I'm currently cleaning raw data for a project in university. I received 262 responses- 22 of these were incomplete. Am I correct to extract these responses from my data or will this negatively effect my statistical power?
Abigail -
Smaller datasets do lead to greater standard errors, but even more here, you could be biasing your results. There could be a reason that those items were left out.
Some would impute results based on mean values. However, if there is a bias, and the results should have a different mean, then you have introduced bias. You would also then artificially decrease your estimate of variance (leading to standard error) because you have put in mean values, not ones that average that way.
So either ignoring data or replacing with means can be a substantial problem.
If some variables are used in regressions, this complicates the problem. However, one of the other imputation methods, besides use of means, involves an application of regression where you could look at some outside data you have on all cases as regressor data, and 'predict' for your missing data.
There are various imputation methods with pros and cons.
Perhaps you could do sensitivity analyses. For one thing, you could take an item from your 262 responses, and say you were interested in its mean response, you could compare that to what you would get with the 240 responses if that item had been incomplete on all the incomplete responses, or however the missing data laid out. Of course some items are more sensitive than others.
Actually you may have a bigger problem. You said 262 responses. How many nonresponded altogether beyond that? Was that to be a census? Otherwise did you consider the sampling methodology?
Your results may have some serious unknown uncertainty.
I think you need an appendix to explain the results you got doing some sensitivity analyses, perhaps with artificial data that varied in different ways, to indicate just how far from reality the results obtained might be.
As for throwing out the data for those 22 cases, if they are for items you can look at separately, it sounds like you should not throw them out, but there may be a lot more to the story than that. It also depends on how your software handles different options.
Best wishes - Jim
- Badges
- Science topic
- Similar topics
- Phytochemistry
- Extracts