3 Techniques for Imputation

It’s possible that you are missing data in your dataset. When that occurs, it is often helpful to find a way to include that data. You need to look for a method that is mathematically sound, allowing you to put the most plausible answers possible into the missing data.

The process is known as imputation, and while it does reduce the accuracy of your data (because there is no way to be certain that the imputed data is accurate), it can help companies analyze their research when that missing data is needed. Here are three techniques that researchers might use for imputation.

Multiple Imputation Many of the basic forms of imputation deal with providing a single number for the dataset based on some technique that is believed to provide an accurate (or close to accurate) numeric value. Multiple imputation uses simulated datasets that allow researchers to not only provide numbers for the dataset, but also provide a confidence interval based on how much the overall research was affected by the imputed data. Last Observation Another imputation technique is to take the previously collected data by the respondent and plug it into the dataset in lieu of actual data.

The idea is that responses do not change by a wide margin at any given time, and providing the last observation to the missing dataset should do a better job at keeping the data close. Incremental Mean Another imputation method is known as the incremental mean.

But placing the data into matrices, researchers can create a calculated step based on the approximate mean values of X over time. The incremental mean can also be adjusted to allow for uncertainty, which helps ensure that those using your data understand the extent to which the imputation affected the overall results.

Imputation always makes your data less reliable, because any time you exclude an actual response, you reduce the accuracy of your data. If research can be conducted without imputation it should be avoided, since while the numbers may be mathematically likely, they cannot be called “accurate” without knowing the actual missing data. But there are mathematical ways to come up with a probable addition to the dataset, and by using an imputation technique above, you can presumably create a dataset that closely matches the actual data and allows for further analysis.