0% found this document useful (0 votes)
22 views72 pages

Statistical Designs in Horticultural Experiments

This document discusses statistical designs for analyzing horticultural experiments. It covers: - Types of experiments including field experiments, greenhouse experiments, and lab experiments. - Statistical designs for treatment factors including completely randomized designs (CRD), randomized complete block designs (RCBD), factorial designs, and designs with covariates. - Examples of SAS code for analyzing experiments with one or two qualitative or quantitative fixed factors in CRD and RCBD, including testing for interactions. - Considerations for experiments with repeated measures and missing data.

Uploaded by

vsuarezf2732
Copyright
© All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
22 views72 pages

Statistical Designs in Horticultural Experiments

This document discusses statistical designs for analyzing horticultural experiments. It covers: - Types of experiments including field experiments, greenhouse experiments, and lab experiments. - Statistical designs for treatment factors including completely randomized designs (CRD), randomized complete block designs (RCBD), factorial designs, and designs with covariates. - Examples of SAS code for analyzing experiments with one or two qualitative or quantitative fixed factors in CRD and RCBD, including testing for interactions. - Considerations for experiments with repeated measures and missing data.

Uploaded by

vsuarezf2732
Copyright
© All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd

Design and Analysis of Horticultural experiments

George Fernandez ASSOCIATE EDITOR/ASHS UNIVERSITY OF NEVADA- RENO ASHS MEETING 2004

Presentation Outline
Horticultural experiments Statistical designs Data collection / preparation Data exploration, analysis / Interpretation Presenting and publishing your results
George Fernandez --University of Nevada- Reno gcjf@[Link]

[Link] experiments
Observational studies Survey Consumer preference

Designed experiments
Field experiments Greenhouse experiments Lab experiments Growth chambers
George Fernandez --University of Nevada- Reno gcjf@[Link]

Field experiments- confounding


effects of soil and climatic variation Multi-site and / or multi year
Fixed or random effects

Independent Experimental Unit


Field Plot or Individual tree

Annual crop
Single / Multiple harvests (Repeated measures)

Perennial crop
Same crop but different year Same field Year variation Within tree variation
George Fernandez --University of Nevada- Reno gcjf@[Link]

Tree crops

Greenhouse Experiments
Minimize soil and climate variation

Artificial growth media Controlled environment: Light, temperature, humidity Experimental Unit Pots, trays Annual Crops
Single / Multiple harvests (Repeated measures)

Perennial crops

George Fernandez --University of Nevada- Reno gcjf@[Link]

Lab Experiments
Seed germination studies
Experimental unit: Petri dishes

Chemical analysis
Randomly selected plant tissue Samples

Post-harvest / storage experiments

George Fernandez --University of Nevada- Reno gcjf@[Link]

Controlled Growth Chambers


Experiments

Controlled environments Limited number of Growth chambers Lack of true replications for main treatments

George Fernandez --University of Nevada- Reno gcjf@[Link]

2. Statistical designs
Treatment designs
Single, two, three factors factorial designs Qualitative (variety, site, brands etc.) or Quantitative (temp, light, fertilizer levels etc.) Large varietals trials (more than 30 genotypes) Fixed versus random treatment effects

Experimental designs
CRD versus RCBD Factorial versus Split plot designs Experiments with sub-samples Repeated measures experiments ( multi-harvest, weekly, monthly measurements. Independent Covariates Missing values

Statistical designs- Treatment design One factor, CRD


COMPLETELY RANDOMIZED DESIGN Treatments = 4 Replicates = 4 Exp. Units=16
T1 T2 T3 T1 TREATMENTS= T1-T4
HOMOGENEOUS EXPERIMENTAL UNITS

T3 T4 T2 T4

T2 T1 T1 T2 T3 T3 T4 T4

George Fernandez --University of Nevada- Reno gcjf@[Link]

Statistical designs- Treatment design One factor, CRD WITH COVARIATE


COMPLETELY RANDOMIZED DESIGN
Treatments = 4 Replicates = 4 Exp. Units=16
WITH COVARITES
RESPONSE

