Looking forward
Date
2015
Authors
Louviere, J.J.
Flynn, T.N.
Marley, A.A.J.
Editors
Louviere, J.J.
Flynn, T.N.
Marley, A.A.J.
Flynn, T.N.
Marley, A.A.J.
Advisors
Journal Title
Journal ISSN
Volume Title
Type:
Book chapter
Citation
Source details - Title: Best-worst scaling: theory, methods and applications, 2015 / Louviere, J.J., Flynn, T.N., Marley, A.A.J. (ed./s), Ch.6, pp.134-146
Statement of Responsibility
Conference Name
Abstract
This chapter is a positioning one, acknowledging the existing limitations of best-worst scaling (BWS), as a theory and/or as a method of data collection. As such, it should alert the reader to current areas of methodological research and provide a research agenda for the future. It touches upon innovative avenues of research that seek to better understand the psychological processes that might underpin random utility theory. Such work is of key importance in providing physiological and clinical justifications for the (statistical) random utility model. In many cases it refers to work that has begun in the field of health: this simply reflects the fact that stated preference data are both commonly used in that field and lacking in revealed preference counterparts to validate them. Thus, the reader should not conclude that the arguments are relevant only to health. Best versus worst Almost all applications of best-worst scaling in the preceding chapters (see Chapter 15 for an exception) and in the wider BWS literature have assumed, implicitly or explicitly, that the best and worst choices made by a particular individual reflect mirror image values (see Chapter 5 for specific meanings of this term); this enables data pooling (“stacking”) of the worst data below (or above) the best data, and (with a sign change) treating the worst data as just more best data. Effectively, this also is the method for the popular maxdiff model of best and worst choices (see, for example, Sawtooth Software [www.sawtoothsoftware.com] and Chapter 5). The maxdiff model assumes that individuals compare every pair of best-worst and worst-best outcomes, and choose the pair with the largest positive difference. Recent empirical studies that include both the best-worst choices made and the time to make them find that the maxdiff model gives excellent fits to the choices, but the response times show that no individual uses a maxdiff process (Hawkins et al., 2014a). The fact that few (if any) individuals use a maxdiff process raises interesting and important questions as to the nature of the true underlying cognitive process that individuals use to make best and worst choices; in other words, it is likely that different individuals make these choices in different ways
School/Discipline
Dissertation Note
Provenance
Description
Access Status
Rights
Copyright 2015 Cambridge University Press