2
$\begingroup$

I am doing an online study and have just started looking at the data. I noticed two of my participants have listed ages that they couldn't possibly be (e.g 450 and 220).

I'm wondering what the appropriate way to handle this is?

Age isn't the main variable for the study so should I use imputation so I don't lose two data points? Or should I treat it as a missing data point and use little MCARs test to determine if listwise deletion is appropriate?

$\endgroup$
  • 1
    $\begingroup$ I'd argue that if you can afford treating it as missing values, then do so. If you can't you can check imputation options, but make sure to clearly state that you did impute these values $\endgroup$ – deemel Sep 30 '19 at 8:56
3
$\begingroup$

Since these are clear errors, I think you have a good case for ignoring the existing values and using multiple imputation to fill them back in.

If there was uncertainty (say you found an age of 110 - possible but unlikely in most datasets), this would be a tougher decision.

$\endgroup$
-1
$\begingroup$

It is a common practice to assign valid data ranges for your variables.

When data is outside of the range, you can do different things: 1. You can eliminate the entire row. 2. You can treat it as a missing value (as a level, not an ordinal number) 3. You can assign it the average value of the variable (in the whole data set), or assign some other function to set the value.

If you have a large data set, with relatively few invalid values, usually the best approach is to delete the row. (This is the default with SAS Enterprise Miner. It's easy to set ranges, and rows of data with a variable out of it's range are filtered out. It's also easy to change this, as above.)

You should only treat an invalid age as a missing value if the data mining algorithm will handle that (Decision Trees and others). Otherwise, you have to assign it a numerical value, and it will be treated as one (bad). If you are using linear regression, setting it to the average age would allow you to use the rest of the data in the row, with minimal impact on the results.

-V. (M.S. Data Analytics)

$\endgroup$

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.