SPSS output for exemplar 6

Black output is produced by SPSS. Other colours are comments.

For comments on how to run code see the html version of SPSS source code

Links in this page

Summary % missing
Patterns of missing data
Means and variance covariance estimates after missing data replaced with estimate from EM algorithm
Graph of imputed data
Exploratory analysis of imputed data
Logistic regression for imputed data
Same thing excluding missing data

MVA

Summary of proportions missing for each variable.

back to top uparrow

N

Mean

Std. Deviation

Missing

No. of Extremes(a)

Count

Percent

Low

High

SASMKY80

4031

.26

.680

297

6.9

0

183

SBSMKY80

4125

.47

1.046

203

4.7

0

346

SCSMKY80

4206

.93

1.413

122

2.8

0

551

SDSMKY80

4114

1.10

1.525

214

4.9

0

0

SESMKY80

3845

1.20

1.596

483

11.2

0

0

SFSMKY80

3487

1.31

1.632

841

19.4

0

0

RAALCY80

4010

.65

.869

318

7.3

0

161

RBALCY80

4130

1.13

1.296

198

4.6

0

274

RCALCY80

4230

2.00

1.368

98

2.3

0

0

RDALCY80

4061

2.32

1.369

267

6.2

0

0

REALCY80

3814

2.91

1.519

514

11.9

0

0

RFALCY80

3502

2.37

1.298

826

19.1

0

39

GENDER

4328

1.49

.500

0

.0

0

0

ETHGP

4328

1.95

.228

0

.0

237

0

SAAGEGP

4328

2.46

.794

0

.0

0

0

HZRESPRE

4328

1.81

.391

0

.0

815

0

SZINDEP

4328

.56

.496

0

.0

0

0

SZABS04

4328

1.46

.499

0

.0

0

0

HZEVREFI

4328

1.89

.311

0

.0

468

0

YZLEAVE

4328

1.71

.456

0

.0

0

0

sector

4328

1.17

.480

0

.0

0

89

dprev1

3899

.73

.444

429

9.9

0

0

dvar1

3899

2.47

2.577

429

9.9

0

217

dprev2

3992

.73

.446

336

7.8

0

0

dvar2

3992

2.84

2.982

336

7.8

0

226

dprev3

4077

.78

.414

251

5.8

0

0

dvar3

4077

3.43

3.343

251

5.8

0

168

dprev4

3997

.74

.437

331

7.6

0

0

dvar4

3997

3.08

3.253

331

7.6

0

227

dprev5

3752

.71

.454

576

13.3

0

0

dvar5

3752

2.21

2.545

576

13.3

0

189

dprev6

3394

.59

.492

934

21.6

0

0

dvar6

3394

1.36

1.901

934

21.6

0

141

drgcode1

4056

.12

.509

272

6.3

0

175

drgcode2

4152

.14

.554

176

4.1

0

160

drgcode3

4218

.39

.878

110

2.5

0

348

drgcode4

4121

.54

.972

207

4.8

0

470

drgcode5

3847

.39

.658

481

11.1

0

129

drgcode6

3510

.64

.985

818

18.9

0

430

a Number of cases outside the range (Mean - 2*SD, Mean + 2*SD).

 

Now patterns of missing data for variables in combination.

back to top uparrow

Tabulated Patterns

 

Number of Cases

Missing Patterns(a)

Complete if ...(b)

dprev3

dvar3

columns omitted here to make table fit

dvar2

drgcode1

dprev6

dvar6

2242

2242

95

X

X

2337

84

2413

57

2299

95

2337

72

X

2460

66

2308

85

X

X

2327

282

X

X

2656

133

X

X

2997

86

X

2328

57

X

X

2678

Patterns with less than 1% cases (43 or fewer) are not displayed.

a Variables are sorted on missing patterns.

b Number of complete cases if variables missing in that pattern (marked with X) are not used.

Now summary statistics once missing data imputed with the EM algorithm.

back to top uparrow

Summary of estimated means

 

SA-SMKY80

SB-SMKY80

SC-SMKY80

SD-SMKY80

SE-SMKY80

SF-SMKY80

those are the smoking variables

further variables in more columns

All Values

.26

.47

.93

1.10

1.20

1.31

EM

.27

.49

.95

1.16

1.31

1.45

And a small piece of the large correlation table

back to top uparrow

SASMKY80

SBSMKY80

SCSMKY80

SDSMKY80

SESMKY80

SFSMKY80

SASMKY80

1

SCSMKY80

.435

.601

1

SDSMKY80

.384

.530

.744

1

SFSMKY80

.327

.434

.588

.698

.811

1
a Little's MCAR test: Chi-Square = 16185.635, DF = 10910, Sig. = .000

The test above is very strong indication that the data are not missing at random, ie the missingness is related to one or more of the observed variables.

Next analyses are run on the output data set created by the EM option of the MVA procedure.

back to top uparrow

Histogram of sweep 6 volume

Graph shows imputed non-integer values

back to top uparrow

This overcome by rounding data to nearest integer

But we get nconsistent results after rounding seen for group where offending prevalence is zero

back to top uparrow


These should all be 0/0 is prevalence is zero

variety of offending sweep 6

Total

0

1

2

volume of offending sweep 6

0

1464

0

0

1464

1

29

16

0

45

2

19

25

0

44

3

2

19

0

21

4

0

11

0

11

5

0

4

0

4

6

0

6

0

6

7

0

4

1

5

8

0

0

3

3

9

0

1

1

2

10

0

0

2

2

20

0

1

0

1

Total

1514

87

7

1608

 

Now part of the logistic regression output complete imputed cases

ack to top uparrow

Categorical Variables Codings

 

Frequency

Parameter coding

(1)

(2)

(3)

Drug use summary code at sweep 6

none

2366

.000

.000

.000

only cannabis

1363

1.000

.000

.000

only glue

155

.000

1.000

.000

other drug or comb

444

.000

.000

1.000

Drug use summary code at sweep 1

none

4071

.000

.000

.000

only cannabis

82

1.000

.000

.000

only glue

106

.000

1.000

.000

other drug or comb

69

.000

.000

1.000

Individual deprivation socio-economi

Manual/high deprivation

1899

1.000

Non-manual/low deprivation

2429

.000

Wh non white

non-white

237

1.000

white

4091

.000

Gender

Male

2189

1.000

Female

2139

.000

Variables in the Equation

 

B

S.E.

