Missing At Random - explained by the observed variables in the dataset and not dependent on any unobserved value

Missing Completely At Random - No relationship between missing data and the underlying values of the dataset

Missing Completely Not At Random -Missing data are related to either variable that was not recorded (unobserved values) or the outcome

All participants who were randomized need to be included in the analysis, regardless of what happened to them afterward(even if they never started the treatment or crossed over to the other group)

Analyses only subjects who did not deviate from the protocol.

Excluding all patients who did not get or did not complete the specified treatment

Lower power

Lower sample size

Increase type-I and -II errors

No manipulation

No assumptions

Simple

Mean and Median Imputation

Last Observation Carried Forward (LOCF)

Regression Imputation

Each missing value will be replaced by a simulated value. This will be done several times, obtaining multiple sets of completed data by imputation.

Estimates have almost optimal properties

Results are mostly consistent and efficient

Can be applied to most data or models.

Uses conventional software

Different results each time will be run

estimate aregression model. It uses several interactions to fit the model.

Always gives the same result

Has fewer decisions to make

more efficient because it has smaller Standard Errors

Requires specialized software

For large samples

Needs parametric model for the joint distribution of all the variables with missing data