Sie sind auf Seite 1von 24

Word version of Cluster Analysis Output

The material in this document is organised in the following sections.


Section 1 Cluster analysis results
Section 2 Cluster Profiles
Section 2a Profiles using target variables
Section2b Profile using nominal use of budget variable
Section2c Profiles using total scale scores

Section 1 Cluster Analysis Results


Table 1 Initial Cluster Centres
Factor

Cluster
1

Ease of use
Social
influence

-2.45897

1.32641

-1.75815

-1.81544

1.98563

.31065

1.58265

.95027

-2.98908

Benefits of
online grocery
shopping

This is just the starting point for the iterative procedure. It describes the three cluster centres in terms of scores on the three
factors.
It is not essential for the analysis

Table 2 Iteration History


Change in Cluster Centers
Iteration

2.242

1.934

2.199

.312

.274

.221

.170

.184

.079

.072

.073

.045

.070

.023

.045

.098

.011

.082

.165

.045

.091

.118

.018

.089

.101

.037

.054

10

.062

.035

.014

This is a summary of the iterative procedure. It is not essential for the analysis.

Table 3 Final Cluster Centres


Factor

Cluster
1

Ease of use
Social
influence

.40421

.10858

-.55197

-1.01152

.88842

-.33981

.60845

.31152

-1.05053

Benefits of
online grocery
shopping

This is essential. It defines the final solution in terms of the location of cluster centres in terms of scores on the three
factors. It is essential for the profile analysis. See later

Table 4 Results of ANOVA Test


Cluster
Factor
Ease of use
Social
influence

Mean Square

Error
df

Mean Square

Sig.

df

21.848

.866

311

25.230

.000

103.551

.341

311

304.105

.000

73.054

.537

311

136.135

.000

Benefits of
online grocery
shopping

This is essential.
The ANOVA tests that there are significant differences between final cluster centres on each of the target variables (factors).
The ANOVA is a test based on the following hypotheses:
H0: The cluster centre values (average factor scores) are equal
H1: The cluster centre values (average factor scores) are not equal
For a specific solution we would expect significant differences between cluster centres on all the target variables (factors)

Table 5 Number of Cases in Each Cluster


Cluster

Number

Percent

88.000

28.0

135.000

43.0

91.000

29.0

Valid

314.000

100.0

Missing

19.000

Cluster

This is essential. This information tells us about cluster membership. The SPSS output only gives the number of online
grocery shoppers in each cluster. It is more useful for inference to express cluster membership in terms of percentage
composition. The Percent column has been added

Section 2 Cluster Profiles


The third element of cluster analysis is to establish the characteristics or profile of each cluster. In this analysis the profiles
are established from
Target variables:
Based upon the average scores for each cluster on the three factors from Table3 Final Cluster Centres
Behavioural variables:
Based upon a nominal measure concerned with use of a budget
Scale variables
Based upon total scores for:
Attitudes to online grocery shopping
Shopping enjoyment in a store
Satisfaction with online grocery shopping
Repeat purchase intentions

Section 2a Profiles using target variables


The profile is based upon the average factor scores for each cluster,
Remember that you need to translate the factor scores with reference to the original data that is based upon a seven-point
score (1 = Not at all, 7 = Completely)
Table 3 Final Cluster Centres
Factor

Cluster
1

Ease of use
Social
influence

.40421

.10858

-.55197

-1.01152

.88842

-.33981

.60845

.31152

-1.05053

Benefits of
online grocery
shopping

Section 2b Profile using nominal measure of use of a budget

Statistical Test: Chi-square contingency test


H0: Cluster identity and profile variable are independent
H1: Cluster identity and profile variable are associated

Sequence of test
Use significance level of 5% (.050)
Check Table 6b Chi-square Statistics table
Check the Pearson Chi-square results in
Decide whether to accept or reject H0 at the 5 percent significance level
If H0 is rejected interpret the association between cluster and profile variable from Table 6a Crosstab for Cluster Identity
and Use of Budget
Interpretation should use percentages and not numerical counts.

