Table 4

Estimated treatment effects for different imputation strategies when missingness is covariate dependent

Imputation level

Imputation strategies

Analysis model

OR4 and 95% CI5 for Complete Data: GEE2 1.14 (0.76 1.70) RE3 1.12 (0.72 1.76)


OR4 and 95% CI5 for Different Percentage of missingness


5%

10%

15%

20%

30%

50%


Within cluster

Logistic regression

GEE2

1.14 (0.76 1.72)

1.14 (0.76 1.72)


RE3

1.12 (0.71 1.78)

1.13 (0.71 1.78)


Propensity score

GEE2

1.14 (0.75 1.72)

1.14 (0.75 1.73)

1.14 (0.74 1.75)

1.14 (0.73 1.78)

1.15 (0.71 1.84)

1.18 (0.68 2.04)


RE3

1.12 (0.70 1.79)

1.12 (0.70 1.79)

1.12 (0.69 1.82)

1.12 (0.68 1.86)

1.12 (0.65 1.93)

1.15 (0.61 2.18)


MCMC1

GEE2

1.13 (0.75 1.71)

1.13 (0.75 1.70)

1.13 (0.74 1.71)

1.12 (0.74 1.72)

1.12 (0.72 1.74)

1.12 (0.69 1.80)


RE3

1.11 (0.70 1.77)

1.11 (0.70 1.76)

1.11 (0.69 1.77)

1.11 (0.69 1.78)

1.10 (0.67 1.81)

1.10 (0.64 1.88)


Across cluster

Propensity score

GEE2

1.14 (0.77 1.68)

1.14 (0.77 1.67)

1.14 (0.78 1.67)

1.14 (0.79 1.67)

1.15 (0.79 1.67)

1.15 (0.76 1.72)


RE3

1.18 (0.88 1.59)

1.18 (0.87 1.59)

1.18 (0.87 1.60)

1.18 (0.86 1.61)

1.18 (0.85 1.64)

1.17 (0.78 1.76)


Random-effects

GEE2

1.15 (0.78 1.69)

1.16 (0.80 1.70)

1.18 (0.81 1.72)

1.19 (0.81 1.75)

1.22 (0.81 1.83)

1.31 (0.83 2.06)


logistic regression

RE3

1.14 (0.75 1.74)

1.16 (0.77 1.74)

1.18 (0.79 1.76)

1.19 (0.80 1.78)

1.22 (0.80 1.86)

1.31 (0.83 2.05)


Fixed-effects

GEE2

1.14 (0.76 1.71)

1.15 (0.76 1.73)

1.15 (0.76 1.76)

1.16 (0.75 1.79)

1.17 (0.73 1.86)

1.17 (0.67 2.04)


Logistic regression

RE4

1.13 (0.72 1.77)

1.14 (0.72 1.79)

1.14 (0.71 1.83)

1.15 (0.71 1.86)

1.15 (0.68 1.94)

1.15 (0.61 2.18)


Ignore cluster

Logistic regression

GEE2

1.14 (0.78 1.67)

1.14 (0.79 1.65)

1.15 (0.80 1.64)

1.15 (0.81 1.64)

1.16 (0.83 1.63)

1.15 (0.81 1.63)


RE3

1.13 (0.74 1.72)

1.14 (0.76 1.70)

1.15 (0.78 1.68)

1.15 (0.80 1.67)

1.16 (0.82 1.65)

1.15 (0.81 1.63)


Propensity score

GEE2

1.14 (0.78 1.67)

1.14 (0.79 1.65)

1.15 (0.81 1.64)

1.15 (0.82 1.63)

1.15 (0.83 1.61)

1.15 (0.82 1.62)


RE3

1.13 (0.75 1.72)

1.14 (0.77 1.69)

1.15 (0.79 1.67)

1.15 (0.80 1.66)

1.15 (0.82 1.63)

1.15 (0.82 1.62)


MCMC1

GEE2

1.14 (0.78 1.67)

1.14 (0.79 1.65)

1.15 (0.80 1.63)

1.15 (0.81 1.62)

1.15 (0.82 1.59)

1.13 (0.82 1.57)


RE3

1.13 (0.74 1.72)

1.14 (0.77 1.69)

1.14 (0.78 1.67)

1.15 (0.80 1.65)

1.15 (0.81 1.61)

1.13 (0.82 1.57)


Complete case analysis

GEE2

1.14 (0.76 1.70)

1.14 (0.76 1.71)

1.14 (0.76 1.72)

1.15 (0.76 1.73)

1.15 (0.75 1.75)

1.15 (0.73 1.80)


RE3

1.13 (0.72 1.75)

1.13 (0.72 1.76)

1.13 (0.72 1.77)

1.14 (0.72 1.78)

1.14 (0.72 1.80)

1.15 (0.71 1.85)


Note:

1. MCMC = Markov chain Monte Carlo. For MCMC methods, we round the imputed values to 1 if it is equal or greater than 0.5 and to 0 otherwise.

2. GEE = Generalized estimation equation method

3. RE = Random-effects logistic regression

4. OR = Odds ratio

5. CI = Confidence interval

Ma et al. BMC Medical Research Methodology 2011 11:18   doi:10.1186/1471-2288-11-18

Open Data