|
Checking for Reversals
(Otherwise
known as checking the directionality of construct relationships).
|
|
When
comparing the ratings on
any two sets of constructs in a Repertory
Grid, it is always necessary to remember that constructs are
bipolar.
The two sets of ratings may be highly related, but this relationship
may
not be apparent unless its directionality is examined.
It is easiest to proceed by example. Take
the following ratings on two simple constructs which describe a
person’s food preferences. Imagine that the individual has rated 7
different foods on each of the two constructs, using a 5-point scale in
each case. Follow the usual convention that the left-hand pole of each
construct anchors the “1” end of
the scale, and the right-hand end of each construct anchors the “5” end
of the scale.
Table 1
"1"
|
food 1
|
food 2
|
food 3
|
food 4
|
food 5
|
food 6
|
food 7
|
"5"
|
Nice
|
5
|
3
|
1
|
4
|
2
|
3
|
5
|
Nasty
|
Sweet
|
1
|
2
|
5
|
2
|
4
|
3
|
1
|
Savoury
|
Now compute a simple Sum of Differences
score,
taking the absolute difference between the ratings given to each
element
on the two constructs and summing across: (5-1) + (3-2) + (5-1)…
(absolute difference, remember),,, + (4-2) + (4-2) … (another absolute
difference)… + (3-3) + (5-1). Which equals 17. Rather a large sum of
differences, given that the maximum possible difference for 7 elements
rated on a 5-point scale is 7 x 4 = 28! One may be tempted to conclude
that the two constructs aren’t very highly related.
But constructs are bipolar! In the second
construct, the pole labelled “Sweet” is only written on the left,
anchoring the “1” end of the scale, because that is the way it was
written down when the construct was elicited, following the usual rule
that the emergent pole is written down on the left. Suppose, however,
that the emergent pole happened to be “Savoury”; in other words, that
in the triadic elicitation, “Savoury” happened to be the characteristic
which two of the three elements being compared had in common.
In which case, the second construct would
have
been written down as follows.
Table 2
"1"
|
food 1
|
food 2
|
food 3
|
food 4
|
food 5
|
food 6
|
food 7
|
"5"
|
Nice
|
5
|
3
|
1
|
4
|
2
|
3
|
5
|
Nasty
|
Savoury
|
5
|
3
|
1
|
4
|
2
|
3
|
5
|
Sweet
|
Look at the ratings on the second construct.
If food 1 was rated “1” on a scale that runs
from “Sweet = 1” to “Savoury = 5” as in Table 1, it should of course
show a rating of “5” when the scale runs from “Savoury = 1” to “Sweet =
5”: the food
is construed as Sweet regardless of the directionality of the scale!
Likewise, If food 4 received a rating of “2” on a scale that runs from
“Sweet = 1” to “Savoury = 5”, when the direction of the second
construct is reversed so that the scale runs from “Savoury = 1” to
“Sweet = 5”, it has to receive a rating of “4” if the meaning is to be
preserved. And similarly for all
the other element ratings.
And of course, once this is done, it is
obvious in Table 2 that the two constructs are in fact maximally
related. The ratings are identical with the poles written down this way
round. The interviewee thinks savoury foods are nice, and sweet foods
are nasty, consistently
throughout, and was saying all along; yet this is not at all obvious
from
Table 1. But Table 1 is only as it is because of the incidental event
during
elicitation that led to “Sweet” being written down on the left and
“Savoury”
on the right.
To identify the relationship between the
meanings being expressed by any two constructs, then, the similarity
between the two constructs must be checked “both ways”: with the pair
of constructs as
they are; and with the labels of one of the constructs reversed, with
its
ratings being reversed to preserve the intended meaning. The value of a
reversed rating is then given by subtracting the original rating from R
+ 1 where R is the maximum possible rating (5 on a 5-point scale, 7 on
a
7-point scale, and so on).
This example uses sums of differences as its
measure of similarity between constructs. The same rationale applies
where the correlation coefficient is used as the measure of similarity. |
|
Devi Jankowicz
|
|
|