deal with missing data + emailing participants?

[bluebluesky]

---

Members do not see this ad.

I recently collected some data and although I trained the assistants to carefully screen for missing items when the participants turn them in, we discovered quite some questionnaires with missing items. I have two questions below:

1. How do you usually deal with missing items?

2. We have a way to track down the students' email addresses. Can we still email them for filling out the missing items? Will that violate confidentiality stated on informed consents? Of course we only contact them if there're missing values because I really don't know how to deal with it effectively. I feel that it's a pity that we have to throw them out just because they missed one item on a whole questionnaire packet.

Any advice is highly appreciated. 🙄

 

[Blue86]

I don't think it is ok to contact participants about missing items.. maybe they skipped them on purpose? Even if they didn't, I don't think it's ok.
Can't you try to run the statistical analysis taking into consideration that some items are missing? I know that there are several ways to do that...
Maybe other SDNers have a different opinion.

 

[ADDICTED2STATS]

Contacting people is the wrong answer. You can use multiple imputation to impute missing data or fiml estimation which is robust to missing data for the analysis (there's a myriad of other things, these are just two popular options). Missing data is common, there are plenty of ways to address it.

 

[KillerDiller]

Unless you informed participants that their data would be identifiable to the researchers, you should destroy any means you have of connecting data to an email address.

Either way, I agree that contacting participants to fill in missing data is not appropriate.

 

[Metta]

It would also add a potential confound to your study; some participants will have completed parts of the questionnaire under different conditions.

 

[JockNerd]

Do not contact participants to get missing data points.

It's easy to handle missing data. Schafer & Graham, 2002, review the best methods. I use NORM (a free imputation program) to replace missing data for participants without substantial missingness; best method available.

 

[PsychPhDone]

I'll echo the others' sentiments: do not contact the participants. There are ways to deal with missing data statistically, and most datasets do have missing data. Though we try to avoid amassing error, it happens.

 

[irish80122]

I agree with the others that you shouldn't contact the participants. The first thing you need to do is figure out if the data is missing completely at random (MCAR), missing at random (MAR), or not missing at random. If you have the Tabachnick and Fidell multivariate textbook it has a great section on this. Also, I gave a presentation about how to handle missing data (I personally prefer multiple imputation). If you PM me your email address I would be happy to send you a copy of the presentation.

 

[bluebluesky]

My SPSS does not have the component of dealing with missing data so I am not sure yet at this point if I can deal with it efficiently. One of my professors said that I may need to buy a [FONT='PrimaSans BT,Verdana,sans-serif']IBM SPSS Statistics module specifically dealing with missing data.

Do you know any way that SAS can do it? Not sure. Thanks.
.

 

[irish80122]

IMHO, the SPSS missing data tool really isn't that great. You should be able to do multiple imputation through SAS (it is pretty sexy).

Here is how you do it in SAS http://www.sociology.ohio-state.edu/people/ptv/faq/mi_mianalyze.htm

Here is all the brain-numbing details http://support.sas.com/rnd/app/papers/miv802.pdf

 

[cognosco]

Although I agree with the others in that it probably isn't ok for you to contact particpants after the fact. The exception to this, of course, is when you obtain permission to do so in the informed consent. The issue about the confound mentioned by another poster does still exist though. If a P didn't answer a question, it's likely that he/she simply did not want to answer it.

The other issue is my primary reason for responding. One option is to replace the missing data with the mean response of the other participants for that particular item. You should of course report this in your write up of the project, and it is important to note that there are more sophisticated methods (like those mentioned earlier) for achieving more exact estimates of participants' scores. However, if your under a budget, you could use this method using SPSS basic. Good luck!

 

[irish80122]

In reference to cognosco's post, I still see no reason why you couldn't use multiple imputation (one of the fancy and preferred methods) since you have SAS. I think most people still view mean imputation as acceptable, but it doesn't impress and it has significant limitations. I say go with the best. It takes a while to learn how to do multiple imputation with SAS, there should not be any additional cost as long as you have SAS 9.1. Using a procedure like multiple imputation will impress some of your reviewers (it impressed my committee members at my defense) and, if anything, only make them look more positively on your research. Plus, once you know how to do it, then it is easy to do for future projects!

 

[JockNerd]

NORM does multiple imputation and is free.

 

[TBIdoc]

you may have to consider many things:

1) are you permitted to contact the participant after the study (as per informed consent)

2) is this a standardized questionnaire: It depends on your questionnaire also. if it is a standardized questionnaire (e.g., SF-36) you may consult the manual for missing data.

3) If you are able to contact those in the study, Are the questionnaire questions time sensitive: for instance, if it pertains to their mood within the past week up to the day they filled out the questions (b/c you can't ask them to "remember" how they "thought" they felt when they filled it out).

4) can this missing data be handled via a statistical technique? if so, this may be your best bet as to not confound the study.