Note that these tests produce two tables for each test. The information is an essential part of the analysis but we need
to find a more efficient way of summarising the information to present it in the main text.
9

Table 6a Crosstab for Cluster Identity and Use of Budget


Use of budget
Yes
Cluster identity

Cluster 1

Count
% within Cluster identity

Cluster 2

Cluster 3

Total

32

88

63.6%

36.4%

100.0%

76

59

135

56.3%

43.7%

100.0%

39

52

91

42.9%

57.1%

100.0%

171

143

314

54.5%

45.5%

100.0%

Count
% within Cluster identity
Count
% within Cluster identity

Total

56

Count
% within Cluster identity

No

Table 6b Chi-Square Statistics


Asymp. Sig. (2Value

df

sided)

Pearson Chi-Square

8.111a

.017

Likelihood Ratio

8.139

.017

Linear-by-Linear Association

7.798

.005

N of Valid Cases

314

10

Section 2c Profiles using total scale scores

Statistical Test: One-way ANOVA


H0: Average scores are equal for the three clusters
H1: Average scores are not equal for the three clusters

Note that there are quite a few tables to work through here. Be patient, read the notes carefully and work your way through
them.
Table 12 shows you how to summarise the whole lot in a single table.

Sequence:
Identify mean scores for each group and total score from Table 7 Descriptive Statistics
Interpret the Levene test for homogeneity of variances from Table 8
Use a significance level of 5% (.050)
Interpret the ANOVA (Table 9) or Browne Forsythe test (Table 10) for the results of the test for mean scores
If the null hypothesis is rejected, interpret the test for multiple comparisons from Tables 11a -11d to identify why scores are
not equal
11

Table 7 Descriptive Statistics


95% Confidence Interval for Mean
N
Total attitude score

Mean

Std. Deviation

Std. Error

Lower Bound

Upper Bound

Minimum

Maximum

Cluster 1

88

29.1705

7.42196

.79118

27.5979

30.7430

10.00

42.00

Cluster 2

135

31.8148

5.05815

.43534

30.9538

32.6758

14.00

42.00

Cluster 3

90

23.1222

5.76862

.60807

21.9140

24.3304

8.00

36.00

313

28.5719

7.00426

.39590

27.7929

29.3509

8.00

42.00

Total shopping enjoyment score Cluster 1

87

19.2414

6.58392

.70587

17.8382

20.6446

5.00

35.00

Cluster 2

133

22.8722

5.54105

.48047

21.9218

23.8226

5.00

32.00

Cluster 3

90

19.9556

5.35883

.56487

18.8332

21.0779

6.00

33.00

310

21.0065

6.01239

.34148

20.3345

21.6784

5.00

35.00

Cluster 1

87

31.4023

6.51199

.69816

30.0144

32.7902

11.00

42.00

Cluster 2

133

32.2406

5.73420

.49722

31.2571

33.2241

6.00

42.00

Cluster 3

90

25.8889

5.13294

.54106

24.8138

26.9640

7.00

40.00

310

30.1613

6.40690

.36389

29.4453

30.8773

6.00

42.00

Cluster 1

87

13.2184

4.27656

.45850

12.3069

14.1298

3.00

21.00

Cluster 2

133

15.1278

3.59590

.31180

14.5110

15.7446

3.00

21.00

Cluster 3

90

11.0778

3.65423

.38519

10.3124

11.8431

3.00

19.00

310

13.4161

4.16229

.23640

12.9510

13.8813

3.00

21.00

Total

Total
Total satisfaction score

Total
Total repeat purchase score

Total

There is a lot of information here. All you need is the average score for each scale for each cluster in the column headed
Mean. Note that the score for the Total row is simply the average score for all groups

12

Table 8 Levene Test for Homogeneity of Variances


Measure
Total attitude score

Levene Statistic

df1

df2

Sig.

