|
|
|
|
|
|
|
|
DID
HINKLE PROVE LADDERED CONSTRUCTS ARE SUPERORDINATE?
A RE-EXAMINATION OF HIS
DATA SUGGESTS NOT
|
|
|
Richard C. Bell
|
|
|
School of Psychological Sciences, University of Melbourne, Melbourne, Australia
|
|
|
|
|
|
|
Abstract
Laddering is
an construct elicitation technique that purports to obtain superordinate
constructs. Hinkle, who invented the laddering technique, used an implication
grid procedure to test whether the constructs elicited by laddering were
superordinate to those used to initiate the laddering procedure. He concluded
that laddering did indeed produce superordinate constructs. However the analysis on which this conclusion
was based was at the broadest level, including all implicative relationships. No
other studies have addressed this issue in this fashion. A re-examination of
Hinkle’s data focussing on just those implicative relationships between the initiating
(subordinate) and the laddered (superordinate) constructs showed each was as
likely to imply the other as the reverse. It is concluded that Hinkle’s data
did not provide support for the superordinacy of laddered constructs and an
appropriate model to describe the relationship between initiating and laddered
constructs remains to be developed.
Keywords:Laddering,
superordinate-subordinate relationships, implication grids
|
|
|
|
|
|
|
|
Laddering
is the most widely used personal construct psychology technique after repertory
grids. It is generally accepted that the laddering process generates
superordinate constructs, although there are views that question this (Butt,
1995). However Hinkle (1965) appears to be the only person in the personal
construct domain who has formally tested whether or not laddered constructs are
superordinate by considering the implicative relationship between constructs.
Hinkle tested his laddered constructs together with the constructs used to
initiate the laddering in an implication grid format and simply contrasted the
implications data between the original and laddered constructs.
Hinkle
also published the 28 implications grids as an appendix to his thesis. There
were 10 generating constructs (derived from repertory grid triadic elicitation)
and 10 constructs laddered from the generating constructs. Hinkle had found in
pilot work that subjects could generate 8 to 12 superordinate constructs for
any basic construct. He also reported that "the chain of superordinate
constructs in the hierarchy generated from the first subordinate was almost
invariably repeated in the hierarchies of the remaining subordinates".
In his
main study of implication grids he did not identify the level at which these
latter constructs were laddered or which generating constructs generated which
laddered constructs. He recorded an 'x' in the column of the implying construct
at the row of the implied construct and left the symmetric matrix cell (implied
construct column and implying construct row) blank. Unrelated constructs were
blank in both (row-column and column-row) cells. Where two constructs mutually
implied each other, an 'r' was recorded in both cells (row-column and
column-row).
The
matrix typed up was a 20 x 20 square matrix containing 380 'live cells' as the
diagonal represented constructs compared with themselves which was not
relevant. However the information in this grid could be represented by a 190
cell half grid (e.g. in lower triangular form) showing each row against each
column as a cell. These cells could contain four kinds of information.
i.
ii.
iii.
iv. |
Null
– no relationship between row and column
Row
implies column
Column
implies row
Row
and column imply each other.
|
Hinkle
only tested (and only could test) distinctions between generating and laddered
constructs. Thus the above four-fold classification was relevant to
distinctions between generating and laddered constructs. Within the subsets of
generating and laddered constructs there was no useful distinction between row
implies column and column implies row so there was a three-fold classification
within those constructs showing each row against each column as a cell. These
cells could contain three kinds of information.
i.
ii.
iii. |
Null
– no relationship between row and column
Row
implies column or column implies row
Row
and column imply each other.
|
Hence
Hinkle's data had 10 scores of interest:
i.
ii.
iii.
iv
v.
vi.
vii. viii. ix.
x.
|
Null
– no relationship between generating row and column
Row
implies column or column implies row for generating constructs.
Row
and column imply each other for generating constructs.
Null
– no relationship between row (generating) and column (laddered) constructs
Row
(generating) implies column (laddered)
Column
(laddered) implies row (generating)
Row
(generating) and column (laddered) imply each other
Null
– no relationship between laddered row and column
Row
implies column or column implies row for laddered constructs.
Row
and column imply each other for laddered constructs.
|
Data
were transcribed from Hinkle's appendices and means and standard deviations
calculated for the ten scores. These are shown in Table 1. Standard deviations
are substantial indicating that the implication relationships varied
substantially among the 28 respondents.
Table 1.
