The Many Measures of
Accuracy: How Are They
Related?
Matt Spangler, Ph.D.
University of Nebraska-Lincoln

“Prediction is hard, especially about
the future”
Why Worry About Accuracy?
Freedom from mistake or error
 Conformity to truth or to a standard or
model
 Degree of conformity of a measure to a
standard or true value

Webster

Yearling EPD of +100 for YW
◦ +100 after 50 progeny?
Static?
 Values
are standard deviations
 Low accuracy means larger possible
change values
 Not static
◦ Different for each breed, trait, and could
differ between evaluations
Possible Change
Possible Change Table
Weaning Wt. EPD = 40
ACC. = 0.60
Possible change +/- 4.60
68% confident his true EPD is between 35.4 and 44.6
How to Use Possible Change
0.09
0.08
0.07
0.06
0.05
0.04
0.03
0.02
0.01
0
25
30
35
40
45
50
55
68%
95%
Confidence Intervals for EPD = 40, Acc = 0.60, SEP = 4.6
Sire
Dam
Mating—Passing of Alleles





Why the low correlations (accuracy) for
yearling bulls?
Uncertainty surrounding what alleles were
received from parents
We begin to understand this when an
animal has a record
Becomes more clear as we see what it is
passing on to its offspring
Commercial producers do not have this
luxury
Mendelian Sampling

Two yearling bulls with a +5 CED EPD
with accuracy of 0.2.
◦ Possible change of 6
With the addition of more information
their EPDs change
 One favorably and the other unfavorably
 More information earlier allows you to
choose animals more accurately

Calving Ease
Bull A
Bull B
◦ +5
+5

Addition of progeny information

Bull A
Bull B

◦ -1


+11
In this extreme case risk was 12% more
calving difficulties
Average is still +5*
Example-CED

More conservative
◦ More progeny to get higher values
The U.S. standard
 Range from 0-1

BIF Accuracy


Correlation between predicted and “true”
breeding values

2
nh
r
2
4  (n  1)h
True Accuracy

r  1 (1 BIF)
Relationship to BIF
2
Accuracy Compared
The proportion of variation due to additive
genetics of a single trait explained (by a
genomic test)
 Range between 0 and 100%


A test explains 49% of the GV for a trait…
Genetic Variation Explained
Genetic Correlation
BIF Accuracy
0.1
0.005
0.2
0.020
0.3
0.046
0.4
0.083
0.5
0.132
0.6
0.2
0.7
0.286
MBV BIF Accuracy
BW EPD -1.0 with accuracy (BIF) of .90
50K Prediction for BW explaining 50% GV
Test
Any of the measures described can be
used
 But not directly compared
 Easiest if they were on the same scale

What To Use?--Genetics
Different ball of wax
 Interested in proportion of phenotypic
variation explained
 No metrics available to do this

What To Use--Management
True
Accuracy
% GV
BIF
Accuracy
• Not used in U.S. Beef Industry
• Used in context of Genomic Tests
• U.S. Beef Standard
• Used with all EPDs
Relative Ranking Of Measures





Accuracy is important
Not much difference between yearling
bulls
MA-EPD could create differences
Bull buyers will have to look at accuracy
Accuracy helps determine the “benefit”
Reducing Risk

“I’ve used high accuracy calving ease
bulls and had big calves”
Environment?
 Cow?
 Within sire variation?

It Doesn’t Work…
Pay attention to accuracy
 There are different measures

◦ They are not “apples to apples”
Right now you have to do the math
 You need to know how to equate BIF
accuracy to %GV
 If a sire has a high accuracy then
additional information does very little

Take Home
Questions?