|
|
|
|
|
|
|
|
A NOTE ON ALIGNING CONSTRUCTS
|
|
|
Richard C. Bell |
|
|
Psychological Sciences, University of Melbourne, Australia
|
|
|
|
|
|
|
Abstract
The
alignment of constructs (with respect to the location of the preferred pole as
consistently to the right or left) is a problem for both the analysis of grids
and some further investigations based on grids. It is suggested here the sign
of loadings on the first principal component can be used to identify constructs
that need to be reflected or reversed in order for there to be a consistent
alignment of poles on constructs.
Keywords:
repertory grids, alignments of constructs
|
|
|
|
|
|
|
|
INTRODUCTION
The nature of the poles defining constructs
in personal construct psychology is important for a number of reasons. The
concept of the ‘preferred’ pole has implications for the techniques of laddering
and resistance-to-change grids (see Fransella, Bell, and Bannister, 2004,
pp.65-73) as well as a more general use in therapeutic intervention. The
alignment of constructs (irrespective of the values attached to the poles) can
affect correlations between elements in grid data, as shown by McKay (1992).
While a “work-around” is available for element correlations (Bell, 2006) it is
not known how this issue might affect other indices as well as representations
of grid data. McKay’s (1992) demonstration of the problem was focussed on
identification, however he also indicated a number of other areas of grid
analysis that were affected by the orientation of the construct poles. These
included the construct intraclass correlation [used as a measure of cognitive simplicity-complexity],
Landfield’s measures of ordination and functionally independent constructs
[FIC], and some principal component analyses. Some implications are relatively
trivial, such as the reversal of signs on construct loadings, but other changes
are more substantial. For example, the problem affects cluster analysis
representations, such as are implemented in Webgrid5 (Shaw & Gaines, 2010).
Figure 1 below shows construct clustering (using SPSS) in grids E & F from
McKay (1992, p.67, Figure 3), where Grid F is identical to Grid E but for three
constructs reversed.
Figure 1: Effect of construct reversal on
construct cluster solutions.
Slow/Quick and Quiet/Talkative are both
reversed in Grid F and retain their link, however the reversal of Calm/Emotional
(to Emotional/Calm) enables a link to be formed with the distant (in Grid E)
construct of Caring/Selfish.
The configurations from correspondence
analysis [here from SPSS, but also the primary grid representation method in
Gridcor (Feixas & Cornejo, 2004)] are shown in Figure 2 and also show
substantive differences in the configurations when some constructs are
reversed.
Figure 2: Effect of construct reversal on
correspondence analysis solutions.
Such differences will also be found in
singular-value-decomposition representations where the grid data is
double-centred (to remove mean effects) and in unfolding solutions, such as
that shown by Leach, Freshwater & Aldridge (2001). The problem is thus not
a trivial one, and requires a solution.
SOLUTIONS
In individual instances it is possible to
align the grids simply by asking the respondent to indicate the preferred pole.
In some cases, particularly with grids set in less personal domains, the
preferred pole may not be readily identified by the respondent. In research
settings with grids collected in a less individually supervised fashion
respondent input may not be possible. Another approach might be to use some
desirable figure, such as ideal self, to indicate the preferred pole. Again, in
less personal settings, there may be no such figure, and indeed even in
clinical settings it may well be that the ideal self is not aligned with either
pole of a construct (see Winter, Bell & Watson, in press).
Particularly in research settings where
multiple grids are involved, it might be valuable to have an approach by which
constructs in different grids can be similarly aligned. Information on the
alignment of constructs is provided by the construct correlations. However such
information is often difficult to interpret. This is illustrated by
consideration of an example, a grid from the study by Haritos, Gindidis, Doan,
and Bell (2004) in which there is no Ideal Self figure to anchor construct
poles
(Figure 3).
Figure 3: The Grid
Although some constructs may seem unaligned
with others [eg relaxed - worried & tense or accept it as it is – loves to
argue go against what appears to be the general trend of negative poles on the
left] it is not obvious how constructs should be aligned. The correlations
among constructs are shown in Figure 4[1].
Figure 4: Construct inter-correlations.
The general trend can be identified by
examining the discrepant signs of loadings on the first principal component.
This can be used to identify constructs that should be reflected [pole labels
and ratings reversed]. The construct loadings on the first principal component
are shown in Figure 5.
Figure 5: Construct loadings on the first
principal component.
Five of the loadings have positive signs
(although a couple are quite small) and four have negative signs. If we reflect
those with negative signs (as shown in Figure 5) we obtain the correlation
matrix shown in Figure 7. The reversals are as shown in Figure 6.