Average relative occurrence of kinds of relationships in Hinkle’s 28 implication
grids (as means and standard deviations)
|
|
Mean
|
Std. Dev.
|
i.
|
Null –
no relationship between generating row and column
|
26.9
|
8.14
|
ii.
|
Row
implies column or column implies row for generating constructs.
|
11.6
|
5.28
|
iii.
|
Row and
column imply each other for generating constructs.
|
6.5
|
5.39
|
iv.
|
Null –
no relationship between row (generating) and column (laddered) constructs
|
46.8
|
16.82
|
v.
|
Row
(generating) implies column (laddered)
|
16.9
|
8.03
|
vi.
|
Column
(laddered) implies row
(generating)
|
17.1
|
10.06
|
vii.
|
Row
(generating) and column
(laddered) imply each other
|
19.2
|
14.11
|
viii.
|
Null –
no relationship between laddered row and column
|
10.3
|
9.40
|
ix.
|
Row
implies column or column implies row for laddered constructs.
|
17.9
|
7.50
|
x.
|
Row and
column imply each other for laddered constructs.
|
16.8
|
10.15
|
Total
|
|
190
|
|
Overall
there were three types of relationship, null or no relationship, one-way
implication, and reciprocal implication. As might be expected, null
relationships were the most common (43.4%), followed by one-way implications
(33.5%). And more common than might be expected, reciprocal relationships
accounted for 23.1%. Hinkle introduced reciprocal relationships in his general
discussion or relationship types but did not refer to this kind of relationship
as part of his hypotheses. In his discussion (point 20) of the results he
suggested that perhaps reciprocal relationships could be examined separately by
factor analysis since they were symmetric data (like correlation coefficients).
In fact, single reciprocal relationships provide no information as to the
super-ordinate / subordinate relationship between constructs. It should be
noted however that subsets of symmetric relationships can be used to infer a
direction of relationship. This is common in applications of path analysis and
in fact is incorporated into the determination of superordinate / subordinate
construct relationships by the Caputi, Breiger and Pattison (1990) method.
The only
information relevant to the task of determining the superordinate and
subordinate status of laddered versus generating constructs are the mean scores
for rows v. and vi. in Table 1. These are not significantly different (t =
0.08, df = 27, p>.05) indicating laddered constructs are no more likely to
be implied by the generating constructs than the converse. However there are
other differences between generating and laddered constructs that tell us
something. Generating constructs are less likely to be related to other
generating constructs (14.1%) than laddered constructs are to each other
(5.4%). This difference is significant (t = 3.71, df = 27, p<.001). Both
one-way and reciprocal relationships were significantly more common among
laddered constructs than generating constructs ( One-way: t = 3.7, df = 27,
p>.001; Reciprocal: t = 4.63, df = 27, p=<.001) although for laddered
constructs there was no difference between one-way (9.4%) and reciprocal
implications (8.8%), t = 0.39, df = 27, p>.05, while for generating
constructs, there were more one-way implications (6.1%) than reciprocal ones
(3.4%), t = 6.40, df = 27, p>.05.
It would
seem thus that laddered constructs are different in a way to constructs
elicited in the traditional triadic element fashion. They appear to be more closely
interrelated than the triadically elicited constructs. This is not surprising
given Hinkle's observation that there was a tendency for the same laddered
constructs to emerge from different generating constructs. However it does
suggest that laddered constructs will be
less cognitively complex since they will be more highly correlated.
Why did
Hinkle not find these results? There are a number of reasons. He used a simple
chi-square test on the sum of implications. Such a test did not take into
account that these sums were not totally independent because reciprocal implications between
laddered and generating constructs were counted for both laddered and
generating constructs. In addition by simply counting the whole columns and
rows of the implications grid he conflated within group (i.e., within laddered
or within generating) implications with between
group (i.e., laddered versus generating) implications.
In the
introduction to this paper, it was noted that Hinkle was the only person in the
Personal Construct domain who has formally tested whether or not laddered
constructs are superordinate. While this is true of the Personal Construct domain,
elsewhere the hierarchical assumption of laddering has been tested with an
implication task (van Rekom & Wierenga, 2007). As found in the present analysis
, they reported a substantial number of reciprocal implications between
subordinate constructs and superordinate constructs. They were investigating
the assumption in a business context and referred to the basic generating
constructs as 'concrete attributes' and the higher order constructs as
'abstract values'. This is a common distinction in the business applications of
laddering. In the Personal Construct Psychology domain, it has been suggested
that the presence of abstract constructs in those laddered is evidence for their superordinate status
(Hardison & Neimeyer, 2007; Neimeyer, Anderson & Stockton, 2001). The
results of van Rekom & Wierenga (2007) did not support this view. And as
noted by Crockett (1982, p.66), implicative relationships need not be
hierarchical. As an alternative model, van Rekom & Wierenga suggested
a network modelling approach which could represent reciprocal relationships as
well as the purely uni-directional relationships of hierarchical modelling.
