Evaluating Imputation Methods for Missing Quality of Life Data
Author Information
Author(s): Shona Fielding, Peter M. Fayers, Alison McDonald, Gladys McPherson, Marion K. Campbell
Primary Institution: University of Aberdeen, UK
Hypothesis
Can simple imputation methods accurately handle missing not at random (MNAR) quality of life data?
Conclusion
Simple imputation methods were found to be inadequate for handling MNAR data, suggesting the need for alternative strategies.
Supporting Evidence
- The study found that simple imputation methods were not sufficiently accurate for MNAR data.
- Reminder-response data indicated that the missing data were likely MNAR.
- Logistic regression showed that previous QoL scores were significant predictors of missingness.
Takeaway
When researchers don't get all the answers they need, they sometimes guess what the missing answers might be. This study found that guessing isn't always good enough, especially when the missing answers are important.
Methodology
The study used hypothesis tests and logistic regression to evaluate missingness mechanisms and compared simple imputation methods against actual observed scores.
Potential Biases
The assumption that reminder data is equivalent to immediate data may introduce bias.
Limitations
The study was based on a single trial involving older people, which may not be representative of other populations or conditions.
Participant Demographics
The majority of participants were female (85%), aged over 70, and most lived independently before and after their index fracture.
Statistical Information
P-Value
p<0.001
Statistical Significance
p<0.001
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website