+ All Categories
Home > Documents > Multiple Comparisons & Factorial...

Multiple Comparisons & Factorial...

Date post: 03-Feb-2018
Category:
Upload: lamquynh
View: 220 times
Download: 0 times
Share this document with a friend
38
Multiple Comparisons & Factorial Designs EdPsych 496 C.J. Anderson Multiple Comparisons & Factorial Designs – p.1/38
Transcript
Page 1: Multiple Comparisons & Factorial Designscourses.education.illinois.edu/EdPsy580/lectures/multiple... · • Tukey HSD. • Simple or Complex ... Multiple Comparisons & Factorial Designs

Multiple Comparisons & Factorial Designs

EdPsych 496

C.J. Anderson

Multiple Comparisons & Factorial Designs – p.1/38

Page 2: Multiple Comparisons & Factorial Designscourses.education.illinois.edu/EdPsy580/lectures/multiple... · • Tukey HSD. • Simple or Complex ... Multiple Comparisons & Factorial Designs

Multiple Comparisons (part 2)

• When there is no interaction.. . . basically whatwe did with 1-Factor ANOVA• Example• Contrasts

• When there is an interaction.• Example• Contrasts

Multiple Comparisons & Factorial Designs – p.2/38

Page 3: Multiple Comparisons & Factorial Designscourses.education.illinois.edu/EdPsy580/lectures/multiple... · • Tukey HSD. • Simple or Complex ... Multiple Comparisons & Factorial Designs

MC Procedures We’ll Cover

• Pairwise (simple):• Fisher LSD.• Tukey HSD.

• Simple or Complex:• Bonferroni• Scheffé

Multiple Comparisons & Factorial Designs – p.3/38

Page 4: Multiple Comparisons & Factorial Designscourses.education.illinois.edu/EdPsy580/lectures/multiple... · • Tukey HSD. • Simple or Complex ... Multiple Comparisons & Factorial Designs

When There is NO Interaction

• Only need to perform multiple comparisonsfor main effects that were significant.

• Do not need to do them if a factor only has 2levels.

• Multiple comparison procedures are the sameas those used in 1-Factor ANOVA, exceptneed to make sure you’re using the correct n.

• To illustrate, need data with at least 1 maineffect and no interaction. . .

Multiple Comparisons & Factorial Designs – p.4/38

Page 5: Multiple Comparisons & Factorial Designscourses.education.illinois.edu/EdPsy580/lectures/multiple... · • Tukey HSD. • Simple or Complex ... Multiple Comparisons & Factorial Designs

NELS88National Education Longitudinal Study:

• Conducted by National Center for Education Statisticsof the US department of Education. These dataconstitute the first in a series of longitudinalmeasurements of students starting in 8th grade. Thesedata were collected Spring 1988.

• We’ll just use data from a sample of N = 225 studentsfrom 23 schools.

• Response variable: math achievement test score.

• Factors: SES (Low, Middle, High) and Location ofschool (rural, urban, sub-urban).

Multiple Comparisons & Factorial Designs – p.5/38

Page 6: Multiple Comparisons & Factorial Designscourses.education.illinois.edu/EdPsy580/lectures/multiple... · • Tukey HSD. • Simple or Complex ... Multiple Comparisons & Factorial Designs

NELS88: Mean Plots

Multiple Comparisons & Factorial Designs – p.6/38

Page 7: Multiple Comparisons & Factorial Designscourses.education.illinois.edu/EdPsy580/lectures/multiple... · • Tukey HSD. • Simple or Complex ... Multiple Comparisons & Factorial Designs

NELS88: Summary StatisticsMeans and (standard deviations). njk = 25

Location

SES Rural Suburban Urban

Low 44.00 44.60 46.48 45.03

(9.36) (6.84) (9.32) (8.54)

Mid 46.52 49.64 51.20 49.45

(8.85) (10.71) (9.17) (9.60)

High 57.00 53.32 57.96 56.10

(9.17) (11.66) (8.86) (10.05)

49.51 49.19 51.88 50.19

(10.57) (10.46) (10.17) (10.42)Multiple Comparisons & Factorial Designs – p.7/38

Page 8: Multiple Comparisons & Factorial Designscourses.education.illinois.edu/EdPsy580/lectures/multiple... · • Tukey HSD. • Simple or Complex ... Multiple Comparisons & Factorial Designs

NELS88: ANOVA Summary Table

Source df SS MS F p > F

SES 2 4653.90 2326.95 26.27 < .001

Urban 2 324.73 162.36 1.83 .16

SES*Urban 4 229.51 57.38 0.65 .63

Error 216 19130.78 88.57

Corrected Total 224 24338.78

• SES is significant.

