Degree Type

Thesis

Date of Award

2009

Degree Name

Master of Science

Department

Animal Science

First Advisor

Thomas J. Baas

Abstract

Even though breeding values (BV) for each pig are estimated accurately, there is a risk that comparison of individual BV from different herds can be negatively biased. The cause of this bias mainly comes from the assumption that genetic means (means of BVs) of each herd are the same. In many cases this assumption is not valid, which may negatively bias the accuracy of BV comparisons across herds. To indicate the degree of bias, many researchers have studied "connectedness", genetic similarity between herds. Few comparisons between many connectedness statistics are available. Most researchers used simulation data to calculate connectedness rather than field data due to computational requirements in large populations.

In addition, no research indicates current connectedness levels of the swine industry in the United States. There were 3 objectives for this study. The first objective was to define several methods of estimating connectedness and compare advantages and disadvantages of each. The second objective was to evaluate levels of connectedness and prediction error of difference of BV between animals (PEVD), using purebred swine industry data of the Duroc breed from 21 herds over 19 years. The third objective was to investigate the relationship between connectedness and bias, and between different methods of estimating connectedness.

In this study, connectedness rating (CR), connectedness correlation (R), common sire % (CS%) were chosen as indicators of connectedness. The results of this study indicate that increasing values for connectedness decreases bias (low PEVD) of comparisons of BV across herds. These 3 connectedness statistics were highly and positively correlated. The correlations between these 3 connectedness statistics and bias were largest for R, and smallest for CS%. Even though R was the most accurate indicator of connectedness, this statistic was computationally demanding to obtain. Therefore, CR was most suitable as an indication of connectedness, due to its ease of computation. The CS% can also be an indication of connectedness, but only when pairs of herds are relatively large (>100). When these connectedness statistics become lower than approximately 10%, the risk of bias significantly increases. These biases decrease accuracy of comparison of BV, which in turn decrease genetic response to selection. Level of connectedness in the current Duroc breed over 19 years (1990 to 2008) was estimated. As a general tendency, connectedness (CR, R and CS %) increased until 2003 but after that, they started to decrease. Because these connectedness levels were less than 10% from 1990 to 2008, there is risk of bias in reporting accuracy of BV prediction in the Droc breeds. Thus, exchanging common reference sires is required to establish well connected herds for accurate comparison of BV across herds.

For further study, in order to investigate the effect of heritability and record density in multiple trait models, analysis using additional traits of interest such as maternal traits which have lower heritability and are sparsely recorded will be required. Different heritability levels can be evaluated with the current dataset (i.e., pH vs. Backfat). An attempt to investigate the effect of progeny number with additional data may be also useful and other breeds should also be studied. Using these results, a more detailed simulation study of the effect of exchanging common sires is warranted. Further studies could evaluate the needed number of common sires for maternal traits where records are obtained only in one sex and later in life, as well as simulation studies that examine the effectiveness of sire sampling programs.

DOI

https://doi.org/10.31274/etd-180810-719

Copyright Owner

Norikazu Soga

Language

en

Date Available

2012-04-30

File Format

application/pdf

File Size

82 pages

Share

COinS