Personal Attribute Judgments - Modeling Users' Experiences with Interactive Systems - page 25

2.3.3 On the Measurement of Meaning

Due to the notion of bipolarity, semantic differential scales (Osgood et al., 1957) are
traditionally employed in RGT rating practices. In (Karapanos and Martens, 2008)
we argued for comparison scales (such as paired comparisons) as an alternative,
since such scales are known to be less sensitive to contextual effects than single-
stimulus scales, such as the semantic differentials. de Ridder (1996) explored this
issue in an experiment where quality assessments were made on a set of images.
The same set of images was embedded in a larger set of images with a positively or
a negatively skewed distribution in quality. The sensitivity of different scales (such
as single-stimulus and comparison scales) to these contextual effects was assessed.
It was found that contextual effects were negligible only when comparison scales
were used. Thus, in the context of the RGT, contextual effects such as individual
differences in prior experiences or order effects during the rating process will ev-
idently make individuals' judgments less reliable when single-stimulus scales are
employed.

2.4 Analyzing Personal Attribute Judgments - An Initial

Exploration

One of the advantages of personal attribute elicitation techniques such as the Reper-
tory Grid, i.e. their ability to provide rich insights into individual's perceptual and
cognitive processes, is also one of their dominant limitations as this complicates the
analysis phase.
Personal attribute elicitation techniques emphasize the idiosyncratic nature of
perception and evaluation of objects. In other words, individuals perceive interactive
products through different, individual “templates”. This in turn leads to diversity in
the obtained attributes and attribute ratings. Some people may use entirely different
attributes to evaluate a product, whereas others may use the same attributes but apply
them differently. Researchers are then confronted with as many idiosyncratic views
as participants. In practice, the consequence is either an idiosyncratic analysis with
a “narrative” summarization (e.g. Hassenzahl and Trautmann, 2001) or the use of
average models.
Idiosyncratic approaches often discard the quantitative structure of elicited at-
tributes and utilize only their semantic information. The researcher attempts to
summarize the semantically distinct attributes and to qualitatively identify relations
among them. Techniques such as laddering and pyramiding, which probe the par-
ticipant to further elaborate on the elicited attribute, provide rich qualitative insight
into the elicited attribute. However, as these approaches do not properly utilize the
quantitative structure of attributes, they render the Repertory Grid as yet another
qualitative technique, and thus undermine its true value which is in quantifying rich
qualitative insights.
Quantitative analysis procedures typically rely on exploratory multivariate tech-
niques such as Principal Components Analysis (PCA) and Multi-Dimensional