T1
COVARIATE

T3 T4 T2 T4

T2 T1 T1 T2 T3 T3 T4 T4

T2 T3 T1 TREATMENTS= T1-T4

HETEROGENEOUS EXP UNITS TOTAL = TRT. EFFECT + COVARIATE + ERROR

George Fernandez --University of Nevada- Reno gcjf@[Link]

Statistical designs- Treatment design One factor, CRD


One-way ANOVA / ANCOVA SAS PROC GLM (Fixed treatment effects). Missing value- No problem Mean comparisons Qualitative trt (Mean option):
Pair wise mean comparison: LSD, Tukey , Duncan Group Comparison - Contrast

Quantitative treatment Regression (GLM) Covariate mean comparison using LSMEANS

George Fernandez --University of Nevada- Reno gcjf@[Link]

Statistical designs- Treatment design One factor, CRD


SAS Code: Qualitative treatment (TRT)
PROC GLM ; CLASS trt ;
MODEL resp=trt; LSMEANS trt /pdif stderr ADJUST=Tukey; RUN;

** ADJUST=Tukey Optional option

George Fernandez --University of Nevada- Reno gcjf@[Link]

One Qualitative factor, with a covariate (COV) in CRD


SAS Code: Testing for interaction between cov x trt PROC GLM ; CLASS trt ; MODEL resp=trt cov trt*cov; RUN; SAS Code: ANCOVA PROC GLM ; CLASS trt ; MODEL resp=trt cov ; LSMEANS trt /pdiff stderr ADJUST=Tukey; RUN;

George Fernandez --University of Nevada- Reno gcjf@[Link]

Statistical designs- Treatment design One factor, CRD


SAS Code: QUANTITATIVE treatment (QT)
PROC GLM ;
MODEL resp=QT QT*QT /solution ss1; RUN;

George Fernandez --University of Nevada- Reno gcjf@[Link]

Statistical designs- Treatment design One factor, RCBD


RANDOMIZED COMPLETE BLOCK DESIGN Treatments = 4 Replicates = 4 Exp. Units=16

T1 T3 T4

T3 T4 T3

T4 T1 T2

T2 T2 T1

B1 B2 B3

T2

T1

T3

T4

B4

TREATMENTS= T1-T4 Blocks B1-B4


HETEROGENOUS EXPERIMENTAL UNITS -

Total variability = Treatment + Block + experimental error George Fernandez --University of


Nevada- Reno gcjf@[Link]

Statistical designs- Treatment design One factor, RCBD


Two-way ANOVA SAS PROC MIXED (Treatment: Fixed effectsBlocks:- random --Mixed model). PROC GLM: CI for treatment means incorrect Mean comparisons (MIXED) (ONLY LSMEANS):
Pair wise mean comparison: LSD, Tukey , BON Group Comparison - Contrast Quantitative treatment Regression Covariate (LSMEANS at the mean level of the covariate)

George Fernandez --University of Nevada- Reno gcjf@[Link]

Statistical designs- Treatment design One factor, RCBD


SAS Code: Qualitative treatment (TRT) and random block (blk)
PROC Mixed ; CLASS trt ;
MODEL resp=trt; RANDOM blk; LSMEANS trt / cl diff ADJUST=Tukey; RUN;

** ADJUST=Tukey Optional option

George Fernandez --University of Nevada- Reno gcjf@[Link]

Statistical designs- Treatment design TWO factors, QL x QL in CRD


Treatments = 2x2 Replicates = 4 Exp. Units=16

TREATMENTS= a1b1, a1b2, a2b1, a2b2


HOMOGENEOUS EXPERIMENTAL UNITS
George Fernandez --University of Nevada- Reno gcjf@[Link]

Statistical designs- Treatment design TWO factors, QL x QL in CRD

Response

A1

A1

A1

A2
Simple effect of A at b1

A2 B1
Non cross-over interaction

A2 B1 B2

B1
No interaction

B2

B2

Cross-over interaction

Interactions in a factorial experiment


George Fernandez --University of Nevada- Reno gcjf@[Link]

Statistical designs- Treatment design TWO factors, QL x QL in CRD


Two-way ANOVA SAS PROC GLM (Treatment: Fixed effects-). Additional Covariates - reduce experimental error Test for significant interactions Mean comparisons (Significant Interactions):
Interaction means:
Pair wise mean comparison: LSD, Tukey , BON (LSMEANS) Group Comparison - Contrast

