Introduction to

Augmented Designs

Applications in Plant Breeding

Jennifer Kling
Oregon State University
Outline – Augmented Designs
 Essential features
• When are they used in plant breeding?
 Design options
• Today - one-way control of heterogeneity
 Augmented Block Design - Example
• Randomization and Field Plan
• Analysis with SAS
• Interpretation of Results
 Overview of variations on the basic design
 Software and Further References
Augmented Designs - Essential Features
 Introduced by Federer (1956)
 Controls (check varieties) are replicated in a
standard experimental design
 New treatments (genotypes) are not
replicated, or have fewer replicates than the
checks – they augment the standard design
When are they used in plant breeding?
 Early generations
• Seed is limited
• Land and other resources are limited
 Want to evaluate as many genotypes as possible
 Can re-evaluate selections in subsequent seasons

• Difficult to maintain homogeneous blocks when

comparing so many genotypes
 Need a mechanism to adjust for field variation
When are they used in plant breeding?
 Adaptation to diverse environments is a primary goal
• Participatory Plant Breeding
 Farmers may prefer to grow a single replication when
there are many genotypes to evaluate
 May not be able to accommodate all entries

• Farming Systems
Research
 Want to evaluate
promising genotypes in
as many environments
as possible
 Before
Augmented Designs
Systematic use of checks
Breeding sorghum for
resistance to Striga at ICRISAT, 1983
Stage I Stage II Stage III
Observation Nursery Preliminary Screening Advanced Screening
(unreplicated) (replicated) (replicated)
susceptible test
check plot

Checkerboard
Augmented Designs - Advantages
 Unreplicated designs can make good use of
scarce resources
• Evaluate more genotypes
• Test in more environments
 Fewer check plots are required than for
designs with systematic repetition of a single
check
Augmented Designs - Advantages
 Provide an estimate of standard error that can
be used for comparisons
• Among the new genotypes
• Between new genotypes and check varieties
 Observations on new genotypes can be
adjusted for field heterogeneity (blocking)
 Flexible – blocks can be of unequal size
Some Disadvantages
 Considerable resources are spent on
production and processing of control plots
 Relatively few degrees of freedom for
experimental error, which reduces the
power to detect differences among
treatments
 Unreplicated experiments are
inherently imprecise, no matter
how sophisticated the design
Design Options
 Choose a design that is appropriate for
controlling the heterogeneity in the
experimental area
• One-way blocking
 Randomized Complete Block Design
 Incomplete Block Designs (e.g. Lattice Design)

• Two-way blocking
 Latin
Square (Complete Blocks)
 Youden Square (Incomplete Blocks)

 Row-Column Designs
Design Options
 Underlying design refers to assignment of
checks to the experimental units
 All augmented designs are incomplete with
respect to the new entries
 Can be replicated in different environments
• Need to consider relative efficiency compared to
other possible designs
Augmented Block Design Example
 Control heterogeneity in one direction
 Simplest case:
• Checks occur once in every block
• New entries occur once in the experiment
 This is an ARCBD
Meadowfoam (Limnanthes alba)
 Native plant first produced as a crop in 1980
 Seed oil with novel long-chain fatty acids
• light-colored and odor free
• exceptional oxidative stability
 Used in personal care products
 Potential uses
• fuel additive
• vehicle lubricants
• pharmaceutical products
Meadowfoam in Oregon
 Good rotation crop in the
Willamette Valley
• Winter annual
• Plant and seed meal are high in
glucosinolates
• Same equipment as grass seed
Pollinators

Honeybees

Blue Orchard Blue Bottle

Bees Flies
The Germplasm
 Diverse breeding populations were inherited from a
retired breeder (Gary Jolliff)
 Populations were regenerated in the greenhouse and
selfed
 S2 lines were transplanted to the field and allowed to
outcross
 Insufficient seed for replicated progeny trials
 Goal - form a broadbased pool for recurrent selection
• Screen S2-testcrosses
• Recombine selected S2 parents
The Experiment (2007-2008)
 New treatments:
• 50 S2-testcross families
 Check varieties:
• Ross (C1)
 Cycle4 of an elite population (OMF58)
 Widely grown commercial variety