10.090

310

.000

3.323

307

.037

Total satisfaction score

2.498

307

.084

Total repeat purchase score

2.888

307

.057

Total shopping enjoyment


score

Nullhypothesisisthatthetruevariancesareequalbetweengroups
Alternativehypothesisisthattruevariancesarenotequalbetweengroups
Assumeasignificancelevelof5%(.050)
IfSignificancestatistic(Sig)inthetableislessthanthesignificancelevelrejectH0
IfSignificancestatistic(Sig)inthetableisgreaterthanthesignificancelevelacceptH0

13

Table 9 ANOVA Results


Sum of Squares
Total attitude score

Between Groups

df

Mean Square

4124.163

2062.082

Within Groups

11182.469

310

36.072

Total

15306.633

312

833.407

416.703
33.670

Total shopping enjoyment

Between Groups

score

Within Groups

10336.580

307

Total

11169.987

309

2351.826

1175.913

Within Groups

10332.109

307

33.655

Total

12683.935

309

885.186

442.593
14.554

Total satisfaction score

Between Groups

Total repeat purchase

Between Groups

score

Within Groups

4468.133

307

Total

5353.319

309

Sig.

57.165

.000

12.376

.000

34.940

.000

30.410

.000

Notethatthistestcanonlybeusedifthetestforthehomogeneityofvarianceindicatesthatgroupvariancesare
equal(SeeTable8TestforHomogeneityofVariances).
FromTable8weseethatthisonlyappliesinthecaseofthescoresTotalsatisfactionscoreandTotalrepeatpurchase
score.
InthecaseofthescoresTotalattitudescoreandTotalshoppingenjoymentscorewehavetousetheBrowne
Forsytherobusttestfortheequalityofmeans.SeeTable10.
14

Nullhypothesisisthatthetruemeanscoresofthesetofdependentvariablesareequalbetweengroups
Alternativehypothesisisthattruemeanscoresofthesetofdependentvariablesarenotequalbetweengroups
IfSignificancestatistic(Sig)inthetableislessthanthesignificancelevelrejectH0
IfSignificancestatistic(Sig)inthetableisgreaterthanthesignificancelevelacceptH0

15

Table 10 Browne-Forsythe Robust Test for the Equality of Means


Statistica
Total attitude score

Brown-Forsythe

Total shopping enjoyment

Brown-Forsythe

score

df1

df2

Sig.

52.971

233.872

.000

12.062

260.846

.000

Total satisfaction score

Brown-Forsythe

34.597

265.280

.000

Total repeat purchase score

Brown-Forsythe

29.491

262.312

.000

ThistestisusedinsteadoftheANOVAtestiftheLevenetest(Table8)indicatesthatthegroupvariancesarenotthe
same.ThisappliestothescoresTotalattitudescoreandTotalshoppingenjoymentscores.
Nullhypothesisisthatthetruemeanscoresofthesetofdependentvariablesareequalbetweengroups
Alternativehypothesisisthattruemeanscoresofthesetofdependentvariablesarenotequalbetweengroups
IfSignificancestatistic(Sig)inthetableislessthanthesignificancelevelrejectsH0
IfSignificancestatistic(Sig)inthetableisgreaterthanthesignificancelevelacceptH0

16

Post Hoc Tests


NotethatANOVAisanomnibustest.Itdetermineswhetheratleasttwomeanscoresinthesetofdependentvariablesare
significantlydifferentbutdoesnotidentifywhichonestheyare.
ThePostHocTestallowsustoidentifywhichgroupmeansdiffer.
Thetestsareconductedforthesignificanceofdifferencesbetweeneachpairofgroupsforthethreeclusters:Cluster1
(C1),Cluster2(C2)andCluster3(C3).Thecomparisonsare:
C1withC2
C1withC3
C2withC3
ThetestresultsthatisuseddependsontheresultsofLevenetestforequalityofvariances(SeeTable8)
UseBonferroniresultsifvariancesareequal
UseGamesHowellifvariancesarenotequal

17

Thehypothesesare
Nullhypothesisisthatthetruemeanscoresareequalbetweenthetwogroups
Alternativehypothesisisthattruemeanscoresarenotequalbetweengroups
Look at the significancestatistic(Sig)inthetable.
Adoptasignificancelevelof5%(.050)
IfSignificancestatistic(Sig)islessthanthesignificancelevelrejectH0
IfSignificancestatistic(Sig)inthetableisgreaterthanthesignificancelevelacceptH0
NotethatthematerialinthissectionhasbeenpresentedinadifferentwaycomparedtotheSPSSOutput.
Ihavepresentedseparatetablesforeachscore.TheSPSSoutputpresentsasingletable.Itsjustthattheoriginal
tableistoobigtofitonasinglepage.

