14.09.2015 Views

Biometry 802 Experimental Design

Blocking and Randomized Complete Block Designs - Department of ...

Blocking and Randomized Complete Block Designs - Department of ...

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

<strong>Biometry</strong> <strong>802</strong><br />

<strong>Experimental</strong> <strong>Design</strong><br />

S. D. Kachman<br />

Department of <strong>Biometry</strong><br />

University of Nebraska–Lincoln<br />

http://www.ianr.unl.edu/ianr/biometry/faculty/steve/<strong>802</strong>/2000


Randomized Complete Block <strong>Design</strong>s –<br />

Introduction<br />

Outline<br />

• Blocking<br />

• <strong>Design</strong><br />

• Model<br />

• Estimation<br />

• ANOVA<br />

• Testing<br />

<strong>Biometry</strong> <strong>802</strong> – <strong>Experimental</strong> <strong>Design</strong> – Fall 2000 64


Blocking<br />

• In a CRD assignment takes place completely at random<br />

– Each treatment has an equal chance to be assigned to<br />

a more or less favorable material<br />

– Some treatments will be assigned to more favorable<br />

material<br />

– The role of the analysis is to determine how much of<br />

the observed differences can be accounted for by<br />

chance<br />

• Many times it is possible for the researcher to<br />

systematically group the experimental material into<br />

“blocks” of similar material<br />

– Continuous sources<br />

∗ Light, Temperature, and Moisture<br />

– Discrete Sources<br />

∗ Locations, Days, and Operators<br />

• By assigning each treatment to each block the distribution<br />

of experimental material will be more equitable<br />

– Each treatment will receive an equal allocation of<br />

favorable material.<br />

– The ability of the experimenter to group the material will<br />

determine the effectiveness of blocking<br />

– A poor job of blocking will result in a weaker experiment<br />

<strong>Biometry</strong> <strong>802</strong> – <strong>Experimental</strong> <strong>Design</strong> – Fall 2000 65


<strong>Design</strong><br />

Divide <strong>Experimental</strong> Units into subsets<br />

• Subsets are Different<br />

• <strong>Experimental</strong> Units within subsets are homogeneous<br />

r Blocks<br />

t Treatments<br />

• Each treatment in every block<br />

• Randomly assign treatments to experimental units within<br />

each block<br />

<strong>Biometry</strong> <strong>802</strong> – <strong>Experimental</strong> <strong>Design</strong> – Fall 2000 66


Block 1 Block 2 Block 3 Block 4<br />

<strong>Biometry</strong> <strong>802</strong> – <strong>Experimental</strong> <strong>Design</strong> – Fall 2000 67


Formula<br />

y ij = µ + τ i + β j + e ij<br />

β j ∼ N(0, σ 2 B )<br />

e ij ∼ N(0, σ 2 )<br />

• β j ’s and e ij ’s are independently distributed<br />

<strong>Biometry</strong> <strong>802</strong> – <strong>Experimental</strong> <strong>Design</strong> – Fall 2000 68


Model<br />

• Expected Value<br />

E(y ij ) = µ + τ i<br />

• Dispersion<br />

var(y ij ) = σ 2 + σ 2 B<br />

cov(y ij , y i ′ j ) = σ 2 B<br />

cov(y ij , y i ′ j ′) = 0<br />

• Distribution – Normal<br />

• Treatment differences<br />

– Estimator ȳ 1· − ȳ 2·<br />

– Standard Error ̂σȳ1·−ȳ 2· =<br />

√<br />

MSE 2 r<br />

<strong>Biometry</strong> <strong>802</strong> – <strong>Experimental</strong> <strong>Design</strong> – Fall 2000 69


ANOVA<br />

Source df SS MS EMS F<br />

Block r − 1 SSBlock MSBlock tσ 2 β + σ2<br />

Treat t − 1 SStrt MStrt rκ 2 τ + σ2 F<br />

Error (r − 1)(t − 1) = rt − r − t + 1 SSerror MSerror σ 2<br />

Of course actual numbers would be used in the df, SS, MS, F, and p-value columns.<br />

