The proportion of variance in the ratings that is explained by the modified exponential
has increased from 0.40 in 1983 to 0.48 in 2003. Therefore, the development of this
statistic points in the same direction as the development of the function coefficients: an
increase in the degree of locational self-preference for the Netherlands as a whole.
Analysis of the function curves for entrepreneurs in the Randstad provinces leads to
results that are less consistent. The rating of distant places by entrepreneurs in the
Randstad has increased, which seems to support the assumption of decreasing self-
preference in that area. On the other hand, just like elsewhere, the starting value has
increased and the turning point has come closer to the firm’s location. The increase of
the starting value is considerably larger than the increase of the end value, so we may
conclude that even in the Randstad area, locational self-preference has actually become
stronger.
Patterns of thought
Studying average ratings, interesting as they may be, does not give much insight into
the differences between the individual survey questionnaires, or into the patterns that
these questionnaires have in common. Individual respondents tend to distinguish groups
of places that they give a common rating to. By calculating mean ratings, these patterns
are hidden from view. Factor analysis is a technique that lends itself well to revealing
patterns that are hidden in the data material.
Holvoet (1981) put the technique to good use in his analysis of the rating of locations in
Belgium. He had a group of economics students rate locations in Belgium as possible
sites for firms, and applied principal components analysis to the data. The three
components that he identified represent important and recognizable oppositions, namely
Flanders versus Wallonia, the old industrial areas along the rivers Meuse and Sambre
versus the rest of Belgium, and the opposition between the large agglomerations
Antwerp and Brussels versus the periphery. These oppositions can be seen as patterns of
thought that are apparently related to location factors.
In our research, we have applied factor analysis as well. For each of our surveys, the
respondents are treated as cases, and the variables are linked to the places subject to
rating. The type of factor analysis that we chose to apply is principal components
analysis with varimax rotation. To determine the optimal number of components to be
rotated, Dirkzwager’s (1966) hierarchy model was used. This model implies that
separate rotations are carried out on successively larger numbers of factors. In this