31 August 2012 (updated 29 January 2013)

- a proposal to reduce confusion and improve understanding -


A year ago, a stud cattle breeder questioned an unusually large rise in the eye muscle area Estimated Breeding Value (EBV) of a foreign AI sire, and recently other breeders reported large falls in the docility EBVs of Limousin Information Nucleus (LIN) sires. Both observations in fact should have been considered unremarkable because they were within the prediction error margin, also known as confidence range, defined by the respective EBV accuracy figures.

Stud breeders have also published EBVs in sale catalogues without accuracy figures. Others have incorrectly advertised unproven bulls as high ranking trait leaders, or described them misleadingly in terms of their EBV and $Index rankings, but without warning that the EBVs, $Indexes, and their rankings, were imprecise and could change significantly.

Common themes are overestimation of Breedplan's capabilities and lack of awareness of the conditions of use of EBVs and $Indexes in promotion and advertising. Some breeders' use of EBVs and $Indexes in sales catalogues might even breach the Competition and Consumer Act, which Angus Australia appears to have addressed in its Regulation 17.4.

Overestimation of Breedplan's capabilities can arise from the authoritative and apparently precise appearance of displayed EBVs and $Indexes. The multi-number format can also be overwhelming. Even more confusing, EBV accuracy figures can exaggerate the true precision of EBVs by more than 40%.

The LMA has developed an alternative method of presenting EBV, $Index and accuracy information with the aim of improving understanding and interpretation of Breedplan figures. Since the original Limousin focused proposal was published, the concept has been expanded to include Angus cattle, and modified to incorporate comments of the Animal Genetics and Breeding Unit (AGBU), which developed and has responsibility for managing and maintaining the Breedplan model.

Breedplan's Strengths and Weaknesses show section

Current Presentation Format show section

An Improved Presentation Format

The LMA's proposal automates the time-consuming processes of determining error margins and rankings of EBVs and $Indexes.

Accuracy figures are automatically converted into a prediction error margin for each EBV. The error margin centred on the EBV is then displayed as a bar on a chart showing ranking relative to calves born in the reference year (currently 2010). Bars representing each published trait span a range of values within which the true breeding value is expected to exist with a probability of about two in three.

A statistical summation process then combines appropriately weighted direct and correlated trait error margins to produce an error margin for each $Index. Error margin bars for $Indexes are also displayed. The default condition is to print only the bar. A selectable option exists to print EBVs and $Indexes on each bar.

Move your cursor over individual items in the following chart to reveal descriptions of key features and the data used to determine the error margins.

The chart represents the EBVs, $Indexes and error margins for a sire described in official literature as having "high accuracy" EBVs as a result of participating in the LIN. The sire obviously does not have EBVs that could reasonably be described as high accuracy, which the current tabular method of presenting EBVs, $Indexes and accuracies fails to show. Here, the frequently overlooked EBV accuracy figure is clearly more important than the EBV it relates to!

If you would like to create charts for your own animals (Limousin or Angus), click here to go to the LMA chart creator page.

Footnote: Observations, Anomalies & Discussion

At the foot of this section are two pairs of charts, respectively for Limousin and Angus. Confidence ranges are displayed for a hypothetical animal with EBVs and $Indexes ranked in the top and bottom 5% for each breed, and with EBV accuracies of 75%.

Most confidence range bars for the first chart for each animal (top 5% in breed) are so long that it would be misleading to describe either as having EBVs and $Indexes in the top 5% of the breed (although common practice in sales catalogues), without clear warnings that the EBVs and $Indexes are imprecise and subject to change (not common practice). In fact, most breeding values and $Indexes would be more accurately described as probably being ranked in the top 20-30% for the breed, with some described no more precisely than probably being above average (in the top 50%).

A clearly observable anomaly is mature cow weight, which is recorded in ABRI percentile charts in the wrong sense - heavy mature cow weights (as with birth weight) are not preferred. Other anomalies are visible in the Limousin charts, in particular for all carcase traits except carcase weight. Inconsistent and irregular confidence range bars are indicative of trait distributions that are displaced in one direction and/or not symmetrical. Theoretically, the top 5% and bottom 5% charts should be mirror images of each other for normal (bell-shaped) trait distributions centred on the 50th percentile. Normal trait distributions are assumed in genetic evaluations.

Indeed, for Limousins, all fat trait averages coincide approximately with the 30th percentile value, the eye muscle area trait average coincides with the 60th percentile value, and the retail beef yield trait average coincides with the 70th percentile value. These skewed distributions are counterintuitive if based on Purebred or French Pure Limousin data, but could be explained by the large clusters of Lim-Flex crossbreds (one with less than 25% Limousin content) noted at extreme values.

