Skip to main content
ARS Home » Northeast Area » Beltsville, Maryland (BARC) » Beltsville Agricultural Research Center » Animal Genomics and Improvement Laboratory » Research » Publications at this Location » Publication #397412
Research Project: Increasing Accuracy of Genomic Prediction, Developing Algorithms, Selecting Markers, and Evaluating New Traits to Improve Dairy Cattle

Location: Animal Genomics and Improvement Laboratory

Title: Does modeling causal relationships improve the accuracy of estimating lactation milk yields?

Author
item WU, XIAO-LIN - Council On Dairy Cattle Breeding
item Miles, Asha
item Van Tassell, Curtis - Curt
item WIGGANS, GEORGE - Council On Dairy Cattle Breeding
item NORMAN, HOWARD - Council On Dairy Cattle Breeding
item Baldwin, Ransom - Randy
item BURCHARD, JAVIER - Council On Dairy Cattle Breeding
item DURR, JOAO - Council On Dairy Cattle Breeding

Submitted to: Journal of Dairy Science Communications
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 4/5/2023
Publication Date: 7/20/2023
Citation: Wu, X., Miles, A.M., Van Tassell, C.P., Wiggans, G.R., Norman, H.D., Baldwin, R.L., Burchard, J., Durr, J. 2023. Does modeling causal relationships improve the accuracy of estimating lactation milk yields? Journal of Dairy Science Communications. https://doi.org/10.3168/jdsc.2022-0343.
DOI: https://doi.org/10.3168/jdsc.2022-0343

Interpretive Summary: The amount of milk a cow gives over the course of her lactation is often affected by events she experiences early in her lactation. Some statistical models can account for the effects of these events on her overall yield and can improve the accuracy of the prediction of her total milk yield. Accurate predictions are an important tool for dairy producers who make many breeding and culling decisions within the first 100 days in milk. This short communication compares several types of models and discusses the trade-offs between simple and more complex model application.

Technical Abstract: An early milk yield or health condition can impact a later yield or condition. Hence, recursive models can be useful for estimating lactation milk yields. In the present study, we compared three correlational and two causality models for estimating lactation milk yields. The models in the former category were best prediction (BP), linear regression (LR), and feed-forward neural networks (FFNN), whereas the latter category included a recursive structural equation model (RSEM) and recurrent neural networks (RNN). Wood lactation curves (WLC) were used to simulate data and served as a benchmark model. Individual WLC had an excellent parametric interpretation of lactation dynamics, yet their prediction accuracies were subject to the coverage of test dates. BP performed slightly better than other methods in the absence of mastitis, but it was suboptimal when mastitis was present and not accounted for. Causality models facilitated inferences about causality underlying lactation. Still, precisely capturing the causal relationships was challenging because the true biology was unknown. Misspecification of recursive effects in RSEM led to a loss of accuracy. RNN had the best accuracies when mastitis was present. Hence, modeling causality relationships did not necessarily guarantee improved accuracies, but the accuracies varied with specific models. In practice, a parsimonious model is often preferred, subject to the tradeoff between the model complexity and accuracy. Relative to the choice of statistical models, appropriately accounting for factors and covariates affecting lactation milk yields properly is equally crucial.