felt that a description was not needed.

1. Dorantes-Coronado, E.; Torres-Hernández, G.; Hernández-Mendo, O.; Rojo-Rubio, R. Zoometric measures and their utilization in prediction of live weight of local goats in southern México. Springerplus 2015, 4, 695.
2. Szymanowska, A.; Patkowski, K.; Miduch, A.; Milerski, M. Correlation between mammary gland morphology and gland cistern size to lactation milk yield in goat. Annales UMCS, Zootechnica 2010, 28.

Line 275 - comma lost

Response: Corrected.

And another one. Most research using PCA has graphical illustrations - QQ-plot and other.

Response: The reason why we decided not to add graphics in the case of Q-Q plots was that the objective of these graphs is to check the normality assumption (which was tested in our paper, see sections 2.1 and 3.1.) and detecting outliers. As our data was processed to remove outliers and we have a huge number of variables included in the analyses (both derived from LAS and zoometry), adding a particular Q-Q plot for each of them would have required much space to present results than in the present way and representing all of them in three (one for primipara, multipara and bucks) per method (LAS and Zoometry) would have resulted in crowded graphs which may not let interpreting results properly. Additionally, eigenvalues could be represented in graphics (which in our paper are represented in Tables 2 and 3 respectively. Anyway, as suggested by reviewer we added a graphic representing eigenvalue evolution across dimensions, which is the most common graphic in PCA in literature.

Reviewer 2 Report

I have liked to read this interesting paper. It is well written although English sometimes reads not properly. I would recommend authors to send the paper to an external Engish native to have it revised.

I only have two concerns about the paper, the first one being that all the covariates involved in the analysis have been related linearly to the response. There is a no clear idea about any other relationship between them. Maybe a representation of the response vs. each one of the covariates would help. In case any non-linear relationship would show up, transformations of the covariates/response should be performed (or either use additive models to incorporate that non-linear relationship).

Secondly, I think that the model selection process should be also be described with more detail. In particular, why authors have not checked among the covariates which ones are the ones included in the final model by comparing them all.

Author Response

Reviewer 2

I have liked to read this interesting paper. It is well written although English sometimes reads not properly. I would recommend authors to send the paper to an external Engish native to have it revised.

Response: The whole manusript was revised by a Cambridge ESOL examination instructor to check and correct grammar mistakes and typos, and to improve readabiltiy.

I only have two concerns about the paper, the first one being that all the covariates involved in the analysis have been related linearly to the response. There is a no clear idea about any other relationship between them. Maybe a representation of the response vs. each one of the covariates would help. In case any non-linear relationship would show up, transformations of the covariates/response should be performed (or either use additive models to incorporate that non-linear relationship).

Response: Spearman’s correlations were performed. A significant Spearman’s correlation between the two variables in a pair may result when the two variables involved in the comparison are monotonically related, even if they share a nonlinear relationship. Hene, nonlinear relationships were accounted for in statistical analyses.

13. Bishara, A.J.; Hittner, J.B. Testing the significance of a correlation with nonnormal data: comparison of Pearson, Spearman, transformation, and resampling approaches. Psychol. Methods 2012, 17, 399.

Secondly, I think that the model selection process should be also be described with more detail. In particular, why authors have not checked among the covariates which ones are the ones included in the final model by comparing them all.

Response: Model selection was further explained. Indeed, we performed CATREG analyses on two complete models, that is comprising the whole of the variables (LAS and traditional zoometry) studied with certain exceptions as follows.

The first CATREG analysis that we performed aimed at computing the ability to describe the variability in the population for the final score for zoometric traits considered as a dependent variable (DV) considering the actual scores for the complete set of seventeen measurements directly taken from the Murciano-Granadina primipara and multipara does and bucks, as the independent variables (IV) in the model.

Then, we performed a second regression analysis to evaluate the variability in the population for the final score for zoometric traits, but using all the 17 LAS variables (9-points scale) routinely evaluated in Murciano-Granadina goats instead. Afterward, the comparison of the determination power or prediction efficiency of both regression models was used to validate the performance of CAPRIGRAN LAS. Regression analyses were separately performed for primipara and multipara does and bucks, given traits comprising the mammary system were excluded from male’s appraisal.

Categorical linear regression (CATREG) models for primipara and multipara does did not comprise the variable of rump angle as it had resulted redundant at the PCA comprising LAS data (Component loadings<|0.5| across dimensions). For this reason, it was also excluded from the zoometric scale regression comparative model. Similarly, for adult and young males, the linear regression model did not comprise the variables included in the mammary system major category (anterior insertion, rear insertion height, median suspensor ligament, udder width, udder depth, nipple placement/insertion, and nipple diameter) as these are not measured in males.

For model validity comparison, the variables of body depth, angulosity, bone quality, rear legs rear view, and side view and mobility were excluded from does and bucks’ comparative regression models. This decision was made based on the fact that even during the regular zoometric assessment, these zoometric parameters may rely on rather descriptive hedonic measurements than provide the actual direct measurement for a certain zoometric trait, hence, no comparison could be performed as they were always scored using their specific LAS.