Dealing with missing data in a multi-question depression scale: a comparison of
imputation methods [electronic article]
Missing data will in the first instance be managed with a «last observation carried forward» approach with additional sensitivity analyses undertaken using multiple
imputation methods.
Lack of information on reasons for missing data, insufficient evidence of effect of missing data on outcome, lack of information on
imputation methods or insufficient detail on intention - to - treat and participant departure from allocation to make a judgement of low or high risk of bias
EM
imputation methods used provided a value for missing data points for each individual participant based on both the estimated sample parameters and the actual responses at completed data points, preserving sample size and making use of all available data, while maintaining the variance within the sample.
Missing values (missing completely at random)(10.5 %) were imputed with single
imputation method.
Not exact matches
Coefficients and SEs for the variability between
imputations were combined using the
method of Rubin.26
We undertook multiple
imputation (MI) of predictor variables to assess the sensitivity of results to missing data using the chained regression
method of MI to generate five imputed data sets.
We used multiple
imputation with the
method of chained equations to account for missing maternal data for children with a mother in the household.24 In addition to the mother's BMI status (missing for 1085 [22 %] of the children),
imputation was conducted for 4 maternal covariates with few (< 1 %) missing cases (education [n = 13], warmth [n = 47], control [n = 49], and irritability [n = 48]-RRB-.
In the case of data
imputation, we will specify the
methods used in the «Characteristics of included studies» tables.
Missing data for specific time points and from loss to follow - up will be dealt with through multivariate multiple
imputation or full information maximum likelihood
methods as appropriate.