Respondent’s Answering Time for Question as a Factor of Data Quality Estimation
DOI:
https://doi.org/10.29038/2306-3971-2017-01-36-43Keywords:
median equality test, median confidence inerval, paradata, vignette, online survey, timestampsAbstract
Computer assisted tools for survey conduction enabled a record a lot of additional information, such as questions answering time as an additional factor to assess data quality. Within the article, the author offers an approach to evaluate the quality of data by evaluating thresholds of temporal characteristics of responses to vignettes in an online poll created with factorial design in program R. We used for this answers to a series of five vignettes describing complex experimental situations and minimal their reading in at least 10 seconds, obtained in a solid survey of students of sociology faculty, Taras Shevchenko National University of Kyiv in on-line shell LimeSurvey (2015).
Construction and analysis of the confidence interval for the sample or the confidence interval for median author used as statistical method for eliminating the outliers. The difference between the model constructed for a specified time interval, and total full model is not too large. Significant difference between the distribution time is observed only between 1st and all other vignettes: vignettes are homogeneous, so the time spent on building understanding experimental situation only for 1st vignette, and everyone else has seen «by analogy». Fullness of time intervals by gender shows that the proportion of women who gave a quick response, with every vignette is reduced, man, on the contrary – are increasing. Perhaps men are less closely to the tasks they become more bored to read vignettes with similar experimental situations.
Because the answer time of question depend of a lot of different factors – the structure and complexity of the questions, the personal qualities of the respondent, etc. – require additional research, including using regression analysis.