Mean comparisons (NON Significant Interactions):


Main effects means:
Pair wise mean comparison: LSD, Tukey , BON (LSMEANS) Group Comparison - Contrast

George Fernandez --University of Nevada- Reno gcjf@[Link]

Two fixed factors (AxB), CRD


SAS Code: Qualitative treatment (A*B)
PROC GLM ; CLASS A B ;
MODEL resp=A B A*B ; LSMEANS A B A*B /pdiff stderr ADJUST=Tukey; RUN;

** ADJUST=Tukey Optional option

George Fernandez --University of Nevada- Reno gcjf@[Link]

Two Qualitative factors (A*B), with a covariate (COV) in CRD


SAS Code: Testing for interaction between cov x A x B PROC GLM ; CLASS A B ; MODEL resp= A*B cov A*B*cov; RUN; SAS Code: ANCOVA PROC GLM ; CLASS A B ; MODEL resp= A B A*B cov ; LSMEANS A B A*B /pdiff stderr ADJUST=Tukey; RUN;

George Fernandez --University of Nevada- Reno gcjf@[Link]

Two fixed factors (AxB) in RCBD (BLK)


SAS Code: Qualitative treatment (A*B)
PROC MIXED ; CLASS A B BLK;
MODEL resp=A B A*B ; RANDOM BLK; LSMEANS A B A*B /diff CL ADJUST=Tukey; RUN;

** ADJUST=Tukey Optional option

George Fernandez --University of Nevada- Reno gcjf@[Link]

Two Qualitative factors (A*B), with a covariate (COV) in RCBD (BLK)


SAS Code: Testing for interaction between cov x A x B PROC MIXED ; CLASS A B BLK; MODEL resp= A*B cov A*B*cov; RANDOM BLK; RUN;

SAS Code: ANCOVA PROC MIXED ; CLASS A B BLK ; MODEL resp= A B A*B cov ; RANDOM BLK; LSMEANS A B A*B / diff CL ADJUST=Tukey; RUN;

George Fernandez --University of Nevada- Reno gcjf@[Link]

Treatment design TWO factors, QL x QT


Regression (QT) with indicator (QL) variable SAS PROC GLM (Treatment: Fixed effects, CRD). SAS PROC Mixed (Treatment: Fixed effects, BLOCKS: RANDOM effects-). Additional Covariates - reduce experimental error Test for significant interactions Mean comparisons (Significant Interactions):
Mean comparison between QL factor levels at a given level of QT (LSMEANS )

Mean comparisons (NON Significant Interactions):


Mean comparison between QL factor levels at any level of QT (LSMEANS ) a1

a1 a2 B1 b2

a1 a2
George Fernandez --University of Nevada- Reno gcjf@[Link]

a2 b1 b2

Two fixed factors (AxQT), CRD


SAS Code: Qualitative x quantitative treatment (A*QT)
PROC GLM ; CLASS A ; /*Linear model*/
MODEL resp=A QT A*QT ; LSMEANS A /pdiff stderr ADJUST=Tukey AT qt=4; RUN;

PROC GLM ; CLASS A ; /*quadratic model*/


MODEL resp=A QT Qt*QT A*QT A*QT*QT /solution ss1; LSMEANS A /pdiff stderr ADJUST=Tukey AT ( qt qt*qt=4 16); RUN;

** ADJUST=Tukey Optional option


at qt=4 Comparing A means at QT level 4

George Fernandez --University of Nevada- Reno gcjf@[Link]

Two fixed factors (AxQT) in RCBD (BLK)


SAS Code: Qualitative x quantitative treatment (A*QT)
PROC MIXED ; CLASS A BLK; /*Linear model*/
MODEL resp=A QT A*QT ; RANDOM BLK; LSMEANS A /diff CL ADJUST=Tukey AT qt=4; RUN; MODEL resp=A QT Qt*QT A*QT A*QT*QT / solution htype=1; RANDOM BLK; LSMEANS A /pdiff stderr ADJUST=Tukey AT ( qt qt*qt=4 16); RUN;

