There are two types of methodology to handle missing data.
1. Removing the portion of the data which has the missing value. 2. Imputation method which is based on close estimation.
Which one is better? if we choose 1, what will be the impact prediction?
Hi,
this came up once ago. Perhaps the thread is informative
https://www.researchgate.net/post/proper_imputation_method_for_my_missing_data_pattern
Short answer: Removing data is never a good idea as you diminish the sample size, power and statistical precision. Worse, you may introduce selection bias.
Best,
Holger
One issue that many people don't realize is that removing cases with missing scores (e.g., by listwise deletion) implies a more restrictive assumption about the missing data mechanism (i.e., missing completely at random or MCAR data) than do modern approaches for handling missing data such as multiple imputation or full information maximum likelihood (FIML) estimation which only require missing at random (MAR) data. FIML doesn't even require you to go through an imputation phase. You simply estimate your statistical model with all available data. See: https://www.youtube.com/watch?v=SoQCSwlC_b4
I recommend checking out a good resource on missing data analysis such as Craig Enders' book "Applied missing data analysis" (2022, Guilford Press) and/or
Schafer, J. L., & Graham, J. W. (2002). Missing data: Our view of the state of the art. Psychological Methods, 7(2), 147–177. https://doi.org/10.1037/1082-989X.7.2.147
See also: https://www.youtube.com/watch?v=j8mpYzfobMo&t=20s
I'm going to take a somewhat different point of view and say it depends on the data and the research question as well. An example from my work is to consider determining disease risk factors (genetic) from diagnosis and genotype of the patient. If I don't use the MCAR assumption then it seems to me that I am assuming the result of the analysis before I have it and hence I use missing value deletion. I would be quite interested in any other points of view. Thanks to all. David Booth BTW the final n is such that 3000
- Badges
- Science topic