Premium
Data approximation strategies between generalized line scales and the influence of labels and spacing
Author(s) -
Kershaw Jonathan C.,
Running Cordelia A.
Publication year - 2019
Publication title -
journal of sensory studies
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.61
H-Index - 53
eISSN - 1745-459X
pISSN - 0887-8250
DOI - 10.1111/joss.12507
Subject(s) - normalization (sociology) , generalized additive model , categorical variable , generalized linear model , generalized estimating equation , statistics , mathematics , scale (ratio) , comparability , sample size determination , generalized linear mixed model , econometrics , geography , cartography , combinatorics , sociology , anthropology
Comparing sensory data gathered using different line scales is challenging. We tested whether adding internal labels to a generalized visual analog scale (gVAS) would improve comparability to a typical generalized labeled magnitude scale (gLMS). Untrained participants evaluated cheeses using one of four randomly assigned scales. Normalization to a cross‐modal standard and/or two gLMS transformations were applied to the data. Response means and distributions were lower for the gLMS than the gVAS, but no difference in resolving power was detected. The presence of labels, with or without line markings, caused categorical‐like lumping of responses. Closer low‐end label spacing for gLMS increased influenced participants to mark near higher intensity labels when they were evaluating low‐intensity samples. Although normalization reduced differences between scales, neither transformation nor normalization was supported as appropriate gLMS/gVAS approximation strategies. This study supports previous observations that neither scale offers a systematic advantage and that participant usage differences limit direct scale comparisons. Practical applications Practitioners should exercise caution when comparing between gVAS and gLMS data, as neither normalization nor transformations make these equivalent. The value of qualitative information from internal labels generalized labeled magnitude scale (gLMS) and the expected intensity of samples should be considered when choosing a scale (generalized visual analog scale may be better for lower intensity samples, gLMS for high intensity). While all scales in this study provide valid information regarding sample intensity, the impact of scale on statistical assumptions, most notably the non‐normal distribution of residuals from gLMS data, should also be considered and corrected when necessary.