Skip to main content

Table 2 Proportion of item-level simulations (ologit) not converging until 1000 valid simulation results were obtained*

From: Multiple imputation for patient reported outcome measures in randomised controlled trials: advantages and disadvantages of imputing at the item, subscale or composite score level

Sample size

Proportion of missing data

OKS

SF-12

EQ-5D-3L

Observed missing data pattern

Unit-nonresponse

70% item missingness

Observed missing data pattern

Observed missing data pattern

100

5

88.2%

89.7%

88.0%

91.9%*

72.1%

10

95.6%*

95.3%*

94.1%*

98.2%*

83.8%

20

99.8%*

99.6%*

99.1%*

99.9%*

94.4%*

40

100%*

100%*

100%*

100%*

99.9%*

200

5

40.1%

22.7%

35.4%

52.1%

19.9%

10

50.8%

25.2%

45.6%

69.6%

25.5%

20

64.5%

32.1%

57.0%

87.2%

45.30%

40

99.7%*

60.0%

78.4%

99.3%*

85.6%

500

5

21.6%

8.84%

20.0%

3.5%

11.2%

10

24.9%

9.3%

23.4%

6.3%

14.2%

20

27.8%

11.0%

27.5%

12.8%

17.3%

40

41.4%

16.0%

28.6%

34.7%

23.6%

Full samplea

5

1.0%

0%

0.7%

0%

0.3%

10

3.9%

0.1%

1.6%

0.2%

1.86%

20

12.5%

0.4%

5.2%

1.1%

6.8%

40

28.0%

3.8%

10.4%

8.4%

17.4%

  1. *For scenarios highlighted *, 1000 valid simulations could not be obtained, and these scenarios are not included in subsequent summaries
  2. aThe full sample includes 1030 observation for the OKS simulation, 797 for the SF-12 and 1160 for the EQ-5D-3L