Day 017 - 25 Jul 94 - Page 21
1 cancer in women".
2 Then he goes on to consider population
studies: "International correlations. We owe to Lea
3 (1966) the observation that mortality from breast cancer
in different countries shows a strong correlation with the
4 corresponding per caput consumption of fat (as shown in
the food balance tables of the Food and Agriculture
5 Organisation)."
6 Dr. Arnott, what exactly are food balance tables?
A. Right. Basically, what food balance tables do is they
7 measure the total production of a substance or such as fat
which is the national production which includes obviously
8 farming animals imports, and they try to make some
allowance for wastage or exports and at the end of that
9 they then end up with an answer which is considered to be
the consumption of that substance in that particular
10 country.
11 However, that is a very crude analysis, as I am sure you
will be aware, from my description of it.
12
Q. It goes on: "Many other workers have confirmed this
13 observation, but none so comprehensively as Armstrong and
Doll" -- that is the same Sir Richard Doll who did the
14 research on smoking?
A. It is, yes.
15
Q. "(1975), who examined cancer incidence and mortality in
16 relation to the average consumption of many different food
items in 37 countries. Total fat consumption and breast
17 cancer mortality gave the highest correlation coefficient
(r = 0.89) of any dietary constituent and any cancer. At
18 face value, this coefficient indicates that fat could
'explain' about 80% of the variation in breast cancer
19 mortality among the countries examined. This relationship,
shown in Fig. 1, is linear". If one casts one's eye
20 across the page, one sees that is right; he has plotted a
series of dots which go up in a straight line?
21 A. Yes.
22 MR. JUSTICE BELL: Mr. Rampton, could we have an explanation of
how the correlation coefficient works? It may be obvious,
23 but I may misunderstand it if I just assume, and when we
come to "P" in statistical terms, if I could be offered
24 some explanation of that so my understanding is correct.
25 MR. RAMPTON: I think I understand correlation and coefficient,
but I am not going to attempt that. "P" in statistical
26 terms has alluded me, I think. Can you do those for us,
please?
27 A. Basically, a correlation coefficient is measuring the
strength of a relationship between one thing and another.
28 One can apply certain statistical tests to this particular
association. Normally we accept that some of these
29 findings may have arisen by chance, and for the
coefficient to be meaningful we would say that we would
30 have to accept that it could have arisen by chance in no
more than five per cent of cases. This is how we get a P
