3. Importance measuresThe second major difficulty facing IPA, and the focus ofthis paper, is the measurement of importance and performance.In this regard, performance has generally been less controversialthan importance: the usual measurement procedure hasbeen to take the mean of the performance ratings obtainedfrom an appropriate group of people by means of a metric orLikert scale. However, a variety of means exist to performimportance measurement. In particular, two quite differentkinds of measure are common in IPA applications: 1) directmeasures based on Likert scale, k-point scale or metric ratingsobtained in the same way as for performance; and 2) indirectmeasures obtained from the performance ratings, either bymultivariate regression of an overall product or service ratingon the ratings given to the individual attributes (Danaher andMattsson, 1994; Dolinsky, 1991; Neslin, 1981; Taylor, 1997;Wittink and Bayer, 1994) or by means of conjoint analysistechniques (Danaher, 1997; DeSarbo et al., 1994; Ostrom andIacobucci, 1995).Although a recent review of these methods (Bacon, 2003)supports earlier studies (e.g. Alpert, 1971; Heeler et al., 1979)in finding that direct measures capture the importance ofattributes better than indirect measures, the former have seriousdrawbacks. In particular, Bacon (2003) himself emphasizesthat direct importance assessment is often misleading becauseratings are uniformly high. The main source of this problem isinherent to Martilla and James' procedure, in which the firststep should be to identify themost salient attributes of a productor service by qualitative studies (focus groups and/orunstructured personal interviews) or by reviewing previousresearch. This procedure has a natural tendency to record highimportance ratings on a metric or Likert scale for all theattributes selected for evaluation, with the result that they allcrowd together at the top of the IPA grid. This defeats thepurpose of the exercise, since the objective of the IPA diagramis not to act simply as a record of absolute importance andperformance values, but to discriminate among attributes as