Questionnaire design considerations with planned missing data
Year: 2009
Type: article
Abstract: One of the unanswered questions that emerge from planned missing data designs is the appropriate distribution of the missing data we plan to introduce by design.All decisions concerning what data not to collect to decrease survey length have to be carefully considered.With the availability of computer-assisted survey techniques, it is relatively straightforward to tell the computer to leave some questions out based on randomization or other criteria.Unfortunately, a computer-assisted design is not always available.Sometimes we are still forced to use the good oldfashioned paper and pencil.In these situations, it makes a difference if we have to design, distribute and code 3 versions of a survey, 15 versions of a survey, or 225 versions of a survey.Conventional wisdom suggests that we should strive for the utmost uniformity of missing information.Unfortunately, for reasons described below in greater detail, this also means the use of many different questionnaire versions.The analysis presented here suggests, counterintuitively, that there is no practical analytical advantage in striving for a great uniformity of missing information by maximizing the number of questionnaire versions.Three versions of a paper and pencil questionnaire is just fine.Following the principles laid out by King (1995), scripts used to conduct the analyses are provided in the online appendix downloadable at: http://levente.littvay.hu/reviewofpsychology/appendix.pdfWith these (and possibly through minor modifications of these when necessary), any researcher has the ability to replicate the results of the analysis presented.Providing all the information for replication is a norm that needs to gain more ground and acceptance in all of the sciences.This information could also provide a foundation for running similar tests for different analytical contexts. Missing dataA planned missing data design is one where the missing data is purposely built into the research design by the researcher.Given that missing data poses a problem in almost every field of social science, it might be difficult to imagine circumstances where the researcher would like to do that on purpose.Thus far, missing data has been seen as a problem.Only recently have researchers begun to think about missing data as a means to an end, a tool they can utilize to improve data quality (Graham, Hofer, & MacKinnon, 1996;Graham, Taylor, Olchowski, & Cumsille, 2006).The key to recognizing that planned missing data can be a blessing, not only a curse, is the missing data problem's foundations.The reason missing data is unwanted is that it rarely follows a mechanism that makes it quantitatively manageable.
Author: Levente Littvay
Institution: Central European University
Language: English
Subfield: Sociology and Political Science
Field: Social Sciences
Domain: Social Sciences
Open Access status: green
Funder: Central European University
OpenAlex ID: W2185277759