SSBlock =<br />

∑ r<br />

j=1 y2·j<br />

t<br />

− y2··<br />

rt<br />

SST rt =<br />

SSError =<br />

∑ t<br />

i=1 y2 i·<br />

r<br />

t∑ r∑<br />

i=1<br />

j=1<br />

− y2··<br />

rt<br />

y 2 ij − ∑ t<br />

i=1 y2 i·<br />

r<br />

−<br />

∑ r<br />

j=1 y2·j<br />

t<br />

+ y2··<br />

rt<br />

<strong>Biometry</strong> <strong>802</strong> – <strong>Experimental</strong> <strong>Design</strong> – Fall 2000 70


Testing<br />

• t-tests for single degree of freedom contrasts<br />

• F-test for multiple degree of freedom contrasts<br />

• Use error degrees of freedom from ANOVA<br />

• Multiple comparison procedures and Confidence Intervals<br />

as before<br />

<strong>Biometry</strong> <strong>802</strong> – <strong>Experimental</strong> <strong>Design</strong> – Fall 2000 71


RCBD–Example Emergence Rate<br />

Outline<br />

• Layout<br />

• <strong>Design</strong><br />

– Treatment and <strong>Experimental</strong><br />

• Formula<br />

• Analysis<br />

• Conclusions<br />

<strong>Biometry</strong> <strong>802</strong> – <strong>Experimental</strong> <strong>Design</strong> – Fall 2000 72


Layout<br />

Comparing 4 treatments of soybeans and an untreated<br />

control on the emergence rate of soybeans. The following<br />

layout was used:<br />

Rep 1 Rep 2 Rep 3 Rep 4 Rep 5<br />

<strong>Biometry</strong> <strong>802</strong> – <strong>Experimental</strong> <strong>Design</strong> – Fall 2000 73


<strong>Design</strong><br />

• Treatment <strong>Design</strong><br />

– Treatment factor – Seed treatment<br />

– Treatment levels (5) – Check, Arasan, Spergon,<br />

Semesan Jr., and Fermate<br />

• <strong>Experimental</strong> <strong>Design</strong><br />

– The field is located on a slope<br />

– Form blocks based on elevation<br />

– 5 plots at each elevation<br />

– 5 replications of a RCBD<br />

• Data consists of the number of plants that emerge out of a<br />

total 100 that were planted<br />

– Binomial may be a better choice for the distribution<br />

<strong>Biometry</strong> <strong>802</strong> – <strong>Experimental</strong> <strong>Design</strong> – Fall 2000 74


Formula<br />

y ij = µ + τ i + β j + e ij<br />

• y ij emergence rate of seeds treated with treatment i in<br />

Block j<br />

• µ overall average emergence rate<br />

• τ i increase in emergence rate for seeds treated with<br />

treatment i<br />

• β j increase in emergence rate for seeds planted in block j<br />

• e ij increase in emergence rate for using treatment i in<br />

block j<br />

<strong>Biometry</strong> <strong>802</strong> – <strong>Experimental</strong> <strong>Design</strong> – Fall 2000 75


Analysis<br />

Source df MS EMS F p-value<br />

Rep 4 12.46 σ 2 + 5σ 2 β<br />

2.30 .1032<br />

Treat 4 20.96 σ 2 + 5κ 2 τ<br />

3.87 .0219<br />

Cont vs rest 1 67.41 σ 2 + Q 1 12.46 .0028<br />

Rest 3 5.53 σ 2 + Q 2 1.02 .4087<br />

Error 16 5.41 σ 2<br />

V ar(ȳ Ch· − .25 ∑<br />

i≠Ch<br />

y i·) = V ar(ē Ch· − .25 ∑<br />

= σ2<br />

5 + σ2<br />

20<br />

= σ 2 5<br />

20<br />

= σ 21 + 4 1<br />

16<br />

5<br />

i≠Ch<br />

ē i·)<br />

<strong>Biometry</strong> <strong>802</strong> – <strong>Experimental</strong> <strong>Design</strong> – Fall 2000 76


