I sent questionnaires to 1000 individuals and in response I received 528 questionnaires. Out of those 528 questionnaire, 12 questionnaires were discarded because they were plagued with consistent response bias, 8 questionnaires had more than 25% missing data so i decided to use listwise deletion, and rest of the 508 questionnaire were complete in all respects, what justification I should mention for using listwise deletion, as I am applying structural equation modeling in my study. Please guide
Your missing values are less than 10 percent of the data so you can take average to address this problem of missing values.
Regards
Rafique
I have been working with data from the IMSS survey for several years (http://www.manufacturingstrategy.net). It is common practice in our field to delete cases that have missing values on the variables of interest (listwise), especially if they are so few as in your case. I do not think you need to provide a justification, if asked, you can cite any previous survey based work that did that, for instance:
Golini, R., Longoni, A., Cagliano, R., 2014. Developing sustainability in global manufacturing networks: the role of site competence on sustainability performance. International Journal of Production Economics 147, Part B, 448-459.
For SEM, the more important issue is having enough observations to get stable estimates for your model. There are various recommendations about how many observations you should have for each "parameter" in your model -- most commonly either 5 or 10 people per item you estimate.
For the missing data, the amounts you describe are quite reasonable, so I would simply report them as is.
Thanks Sir Morgan for your valuable input, i was worried because some of the researcher e.g., Peugh & Enders (2004), Graham, 2009 have termed listwise deletion as one of worst methods but in my study I don't have any other option,
Thanks for your guidance again
Thanks Sir Zainudin, but excluding these 8 questionnaires which had more than 25 % missing values, the rest of the questionnaires are complete in all respects
Sir Alexander, thanks for your insightful post, it means that in my case I should not use listwise deletion, so what's is the solution, since 8 questionnaire had more than 25% missing values
I agree with the previous posters, especially with Alexander Roppelt. What you should really be concerned of here is the large number (almost 50%) of nonrespondents in your study, introducing the possibility of nonresponse bias. If those who do not participate in the study systematically differ from your study participants, your results are likely biased, i.e. the validity/generalizability is limited. A good option would be to have at least some information on the nonresponders (main study outcome and/or sociodemographic: age gender etc.) and to compare study participants and nonparticipants with respect to these data. If this is not possible, another option would be to compare early and late responders in your study. It is assumed that the late responders are more similar to nonparticipants than to participants. Hence, if you do not find differences in response patterns between early and late responders, this is a good indication of the absence of nonresponse bias in your data.
Also see
Groves, R., Fowler, F. J., Couper, M. P., Lepkowski, J. M., Singer, E., & Tourangeau, R. (2009). Survey methodology (2. ed.). Wiley series in survey methodology. Hoboken, NJ: Wiley.
Sir Alexander, Thank you very much for your detailed input, in fact it is a great help and a valuable guideline, thanks again
Dear Razia,
list wise deletion suffers from (1) a loss of efficiency of your estimation (=less statistical power) and, maybe, (2) an introduction of bias if missing patterns are not completely at random (MCAR). You could go for FIML (full Information maximum likelihood) or MI (multiple imputation) to exploit the data best instead of throwing it away by list wise deletion.
Practically, I think in your case (we are talking about 8 cases right?) that won't matter much (the 50% non-response is the bull in the china shop) but could be worth trying.
Best
Yes Razia. Throw away the non response questionnaire (more than 20% non response). Also throw away the "Straight Lining" response - the same response throughout, meaning the respondents were responding without reading your item statement.
This site offers good practical advice
http://www.missingdata.org.uk/
I like other correspondents are most worried about those that did not respond at all and you may have a biased sample. If you have any data on them consider a logit model of respond / not as the response and then any predictors that you have for both groups - do they systematically differ? Even urban/ rural of their address would be a start.
While there is good reason to worry about any substantive bias due to systematic non-response, it seems to me that a response rate of 50% on a mail out survey is certainly within current expectations -- and better than many examples in published work.
Don Dillman is a well recognized expert on response rates for mail surveys, so if you have any worries about your response rate, I would suggest consulting his work.
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Techniques
- Structural Equation Modeling