Topic 32: Two-Way Mixed Effects Modelbacraig/notes512/Topic_32.pdf · Two-way mixed model •...

29
Topic 32: Two-Way Mixed Effects Model

Transcript of Topic 32: Two-Way Mixed Effects Modelbacraig/notes512/Topic_32.pdf · Two-way mixed model •...

Page 1: Topic 32: Two-Way Mixed Effects Modelbacraig/notes512/Topic_32.pdf · Two-way mixed model • Two-way mixed model has –One fixed effect –One random effect • Tests: –Again

Topic 32: Two-Way Mixed Effects Model

Page 2: Topic 32: Two-Way Mixed Effects Modelbacraig/notes512/Topic_32.pdf · Two-way mixed model • Two-way mixed model has –One fixed effect –One random effect • Tests: –Again

Outline

• Two-way mixed models • Three-way mixed models

Page 3: Topic 32: Two-Way Mixed Effects Modelbacraig/notes512/Topic_32.pdf · Two-way mixed model • Two-way mixed model has –One fixed effect –One random effect • Tests: –Again

Data for two-way design• Y is the response variable• Factor A with levels i = 1 to a• Factor B with levels j = 1 to b• Yijk is the kth observation in cell (i, j)

k = 1 to nij

• Have balanced designs with n = nij

Page 4: Topic 32: Two-Way Mixed Effects Modelbacraig/notes512/Topic_32.pdf · Two-way mixed model • Two-way mixed model has –One fixed effect –One random effect • Tests: –Again

Two-way mixed model• Two-way mixed model has

– One fixed effect– One random effect

• Tests: – Again use EMS as guide– Two possible models

• Unrestricted mixed model (SAS)• Restricted mixed model (Text)

Page 5: Topic 32: Two-Way Mixed Effects Modelbacraig/notes512/Topic_32.pdf · Two-way mixed model • Two-way mixed model has –One fixed effect –One random effect • Tests: –Again

KNNL Example• KNNL Problem 25.15, p 1080• Y is fuel efficiency in miles per gallon • Factor A represents four different

drivers, a=4 levels• Factor B represents five different cars

of the same model , b=5 • Each driver drove each car twice over

the same 40-mile test course

Page 6: Topic 32: Two-Way Mixed Effects Modelbacraig/notes512/Topic_32.pdf · Two-way mixed model • Two-way mixed model has –One fixed effect –One random effect • Tests: –Again

Read and check the datadata a1; infile 'c:\...\CH25PR15.TXT';input mpg driver car;

proc print data=a1; run;

Page 7: Topic 32: Two-Way Mixed Effects Modelbacraig/notes512/Topic_32.pdf · Two-way mixed model • Two-way mixed model has –One fixed effect –One random effect • Tests: –Again

The dataObs mpg driver car1 25.3 1 12 25.2 1 13 28.9 1 24 30.0 1 25 24.8 1 36 25.1 1 37 28.4 1 48 27.9 1 49 27.1 1 510 26.6 1 5

Page 8: Topic 32: Two-Way Mixed Effects Modelbacraig/notes512/Topic_32.pdf · Two-way mixed model • Two-way mixed model has –One fixed effect –One random effect • Tests: –Again

Prepare the data for a scatterplot

data a1; set a1;if (driver eq 1)*(car eq 1) then dc='01_1A';

if (driver eq 1)*(car eq 2) then dc='02_1B';

⋮if (driver eq 4)*(car eq 5)then dc='20_4E';

Page 9: Topic 32: Two-Way Mixed Effects Modelbacraig/notes512/Topic_32.pdf · Two-way mixed model • Two-way mixed model has –One fixed effect –One random effect • Tests: –Again

Plot the datatitle1 'Plot of the data';symbol1 v=circle i=none c=black;proc gplot data=a1;

plot mpg*dc/frame;run;

Page 10: Topic 32: Two-Way Mixed Effects Modelbacraig/notes512/Topic_32.pdf · Two-way mixed model • Two-way mixed model has –One fixed effect –One random effect • Tests: –Again
Page 11: Topic 32: Two-Way Mixed Effects Modelbacraig/notes512/Topic_32.pdf · Two-way mixed model • Two-way mixed model has –One fixed effect –One random effect • Tests: –Again

Find the meansproc means data=a1;

output out=a2 mean=avmpg;var mpg;by driver car;

Page 12: Topic 32: Two-Way Mixed Effects Modelbacraig/notes512/Topic_32.pdf · Two-way mixed model • Two-way mixed model has –One fixed effect –One random effect • Tests: –Again