PROC MIXED ; CLASS A BLK; /*quadratic model*/


** ADJUST=Tukey Optional option


at qt=4 Comparing A means at QT level 4

George Fernandez --University of Nevada- Reno gcjf@[Link]

Statistical designs- Treatment design TWO factors, QT x QT


Response surface Regression SAS PROC RSREG (Treatment: Fixed effects, CRD). SAS PROC Mixed (Treatment: Fixed effects, BLOCKS: RANDOM effects-). Additional Covariates - reduce experimental error Study the response surface
George Fernandez --University of Nevada- Reno gcjf@[Link]

Two fixed QT factors (AxB), CRD Response surface


SAS Code: Quantitative Factors (A*B)
PROC RSREG ; MODEL resp=A B ; RUN;

George Fernandez --University of Nevada- Reno gcjf@[Link]

Statistical designs- Treatment design TWO factors, QT x QT

George Fernandez --University of Nevada- Reno gcjf@[Link]

Statistical designs- Treatment design Three factors, QLx QL x QL


Three-way ANOVA (main factors, two-factor and three factor interactions) SAS PROC GLM (Treatment: Fixed effects-). SAS PROC Mixed (Treatment: Fixed effects- Blocks Random ). Additional Covariates - reduce experimental error Test for significant interactions Mean comparisons (Significant Interactions):
Interaction means (Comparing one factor levels- while keeping other factor levels constant):
Pair wise mean comparison: LSD, Tukey , BON (LSMEANS) Group Comparison - Contrast

Mean comparisons (NON Significant Interactions):


Main effects means:
Pair wise mean comparison: LSD, Tukey , BON (LSMEANS) Group Comparison - Contrast
George Fernandez --University of Nevada- Reno gcjf@[Link]

Three fixed factors (AxBxC), CRD


SAS Code: Qualitative treatment (A B C)
PROC GLM ; CLASS A B C;
MODEL resp= A B C A*B A*C B*C A*B*C; LSMEANS A B C A*B A*C B*C A*B*C /pdiff stderr ADJUST=Tukey; RUN;

** ADJUST=Tukey Optional option

George Fernandez --University of Nevada- Reno gcjf@[Link]

Three fixed factors (AxBxC), RCBD (BLK)


SAS Code: Qualitative treatment (A B C)
PROC MIXED ; CLASS A B C BLK;
MODEL resp= A B C A*B A*C B*C A*B*C; RANDOM BLK; LSMEANS A B C A*B A*C B*C A*B*C /diff CL ADJUST=Tukey; RUN;

** ADJUST=Tukey Optional option

George Fernandez --University of Nevada- Reno gcjf@[Link]

Three Qualitative factors (A, B, C), with a covariate (COV) in CRD


SAS Code: Testing for interaction between cov * A * B*C PROC GLM ; CLASS A B C; MODEL resp= A*B*C cov A*B*C*cov; RUN;

SAS Code: ANCOVA (No interaction between covariate and treatment) PROC GLM ; CLASS A B C ; MODEL resp= A B C A*B A*C B*C A*B*C cov ; LSMEANS A B C A*B A*C B*C A*B*C /pdiff stderr ADJUST=Tukey; RUN;

George Fernandez --University of Nevada- Reno gcjf@[Link]

Three Qualitative factors (A, B, C), with a covariate (COV) in RCBD (BLK)
SAS Code: Testing for interaction between cov and A * B*C PROC MIXED ; CLASS A B C BLK; MODEL resp= A*B*C cov A*B*C*cov; RANDOM BLK; RUN; SAS Code: ANCOVA (No interaction between covariate and treatment) PROC MIXED ; CLASS A B C BLK; MODEL resp= A B C A*B A*C B*C A*B*C cov ; RANDOM BLK; LSMEANS A B C A*B A*C B*C A*B*C /diff CL ADJUST=Tukey; RUN;
George Fernandez --University of Nevada- Reno gcjf@[Link]

Statistical designs- Treatment design Three factors, (QL x QL x QT )


Regression (QT) with indicator (QL) variables SAS PROC GLM (Treatment: Fixed effects, CRD). SAS PROC Mixed (Treatment: Fixed effects, BLOCKS: RANDOM effects-). Additional Covariates - reduce experimental error Test for significant interactions Mean comparisons (Significant Interactions):
Mean comparison between QL factor levels at a given level of QT (LSMEANS )