• Urban and Interaction are not significant.

Multiple Comparisons & Factorial Designs – p.8/38

Page 9: Multiple Comparisons & Factorial Designscourses.education.illinois.edu/EdPsy580/lectures/multiple... · • Tukey HSD. • Simple or Complex ... Multiple Comparisons & Factorial Designs

NELS88: Pairwise

• Since we have balanced design, we’ll useminimum significant differences.

Null Observed Differences

hypothesis between sample means Conclusion

µhigh = µmid 56.10 − 49.45 = 6.64

µmid = µlow 49.45 − 45.03 = 4.42

µhigh = µlow 56.10 − 45.03 = 11.07

Multiple Comparisons & Factorial Designs – p.9/38

Page 10: Multiple Comparisons & Factorial Designscourses.education.illinois.edu/EdPsy580/lectures/multiple... · • Tukey HSD. • Simple or Complex ... Multiple Comparisons & Factorial Designs

NELS88: Fisher’s LSD• Controls type I error rate per comparison, α.

• Least significant difference is

LSD =(1−α/2) tνesy−y

• For α = .05 and our example, .975t216 = 1.971 and

sy−y =

MSe

(

1

nK+

1

nK

)

=

88.56778

(

2

25(3)

)

= 1.54

• So LSD = 1.971(1.54) = 3.03 −→ Reject all Ho’s.Multiple Comparisons & Factorial Designs – p.10/38

Page 11: Multiple Comparisons & Factorial Designscourses.education.illinois.edu/EdPsy580/lectures/multiple... · • Tukey HSD. • Simple or Complex ... Multiple Comparisons & Factorial Designs

NELS88: Tukey’s HSD• Controls familywise type I error rate per, ασ — the

family being pairs of means for SES.

• Honest significant difference is HSD = qνe,ασ ,Jsy−y.

• For α = .05, νe = 216 and J = 3, q = 3.338.

• The standard error

sy−y =

MSe

2

(

1

nK+

1

nK

)

=

88.56778

2

(

2

25(3)

)

= 1.09

• So HSD = 3.338(1.087) = 3.63 −→ Reject all Ho’s.Multiple Comparisons & Factorial Designs – p.11/38

Page 12: Multiple Comparisons & Factorial Designscourses.education.illinois.edu/EdPsy580/lectures/multiple... · • Tukey HSD. • Simple or Complex ... Multiple Comparisons & Factorial Designs

Complex Comparisons• For Complex (and simple) comparisons.• For a few planned comparisons −→

Bonferroni.• For post hoc and/or more than (J − 1) −→

Scheffé.• Consider the following two comparisons for

NELS data:

ψ1 = µlow + µmid − 2µhigh

ψ2 = µhigh + µmid − 2µlow

Multiple Comparisons & Factorial Designs – p.12/38

Page 13: Multiple Comparisons & Factorial Designscourses.education.illinois.edu/EdPsy580/lectures/multiple... · • Tukey HSD. • Simple or Complex ... Multiple Comparisons & Factorial Designs

Complex Comparisons• The sums of squares for comparisons:

SSψ =nK(

∑Jj=1 cjYj·)

2

∑Jj=1 c2

j

=nK(ψ)2

∑Jj=1 c2

j

• For comparison 1:

ψ21 = (45.027 + 49.453 − 2(56.093))2 = 313.502

SSψ1=

25(3)(313.502)

12 + 12 + (−2)2=

25(3)(313.502)

6= 3, 919.07

• For comparison 2:

ψ22 = (−2(45.027) + 49.453 + 56.093)2 = 240.002

SSψ2=

25(3)(240.002)

6= 3, 000.03

Multiple Comparisons & Factorial Designs – p.13/38

Page 14: Multiple Comparisons & Factorial Designscourses.education.illinois.edu/EdPsy580/lectures/multiple... · • Tukey HSD. • Simple or Complex ... Multiple Comparisons & Factorial Designs

Complex Comparisons• Test statistics:

Contrast df SS MS F

(low & middle) vs high 1 3,919.07 3919.07 44.25

low vs (middle & high) 1 3,000.03 3,000.03 33.87

Error 216 19,130.64 88.568

• For Bonferroni, αΣ = .05, (.975/2)t216 = 2.413, and for theF ratio, the critical value is (2.413)2 = 5.82

• For Scheffé, (J − 1)(1−αΣ)F(J−1),νe= 2(3.0376) = 6.08.