• OMF183 (C2)
 Cycle 5 from OMF58
• Starlight (C3)
 released variety derived from this germplasm collection
The Design
 Block in one direction
 3 checks (c=3)
 6 blocks (r=6)
 Error degrees of freedom = (r-1)(c-1) = 10
 50 new entries (n=50)
 Number of plots = n + r*c = 50 + 6*3 = 68
 Plots per block = 68/6 = 11.3  12 (c=3, n=9)
 Last block has only 8 plots (c=3, n=5)
• Unequal numbers of new entries per block is OK
Statistical Model
Yij  μ + βi + c j + τ k(i) + εij
mean + blocks + checks + new entries + error
Field Plan
 Include a sufficient number of checks and replicates
to provide a good estimate of experimental error
and adequate power for detecting differences
among varieties
 Arrange blocks along field gradient to maximize
variation among blocks and minimize variation
within blocks
 Assign each of the checks at random to each block
 Assign new entries at random to remaining plots
 On-line tool for randomization:
http://www.iasri.res.in/design/Augmented%20Designs/home.htm
Designation of plots in blocks
168 167 166 165 164 163 162 161
Block 6

149 150 151 152 153 154 155 156 157 158 159 160
Block 5

148 147 146 145 144 143 142 141 140 139 138 137
Block 4

125 126 127 128 129 130 131 132 133 134 135 136
Block 3

124 123 122 121 120 119 118 117 116 115 114 113
Block 2

101 102 103 104 105 106 107 108 109 110 111 112
Block 1
Checks assigned to plots in each block
168 167 166 165 164 163 162 161
Block 6 C2 C1 C3

149 150 151 152 153 154 155 156 157 158 159 160
Block 5 C1 C3 C2

148 147 146 145 144 143 142 141 140 139 138 137
Block 4 C3 C2 C1

125 126 127 128 129 130 131 132 133 134 135 136
Block 3 C1 C2 C3

124 123 122 121 120 119 118 117 116 115 114 113
Block 2 C1 C3 C2

101 102 103 104 105 106 107 108 109 110 111 112
Block 1 C2 C1 C3
New entries included in final plan
168 167 166 165 164 163 162 161
Block 6 17 33 26 C2 46 25 C1 C3

149 150 151 152 153 154 155 156 157 158 159 160
Block 5 6 12 C1 43 39 29 38 37 C3 10 21 C2

148 147 146 145 144 143 142 141 140 139 138 137
Block 4 C3 1 23 15 C2 40 48 C1 22 16 32 47

125 126 127 128 129 130 131 132 133 134 135 136
Block 3 24 34 50 49 4 C1 20 2 35 19 C2 C3

124 123 122 121 120 119 118 117 116 115 114 113
Block 2 C1 11 C3 7 13 5 30 44 36 C2 3 31

