I have a dataset with approximately 1% of values missing completely at random.

I have thought about using the Multiple Imputation technique but I am not sure if this would be the best solution.

Can anyone please explain what the best technique is for handling missing values on a dataset as described above?


If it is 1% missing on any variable -- that is, you will have 99% of your data if you just delete any observation with missing data on any variable -- and if it is truly MCAR, then you can probably just ignore it. The only loss will be a very small reduction in power.

If it is 1% on each of many variables, then, if you have a lot of variables, and the missingness is independent, then you may be missing a lot of data. In that case, I'd go with some form of multiple imputation.

| cite | improve this answer | |
  • $\begingroup$ Thanks so much @PeterFlom-ReinstateMonica♦! $\endgroup$ – Jord Nov 19 '19 at 13:02
  • $\begingroup$ In paragraph 2 when you say "then you may be missing a lot of data" do you mean "then if you perform listwise deletion (complete-case analysis) you may be missing a lot of data"? Because the question already specifies that 1% of data values are missing. $\endgroup$ – rolando2 Nov 20 '19 at 12:55
  • $\begingroup$ @rolando2 "1% missing" is a little ambiguous. If you have 50 variables and 1% missing on each, you could be missing a lot of data. You are missing that data regardless of what analytic method you choose, although it will affect different analyses differently. $\endgroup$ – Peter Flom Nov 20 '19 at 14:34

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.