7.3.1 Rationale

К оглавлению
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 
17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 
34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 
51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 
68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 
85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 
102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 
119 120 121 122 123 124 125 126 127 128 129 130 131 

However, there is a technique, first developed by Honey (1979), which

achieves what, at first sight, appears to be impossible. It aggregates different

constructs across a sample and provides a way in which we can make use of

some of the individual meanings being conveyed by each person’s ratings.

The technique assumes that what we’re interested in is each individual’s

personal understanding of the topic in question, and treats each construct

offered by the individual as more closely related, or less closely related, to the

overall issue s/he has in mind when thinking about the topic.

Thismakes a lot of sense.Kelly’s theory asserts that constructs are organised into a

system, with some constructs being superordinate to others (the Organisation

Corollary), and, you’ll recall, we make use of this property when we seek to elicit

more specific constructs by laddering downwards (see Section 4.4.1). Some

constructs are crucial and central to the individual’s knowledge and views about the

topic, while others, while relevant, are somewhat more peripheral. Moreover, we

expect some kind of consistency among the constructs a person uses. So, for

example, if an interviewee were to characterise an individual as‘unreliable’and at the

same time ‘dependable’, we would be surprised, and look for a rationale.We expect

constructs to hang together, to express a coherent point of view, and to tell a

consistent story.

But Kelly also mentions that the different ways we have of making sense of different

situations may not necessarily be compatible with each another (Fragmentation

Corollary). A person’s construct system isn’t a monolithic structure. And so it is very

reasonable to expect that some constructs are particularly related to any issue

being construed, carrying more meaning about the topic in question. Others, while

relevant, might flesh out the meaning, adding grace notes, as it were, without being

‘what the whole thingis reallyall about’.

In Honey’s technique, ‘what the individual has in mind when thinking about

the topic’ is summarised by supplying a construct: an ‘overall summary’

construct designed to sum up the interviewee’s individual stance to the topic

as a whole. (See Section 4.2.7 on supplied constructs and their uses. Refresh

your memory now, before reading on!)

Thus, a grid designed to capture the reactions of a set of trainees to the

different sessions of a course which they attended, in order to make

improvements in the course, might involve a supplied construct which

assessed the extent to which they felt they learnt: ‘overall, learnt a lot – overall,

didn’t learn much’. A grid which assessed the knowledge in a company’s sales

team about which approaches worked best in selling to their customers might

conclude by asking each salesperson to rate each approach on ‘overall, more

effective – overall, less effective’. A grid about personal friendship using the

names of friends and acquaintances as the elements might require the

interviewee to rate all of them on the supplied construct, ‘overall, a closer

relationship – overall, a more distant relationship’. Table 7.8 provides some

further examples.

Honey’s content analysis aggregates constructs across the sample, as in Section

7.2.1, but labels each construct with two indices reflecting the extent to which

the ratings on the particular construct match the ratings on the ‘overall’

construct. The first index is our familiar % similarity score, as outlined in

Section 6.1.2. One hundred per cent means that the ratings on the individual

construct are identical to the ratings on the overall construct; 70% means that

the ratings are less similar, and so on.

The second index reflects the fact that people differ in their typical % similarity

scores. When we compare the ratings elicited from one interviewee on all of

the constructs in the grid, we may find that they typically reveal a relatively

narrow range of % similarity scores. The highest may be 100% and the lowest

perhaps 80%. A second person, however, might be inclined to see many

different, and relatively unrelated, aspects of the issue when thinking about

the topic, and so her highest % similarity score may be 85% and the lowest,

55%. To the extent that different people have different ranges of % similarity

scores for any topic, we can talk of different personal metrics. At 85%, what

feels very similar for the second person may lie towards the bottom of the

range of similarity for the first person.

And so, Honey’s procedure acknowledges that % similarity scores are relative,

and as well as noting their actual percentage value, notes whether that value is

placed among the high, the intermediate or the low (H-I-L) values (what

Honey calls the ‘top-and-tail data’ for that particular individual.

Okay. So, taking both indices (% similarity and the H-I-L index) into account

(see below for the detail), it’s clear that some constructs will match highly with

the supplied ‘overall’ construct. In other words, they will represent what that

particular individual felt and thought, overall, very well. Others, however,

will match less highly with the ‘overall’ construct; that is, they will represent

what the individual felt and thought about the topic, overall, somewhat less


The aggregated set of constructs for the sample as a whole will, in other words,

represent the categorised views of all the individuals in the sample, but will

also preserve information about each individual’s views in terms of how he or

she severally, personally, idiosyncratically if you will, thought about the topic.

Table 7.8 Examples of ‘overall’ constructs in Honey’s content-analysis technique

Topic Elements

Possible qualifying phrase

‘. . . from the point of view of . . .’

Supplied construct

‘overall . . .’

Lecturer effectiveness Lecturers, rated by a sample

of students

. . . What they do that makes

them more, or less, effective

as lecturers

. . . More effective – less


Counselling style Different types of ‘helper’:

‘a friend’, ‘a counsellor’,

‘a parent’, ‘a priest’

. . . What they do and how they

do it that makes them more,

or less, approachable to a

person seeking help

. . . More approachable – less


Being an effective clerical


Different clerical officers,

including some more

effective and some less

. . .How they do their job

which makes them good

at it, or less good at it

. . . More effective clerical

officer – less effective clerical


Understanding creative block

in the figurative arts

Several of their own paintings

as nominated by each artist

. . .What was going on which

made them easy to paint,

or led to getting ‘stuck’

. . . More straightforward –

more problematic

Thinking about what

friendship means to you

People you know, ranging

from ‘best friend’ to

‘disliked acquaintance’

. . .What it is about them that

makes for friendship – or


. . .A real friend – rather less so

1. In each case, your analysis will identify the constructs, in their categories, which each interviewee has particularly in mind when thinking about the

issue involved in the topic, as that issue is summarised by the ‘overall’ construct. Thereby specifying what ‘effectiveness’, ‘approachability’, and so

on mean in particular to him or her.

2. Notice how the ‘overall’ construct summarises the issue at stake, as previously highlighted by the qualifying phrase. (Glance again at the rationale

for the use of qualifying phrases in Sections 3.2.1 and 3.2.3. Table 3.2 is worth another look too.)

And it will take into account both the % similarity value and the individual’s

personal metric!

This is tremendous. It’s pure gold. I have taken my hat off to Peter Honey for a

quarter-century now, and shall continue to do so for another 50 years at least. I

hope you can see why. While strongly advocating procedures for ensuring

reliability, he doesn’t go into reliability measurement in any great detail; I have

added that, below, together with some grace notes of my own. But the overall

thrust is his; he gives us exactly what we need when we seek to aggregate

large samples of respondents.