Missing data completely for one item in an 18 item questionnaire I have been preparing a data set for analysis and came across an unusual problem. For one of the stigma questionnaires used 1 item out of 18 reports was missing for all subjects. The PI wasn't sure what could have done it, and thinks it may have been a coding issue when the survey was ran in 2014. For the rest of the items in the questionnaire we have a nearly 100% response rate. What options do we have? Is there anything that could be done to "approximate" what their response may have been given responses to the rest of the questionnaire or should we just throw it out? 
 A: From what you are describing, there is no chance correctly 'approximate' these values using your dataset.
Since it was missing for all reports....this is impossible.
You have 2 options:


*

*throw this item completely out of your analysis

*use other datasets to impute these values.
(a dataset that contains correlations between variables you do have and the lost variable)
A: If this is one of those multi item scales in the social sciences where respondents are asked many very similar questions, you might just use it without the missing item. (I'm talking about those scales that have many more questions than underlying dimensions where you typically perform cronbach's alpha for internal consistency and a PCA with varimax rotation.)
There is no hard reason why the scale would only work with 18 items but not benefit from item number 19 and break down as soon as there are only 17 items. One researcher has come up with the 18 items at one point in time and tested their consistency. Since then, other researchers use the same scale and cite him.
The only problem I see with dropping an item is replicability. Future researchers will want to use the original scale and not randomly drop an item like you did.
