Imputation vs Deleting Data

Recently we have had several posts about imputation, and how it can be used to replace missing data points in your survey. But for those that are new to advanced statistical research, imputation seems like a needlessly complicated step. In the past, researchers may have simply deleted or ignored the missing data, assuming that it didn’t bring any value.

In fact, for studies with few missing data points, this may actually be the case, since the effects of a few pieces of missing data will be unlikely to affect analysis to any large degree, and adding the imputed data points back in will be unlikely to make any major differences. In addition, if it can be found that the missing data is truly random – that is, there is no systematic reason that this data is missing and that the respondents that were unable to complete or find this data was chance – then deleting data shouldn’t have any effect on your analysis provided your sample is sufficiently sized.

But for larger studies, where a great deal of data may be missing, researchers have found that, in general, it cannot be assumed that missing data won’t affect the dataset or is entirely random . That is because the more data that is missing, the more likely it is that these respondents have something in common, and those that are excluded from the dataset completely because of these commonalities will not be represented in the data.

In addition, the more data that is deleted, the more it may represent sample size issues that – while not fully addressed by imputation – can improve the likelihood of collecting an adequate sample of data. Imputation is designed to help correct for these issues. By utilizing mathematically based imputation techniques that provide a reasonable value (or values) for the missing data, you will have an easier time performing analysis and drawing meaningful conclusions.

If the imputed information is accurate based on the information you have available, your data should be able to provide you with the conclusions you were hoping to discover, and the missing data shouldn’t have as many consequences.

No matter what type of study you are running, it is possible for you to be missing important data. While imputation is not going to create a perfect dataset by any means, it can at least get you closer to arriving at more reliable conclusions.