Hello, I am a graduate student and I am working on cleaning survey data. When cleaning the data I realized that a few respondents completed the survey twice. I reviewed the survey data to see if one of the submissions from the respondent was only partially completed. However, there are a few respondents who answered and submitted 100% of the survey questions twice (we have since figured out how to prevent respondents from taking the survey multiple times). My initial thought is to use the survey submission that the respondent submitted first, as it was their first interaction with the survey, but I wanted to confirm that this is methodologically correct. I have been searching resources online, talking with colleagues, and reaching out to groups at my university but have not been able to find an answer to this question, (I may be phrasing the question incorrectly). Any resources or answers to this question would be greatly appreciated! Thank you for your help!
There are a few reasons why it may appear someone does it more than once, but your assumption to use only the first is consistent with what I do. Here are a couple of the reasons.
1. It is the same IP address but may not be the same person. Still, the people likely talked so the data are not independent so the second person's data should not be used.
2. It is the same person, you are paying them, and they thought get more money (again, you don't want multiple ones).
3. They started, stopped, so have come back to complete. Here there might be reasons to combine them. Say one question required them to look up something, so they did that and then came back. Do you have any info that this could have happened? If it is a study that would require that, the increased time/the break adds a potential confound. You might get rid of both.
There are other reasons, and really to provide guidance it would be important to know the reasons. Also, how many cases was this for?
There are a few reasons why it may appear someone does it more than once, but your assumption to use only the first is consistent with what I do. Here are a couple of the reasons.
1. It is the same IP address but may not be the same person. Still, the people likely talked so the data are not independent so the second person's data should not be used.
2. It is the same person, you are paying them, and they thought get more money (again, you don't want multiple ones).
3. They started, stopped, so have come back to complete. Here there might be reasons to combine them. Say one question required them to look up something, so they did that and then came back. Do you have any info that this could have happened? If it is a study that would require that, the increased time/the break adds a potential confound. You might get rid of both.
There are other reasons, and really to provide guidance it would be important to know the reasons. Also, how many cases was this for?
I'd like to add to Daniel's great answer:
Given these were really several complete submissions from the same person, I would randomly select one of them for the analysis and use these replicate submissions to investigate the consistency or variability of submissions that I might expect from a single person. Knowing this might help to judge the quality of the data and the relevance of effects seen in the actual analysis.
Data cleaning is for that only you can just delete the multiple survey questionnaire retain one which has full information. Indirectly its a method of validating your questionnaire right, replicability.
Thank you all so much for your responses! Daniel Wright- In response to your questions/thoughts. We had this incidence happen for 4 respondents. The names and email address they submitted with the two submissions are the same and they completed the survey 100% both times they took it, however the responses vary slightly (though the demographic data and much of the other data is consistent). We are not paying the respondents for their participation. Jochen Wilhelm in this case would you randomly select the case to include in the sample, since both cases have complete information? Thank you again for your help!
Yes, toss a coin - if there is no external reason to prefer one (e.g. because one was answered timely closer with the rest, or because answers should be made as spontaneous as passible, in which case I would take the first one returned, same if I want to make sure that the questions/possible answers are not discussed among participants... things like that).
Jochen Wilhelm Thank you very much for following up, your feedback was very helpful. We ended up randomly selecting one of the cases to include in the sample.
You have had some very savvy advice. Make sure that you retain something that identifies these duplicates, just in case they turn out to be very influential outliers.
- Badges
- Science topic