Plot the meanstitle1 'Plot of the means';symbol1 v='A' i=join c=black;symbol2 v='B' i=join c=black;symbol3 v='C' i=join c=black;symbol4 v='D' i=join c=black;symbol5 v='E' i=join c=black;proc gplot data=a2;

plot avmpg*driver=car/frame;run;

Page 13: Topic 32: Two-Way Mixed Effects Modelbacraig/notes512/Topic_32.pdf · Two-way mixed model • Two-way mixed model has –One fixed effect –One random effect • Tests: –Again
Page 14: Topic 32: Two-Way Mixed Effects Modelbacraig/notes512/Topic_32.pdf · Two-way mixed model • Two-way mixed model has –One fixed effect –One random effect • Tests: –Again

Example Revision• Suppose that the four drivers were not

randomly selected and there is interest in comparing the four drivers in the study

• Driver (A) is now a fixed effect• Still consider Car (B) to be a random

effect

Page 15: Topic 32: Two-Way Mixed Effects Modelbacraig/notes512/Topic_32.pdf · Two-way mixed model • Two-way mixed model has –One fixed effect –One random effect • Tests: –Again

Mixed effects model(unrestricted)

• Yijk = μ + αi + βj + (αβ)ij + εijk

• αi are unknown constants• βj ~ N(0, σβ2)• (αβ)ij ~ N(0, σαβ2)• εij ~ N(0, σ2)• σY

2 = σβ2 + σαβ2 + σ2

• Σαi =0 (common restriction)

Page 16: Topic 32: Two-Way Mixed Effects Modelbacraig/notes512/Topic_32.pdf · Two-way mixed model • Two-way mixed model has –One fixed effect –One random effect • Tests: –Again

Mixed effects model(restricted)

• Yijk = μ + αi + βj + (αβ)ij + εijk

•• εij ~ N(0, σ2)• σY

2 = σβ2 + ((a-1)/a)σαβ2 + σ2

• Common restrictions on “fixed” effects – Σαi =0– Σ(αb)ij =0 for all j

2 2j ij

1~ (0, ) and ( ) ~ (0, )aN Naβ αββ σ αβ σ−

Page 17: Topic 32: Two-Way Mixed Effects Modelbacraig/notes512/Topic_32.pdf · Two-way mixed model • Two-way mixed model has –One fixed effect –One random effect • Tests: –Again

Parameters• There are a+3 parameters in this

model– a fixed effects means– σβ

2

– σαβ2

– σ2

Page 18: Topic 32: Two-Way Mixed Effects Modelbacraig/notes512/Topic_32.pdf · Two-way mixed model • Two-way mixed model has –One fixed effect –One random effect • Tests: –Again

ANOVA table• The terms and layout of the ANOVA

table are the same as what we used for the fixed effects model

• The expected mean squares (EMS) are different and vary based on the choice of unrestricted or restricted mixed model

Page 19: Topic 32: Two-Way Mixed Effects Modelbacraig/notes512/Topic_32.pdf · Two-way mixed model • Two-way mixed model has –One fixed effect –One random effect • Tests: –Again

EMS (unrestricted)• E(MSA) = σ2 + bnΣαi

2 /(a-1)+ nσαβ2

• E(MSB) = σ2 + anσβ2 + nσαβ2

• E(MSAB) = σ2 + nσαβ2

• E(MSE) = σ2

• Estimates of the variance components can be obtained from these equations, replacing E(MS) with table value, or other methods such as ML

Page 20: Topic 32: Two-Way Mixed Effects Modelbacraig/notes512/Topic_32.pdf · Two-way mixed model • Two-way mixed model has –One fixed effect –One random effect • Tests: –Again

EMS (restricted)• E(MSA) = σ2 + bnΣαi

2 /(a-1)+ nσαβ2

• E(MSB) = σ2 + anσβ2

• E(MSAB) = σ2 + nσαβ2

• E(MSE) = σ2

• Estimates of the variance components can be obtained from these equations, replacing E(MS) with table value, or other methods such as ML

Different here

Page 21: Topic 32: Two-Way Mixed Effects Modelbacraig/notes512/Topic_32.pdf · Two-way mixed model • Two-way mixed model has –One fixed effect –One random effect • Tests: –Again

Hypotheses (unrestricted)• H0A: σα2 = 0; H1A: σα2 ≠ 0

– H0A is tested by F = MSA/MSAB with dfa-1 and (a-1)(b-1)

• H0B: σβ2 = 0; H1B : σβ2 ≠ 0– H0B is tested by F = MSB/MSAB with df

b-1 and (a-1)(b-1)• H0AB : σαβ2 = 0; H1AB : σαβ2 ≠ 0

– H0AB is tested by F = MSAB/MSE with df(a-1)(b-1) and ab(n-1)