A review of the published rib fat EBVs of 2010 born calves (556 records) identified 46 Lim-Flex crossbreds with rib fat EBVs ranging from +1.1 mm to +2.9 mm (top value for all records), with an average of +1.93 mm. The range spread for all 556 published records was -1.2 mm to +2.9 mm, averaging +0.07 mm (compared with the published Limousin breed average for 2010 born calves of +0.0 mm), and the 50th percentile value was -0.1 mm (which equalled the published value). When the 46 Lim-Flex crossbred records were removed, the average rib fat EBV of the 510 Limousin records reduced by nearly 0.2 mm to -0.1 mm, which equalled an unchanged 50th percentile value of -0.1 mm. Presumably, removal of Lim-Flex crossbred records from the other anomalous Limousin carcase trait records would result in corrections to averages that would place them closer to the respective 50th percentile values.

In order to illustrate the above points more clearly (using more records), the distribution of carcase trait records for "Limousin" calves born in the period 2008 - 2011, and with a full set of records, were analysed. Total records available were for 1,247 Purebred and French Pure Limousins, and for 80 Lim-Flex crossbreds. The following chart clearly indicates a bell-shaped distribution for Limousin centred on a Rib Fat EBV of about 0.0, and a second distibution for Lim-Flex centred on about +1.8.

Rib Fat EBV Distribution Chart

Next, the carcase weight EBVs for 2010 born calves (1360 records) were analysed. The records included 30 for Lim-Flex crossbreds. The carcase weight EBVs for all records ranged from +3 kg to +50 kg, averaging +26.8 kg (compared with the published +22 kg), and the 50th percentile was +27 kg (compared with the published +22 kg). Lim-Flex crossbreds were relatively evenly distributed throughout the records. When the 30 Lim-Flex crossbred records were removed, the average of the remaining 1330 increased slightly to +26.9 kg, and the 50th percentile remained unchanged at +27 kg.

The same records for 2008-2011 born calves were analysed to produce a carcase weight EBV distribution chart. Here, the Limousin distribution can be seen to be centred on an EBV of about +26, and the Lim-Flex distribution centred on about +21, but with all Lim-Flex values falling within the Limousin spread.

Carcase Weight EBV Distribution Chart

For completeness, the following two distribution charts are included for IMF and Retail Beef Yield.

IMF EBV Distribution Chart Retail Beef Yield EBV Distribution Chart

Distortion of the Limousin confidence ranges can thus be explained by the inclusion of Lim-Flex crossbred records in the Limousin Breedplan analysis. The distortion is not surprising given that many Lim-Flex crossbreds are influenced more by Angus than Limousin genetics. Furthermore, it is likely that the majority of Lim-Flex crossbreds were heterozygous for the F94L myostatin gene variant, which clearly shifts carcase trait EBV peaks and causes inconsistency in Lim-Flex progeny's carcase traits depending upon whether one or no gene is inherited. Such circumstances would be expected to require publication of two sets of carcase trait EBVs for Lim-Flex for each case of myostatin gene inheritance. Research published over the past decade (and summarised in a flyer on the ALBS website) indicates that variations in carcase traits arising from inconsistent F94L gene inheritance may even be more significant than the error margins predicted by current EBV accuracy figures!

Discrepancies between detailed records downloaded from the ABRI database and averages, percentile values, low and high values published by ABRI suggest that the latter is based on additional information not evident in the downloaded data. However, conclusions about the adverse effects of Lim-Flex crossbreds on the Limousin Breedplan analyses are not expected to change significantly with additional information. A justification for continued inclusion of Lim-Flex crossbreds in Limousin Breedplan analyses therefore needs to be made.

The note at the bottom of ABRI EBV Graphs ("50th percentile is the Breed Avg. EBVs for 2010 Born Calves") is clearly incorrect for Limousin's carcase traits as they are currently presented.

All Angus trait averages are within about 5% of the 50th percentile value, which is reflected in the close symmetry of the breed's two bar charts.

Adverse correlations between some traits, for example between fat and yield, and between growth and birth weight, conspire to reduce the likelihood of breeding an animal with all EBVs in the top 5% for the breed. Furthermore, most stud animals, especially animals with no progeny, are unlikely to have all EBVs that equal or exceed 75% accuracy. Confidence range bars for these animals would thus be longer than depicted below.



Post a comment

Click here.

Principal References


Wikipedia Limousin (cattle)