• For this NELS example, both comparisons aresignificant.

Multiple Comparisons & Factorial Designs – p.14/38

Page 15: Multiple Comparisons & Factorial Designscourses.education.illinois.edu/EdPsy580/lectures/multiple... · • Tukey HSD. • Simple or Complex ... Multiple Comparisons & Factorial Designs

Complex Comparisons• For the complex comparison, we implicitly

tested (for example)Ho(1) : ψ1 = 0 vs Ha(1) : ψ1 < 0

• For directional alternatives, e.g.,Ha(1) : ψ1 < 0, use t-ratio,

t =ψ

=

j cjYj·√

MSe

(

j c2j/(nK)

)

• The critical value is the square root of theBonferroni or Scheffé (depending on whichyou’re doing).

Multiple Comparisons & Factorial Designs – p.15/38

Page 16: Multiple Comparisons & Factorial Designscourses.education.illinois.edu/EdPsy580/lectures/multiple... · • Tukey HSD. • Simple or Complex ... Multiple Comparisons & Factorial Designs

Complex Comparisons• For our comparison 1 with Ha(1) : ψ1 6= 0

t =45.027 + 49.453 − 2(56.093)

88.568(

625(3)

)

=−17.705

2.662= −6.65

• Note: observed test statistic is

t2 = (−6.65)2 = 44.25.

• The (negative) square root of critical values

given previously.

• Bonferroni: tcrit = −2.41.

• Scheffé: −2.47Multiple Comparisons & Factorial Designs – p.16/38

Page 17: Multiple Comparisons & Factorial Designscourses.education.illinois.edu/EdPsy580/lectures/multiple... · • Tukey HSD. • Simple or Complex ... Multiple Comparisons & Factorial Designs

When There is an Interaction

• The basic idea and methodology isessentially the same as multiple comparisonsfor main effects.

• Types of multiple comparisons:• Simple Effect• Mini-Factorials• Specific Cell Means

• We need another data set — a factorial withan interaction.. . .

Multiple Comparisons & Factorial Designs – p.17/38

Page 18: Multiple Comparisons & Factorial Designscourses.education.illinois.edu/EdPsy580/lectures/multiple... · • Tukey HSD. • Simple or Complex ... Multiple Comparisons & Factorial Designs

When There is an InteractionModified hypothetical example from Tatsuoka & Lohnes(1988, p273-279):10 subjects were randomly assigned to one of twomethods (A and B) of teaching keyboarding to one of thefollowing 3 conditions:

• C1: 2 hours of instruction per day for 6 weeks.

• C2: 3 hours of instruction per day for 4 weeks.

• C3: 4 hours of instruction per day for 3 weeks.

At the end of the course, each of the 60 subjects was given

a standardized test measuring accuracy.

Multiple Comparisons & Factorial Designs – p.18/38

Page 19: Multiple Comparisons & Factorial Designscourses.education.illinois.edu/EdPsy580/lectures/multiple... · • Tukey HSD. • Simple or Complex ... Multiple Comparisons & Factorial Designs

The DataThe “raw” data:

Method

Condition A B

C1 26, 22, 21, 23, 21, 25, 24, 29, 25, 26,

19, 25, 20, 21, 20 28, 25, 24, 24, 26

C2 17, 21, 17, 18, 23, 18, 19, 17, 17, 19,

19, 16, 19, 15, 14 19, 17, 21, 18, 20

C3 14, 14, 16, 16, 13, 11, 10, 10, 12, 11,

13, 12, 15, 14, 14 12, 14, 13, 12, 11

Multiple Comparisons & Factorial Designs – p.19/38

Page 20: Multiple Comparisons & Factorial Designscourses.education.illinois.edu/EdPsy580/lectures/multiple... · • Tukey HSD. • Simple or Complex ... Multiple Comparisons & Factorial Designs