101 102 103 104 105 106 107 108 109 110 111 112
Block 1 8 45 C2 9 27 28 C1 42 18 14 41 C3
Meadowfoam progeny trials
Data Collection
 Flowering dates, plant height, disease
resistance, lodging, seed yield, 1000-seed
weight, (oil content)
 For this example: 1000-seed weight (TSW)
• Relatively high heritability
• Positively correlated with oil content and oil yield
Fixed or Random?
 Generally assume that blocks are random
• They represent a larger group of potential blocks
• We want to make inferences beyond the particular
sample of blocks in the experiment
 Genotypes
• Checks are fixed effects – we are interested in
making comparisons with these specific varieties
• New entries could be fixed or random
 May be fixed in this case – want to select the winners
 Most commonly random – particularly in genetic studies
SAS data input – genotypes fixed
DATA anyname;
INPUT Plot Entry Name$ Block TSW;
datalines;
101 8 31 1 9.25
102 45 126 1 9.43
103 90 MF183 1 10.30
104 9 34 1 9.02
105 27 89 1 10.97
106 28 90 1 10.52
107 89 Ross 1 9.86 Other options for data input
108 42 121 1 10.56
109 18 68 1 10.39
 Import Wizard
110 14 57 1 10.97  Infile statements
111 41 118 1 10.32
112 91 Starlight 1 10.44
 SAS libraries
. . . . .
. . . . .
161 91 Starlight 6 10.15
162 89 Ross 6 10.44
163 25 85 6 10.85
164 46 132 6 9.52
165 90 MF183 6 10.14
166 26 87 6 10.73
167 33 101 6 9.24
168 17 66 6 8.79
;
Analysis #1 – new entries fixed
PROC MIXED;
TITLE 'Augmented Design using PROC MIXED – entries fixed';
CLASS block entry;
MODEL TSW=entry;
RANDOM block;
/*compare all means using Tukey's test*/
LSMEANS entry/pdiff adjust=tukey;
/*test for entries that exceed Starlight*/
LSMEANS entry/pdiff=CONTROLU(‘91') adjust=dunnett;
ods output lsmeans=TSWadj diffs=TSWpdiff;
RUN;

Use export wizard to export

 TSWadj
 TSWpdiff
Results for Analysis #1 (fixed entries)
The Mixed Procedure
Covariance Parameter Estimates
Cov Parm Estimate
Block 0.1381
Residual 0.06981
Fit Statistics
-2 Res Log Likelihood 17.7
AIC (smaller is better) 21.7
AICC (smaller is better) 22.7
BIC (smaller is better) 21.3
Type 3 Tests of Fixed Effects
Num Den
Effect DF DF F Value Pr > F
Entry 52 10 7.51 0.0008
Output from Dunnett Test
TSWpdiff.xls
Effect Entry _E n t r y E s t im a t e StdErr DF t V a lu e Probt A d ju s t m e n t A d jp
Entry 1 91 0 0 0 .3 8 7 9 0 0 0 .3 1 3 1 10 1 .2 4 0 .1 2 1 9 D u n n e t t -H s u 0 .9 3 5 1
Entry 2 91 -0 0 0 .5 8 6 2 0 0 0 .3 1 3 1 10 -1 .8 7 0 .9 5 4 7 D u n n e t t -H s u 1 .0 0 0 0
Entry 3 91 0 0 0 .4 3 6 6 0 0 0 .3 1 3 1 10 1 .3 9 0 .0 9 6 7 D u n n e t t -H s u 0 .8 8 7 9
Entry 4 91 -0 0 0 .1 3 6 2 0 0 0 .3 1 3 1 10 -0 .4 4 0 .6 6 3 6 D u n n e t t -H s u 1 .0 0 0 0
Entry 5 91 0 0 0 .5 3 6 6 0 0 0 .3 1 3 1 10 1 .7 1 0 .0 5 8 7 D u n n e t t -H s u 0 .7 4 5 9
Entry 6 91 -0 0 0 .4 7 8 2 0 0 0 .3 1 3 1 10 -1 .5 3 0 .9 2 1 2 D u n n e t t -H s u 1 .0 0 0 0
Entry 7 91 -0 0 0 .7 0 3 4 0 0 0 .3 1 3 1 10 -2 .2 5 0 .9 7 5 8 D u n n e t t -H s u 1 .0 0 0 0
Entry 8 91 -0 0 1 .0 5 6 5 0 0 0 .3 1 3 1 10 -3 .3 7 0 .9 9 6 5 D u n n e t t -H s u 1 .0 0 0 0
Entry 9 91 -0 0 1 .2 8 6 5 0 0 0 .3 1 3 1 10 -4 .1 1 0 .9 9 8 9 D u n n e t t -H s u 1 .0 0 0 0
Entry 10 91 -0 0 0 .0 2 8 2 0 0 0 .3 1 3 1 10 -0 .0 9 0 .5 3 5 0 D u n n e t t -H s u 1 .0 0 0 0
Entry 11 91 0 0 1 .4 6 6 6 0 0 0 .3 1 3 1 10 4 .6 8 0 .0 0 0 4 D u n n e t t -H s u 0 .0 1 4 9
Entry 12 91 0 0 0 .4 5 1 8 0 0 0 .3 1 3 1 10 1 .4 4 0 .0 8 9 8 D u n n e t t -H s u 0 .8 7 0 0
Entry 13 91 0 0 0 .3 2 6 6 0 0 0 .3 1 3 1 10 1 .0 4 0 .1 6 0 7 D u n n e t t -H s u 0 .9 7 2 6