Check versus rest<br />

̂γ ± t .05,16̂σ̂γ<br />

̂γ = 1(10.8) −<br />

̂σ̂γ =<br />

= −4.1<br />

√<br />

= 1.16<br />

t .05,16 = 2.12<br />

−6.5 < γ < −1.65<br />

6.2 + 8.2 + 6.6 + 5.8<br />

4<br />

5.41 12 + 4 × (1/4) 2<br />

5<br />

We are 95% percent confident that on average the<br />

emergence rate of the treated seeds is between 6.5% and<br />

1.7% more than the emergence rate of the untreated control.<br />

<strong>Biometry</strong> <strong>802</strong> – <strong>Experimental</strong> <strong>Design</strong> – Fall 2000 77


Conclusions<br />

• Treating seeds does have an effect on emergence rate<br />

• There was not a significant difference between the four<br />

treatments<br />

• It might be of interest to check if the four treatment are<br />

“equivalent”<br />

<strong>Biometry</strong> <strong>802</strong> – <strong>Experimental</strong> <strong>Design</strong> – Fall 2000 78


Summary RCBD<br />

• CRD versus RCBD<br />

• Building with blocks<br />

• Analysis<br />

<strong>Biometry</strong> <strong>802</strong> – <strong>Experimental</strong> <strong>Design</strong> – Fall 2000 79


CRD versus RCBD<br />

• The biggest advantage of a CRD is its simplicity<br />

• If we do good job constructing blocks, the advantage of a<br />

RCBD is greater precision<br />

• With a large number of treatments it can be difficult to<br />

form effective blocks<br />

• The CRD and RCBD are both examples of experimental<br />

designs<br />

• At the analysis stage: It is the randomization used that<br />

drives what analysis to use<br />

• There are often times when a CRD is a better design<br />

– Forming blocks for the sake of forming blocks is<br />

generally not a good idea<br />

– Blocking incorrectly will make matters worse<br />

<strong>Biometry</strong> <strong>802</strong> – <strong>Experimental</strong> <strong>Design</strong> – Fall 2000 80


Relative efficiency<br />

• Was blocking effective?<br />

• Based on looking at the number of replications needed to<br />

obtain the same standard for a difference<br />

n CRD = 2σ2 CR<br />

δ 2<br />

n RCBD = 2σ2 RCBD<br />

δ 2<br />

n CRD<br />

n RCBD<br />

= σ2 CRD<br />

σ 2 RCBD<br />

• From the RCBD an ̂σ 2 RCBD = MSE<br />

• After some algebra,<br />

̂σ 2 CRD<br />

=<br />

(r − 1)MSBlock + r(t − 1)MSE<br />

rt − 1<br />

• For the emergence rate experiment<br />

re =<br />

(5 − 1)12.46 + 5(5 − 1)5.41<br />

(5 ∗ 5 − 1)5.41<br />

= 1.22<br />

<strong>Biometry</strong> <strong>802</strong> – <strong>Experimental</strong> <strong>Design</strong> – Fall 2000 81


• Therefore to obtain the same precision we would need<br />

about 6 replications of a CRD versus 5 replications of a<br />

RCBD<br />

<strong>Biometry</strong> <strong>802</strong> – <strong>Experimental</strong> <strong>Design</strong> – Fall 2000 82


Building with blocks<br />

• Subjects/Plots/Animals within a block should be as similar<br />

as possible<br />

• Subjects/Plots/Animals is different blocks should be as<br />

different as possible<br />

• Common blocking factors<br />

– Time<br />

– People/Animals/Plants<br />

– Sources of materials<br />

– Locations<br />

– Machines<br />

<strong>Biometry</strong> <strong>802</strong> – <strong>Experimental</strong> <strong>Design</strong> – Fall 2000 83


Analysis<br />

<strong>Biometry</strong> <strong>802</strong> – <strong>Experimental</strong> <strong>Design</strong> – Fall 2000 84


ANOVA<br />

Source df SS MS EMS F<br />

Block r − 1 SSBlock MSBlock tσ 2 β + σ2<br />

Treat t − 1 SStrt MStrt rκ 2 τ + σ2 F<br />

Error (r − 1)(t − 1) = rt − r − t + 1 SSerror MSerror σ 2<br />

<strong>Biometry</strong> <strong>802</strong> – <strong>Experimental</strong> <strong>Design</strong> – Fall 2000 85


Estimates<br />

• As before<br />

<strong>Biometry</strong> <strong>802</strong> – <strong>Experimental</strong> <strong>Design</strong> – Fall 2000 86

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!