Skip to main content

Table 2 Percentages selection frequency of variables after backward selection in Multiply Imputed datasets using four different pooling methods and in the complete dataset

From: A simple pooling method for variable selection in multiply imputed datasets outperformed complex methods

Dataset

Variable

D1

D2

MR

MPR

Comp*

N = 200. corr 0.2 P-out < 0.1

Noise

12.2

12.2

12.4

11.4#

11.4

Cont4

63.8

64.6

64.6

67#

79.6

Cat1

63.8

66.6

67.2

84.2#

87

Cat2

76.6

83

84.8

92.8#

95.4

Dich1

35.8

35.8

36.2

37.6#

38.2

N = 200. corr 0.6 P-out < 0.1

Noise

11.2#

11.4

11.2#

11.8

10.2

Cont4

49.8

51.6

51.6

52.8#

65.2

Cat1

51.8

51.8

52.4

73.6#

74.4

Cat2

76.8

79.6

82

86.2#

86.8

Dich1

32.4

33

32.8

34.6#

34.6

N = 200. corr 0.2 P-out < 0.05

Noise

6#

6.6

6.2

6.2

5.2

Cont4

5

50.6

50.6

52.8#

67.2

Cat1

53.2

54.6

54.2

74.6#

75.2

Cat2

65

68.8

73

88#

86.2

Dich1

27.2

27.2

27.4#

29

27.8

N = 200. corr 0.6 P-out < 0.05

Noise

6.4

6.8

6.2#

6.2#

4.8

Cont4

39.4

39.2

40.2

41.4#

49.2

Cat1

38.2

38.8

38.2

61#

53.2

Cat2

65

65.4

70

78.6#

74.6

Dich1

22.8#

23.6

23.2

24.6

22.2

N = 500. corr 0.2 P-out < 0.1

Noise

12

11.6#

11.6#

11.6#

10

Cont4

94.6

94.8#

94.8#

94.8#

99

Cat1

96.2

98.8

98.6

99.4#

100

Cat2

99.2

100#

100#

100#

100

Dich1

64.8

65#

65#

65#

69.2

N = 500. corr 0.6 P-out < 0.1

Noise

10.4

10#

10#

10#

10

Cont4

82#

82#

82.2#

82#

91.2

Cat1

89.6

93.6

93.2

97.8#

98.8

Cat2

98.4

99.8

99.8

100#

100

Dich1

57.4

58

58

58.4#

61.2

N = 500. corr 0.2 P-out < 0.05

Noise

6.2

6

5.8#

6.2

05.2

Cont4

92

91.8

91.8

92.2#

97.8

Cat1

92.6

96.2

95.4

99#

99.8

Cat2

97.2

99.8

99.8

100#

100

Dich1

52

53

53

53.2#

58.8

N = 500. corr 0.6 P-out < 0.05

Noise

6.4

6.8

6.2#

6.2#

4.8

Cont4

39.4

39.2

40.2

41.4#

49.2

Cat1

38.2

38.8

38.2

61#

53.2

Cat2

65

65.4

70#

78.6#

74.6

Dich1

22.8#

23.6

23.2

24.6

22.2

  1. Number of observations, corr Correlation, P-out P-value for excluding a variable out of the prognostic model, Noise Noise variable, Cont4 Continuous variable 4, Cat1 Categorical variable, Cat2 Categorical variable 2, Dich1 Dichotomous variable 1, D1 D1 method, D2 D2 method, D3 D3 method, MPR Median-P-rule, comp analyses in complete dataset (reference values for the pooling methods)
  2. The selection frequency of variables in the complete dataset act as the reference standard: * = reference values for comparison the pooling methods with the complete data; # = value that is closest to the reference value