Last modified: 4/17/25
ANESTHESIOLOGY
Missing Data Part 2: Methods
Type/definition
Pros
Cons
Used for
Listwise deletion
removes any case
with missing data.
Complete case
analysis
-easy to implement
and analyze
-minimal bias if
models specified
properly
-loss of statistical
power
-some bias if MAR but
not MCAR
-MAR, MCAR
-MNAR (robust for
predictors in
regression)
Pairwise deletion
uses non-missing
values from cases
with missing data.
Available case
analysis
-uses all available
information
-approximately
unbiased if MCAR
-different subsets for
analyses based on
availability
-inaccurate standard
errors
-MCAR
-MAR (may be biased)
Dummy variable
adjustment adds a
missing category to a
categorical variable.
-doesn’t drop cases
-easy to implement
-retains sample size
-produces biased
coefficient estimates
-not recommended
Mean substitution
replaces missing
values with the mean
for the given variable.
-intuitive on the
surface
-easy to implement
-retains sample size
-estimates are less
precise and too close
to the mean
-can be biased
-not recommended
Last observation
carried forward
(LOCF) uses the last
non-missing value to
complete
longitudinal data
-intuitive on the
surface
-retains sample size
- biased estimates of
treatment effects
-values carried forward
for dropouts may
obscure the effects
-no recommended in
general
-may be acceptable
for filling some gaps, if
it makes sense
clinically
Multiple imputation
completes the data
by algorithm, the
most common being
MCMC based on
linear regression,
-retains sample size
-estimates are
consistent and
unbiased
-complete case
analysis can be used
after imputation
-analysis takes longer
-interpretation of
pooled results is more
complicated
-different methods
produce different
results
-MCAR
-MAR
-MNAR
Last modified: 4/17/25
References
1. Allison, PD. (2009). Missing Data. SAGE handbook of Quantitative Methods in Psychology.
Thousand Oaks, CA.
2. Graham, JW. (2009). Missing Data Analysis: Making it Work in the Real World. Annu. Rev.
Psychol. 60:549-576.
3. Kenward MG, Carpentr J. (2007) Multiple imputation: Current perspectives. Statistical Methods
in Medical Research; 16: 199-218