Mean comparisons (NON Significant Interactions):


Mean comparison between QL factor levels at any level of QT (LSMEANS )
George Fernandez --University of Nevada- Reno gcjf@[Link]

Three fixed factors (AxBxQT), CRD


SAS Code: Qualitative x Qualitative xquantitative treatment (A*B*QT)
PROC GLM ; CLASS A B; /*Linear model*/
MODEL resp = A B A*B QT A*QT B*QT A*B*QT / solution ss1; LSMEANS A B A*B / pdiff stderr ADJUST=Tukey AT qt=4; RUN;

PROC GLM ; CLASS A B; /*quadratic model*/


MODEL resp= A B A*B QT Qt*QT A*QT B*QT A*B*QT A*QT*QT B*QT*QT A*B*QT*QT /solution ss1; LSMEANS A /pdiff stderr ADJUST=Tukey AT ( qt qt*qt=4 16); RUN;

** ADJUST=Tukey Optional option


at qt=4 Comparing A means at QT level 4

George Fernandez --University of Nevada- Reno gcjf@[Link]

Tree fixed factors (AxBxQT), RCBD


SAS Code: Qualitative x Qualitative xquantitative treatment (A*B*QT)
PROC MIXED ; CLASS A B BLK; /*Linear model*/
MODEL resp = A B A*B QT A*QT B*QT A*B*QT / solution htype=1; RANDOM BLK; LSMEANS A B A*B / diff CL ADJUST=Tukey AT qt=4; RUN;

PROC MIXED ; CLASS A B BLK; /*quadratic model*/


MODEL resp= A B A*B QT Qt*QT A*QT B*QT A*B*QT A*QT*QT B*QT*QT A*B*QT*QT /solution htype=1; RANDOM BLK; LSMEANS A /diff CL ADJUST=Tukey AT ( qt qt*qt=4 16); RUN;

** ADJUST=Tukey Optional option


at qt=4 Comparing A means at QT level 4

George Fernandez --University of Nevada- Reno gcjf@[Link]

Statistical designs- Experiment design with sub samples


COMPLETELY RANDOMIZED DESIGN
Treatments = 2x2 Replicates = 4 Exp. Units=16 subsamples = 2 ; observatianal units = 32
resp1
a1b1 a2b2 a1b2 a2b1

resp2
a2b2 a1b1 a2b1 a1b2

a2b1

a1b2

a2b2

a1b21

a2b1

a1b1

a2b2

a1b2

TREATMENTS= a1b1, a1b2, a2b1, a2b2


HOMOGENEOUS EXPERIMENTAL UNITS
George Fernandez --University of Nevada- Reno gcjf@[Link]

Statistical designs- Experiment design with sub samples


Experimental and observational units Experimental and Sampling error SAS PROC Mixed (USE RANDOM statement to specify the random experimental error, block effects etc.-). Single factor: Lsmeans or Contrasts for mean comparisons TWO or more factors: Test for significant interactions
Mean comparisons (Significant Interactions):
Mean comparison between one factor levels at a given level of other factor (LSMEANS )

Mean comparisons (NON Significant Interactions):


Mean comparison between one factor levels at any level of other factor (LSMEANS )
George Fernandez --University of Nevada- Reno gcjf@[Link]

Treatment design One factor (TRT), with replications (REP) and sub samples (Sub) - CRD
SAS Code: Qualitative treatment (TRT) and random replicates (rep) and subsamples
PROC Mixed ; CLASS trt Rep;
MODEL resp= trt; RANDOM REP(TRT); LSMEANS trt / cl diff ADJUST=Tukey; RUN;

** ADJUST=Tukey Optional option

George Fernandez --University of Nevada- Reno gcjf@[Link]

Treatment design One factor (TRT), with replications (REP) and sub samples (Sub) - RCBD
SAS Code: Qualitative treatment (TRT) and random replicates (rep) and subsamples
PROC Mixed ; CLASS trt Rep;
MODEL resp= trt; RANDOM REP REP*(TRT); LSMEANS trt / cl diff ADJUST=Tukey; RUN;