Summary Statistics

Condition

C1 C2 C3 Over Conditions

n11 = 10 n12 = 10 n13 = 10 n1• = 30

Method Y11 = 21.8 Y12 = 17.9 Y13 = 14.1 Y1· = 17.93

A s2

11= 5.07 s2

12= 7.43 s2

13= 1.66 s2

A= 14.62

n21 = 10 n22 = 10 n23 = 10 n2• = 30

Method Y21 = 25.6 Y22 = 18.5 Y23 = 11.6 Y2· = 18.57

B s2

21= 2.93 s2

22= 1.83 s2

23= 1.60 s2

B= 35.77

Over Methods n•1 = 20 n•2 = 20 n•2 = 20 n•• = 60

Y·1 = 23.7 Y·2 = 18.2 Y·3 = 12.85 Y·· = 18.25

s2

C1= 7.59 s2

C2= 4.48 s2

C3= 3.19 s2 = 24.87

Multiple Comparisons & Factorial Designs – p.20/38

Page 21: Multiple Comparisons & Factorial Designscourses.education.illinois.edu/EdPsy580/lectures/multiple... · • Tukey HSD. • Simple or Complex ... Multiple Comparisons & Factorial Designs

Computing Sums of Squares• Total Sum of Squares: (Corrected for total mean)

SStot =J

j=1

K∑

k=1

n∑

i=1

(Yijk − Y··)2 = (nJK − 1)s2

= (60 − 1)(24.87) = 1467.25

• Sum of Squares Within (error):

SSe =J

j=1

K∑

k=1

n∑

i=1

(Yijk − Yjk)2 = (n − 1)

2∑

j=1

3∑

k=1

s2jk

= (10 − 1)(5.07 + 7.43 + 1.66 + 2.93 + 1.83 + 1.60)

= 184.70

Multiple Comparisons & Factorial Designs – p.21/38

Page 22: Multiple Comparisons & Factorial Designscourses.education.illinois.edu/EdPsy580/lectures/multiple... · • Tukey HSD. • Simple or Complex ... Multiple Comparisons & Factorial Designs

Computing Sums of Squares• Sum of Squares Between Methods:

SSM = nK

J∑

j=1

(Yj· − Y··)2

= (10)(3)[(17.93 − 18.25)2 + (18.57 − 18.25)2]

= 6.02

• Sum of Squares Between Conditions:

SSC = nJK

k=1

(Y·k − Y

··)2

= (10)(2)[(23.7 − 18.25)2 + (18.2 − 18.25)2 + (12.85 − 18.25)2]

= 1, 177.3

Multiple Comparisons & Factorial Designs – p.22/38

Page 23: Multiple Comparisons & Factorial Designscourses.education.illinois.edu/EdPsy580/lectures/multiple... · • Tukey HSD. • Simple or Complex ... Multiple Comparisons & Factorial Designs

Computing Sums of Squares• Sum of Squares for Method × Condition Interaction:

SSMC = n

J∑

j=1

K∑

k=1

(Yjk − Yj· − Y·k + Y

··)2

= 10[(21.80 − 17.93 − 23.70 + 18.25)2 + . . .

+(11.60 − 18.57 − 12.85 + 18.25)2]

= 99.23

• or much easier

SSMC = SStot − SSM − SSC − SSe

= 1467.25 − 6.02 − 1177.30 − 184.70 = 99.23

Multiple Comparisons & Factorial Designs – p.23/38

Page 24: Multiple Comparisons & Factorial Designscourses.education.illinois.edu/EdPsy580/lectures/multiple... · • Tukey HSD. • Simple or Complex ... Multiple Comparisons & Factorial Designs

ANOVA Summary Table

Source df SS MS F p–value

Method 1 6.02 6.02 1.76 .1903

Condition 2 1177.30 588.65 172.12 .0001

M × C Interaction 2 99.23 49.62 14.51 .0001

Within (error) 54 184.70 3.420

(corrected) total 59 1467.25

R2 =SSmodel

SStot

= (6.02 + 1177.30 + 99.23)/1467.25 = .8741

Multiple Comparisons & Factorial Designs – p.24/38

Page 25: Multiple Comparisons & Factorial Designscourses.education.illinois.edu/EdPsy580/lectures/multiple... · • Tukey HSD. • Simple or Complex ... Multiple Comparisons & Factorial Designs

Let’s Take a Look. . .