Figure 6: Reversal of negative loading
constructs.
Figure 7: Construct inter-correlations with
some constructs reversed.
The original correlation matrix (Figure 4)
had 18 pairs of constructs with negative correlations with eight being greater
than -0.40. The correlations of reflected constructs show six negative correlations,
the greatest being -0.29. Simply trying to identify constructs that need to be
reflected by negative correlations would not ensure that in general constructs
are positively correlated as the first principal component of this matrix (with
these few negative correlations) shows as in Figure 8.
Figure 8: Construct loadings (with some
constructs reversed) on the first principal component.
CONCLUSIONS
The problem of the effect of the alignment
of construct poles on grid statistics has never been dealt with in a
comprehensive way. It has been shown to have an effect on grid structure and
statistics (Bell, 2006; McKay, 1992) and while some solutions have been
proposed (eg restricting element comparisons to distances (McKay,1992), or
computing construct-invariant element correlations (Bell, 2006)] these have
been restricted to the context of element comparisons and there has been no
comprehensive solution.
The problem is analogous to the problem in
factor analysis where the initial factor extraction does not result in a unique
solution. There the problem has been solved (so to speak) by adopting the
simple structure criterion of Thurstone (1945) and rotating the solution to
best approximate that. Here it is suggested that grid analysis adopt an analogous
criterion, that all constructs be aligned so as to have a similar sign in
loadings on the first principal component. While this will not impact on some
representations of grids (apart from reversing poles) and some indices such as
intensity or PVAFF from construct correlations, it does not distort the grid
data (since construct orientation is arbitrary) and might as well be routinely
applied.
This and
subsequent figures are taken from the output of the current version of Gridstat
(Bell, 2009) which contains an automatic detection of misaligned constructs and
the option to automatically reverse them as indicated in this paper.
| |
|
|
|
|
REFERENCES |
|
|
Bell, R.C. (2006) A note on the correlation
of elements in repertory grids: How to and why. Journal of Constructivist
Psychology, 19, 273-279.
Bell, R. C. Gridstat: A program for
analyzing the data of a repertory grid (Computer software version 5. 0).
Melbourne: Psychological Sciences, University of Melbourne, 2009
Feixas, G., & Cornejo-Alvarez, J. M.
(2004). Gridcor (Version 4.0) [Computer software and manual]. Retrieved from
http://www.terapiacognitiva.net/record/pag /index.htm.
Fransella, F., Bell, R., & Bannister,
D. (2004). A Manual for Repertory Grid Technique. Second Edition Chichester:
Wiley.
Haritos, A., Gindidis, A., Doan, C., &
Bell, R.C. (2004) The effect of element role titles on construct structure and
content. Journal of Constructivist Psychology, 17, 221-236.
Leach, C., Freshwater, K., Aldridge, J.
& Sunderland, J. (2001). Analysis of repertory grids in clinical practice. British
Journal of Clinical Psychology, 40, 225-248
Mackay, N. (1992) Identification,
reflection, and correlation: Problems in the bases of repertory grid measures. International
Journal of Personal Construct Psychology, 5, 57-75.
Shaw, M. G., & Gaines, B. R. (2010). WebGrid5
[Computer software]. Retrieved from http://gigi.cpsc.ucalgary.ca:2000/
Thurstone, L.L. (1945) Multiple factor
analysis. Chicago: University of Chicago Press.
Winter, D.A., Bell, R.C., & Watson,
S.B. (in press) Midpoint ratings on personal constructs: constriction or the
middle way? Journal of Constructivist Psychology.
| |
|
|
|
|
ABOUT
THE
AUTHOR
Richard Bell is an Associate Professor in the School of Behavioural Science at the
University of Melbourne. He has a slight interest in analysing the data of
repertory grids. Email: rcbell@unimelb.edu.au
Web:
http://www.psych.unimelb.edu.au/people/staff/BellR.html
Address for correspondence: Assoc Prof. Richard C. Bell Psychological Sciences University of Melbourne Vic 3010, Australia
|
|
|
|
|
|
REFERENCE
Bell, R. C. (2010). A note on aligning constructs. Personal
Construct Theory & Practice, 7, 42-48, 2010
(Retrieved from http://www.pcp-net.org/journal/pctp10/bell10.html)
|
|
|
|
|
|
Received: 5 June 2010 – Accepted: 29 July 2010 –
Published: 7 August 2010 |
|
|
|
|
|