www.frederikfunke.de

Improving Data Quality in Web Surveys with Visual Analogue Scales
Frederik Funke & Ulf-Dietrich Reips

Paper presented at the 2^nd conference of the European Survey Research Association (ESRA)
June 25-29, 2007 in Prague (CZ)

Slides

[PDF - 570 KB]

Abstract
 
A Visual Analogue Scale (VAS) is a graphical rating scale, a plain horizontal line with both ends anchored. Respondents indicate the extent of a subjective impression by placing a mark on the line. VAS lead a shadowy existence. This is mainly because of two reasons. First, VAS are used rarely in paper and pencil studies because of the extra work required in measuring the respondents' markings. Second, only little is known about the impact on data quality and the measurement error when using VAS.
 
With computerized surveys, the first reason for not using VAS does not longer apply. Read out happens automatically, fast and accurate. Using Web surveys the surveying of large samples with VAS is feasible. To explore the impact on data quality and the measurement error of VAS, we conducted four Web experiments, evaluating the quality of data obtained with VAS (also see Reips & Funke, in press).
 
In the first experiment, we were able to show that data from VAS meet the criteria of an interval scale, e.g. equidistance. Thus, a range of important statistical procedures may safely be applied to data from VAS (e.g. the computation of means, parametric procedures like the t-test, Pearson correlation and linear regression models).
 
In experiments two and three, we compared different categorical (Likert type) scales to VAS. Our main finding is: Categorical scales do differ systematically from interval level, producing data on an ordinal level only. Therefore, to compare data from VAS with categorical scales, a special transformation is needed that will be described in the presentation.
 
Experiment four examined retest reliability of VAS and categorical scales. A 40 item personality inventory was repeatedly administered. Result: Although VAS facilitate a far more precise judgment, there was no negative influence on retest reliability. The implications of this series of experiments will be discussed, particularly for using VAS as a way of collecting data on the Web.