Multiple Comparisons & Factorial Designs – p.25/38

Page 26: Multiple Comparisons & Factorial Designscourses.education.illinois.edu/EdPsy580/lectures/multiple... · • Tukey HSD. • Simple or Complex ... Multiple Comparisons & Factorial Designs

MC of Simple Effects. . . by Example

Our design can be represented as

Practice

Conditions

C1 C2 C3

Teaching A

Method B

Multiple Comparisons & Factorial Designs – p.26/38

Page 27: Multiple Comparisons & Factorial Designscourses.education.illinois.edu/EdPsy580/lectures/multiple... · • Tukey HSD. • Simple or Complex ... Multiple Comparisons & Factorial Designs

MC of Simple Effects (continued)

• To investigate simple effects of practice given teachingmethod A, test contrasts involving means in thesecells:

C1 C2 C3

A

• To investigate simple effects of practice given teachingmethod is B, test contrasts involving means in thesecells:

C1 C2 C3

B

Multiple Comparisons & Factorial Designs – p.27/38

Page 28: Multiple Comparisons & Factorial Designscourses.education.illinois.edu/EdPsy580/lectures/multiple... · • Tukey HSD. • Simple or Complex ... Multiple Comparisons & Factorial Designs

Alternatively (or Additionally)

Investigate the simple effects of teaching method givenconditions are test comparisons of means from thefollowing cells:

C1 C2 C3

A A A

B B B

Ho : µAC1= µBC1

Ho : µAC2= µBC2

Ho : µAC3= µBC3

Multiple Comparisons & Factorial Designs – p.28/38

Page 29: Multiple Comparisons & Factorial Designscourses.education.illinois.edu/EdPsy580/lectures/multiple... · • Tukey HSD. • Simple or Complex ... Multiple Comparisons & Factorial Designs

Mini-Factorials: (2 × 2) Sub-TablesOne possible way to break our 2 × 3 design into mini 2 × 2

factorials is as follows:

C1 C2 C2 C3

A A

B B

Sub-design on the left: Test whether there is an interactionfor teaching method and practice conditions C1 and C2 bytesting the hypothesis

Ho : (µAC1− µBC1

) = (µAC2− µBC2

)

or equivalently,

Ho : (µAC1− µAC2

) = (µBC1− µBC2

)

Recall that if there is no interaction, then “curves” shouldMultiple Comparisons & Factorial Designs – p.29/38

Page 30: Multiple Comparisons & Factorial Designscourses.education.illinois.edu/EdPsy580/lectures/multiple... · • Tukey HSD. • Simple or Complex ... Multiple Comparisons & Factorial Designs

Mini-Factorials: (2 × 2) Sub-TablesGiven that pattern of the cell means. . .

C1 C3

A

B

Ho : (µAC1− µBC1

) = (µAC3− µBC3

)

or equivalently,

Ho : (µAC1− µAC3

) = (µBC1− µBC3

)

Recall that if there is no interaction, then “curves” shouldbe parallel; that is, the differences between pairs of meansshould be the same.

Multiple Comparisons & Factorial Designs – p.30/38

Page 31: Multiple Comparisons & Factorial Designscourses.education.illinois.edu/EdPsy580/lectures/multiple... · • Tukey HSD. • Simple or Complex ... Multiple Comparisons & Factorial Designs

Mini-Factorials: (2 × 2) Sub-TablesWhat is the contrast for

Ho : (µAC1− µBC1

) = (µAC3− µBC3

)?

• Contrast

Ho : ψ = 1µAC1+ 0µAC2

− 1µAC3− 1µBC1

+ 1µBC3= 0

• Estimated value

ψ2 = (21.8 − 14.1 − 25.6 + 11.6)2 = (−6.3)2 = 39.69

• Needed for the sum of squares for the contrast∑

j

k

c2jk/njk = 1/10 + 1/10 + 1/10 + 1/10 = .4

Multiple Comparisons & Factorial Designs – p.31/38

Page 32: Multiple Comparisons & Factorial Designscourses.education.illinois.edu/EdPsy580/lectures/multiple... · • Tukey HSD. • Simple or Complex ... Multiple Comparisons & Factorial Designs

Mini-Factorials: (2 × 2) Sub-TablesTesting (continued)

Ho : (µAC1− µBC1

) = (µAC3− µBC3

)?

