Dear Gbolahan,

Basically there is no specific answer to your question. Based on the type of your variable, you need to take decision. Suppose, you have some missing information on expenditure for purchasing healthcare services. In this situation, you could impute the values from similar category (could be disease, socio-economic background of the patient or combination of all these). However, this method would not be applicable if there is missing data on type of disease an individual is suffering from.

Thanks,

Montu

Dear Aramide

Can you please make some details on your type of data you are collecting and the nature of missing. Is the missing due to non-response (all data on certain participants are total missing)? or only few of the data on certain variables are missing?

Generally, non response is excluded from the analysis but must be acknowledged in the methods description. for other item missing, you can ignore the missing in univariate analysis. However if logistic regression or other multivariate analyses are required you have to read some relevant literature. 

The concept of "missing value" is multifold - by best practice approaches

First approach

If missing data means:" there are values missing for one or more variable" you can estimate this missing datapoint  by inserting the mean/average of other probes of overall, or of the same target group or filtered by defined categories. Even an estimation by regression is a good approach to get reasonable data.  That is a classical one that blurs data a bit but gets your results to work with. 

Second approach

Leave it as is. SPSS et al use this concept of implicit using of missing data. That is because "missing" can mean a thing. A sensor can be broken, a question wasn't clear enough etc. So every time the missing value can't be used (e.g. regression, crosstabs) it wont. Only the missing variable value  is excluded from analysis, not the person / patient / probe with all it's data.  

But what' about drop outs in a long term study, patients that move away, get sick and are lost or even die during the experiment or testing?

Try this: Third approach

One of the main reasons that clinical studies do not meet scientific standards are errors due to the "intention to treat" (intentional or not intentional).

The concept of "intention to treat" is based on the assumption that when you start a treatment in a randomised sample  everything that happens can influence the results (and success) of the treatment. E.g If pills to swallow are bitter or to large to swallow, the drop out of patients means: Have look, what happened? Or if people die during the treatment it would be more than misleading to drop those data because it is now missing for the ongoing study.

The "intention to treat" concept seems to be weird on first sight - because for example  you want to see the effect of a new medicine on morbidity or mortality of those who have been treated correctly and behaved in a compliant way. But that is a artificial sight on reality. "Intention to treat" shows all effects of the applied treatments. Don't drop data.

I've mentioned the "intention and non intention" dropping of data. "Intentionally" dropping data can also sometimes smell a little bit as if there has been also an intention to get proper results to verify the positive effect of a new treatment, medicine, product and exclude "interfering" results. 

Hope that helps.

Hans-Werner

PS:

Very interesting to compare the german and english version of wikipedia on this topic. In the german scientific sphere the "intention to treat" is the gold standard of medical research. 

https://de.wikipedia.org/wiki/Intention-to-treat-Analyse

Some other emphasis you can find here:

 https://en.wikipedia.org/wiki/Intention-to-treat_analysis 

Thanks all for your valuable contribution

Hans- compliments for useful information