query re Ravenette & Honey

Sun, 8 Jan 1995 00:31:36 +0000

Bob Green asks for a reply to his query via the pcp group; thought I'd do
so anyway, since it may be useful to other colleagues.

But first: Bob: re the Ravenette material: please e-mail me your address
and I'll put the Ravenette items in the (snail-mail) post; no point in my
giving you a ref. as it's to the EPCA Newsletter, which isn't widely
available. If you want to join EPCA, best write to the Membership
Secretary, Ann Harwood, Kent Cottage, The Drive, Belmont, Surrey SM2 7DH,

Bob et al.:
re Bob's

>Regarding the piece on Honey's content analysis...I would find it useful if you
>could work through step 2, for example, compute the formula using
>the excerpt which you provided

Er, well, it's actually been done for you in the bottom right-hand corner
of the first page of Cookery Corner, and if you look at the sample grid
sheet on the left of the first page of Cookery Corner, you'll see that the
resulting % matching Scores are given below each row of ratings. Try
applying the formula to the Sigma d values yourself and see if you get the
same values of % Matching Score!
(Oh, okay, you need to know that R, the maximum rating possible, is 5 in
that example, and there are, of course, 6 elements so E = 6. Sorry)

>This might also explain what you meant by reversing the construct polarity....
>>I can understand this in general terms, but could not figure out what this
>>meant in terms of the calculations.

As it happens, reversal makes no difference in _these_ data, so they're
straightforward enough. However, here's the rationale and method for
reversal (it's quite general for all grids involving ratings, and isn't
confined solely to Honey's content analysis)

Constructs are bipolar, thus reversal is required when comparing ratings on
any two of them.

Consider ratings of 5 elements A to E (which happen to be wines: go on,
it's Christmas) (-ish) on two constructs as follows:

Nice 1 3 5 4 2 Nasty
Sweet 5 3 1 2 4 Dry
d 4 0 4 2 2
Sigma diff = 12
A _high_ sum of differences, therefore a low match between the two
constructs. On the face of it, no relationship between the two constructs
even without having to compute the % formula.

However, suppose the poles of the first construct had been written down the
other way round in the first place; then the meaning ascribed to the
elements would have been expressed as:

Nasty 5 3 1 2 4 Nice
Sweet 5 3 1 2 4 Dry

Aha! you don't even have to calculate Sigma diff: you can see that the two
sets of ratings are identical, viz., whenever this person says s/he thinks
a wine is sweet, s/he thinks it's nasty, and whenever s/he construes a wine
as dry, s/he thinks it's nice.

Notice, to preserve the meaning expressed in the ratings of "Nice - Nasty",
we've had to subtract each rating from 6 (R+1 where R is the maximum rating
on the scale being used).

Apologies for this level of exposition as you mentioned you understand the
idea of reversal. But it exemplifies what you have to do in terms of
calculations, as you requested, viz.:

a) Work out matching scores with ratings unreversed
b) Then reverse the poles in one of each pair of constructs being compared
and, to preserve the meaning the ratings express, subtract each rating in
the construct whose poles you reversed from R+1 where R is the maximum
rating possible: (6, for a 5-point scale).

In Honey's technique, which involves matches against just one supplied
construct, step b) boils down to reversing the ratings on the supplied
construct and jut working out the matching scores between the supplied
construct and each of the others. Other techniques, e.g. cluster analysis,
involve you in reversing ratings on _all_ constructs, so count yourself

Hope that's clear; meet you all at the International Congress in Barcelona
for more explanations and a drink, or maybe just the drink?

Kindest regards,

Devi Jankowicz