NO IT IS NOT. "Filling missing data"---by whatever method--- is using fake data that never existed. The fact that homogeneity test or any test to show that the fake data does no harm to the process does not make the effort blameless nor the data set free from question. A data set with missing data is a defective data set. It is pure an simple that any "filled" data is not data; any conclusion drawn from it is illusory. Any publication made from it is a potential fraud. It is a serious ethical issue.

Paul Louangrath thanks for your answer. Are you saying replacing missing data is not scientific approach at all ?? I think not

Solomon -

I would not want to do this kind of thing if not necessary. However, imputations of various kinds are often necessary. You took a linear interpolation which sounds reasonable if needed. You have to remember that that is not real data, and consider the impact on results. There is no ethical issue if your report notes what you did and why, and the possible ramifications. I would think variance was artificially reduced, just as when someone imputes a mean value in a finite population estimate of a total. That assumes what is called "ignorable nonresponse," and artificially reduces variance if you don't do anything about that. I prefer "predicting" (using regression) for such missing data.

The point is that we fill in for missing data all the time. When we sample and estimate or predict for the total of a finite population or the mean for any population, we are doing our best to deal with missing data. And it is often likely to be more accurate to estimate or predict (using regression) for a finite population from a good sample than to try to census all the time, because of data quality/nonsampling error problems.

I don't know why it is necessary to do what you are doing to be able to complete your project. But if it is necessary, say so, say why, and evaluate possible impacts. Report it all.

Best wishes - Jim