You should remove the answer completely since no statistical program will be able to measure this, unless you can divide it into two seperate questions, which shouldn't be done. Be sure to inform your participants there should be only one response per question in the beginning of the survey. Good luck!

Edit: Did all of your participants select two options on the questions? In that case you should re-do the whole survey, clearly it's faulty in some way with it not being clear about the guidelines. IF it's a few, then you should like I said earlier remove these from your data, assuming you'll still have enough answers for an analysis.

Thank you, Mattias! Yes, removing it sounds like a good option -- one that I had been considering as well. There are only a few participants who did this -- not all, so removing these answers shouldn't do too much harm hopefully.

I agree that you need to treat those respondents as missing data, because there is no way to assign them to one category or the other.

If only a few participants did this, then i would code those people as missing. if a lot of participants did this, then, you have to set up the variable in multiple choice format which allows you to select more than one response.. In SPSS, each possible response is set up as a separate variable. if the person chose the response, you code as 1 and leave blank otherwise. You then use the multiple choice analysis tool to group the responses. This gives the percentage of cases and the percentage of responses for the separate variables that comprise the response choices (sing define variable sets function). I can provide more detailed instructions if required. Hope this helps. Best of luck wit it.. david

Thanks everyone. I have taken your suggestions and treated those as missing!

I guess, what you are asking should have been sorted out from the development of the survey tool stage. If it was a questionnaire, then you should have provided for multi-response option...which it seems you did not. Moving forward, removing them may compromise the quality or end product, that is the result, unless you do away with the question in all the responses/all the tools used to capture the responses

This kind of problem is usually detected during the pilot study phase, if the researcher opts for pilot study at all. However, sometimes small "n" in pilot study may not be able to detect the ambiguity in the question. Just to clarify at the beginning, it is possible to have multiple response questions in a survey. For example, you can ask respondents whether they have ever taken any of the following foods (a, b, c, d, e). Now you can have one respondent choosing b and e, this is quite possible. This is the kind of multiple response question that you define by the data structure that you have in mind. This is not something happens out of ambiguity. This kind of question can be coded and analysed, please look at the stata link below. I assume your question is not of this type, rather some respondents might have problem in understanding the wording or response category. In this case, as other researchers have pointed out, you have no option but to treat it as missing values since the number of such responses is low.

http://www.stata.com/support/faqs/data-management/multiple-responses/

This will depend on the survey quesfion asks. If it requires multiple answers then you can not remove those responses and might affect the result in the end. For multiple responses encode them separately. Or you must revalidate your questionnaire.

Depending on the nature of your survey, The Golden Mean should suffice.

I don't think I could give you a better answer than the one already provided by Mr. Khandoker Mahmudur Rahman.

Bonne chance!

Bernard Tétreau, Ed.D.

If you are using printed questionnaires, the options are found in the answers to date. If you are using an online questionnaire, Likert scales should be set up as radio buttons so that only one answer can be chosen. 

Albert Werner 's answer seems quite good as far as I can understand.

Bernard Tétreau

Multiple answers for a question are acceptable and you can design your questionnaire or data set to capture this. Just ensure that your data set captures such a question as a multi response.

I agree with all the responses. I also did this error i have a 5 pointer likert scale and some participants initially have marked more than one response for a item as i was unaware of the option to make change in google form to allow only 1 item. Now can i take mean of their responses for those particular items in order to analyse?