Page 22: Topic 32: Two-Way Mixed Effects Modelbacraig/notes512/Topic_32.pdf · Two-way mixed model • Two-way mixed model has –One fixed effect –One random effect • Tests: –Again

Hypotheses (restricted)• H0A: σα2 = 0; H1A: σα2 ≠ 0

– H0A is tested by F = MSA/MSAB with dfa-1 and (a-1)(b-1)

• H0B: σβ2 = 0; H1B : σβ2 ≠ 0– H0B is tested by F = MSB/MSE with df

b-1 and ab(n-1)• H0AB : σαβ2 = 0; H1AB : σαβ2 ≠ 0

– H0AB is tested by F = MSAB/MSE with df(a-1)(b-1) and ab(n-1)

Page 23: Topic 32: Two-Way Mixed Effects Modelbacraig/notes512/Topic_32.pdf · Two-way mixed model • Two-way mixed model has –One fixed effect –One random effect • Tests: –Again

Comparison of Means• To compare fixed levels of A, std

error is

• Degrees of freedom for t tests and CIs are then (a-1)(b-1)

• This is true for both unrestricted and restricted mixed models

2MSAB / bn

Page 24: Topic 32: Two-Way Mixed Effects Modelbacraig/notes512/Topic_32.pdf · Two-way mixed model • Two-way mixed model has –One fixed effect –One random effect • Tests: –Again

Using Proc Mixedproc mixed data=a1;

class car driver;model mpg=driver;random car car*driver / vcorr;lsmeans driver / adjust=tukey;

run;SAS considers unrestricted model only…results in slightly different variance estimates. Not crucial if goal is to compare fixed effect means

Page 25: Topic 32: Two-Way Mixed Effects Modelbacraig/notes512/Topic_32.pdf · Two-way mixed model • Two-way mixed model has –One fixed effect –One random effect • Tests: –Again

SAS OutputCovariance Parameter

EstimatesCov Parm Estimatecar 2.9343car*driver 0.01406Residual 0.1757

Type 3 Tests of Fixed Effects

EffectNum

DFDen DF F Value Pr > F

driver 3 12 458.26 <.0001

Page 26: Topic 32: Two-Way Mixed Effects Modelbacraig/notes512/Topic_32.pdf · Two-way mixed model • Two-way mixed model has –One fixed effect –One random effect • Tests: –Again

SAS OutputLeast Squares Means

Effect driver EstimateStandard

Error DF t Value Pr > |t|driver 1 26.9300 0.7793 12 34.56 <.0001driver 2 34.1500 0.7793 12 43.82 <.0001driver 3 28.8500 0.7793 12 37.02 <.0001driver 4 30.2600 0.7793 12 38.83 <.0001

Proc GLM provides wrong SE here

Page 27: Topic 32: Two-Way Mixed Effects Modelbacraig/notes512/Topic_32.pdf · Two-way mixed model • Two-way mixed model has –One fixed effect –One random effect • Tests: –Again

SAS OutputDifferences of Least Squares Means

Effect driver _driver EstiateStandard

Error DF t Value Pr > |t| Adjustment Adj Pdriver 1 2 -7.2200 0.2019 12 -35.76 <.0001 Tukey-

Kramer<.0001

driver 1 3 -1.9200 0.2019 12 -9.51 <.0001 Tukey-Kramer

<.0001

driver 1 4 -3.3300 0.2019 12 -16.49 <.0001 Tukey-Kramer

<.0001

driver 2 3 5.3000 0.2019 12 26.25 <.0001 Tukey-Kramer

<.0001

driver 2 4 3.8900 0.2019 12 19.26 <.0001 Tukey-Kramer

<.0001

driver 3 4 -1.4100 0.2019 12 -6.98 <.0001 Tukey-Kramer

<.0001

Page 28: Topic 32: Two-Way Mixed Effects Modelbacraig/notes512/Topic_32.pdf · Two-way mixed model • Two-way mixed model has –One fixed effect –One random effect • Tests: –Again

Three-way models• We can have zero, one, two, or three

random effects• EMS indicate how to do tests• In some cases the situation is

complicated and we need approximations of an F test– Example: when all are random, use

MS(AB)+MS(AC)-MS(ABC) to test A

Page 29: Topic 32: Two-Way Mixed Effects Modelbacraig/notes512/Topic_32.pdf · Two-way mixed model • Two-way mixed model has –One fixed effect –One random effect • Tests: –Again

Last slide• Finish reading KNNL Chapter 25• We used program topic32.sas to

generate the output for today