** ADJUST=Tukey Optional option

George Fernandez --University of Nevada- Reno gcjf@[Link]

Two fixed factors (AxB) with replications (rep) and subsamples (Sub) - CRD
SAS Code: Qualitative treatment (A*B)
PROC MIXED ; CLASS A B rep;
MODEL resp=A B A*B ; RANDOM rep(A*B); LSMEANS A B A*B /diff CL ADJUST=Tukey; RUN;

** ADJUST=Tukey Optional option

George Fernandez --University of Nevada- Reno gcjf@[Link]

Two fixed factors (AxB) with replications (rep) and subsamples (Sub) - RCBD
SAS Code: Qualitative treatment (A*B)
PROC MIXED ; CLASS A B rep;
MODEL resp=A B A*B ; RANDOM Rep rep*(A*B); LSMEANS A B A*B /diff CL ADJUST=Tukey; RUN;

** ADJUST=Tukey Optional option

George Fernandez --University of Nevada- Reno gcjf@[Link]

Statistical designs- Experiment design Split-Plot


SPLIT -PLOT 2x3 factorial arrangement
a1
b1

a2
b2

a2
b1

a1
b3

a1
b1

a2
b2

a2
b1 b3

a1

b2

b1

b3

b2

b2

b1

b 3

b2

b3

b3

b2

b1

b3

b3

b2

b1

CRD

RCBD with 2 blocks


George Fernandez --University of Nevada- Reno gcjf@[Link]

Statistical designsExperiment design Split-Plot


Two types of experimental units: Main Plot and Split plot Main plot and Split Plot error ( Mixed Model) SAS PROC Mixed (USE RANDOM statement to specify the mainPlot error, block effects etc.-). Test for significant interactions
Mean comparisons (Significant Interactions):
Mean comparison between main plot factor at a given level of split-plot factor (LSMEANS ) using Main-plot and Split plot error Mean comparison between split plot factor at a given level of main-plot factor (LSMEANS ) using Split plot error

Mean comparisons (NON Significant Interactions):


Mean comparison between main plot factor levels main plot error Mean comparison between split plot factor levels split plot error (LSMEANS )
George Fernandez --University of Nevada- Reno gcjf@[Link]

Two fixed factors (AxB) in Split-Plot arrangement - CRD (REP)


SAS Code: Qualitative treatment (A*B)
PROC MIXED ; CLASS A B rep;
MODEL resp=A B A*B ; RANDOM rep(A); LSMEANS A B A*B /diff CL ADJUST=Tukey; RUN;

** ADJUST=Tukey Optional option

George Fernandez --University of Nevada- Reno gcjf@[Link]

Two fixed factors (AxB) in Split-Plot arrangement - RCBD (REP)


SAS Code: Qualitative treatment (A*B)
PROC MIXED ; CLASS A B rep;
MODEL resp=A B A*B ; RANDOM Rep rep*(A); LSMEANS A B A*B /diff CL ADJUST=Tukey; RUN;

** ADJUST=Tukey Optional option

George Fernandez --University of Nevada- Reno gcjf@[Link]

Statistical designs- Experiment design Repeated Measures ( One or two factors)


Repeated measurements 2x3 factorial arrangement
a1
b1

a2
b1

a2
b1

a1
b1

a1
b1

a2
b1

a2
b1

a1
b1

b2

b2

b2

b2

b2

b2

b2

b2

b3

b3

b3

b3

b3

b3

b3

b3

CRD

RCBD with 2 blocks b1-b3 - repeated measurements


George Fernandez --University of Nevada- Reno gcjf@[Link]

Statistical designs- Experiment design Repeated Measures ( One or two factors)


Between Experimental and within experimental units (Repeated Measures) Repeated measurements are not independent. Model the covariance structures SAS PROC Mixed (USE REPEATED statement to specify the Correlation structures). Single Repeated M factor: Lsmeans or Contrasts for mean comparisons Test for significant interactions Mean comparisons (Significant Interactions): Mean comparison between subject factor at a given level of RM (LSMEANS ) Mean comparison between RM factor at a given level of Main plot (LSMEANS ) Mean comparisons (NON Significant Interactions): Mean comparison Between subject factor levels Mean comparison within RM factor levels
George Fernandez --University of Nevada- Reno gcjf@[Link]