 Standard error of a difference depends on the comparison

 check vs check
 new entry vs check
 new entries in the same block
 new entries in different blocks
Analysis #2 – new entries random
Plot Entry Name Block TSW new entryc
DATA arcbd; 101 8 31 1 9.25 1 999
set anyname; 102 45 126 1 9.43 1 999
103 90 MF183 1 10.3 0 90
if(entry>50) then new=0; 104 9 34 1 9.02 1 999
else new=1; 105 27 89 1 10.97 1 999
if(new) then entryc=999; 106 28 90 1 10.52 1 999
107 89 Ross 1 9.86 0 89
else entryc=entry; 108 42 121 1 10.56 1 999
RUN; 109 18 68 1 10.39 1 999
110 14 57 1 10.97 1 999
111 41 118 1 10.32 1 999
112 91 Starlight 1 10.44 0 91
113 31 96 2 11.58 1 999
114 3 15 2 9.95 1 999
115 90 MF183 2 9.24 0 90

Wolfinger, R.D., W.T. Federer, and O. Cordero-Brana, 1997

Analysis #2 – ANOVA
PROC GLM;
TITLE 'Augmented Design using GLM – entries random';
CLASS block entry entryc;
MODEL TSW = block entryc entry*new/solution;
RANDOM block/test;
RUN;

The GLM Procedure

Tests of Hypotheses for Mixed Model Analysis of Variance

Dependent Variable: TSW

Source DF Type III SS Mean Square F Value Pr > F

Block 5 2.420228 0.484046 6.93 0.0048

entryc 2 0.239211 0.119606 1.71 0.2292
new*Entry 49 26.976052 0.550532 7.89 0.0007

Error: MS(Error) 10 0.698056 0.069806

Analysis #2 – new entries random
PROC MIXED;
TITLE 'Augmented Design using PROC MIXED – entries random';
CLASS block entry entryc;
MODEL TSW=entryc/solution;
RANDOM block entry*new/solution;
LSMEANS entryc;
ods output solutionr=eblups;
RUN;

Use export wizard to export

 eblups
Results for Analysis #2 (random entries)
The Mixed Procedure
Covariance Parameter Estimates
Cov Parm Estimate
Block 0.01418
new*Entry 0.3659
Residual 0.1748
Fit Statistics
-2 Res Log Likelihood 136.5
AIC (smaller is better) 142.5
AICC (smaller is better) 142.9
BIC (smaller is better) 141.9
Type 3 Tests of Fixed Effects
Num Den
Effect DF DF F Value Pr > F
entryc 3 10 0.89 0.4779
Estimated Best Linear Unbiased Predictors
Solution for Fixed Effects
Standard
Effect entryc Estimate Error DF t Value Pr > |t|
Intercept 10.2048 0.1150 5 88.72 <.0001
entryc 89 -0.3148 0.2000 10 -1.57 0.1466
entryc 90 -0.1431 0.2000 10 -0.72 0.4906
entryc 91 -0.03478 0.2000 10 -0.17 0.8654
entryc 999 0 . . . .