18

Post Hoc Test for Total Attitude Score

Table 11a Multiple Comparisons for Total Attitude Score


95% Confidence Interval

Mean Difference

Bonferroni

(I) Cluster identity

(J) Cluster identity

Cluster 1

Cluster 2

-2.64436*

.82287

.004

-4.6250

-.6637

Cluster 3

6.04823*

.90040

.000

3.8809

8.2155

Cluster 1

2.64436*

.82287

.004

.6637

4.6250

Cluster 3

8.69259*

.81732

.000

6.7253

10.6599

Cluster 1

-6.04823*

.90040

.000

-8.2155

-3.8809

Cluster 2

-8.69259*

.81732

.000

-10.6599

-6.7253

Cluster 2

-2.64436*

.90304

.011

-4.7837

-.5050

Cluster 3

6.04823

.99786

.000

3.6881

8.4083

Cluster 1

2.64436*

.90304

.011

.5050

4.7837

Cluster 3

8.69259

.74784

.000

6.9247

10.4605

Cluster 1

-6.04823*

.99786

.000

-8.4083

-3.6881

Cluster 2

.74784

.000

-10.4605

-6.9247

Cluster 2

Cluster 3

Games-Howell

Cluster 1

Cluster 2

Cluster 3

(I-J)

Std. Error

-8.69259

See results from Table 8: H0 is rejected


Covariances are not equal
Use Games-Howell Results
19

Sig.

Lower Bound

Upper Bound

Post Hoc Test for Shopping Enjoyment Score

Table 11b Multiple Comparisons for Total Shopping Enjoyment Score


95% Confidence Interval

Mean Difference

Bonferroni

(I) Cluster identity

(J) Cluster identity

Cluster 1

Cluster 2

-3.63080*

.80010

.000

-5.5568

-1.7048

Cluster 3

-.71418

.87242

1.000

-2.8142

1.3859

Cluster 1

3.63080*

.80010

.000

1.7048

5.5568

Cluster 3

2.91662

.79200

.001

1.0101

4.8231

Cluster 1

.71418

.87242

1.000

-1.3859

2.8142

Cluster 2

-2.91662

.79200

.001

-4.8231

-1.0101

Cluster 2

-3.63080*

.85388

.000

-5.6507

-1.6109

Cluster 3

-.71418

.90406

.710

-2.8523

1.4239

Cluster 1

3.63080*

.85388

.000

1.6109

5.6507

Cluster 3

2.91662

.74157

.000

1.1652

4.6680

Cluster 1

.71418

.90406

.710

-1.4239

2.8523

Cluster 2

.74157

.000

-4.6680

-1.1652

Cluster 2

Cluster 3

Games-Howell

Cluster 1

Cluster 2

Cluster 3

(I-J)

Std. Error

-2.91662

See results from Table 8: H0 is rejected


Covariances are not equal
Use Games-Howell Results
20

Sig.

Lower Bound

Upper Bound

Post Hoc Test for Satisfaction Score

Table 11c Multiple Comparisons for Total Satisfaction Score


95% Confidence Interval

Mean Difference

Bonferroni

(I) Cluster identity

(J) Cluster identity

Cluster 1

Cluster 2

-.83830

.79993

.886

-2.7639

1.0873

Cluster 3

5.51341*

.87223

.000

3.4138

7.6130

Cluster 1

.83830

.79993

.886

-1.0873

2.7639

Cluster 3

6.35171*

.79183

.000

4.4456

8.2578

Cluster 1

-5.51341*

.87223

.000

-7.6130

-3.4138

Cluster 2

-6.35171*

.79183

.000

-8.2578

-4.4456

Cluster 2

-.83830

.85712

.592

-2.8652

1.1886

Cluster 3

5.51341

.88327

.000

3.4242

7.6026

Cluster 1

.83830

.85712

.592

-1.1886

2.8652

Cluster 3

6.35171

.73483

.000

4.6169

8.0866

Cluster 1

-5.51341*

.88327

.000

-7.6026

-3.4242

Cluster 2

.73483

.000

-8.0866

-4.6169

Cluster 2

Cluster 3

Games-Howell

Cluster 1

Cluster 2

Cluster 3

(I-J)

Std. Error

-6.35171

See results from Table 8: H0 is accepted


Covariances are equal
Use Bonferroni Results
21

Sig.