Treatment design One Repeated measures factor (TRT), - with Known subjects or experimental unit (SUB) in CRD
PROC Mixed ; CLASS trt SUB;
MODEL resp= trt /ddfM=KR; REPEATED TRT /type=CS sub=SUB; LSMEANS trt / cl diff ADJUST=Tukey; RUN;

SAS Code: Qualitative Repeated treatment (TRT)

** ADJUST=Tukey Optional option Try other covariance structures (AR(1), TOEP, UN) and selct the best model with the smallest AICC value

George Fernandez --University of Nevada- Reno gcjf@[Link]

Treatment design One Repeated measures factor (TRT), - with known subjects or experimental unit (SUB) in RCBD (BLK)
SAS Code: Qualitative Repeated treatment (TRT)
PROC Mixed ; CLASS trt SUB BLK;
MODEL resp= trt /ddfM=KR; RANDOM BLK; REPEATED TRT /type=CS sub=SUB; LSMEANS trt / cl diff ADJUST=Tukey; RUN;

** ADJUST=Tukey Optional option Try other covariance structures (AR(1), TOEP, UN) and selct the best model with the smallest AICC value

George Fernandez --University of Nevada- Reno gcjf@[Link]

Two factor (AxB) with one Repeated measures factor (B), - with Known subjects or experimental unit (SUB) in CRD
SAS Code: Qualitative Repeated treatment (B)
PROC Mixed ; CLASS A B SUB;
MODEL resp= A B A*B /ddfM=KR; REPEATED B /type=CS sub=SUB; LSMEANS A B A*B / cl diff ADJUST=Tukey; RUN;

** ADJUST=Tukey Optional option Try other covariance structures, Type= (AR(1), TOEP, UN) and selct the best model with the smallest AICC value

George Fernandez --University of Nevada- Reno gcjf@[Link]

Two factor (AxB) with one Repeated measures factor (B), - with Known subjects or experimental unit (SUB) in RCBD (BLK)
SAS Code: Qualitative Repeated treatment (B)
PROC Mixed ; CLASS A B BLK SUB;
MODEL resp= A B A*B /ddfM=KR; RANDOM BLK; REPEATED B /type=CS sub=SUB; LSMEANS A B A*B / cl diff ADJUST=Tukey; RUN;

** ADJUST=Tukey Optional option Try other covariance structures, Type= (AR(1), TOEP, UN) and selct the best model with the smallest AICC value

George Fernandez --University of Nevada- Reno gcjf@[Link]

Two factor (AxB) with one Quantitative Repeated measures factor (B), - with Known subjects or experimental unit (SUB) in CRD
SAS Code: Quantitaive Repeated treatment (B)
PROC Mixed ; CLASS A SUB;
MODEL resp= A B A*B /ddfM=KR solution htype=1; REPEATED B /type=CS sub=SUB; LSMEANS A / cl diff ADJUST=Tukey at (b=4); RUN;

** ADJUST=Tukey Optional option Try other covariance structures, Type= (AR(1), TOEP, UN) and selct the best model with the smallest AICC value

George Fernandez --University of Nevada- Reno gcjf@[Link]

Two factor (AxB) with one Repeated Quantitative measures factor (B), - with Known subjects or experimental unit (SUB) in RCBD (BLK) SAS Code: Quantitative Repeated treatment (B)
PROC Mixed ; CLASS A BLK SUB;
MODEL resp= A B A*B /ddfM=KR solution htype=1; RANDOM BLK; REPEATED B /type=CS sub=SUB; LSMEANS A / cl diff ADJUST=Tukey at b=4; RUN;

** ADJUST=Tukey Optional option Try other covariance structures, Type= (AR(1), TOEP, UN) and selct the best model with the smallest AICC value

George Fernandez --University of Nevada- Reno gcjf@[Link]