• Sum of squares,

SSψ =ψ2

j

k c2jk/njk

=39.69

.4= 99.225

• Test statistic,

F =SSψ/1

MSerror

=99.225

3.420= 29.01

• What should the critical value be?Multiple Comparisons & Factorial Designs – p.32/38

Page 33: Multiple Comparisons & Factorial Designscourses.education.illinois.edu/EdPsy580/lectures/multiple... · • Tukey HSD. • Simple or Complex ... Multiple Comparisons & Factorial Designs

Mini-Factorials: (2 × 2) Sub-TablesTesting (continued)

• The critical value for our test. . .• Post hoc −→ Complex −→ Scheffé.• Scheffé critical value is (νAC)(1−αΣ)FνAC ,νerror

;that is,

(J − 1)(K − 1).95F(J−1)(K−1),N−JK

• Our example,Fcrit = 2(.95F4,54) = 2(2.54) = 5.08

• Since the test statistic F = 29.01, reject Ho.Multiple Comparisons & Factorial Designs – p.33/38

Page 34: Multiple Comparisons & Factorial Designscourses.education.illinois.edu/EdPsy580/lectures/multiple... · • Tukey HSD. • Simple or Complex ... Multiple Comparisons & Factorial Designs

Testing Specific Means

• Contrasts and comparisons between any setof means that makes sense can beperformed.

• In our example, from the interaction plot itlooks like• µAC2

= µBC2

• µBC1> µAC1

• µAC3> µBC3

Multiple Comparisons & Factorial Designs – p.34/38

Page 35: Multiple Comparisons & Factorial Designscourses.education.illinois.edu/EdPsy580/lectures/multiple... · • Tukey HSD. • Simple or Complex ... Multiple Comparisons & Factorial Designs

Testing Specific Means• Our example, our null and alternative

hypotheses would be• Ho(1) : µAC2

= µBC2versus Ha(1) : µAC2

6= µBC2

• Ho(2) : µBC1= µAC1

versus Ha(2) : µBC1> µAC1

• Ho(3) : µAC3= µBC3

versus Ha(3) : µAC3> µBC3

• We need t-ratios for Ho(2) and Ho(3) to test

directional hypotheses.

• For simplicity & a bit conservative, I’ll take the

alternatives to be non-directional.• Post hoc −→ Pairwise −→ Tukey HSD.

Multiple Comparisons & Factorial Designs – p.35/38

Page 36: Multiple Comparisons & Factorial Designscourses.education.illinois.edu/EdPsy580/lectures/multiple... · • Tukey HSD. • Simple or Complex ... Multiple Comparisons & Factorial Designs

Testing Specific MeansResults for Tukey’s Studentized Range (HSD) Test for Y,αΣ = .05, νerror = 54, MSerror = 3.42, qcrit = 4.178 andMinimum significant difference = 2.44Means with the same letter are not significantly different.

Tukey Grouping Mean N group

A 25.6000 10 2 (B,C1)

B 21.8000 10 1 (A,C1)

C 18.5000 10 4 (B,C2)

C

C 17.9000 10 3 (A,C2)

D 14.1000 10 5 (A,C3)

E 11.6000 10 6 (B,C3)

Multiple Comparisons & Factorial Designs – p.36/38

Page 37: Multiple Comparisons & Factorial Designscourses.education.illinois.edu/EdPsy580/lectures/multiple... · • Tukey HSD. • Simple or Complex ... Multiple Comparisons & Factorial Designs

Testing Specific Means• For Tukey, the HSD for a main effect

(marginal means), say factor A, use

HSDA = qα,J,N−JK

MSerror

nK

• For Tukey, the HSD for interaction(cells means) use

HSDAB = qα,JK,N−JK

MSerror

n

Multiple Comparisons & Factorial Designs – p.37/38

Page 38: Multiple Comparisons & Factorial Designscourses.education.illinois.edu/EdPsy580/lectures/multiple... · • Tukey HSD. • Simple or Complex ... Multiple Comparisons & Factorial Designs

SAS & Testing Specific Means• SAS will compute pairwise tests for main effects by the

command

MEANS factorA / tukey;

• This doesn’t work for for cell means with factorialdesigns.

• Solutions:

• Re-run ANOVA but treat cells as a 1-Factor ANOVAand use the MEANS command.

• Run as factorial, but specify pairwise contrasts forall interaction effects using the CONSTRAST command.You need to then find the appropriate critical value.

Multiple Comparisons & Factorial Designs – p.38/38


Recommended