Effect Block Entry Estimate StdErrPred DF tValue Probt

Block 1 000.0189 000.1018 10 0.19 0.8564
Block 2 -000.1264 000.1018 10 -1.24 0.2427
Block 3 000.0188 000.1018 10 0.18 0.8569
Block 4 000.0733 000.1018 10 0.72 0.4882
Block 5 000.0156 000.1018 10 0.15 0.8811
eblups.xls Block 6 -000.0002 000.1044 10 0.00 0.9983
new*Entry 1 000.2991 000.3561 10 0.84 0.4206
new*Entry 2 -000.2461 000.3561 10 -0.69 0.5052
new*Entry 3 -000.0869 000.3561 10 -0.24 0.8122
new*Entry 4 000.0585 000.3561 10 0.16 0.8729
new*Entry 5 -000.0192 000.3561 10 -0.05 0.9581
new*Entry 6 -000.4199 000.3561 10 -1.18 0.2656
new*Entry 7 -000.8584 000.3561 10 -2.41 0.0366
new*Entry 8 -000.6590 000.3561 10 -1.85 0.0939
new*Entry 9 -000.8146 000.3561 10 -2.29 0.0452
new*Entry 10 -000.1153 000.3561 10 -0.32 0.7527
new*Entry 11 000.6102 000.3561 10 1.71 0.1173
Variations – two-way control of heterogeneity
Design Features/Applications Reference
Modified Youden Square Two-way control of heterogeneity Federer and
Raghavarao, 1975
Augmented Row-column Entries adjusted for adjacent Federer, Nair and
checks; requires many checks Raghavarao, 1975

Modified Augmented Systematic placement of controls; Lin and Poushinsky,

(MAD Type 1) Requires square plots 1983
Modified Augmented Systematic placement of controls; Lin and Poushinsky,
(MAD Type 2) Long, rectangular plots 1985
More Variations
Design Features/Applications Reference
Factorial treatments
Augmented Split Block Intercropping, stress adaptation Federer, 2005

Incomplete Blocks
Augmented Lattice Accommodates larger number of Federer, 2002
Square checks
- Lattice Accommodates larger number of Williams and John,
checks 2003
Augmented p-rep Replicates of entries used to make Williams, Piepho, and
(partially replicated) block adjustments and estimate error Whitaker, 2011
Multiple locations
RCBD, ICBD, SAS code available Federer, Reynolds,
row-column, etc. and Crossa, 2001
Multiple Locations – Augmented or Lattice Design?

 Lattice Designs
• Sites can be complete replications
• Information from all plots are used to adjust means for field
effects
 May have greater precision and power to detect
differences among entries
 Augmented Designs
• Flexible arrangement of new entries in field plans
• Estimate of experimental error obtained from each location
 Assess site quality
 Evaluate GXE interactions

 More flexibility in combining information across sites

Software for Augmented Designs
 AGROBASE GenII by Agronomix, Inc.
• Randomize and analyze modified augmented type-2 designs
 Indian Agricultural Research Institute, New Delhi
• Statistical Package for Augmented Designs (SPAD)
www.iasri.res.in/iasrisoftware/SPAD.ppt
• SAS macro called augment.sas
web.iasri.res.in/sscnars/Macros/ABD/use_augment.sas.doc

 CIMMYT – SAS macro called UNREPLICATE

• Developed in 2000 – uses some older SAS syntax
http://apps.cimmyt.org/english/wps/biometrics/
References
 Full list of references provided as supplement
 Federer, W.T. 1961. Augmented designs with one-
way elimination of heterogeneity. Biometrics 17(3):
447-473
 Wolfinger, R.D., W.T. Federer, and O. Cordero-Brana.
1997. SAS PROC GLM and PROC MIXED for
recovering blocking and variety information in
augmented designs. Agronomy Journal 89: 856-859
Acknowledgements
 USDA-CSREES special grant for meadowfoam research
 OMG Meadowfoam Oil Seed Growers Cooperative
 Paul C. Berger Professorship Endowment
 Crop and Soil Science Department, OSU
 Meadowfoam staff
• Mary Slabaugh
• Joanna Rosinska
• Trisha King
• Alicia Wilson
Questions?