Two factor (AxB) with one Repeated Quantitative measures factor (B), - with Known subjects or experimental unit (SUB) in RCBD (BLK) SAS Code: Quantitative Repeated treatment (B) with quadratic effect
PROC Mixed ; CLASS A BLK SUB;
MODEL resp= A B B*B A*B A*B*B/ddfM=KR solution htype=1; RANDOM BLK; REPEATED B /type=CS sub=SUB; LSMEANS A / cl diff ADJUST=Tukey at b=4; RUN;

** ADJUST=Tukey Optional option Try other covariance structures, Type= (AR(1), TOEP, UN) and selct the best model with the smallest AICC value
George Fernandez --University of Nevada- Reno gcjf@[Link]

Statistical designs- Experiment design Pseudo Replication

Growth Chambers study

A1
BLOCK1

A2
BLOCK2

B1

BLOCK2

BLOCK1
George Fernandez --University of Nevada- Reno gcjf@[Link]

iii. Data collection


Minimize measurement errors Calibrate your instruments and verify your accuracy Use same measurement scales Record Plot , sample, experimental unit ID Record the collection date Define your sampling unit ( experimental unit versus sampling unit) Identify blocks, treatment factors, and the response measurements

George Fernandez --University of Nevada- Reno gcjf@[Link]

iv Data preparation
Use a spread sheet (Excel) to enter the data and save as excel worksheet. Most of the stat software's can import excel data directly. One data table per work book Use short column names (Treatment, rep, response) Missing values blank Do not include title, comments etc. No empty rows Consistent units Do not mix numeric and character variable
George Fernandez --University of Nevada- Reno gcjf@[Link]

v. Data exploration
Descriptive statistics response variables Frequency tables Treatment, blocks Box blots, comparative histograms response variables Scatter plot matrix correlation among the responses repeated measures Presence of outliers Distribution issues, heteroscedasticity
George Fernandez --University of Nevada- Reno gcjf@[Link]

Data exploration-box plots

George Fernandez --University of Nevada- Reno gcjf@[Link]

Data exploration -scatter plot matrix

George Fernandez --University of Nevada- Reno gcjf@[Link]

Data exploration (GXE)

George Fernandez --University of Nevada- Reno gcjf@[Link]

Data exploration (AxB interaction)

George Fernandez --University of Nevada- Reno gcjf@[Link]

Statistical analysis
Measurement scale Response variable
Nominal, discrete, interval scales

Variable transformation
Square-root ( # of insects, disease spots, seeds/pot) Log ( Unequal variances among treatment) Arc-sin ( Percentage data from proportions)

Statistical Methods
ANOVA (Balanced ANOVA) General Linear models (Fixed effects, ANOVA, ANCOVA) MIXED (Mixed models, repeated measures, unequal variances) Chi-square analysis (categorical analysis) Logistic regression

George Fernandez --University of Nevada- Reno gcjf@[Link]

Statistical Procedure
PROC UNIVARIATE Exploration Numeric data PROC FREQ Exploring categorical PROC TTEST (Control TRETMENT) PROC ANOVA ( Fixed Effect- One-way ANOVA) PROC GLM ( Fixed effects ANOVA, ANCOVA) PROC MIXED (Mixed effects, Repeated M, Unequal variance) PROC CORR Simple correlations PROC REG Multiple linear Regressions PROC LOGISTIC Logistic regression PROC GENMOD Generalized Linear Models PROC BOXPLOT- Box plots
George Fernandez --University of Nevada- Reno gcjf@[Link]

Checking for violations of Assumptions


Equal variance Absence of influential outliers Normal distribution -Residual

George Fernandez --University of Nevada- Reno gcjf@[Link]

Checking for violations of Assumptions

George Fernandez --University of Nevada- Reno gcjf@[Link]

Describing your experiments M&M


Statistical analysis was performed using SAS Treatments effects were tested for significance using ANOVA LSD was used to detect treatment differences.

George Fernandez --University of Nevada- Reno gcjf@[Link]

Describing your experiments M&M


Treatment and experimental design Statistical Analysis Mean comparison methods Checking model violations Software Procedure - Version

George Fernandez --University of Nevada- Reno gcjf@[Link]

Presenting your results


One-way analysis Showing interactions
Qualitative treatments Regression models

ANCOVA Response surfaces

George Fernandez --University of Nevada- Reno gcjf@[Link]

You might also like