Wald

df

Sig.

Exp(B)

Step 1(a)

drgcode6

237.670

3

.000

drgcode6(1)

1.270

.131

94.659

1

.000

3.562

drgcode6(2)

3.024

.207

212.726

1

.000

20.564

drgcode6(3)

1.496

.173

75.127

1

.000

4.466

drgcode1

37.450

3

.000

drgcode1(1)

1.473

.258

32.710

1

.000

4.363

drgcode1(2)

.055

.325

.029

1

.865

1.057

drgcode1(3)

.754

.301

6.290

1

.012

2.125

GENDER(1)

.983

.117

70.756

1

.000

2.671

ETHGP(1)

-.792

.340

5.423

1

.020

.453

SZINDEP(1)

1.486

.118

157.356

1

.000

4.419

Constant

-4.523

.162

775.552

1

.000

.011

a Variable(s) entered on step 1: drgcode6, drgcode1, GENDER, ETHGP, SZINDEP.

 

Now some tables to check out model

ack to top uparrow

 

Drug use summary code at sweep 6 * anyoff Crosstabulation

 

anyoff

Total

.00

1.00

Drug use summary code at sweep 6

none

Count

2262

104

2366

% within Drug use summary code at sweep 6

95.6%

4.4%

100.0%

only cannabis

Count

1155

208

1363

% within Drug use summary code at sweep 6

84.7%

15.3%

100.0%

only glue

Count

73

82

155

% within Drug use summary code at sweep 6

47.1%

52.9%

100.0%

other drug or comb

Count

370

74

444

% within Drug use summary code at sweep 6

83.3%

16.7%

100.0%

Total

Count

3860

468

4328

% within Drug use summary code at sweep 6

89.2%

10.8%

100.0%

 

Drug use summary code at sweep 1 * anyoff Crosstabulation

 

anyoff

Total

.000000

1.000

Drug use summary code at sweep 1

none

Count

3679

392

4071

% within Drug use summary code at sweep 1

90.4%

9.6%

100.0%

only cannabis

Count

43

39

82

% within Drug use summary code at sweep 1

52.4%

47.6%

100.0%

only glue

Count

93

13

106

% within Drug use summary code at sweep 1

87.7%

12.3%

100.0%

other drug or comb

Count

45

24

69

% within Drug use summary code at sweep 1

65.2%

34.8%

100.0%

Total

Count

3860

468

4328

% within Drug use summary code at sweep 1

89.2%

10.8%

100.0%

 
Now part of the logistic regression output listwise deletion of missing datacases

ack to top uparrow

 

Case Processing Summary

 

Unweighted Cases(a)

N

Percent

Selected Cases

Included in Analysis

3314

76.6

Missing Cases

1014

23.4

Total

4328

100.0

Unselected Cases

0

.0

Total

4328

100.0

a If weight is in effect, see classification table for the total number of cases.

Variables in the Equation
Notice the extreme coefficient for the second drug use dummy at sweep 6. We can see in the table below that there were only 5 people not missing at sweep 6 who answered the glue question positve and none of them had offeneded. Hence the extreme negative coefficient, very different from the one for imputed data.

B

S.E.

Wald

df

Sig.

Exp(B)

Step 1(a)

drgcode6

54.607

3

.000

drgcode6(1)

.608

.177

11.831

1

.001

1.837

drgcode6(2)

-17.884

17665.705

.000

1

.999

.000

drgcode6(3)

1.417

.192

54.420

1

.000

4.126

drgcode1

25.443

3

.000

drgcode1(1)

1.829

.386

22.410

1

.000

6.229

drgcode1(2)

.127

.420

.091

1

.762

1.135

drgcode1(3)

.830

.414

4.029

1

.045

2.294

GENDER(1)

.885

.157

31.587

1

.000

2.423

ETHGP(1)

-.882

.527

2.806

1

.094

.414

SZINDEP(1)

1.635

.167

95.779

1

.000

5.132

Constant

-4.610

.204

510.533

1

.000

.010

a Variable(s) entered on step 1: drgcode6, drgcode1, GENDER, ETHGP, SZINDEP.

Drug use summary code at sweep 6 * anyoff Crosstabulation

 

anyoff

Total

.000000

1.000

Drug use summary code at sweep 6

none

Count

2054

89

2143

% within Drug use summary code at sweep 6

95.8%

4.2%

100.0%

only cannabis

Count

859

73

932

% within Drug use summary code at sweep 6

92.2%

7.8%

100.0%

only glue

Count

5

0

5

% within Drug use summary code at sweep 6

100.0%

.0%

100.0%

other drug or comb

Count

360

70

430

% within Drug use summary code at sweep 6

83.7%

16.3%

100.0%

Total

Count

3278

232

3510

% within Drug use summary code at sweep 6

93.4%

6.6%

100.0%