Lower Bound

Upper Bound

Post Hoc Tests for Repeat Purchase Score

Table 11d Multiple Comparisons for Repeat Purchase Score


95% Confidence Interval

Mean Difference

Bonferroni

(I) Cluster identity

(J) Cluster identity

Cluster 1

Cluster 2

-1.90943*

.52604

.001

-3.1757

-.6432

Cluster 3

2.14061*

.57359

.001

.7599

3.5213

Cluster 1

1.90943*

.52604

.001

.6432

3.1757

Cluster 3

4.05004*

.52071

.000

2.7966

5.3035

Cluster 1

-2.14061*

.57359

.001

-3.5213

-.7599

Cluster 2

-4.05004*

.52071

.000

-5.3035

-2.7966

Cluster 2

-1.90943*

.55447

.002

-3.2211

-.5978

Cluster 3

2.14061

.59882

.001

.7247

3.5566

Cluster 1

1.90943*

.55447

.002

.5978

3.2211

Cluster 3

4.05004

.49557

.000

2.8793

5.2207

Cluster 1

-2.14061*

.59882

.001

-3.5566

-.7247

Cluster 2

.49557

.000

-5.2207

-2.8793

Cluster 2

Cluster 3

Games-Howell

Cluster 1

Cluster 2

Cluster 3

(I-J)

Std. Error

-4.05004

See results from Table 8: H0 is rejected


Covariances are equal
Use Bonferroni Results
22

Sig.

Lower Bound

Upper Bound

Summary Table for One-Way ANOVA Results


This is to demonstrate an efficient way of summarising the tests in a single table.
Read table notes 1 and 2 first

Table 12. Summary Table for One-Way ANOVA Results

Total attitude

Cluster1
29.170ab

Mean Scores1
Cluster 2 Cluster 3
31.815ac 23.122bc

Total
28.572 F(2,233.9)= 52.971, sig = .000

Total shopping enjoyment

19.241a

22.872ab

19.956b

21.006 F(2,260.8)= 12.062, sig = .000

Total satisfaction

31.402a

32.241b

25.889ab

30.161 F(2.307) = 34.940, sig = .000

Total repeat purchase

13.218ab

15.128ac

11.078bc

13.416 F(2,307) = 30.410, sig = .000

Total score

Test result2

Notes
1. In comparing any pair of means, if each mean has the same letter as a superscript, the difference is statistically
significant. If not, the difference is not statistically significant.
2. Test results for Total attitude and Total shopping enjoyment are based on the Browne-Forsythe Robust Test. The test
results for Total satisfaction and Total repeat purchase are based upon the ANOVA test.
23

Writing about ANOVA in the text


ReporttheresultsoftheANOVAtestFstatistics.
F(df1,df2)=Fvalue,p=sigvalue
ReportandinterprettheresultsoftheposthoctestsfortheequalityofmeansforgrouppairsfromthetableMultiple
Comparisonsincludinggroupmeans.

ExampleforTotalAttitude
Onewayanalysisofvarianceresultedintherejectionofthenullhypothesesthatthegroupmeanswereequal(F(2, 233.9)=
52.971, sig = .000). The Games-Howell post hoc test reveal significant differences in the means of all three group
comparisons. Cluster 1(29.170)haslesspositiveattitudestoonlinegroceryshoppingthanCluster2(31.815)andmore
positiveattitudesthanCluster3(23.122).Cluster2hasmorepositiveattitudestoonlinegroceryshoppingthanCluster3.In
absolutetermsCluster2hasthemostpositiveattitudewhileCluster3hastheleastpositiveattitude.

24

Das könnte Ihnen auch gefallen