If possible I contact subject for clarification.

Only other choices:

1-higher Ra

Article Exploring cognitive complaints in schizophrenia: The Subject...

In my view you need to treat this as a missing value.  If you are set on using the response, use the value nearest to the median: in this case 'difficult'.

You need to give very clear instructions that respondents should only circle one answer.   You can force this by using an electronic questionnaire that only allows one box to be clicked.

Hope this helps

Nick

If participants selected more than one response, picking one of the selected responses simply introduces error into the data. Furthermore, picking the highest response introduces bias.

You will just have to take a deep breath and omit the data. Otherwise you will degrade the overall quality of your data by including both random error or systematic error, depending on which replacement strategy you use.

If there is a widespread problem with participant response it argues that the measure is inappropriate (for instance too hard to understand, or containing  items not applicable to many participants) in which case you should think of dropping the measure.

By the way, the reference to the Stata FAQ in the thread you link is wrong – the FAQ is dealing with polytomous items in which the person may legitimately select more than one response (eg "Which of the following have you suffered from in the last year?")

 Thank you all for your reponses. 

Dear Keira, 

I fully agree with the excellent comments made by Ronán and Nick. Omit the data would be the best solution in the scenario you proposed. 

There is a review from Kalton and Schuman that could be of interest to you:

Kalton, Graham; Schuman, Howard (1982). "The Effect of the Question on Survey Responses: A Review". Journal of the Royal Statistical Society. Series A (General). 145 (1): 42–73. doi:10.2307/2981421

With kind regards, 

Julio 

The best solution is omit the data

I concur with omitting the data. This way you do not "mindread". if this happens frequently, then you need to modify the questionnaire. If not, then the missing data should not effect results, unless you have too small a sample.

You need to give very clear instructions that respondents should only circle one answer.   You can force this by using an electronic questionnaire that only allows one box to be clicked.

There is a specyfic situation for scale 1-6 when somebody circle 3 and 4. I sometimes treat is as a 3.5, but it can be difficut to interprate after that. I use it only when I really need to have mean of the whole participants :)

I agree with the previous advice of ways to try to prevent the problem. But it happens anyway with paper / pencil questionnaires. I remember being taught that you alternate how you code it. The first time you use the higher value; the second instance you use the lower value, and so on (or the opposite ... you are just trying to counter-balance the effects). If there is only one occurrence, omit the data. I don't have a specific reference unfortunately, but it makes sense to me. I think you have to track your alternations for every question (in case, for example, you have two questions that provoke double responses you don't want to be always 'increasing' on one of the questions and 'decreasing' on the second. 

Depending on what analyses you are running, another option could be to run separate analyses for 1) treating the data as missing, 2) taking the lower score, 3) taking the higher score, 4) taking the average score, 5) replacing the value with the mean for the scale of the item. If you get the same results with all methods then it doesn't really matter what you do; you can just report what you did and that results were the same either way. It's less clear what to do when you get different results! That kind of procedure is often used when dealing with outliers; consider searching the literature for references on how to handle outliers and then apply the same logic to your situation.

Kiara - just a suggestion:

Have you either heard of or tried actually going out to your  local library to find this very helpfully specific book: Malcom Gladwellʻs "Outliers", before? If you do not know of him, he is a high quality and very prolific author. Malcolm is a somewhat famous American (USA) psychologist whom writes in this rather insightful and useful book; of the same name. Published in November 18th, 200.*

It should explain a few things out to you or further if your stuck.

He also wrote "Blink: The Power of Thinking Without Thinking!" Published in January 11, 2005.

That because  found interesting and intriguing good read and use it many times before, even purchased a copy of his great psychological reported book, personally especially with my previous studies at TAFE college.

It has been published worldwide, so it is available out there in the public domain.

Hope that helps, Kiara with the rest of your honours degree thesis research. (which is great anyways). Good luck.

Yours Sincerely and truly, from:

Patricia Massarotto

(GU ID: 5064377)

In practice, I have chosen the lower or more negative response - very difficult. This will not positively skew your data. The second choice would be to consider it a missing value, taking into account your power analysis and the percent missing data you had estimated.

Hi, Ester Salmán

Yes very consciously true.

Agreeably perspective and good real life analysis too.

i prefer to treat it as a missing data and integrate that knowledge into my analysis