They concluded that a better representation of more important concepts (core
constructs in a personal construct framework) was as more central (i.e. with
more links to other constructs) rather than more hierarchical. Hays (1958)
adopted a similar viewpoint. In an implications modelling framework, he
suggested that more important attributes were more central.
In
conclusion it seems likely that laddering of constructs produces more abstract
values as constructs, although this has not been examined in Hinkle’s data. On
the other hand it seems unlikely that the relationships between laddered
constructs and those used to initiate laddering is hierarchical in nature, and
that an appropriate model for these relationships is yet to be identified.
Postscript:
During the time this
article was in the editorial process, Dennis Hinkle died. There will be a
number of perspectives on his contribution over the following months and
longer, and they all will hopefully differ, since his contribution is more multi-faceted than one might expect from a
quick read of secondary sources. For me, Hinkle’s contributions were twofold.
One was to recognise and try to measure change, the other was to acknowledge
the importance of the major corollary dealing with the relationships between
constructs, the organiation corollary. This was a crucial corollary since you
can’t have a construct system without saying how the components might relate to
one another. We have not succeeded in finding a way of capturing change.
Hinkle’s resistance to change grid was time-consuming and has never been taken
up but alas neither has it spurred others into modifying it. And unfortunately
for Hinkle, the hierarchical organisation of constructs devised by Kelly is
problematic. So his ingenious developments to capture this were bound to encounter
problems. Hinkle was more aware of these problems than many subsequent accounts
indicate. He speculated on the possibility of laddering from the perspective of
another in addition to the self. He was aware of the issue of reciprocal
relationships in his implications grids and the need for such data to be
modelled differently (he suggested factor analysis). Perhaps if he had
continued in research he would have further explored these issues. But in
reading his thesis I got the impression that his subjects were more interesting
than the research questions, and, as for many therapists, the thesis was
primarily a diversion to be dealt with on his way to practice. We are fortunate
he was diverted.
|
|
|
|
|
|
REFERENCES |
|
|
Butt, T.
(1995) Ordinal relationships between constructs. Journal of Constructivist Psychology, 8, 227-236.
Caputi, P., Breiger, R., & Pattison, P. (1990). Analyzing implications grids
using hierarchical models. International
Journal of Personal Construct Psychology, 3, 77-90.
Crockett,
W.H. (1982). The organization of construct systems: The organization corollary.
In J.C. Mancuso and J.R. Adams-Webber (Eds.) The construing person. (pp.
62-95) New York: Praeger.
Hardison,
H. G., &
Neimeyer, R.A. (2007). Numbers and narratives:
Quantitative and qualitative convergence across constructivist assessments. Journal
of Constructivist Psychology, 20, 285-308.
Hays, W.L. (1958). An approach to the study of trait
implication and trait similarity. In R. Tagiuri and L. Petrullo (Eds.) Person perception and interpersonal behavior. (pp.
289-299) Stanford, CA: Stanford University Press.
Hinkle, D.N. (1965). The change of personal
constructs from the viewpoint of a theory of construct implications. Unpublished
PhD thesis, Ohio State University. Retrieved from http://www.pcp-net.org/journal/pctp10/hinkle1965.html. (Retrieved 24 September 2013)
Neimeyer, R.A., Anderson, A., & Stockton, L.
(2001). Snakes versus ladders: A
validation of laddering technique as a measure of hierarchical structure. Journal
of Constructivist Psychology, 14,
85-105.
van Rekom, J. & Wierenga, B. (2007). On the hierarchical nature of
means–end relationships in laddering data. Journal
of Business Research 60, 401–410.
|
|
|
|
|
|
ABOUT
THE
AUTHOR
|
|
|
Richard
Bell is a Principal Research Fellow in the Melbourne School of Psychological
Sciences at the University of Melbourne. He has published widely in the
analysis of repertory grid data.
Email: rcbell12345@gmail.com
|
|
|
|
|
|
REFERENCE
Bell, R. C. (2014). Did Hinkle prove laddered
constructs are superordinate? A re-examination of his data suggests not.
Personal Construct Theory & Practice, 11, 1-5, 2014
(Retrieved from http://www.pcp-net.org/journal/pctp14/bell14.html)
|
|
|
|
|
|
Received: 20 November 2013 – Accepted: 24 December 2013 –
Published: 27 March 2014
|
|
|
|
|
|