Reader Comments

Post a new comment on this article

Did the authors check the effectiveness of their matching?

Posted by astatistician on 26 Mar 2014 at 13:57 GMT

As a statistician, I am surprised to note that the authors do
not appear to have checked the effectiveness of their matching (or, if
they have, they do not report the outcome). Subjects were matched
"according to their sex, age (in age-groups spanning 5 years, e.g. 20-
to 24-year-olds), and socioeconomic status (SES)". The highest age
group was 80+ years, which included 8 vegetarians (all female) and the
same number of women from each of the other 3 meat-eating diet groups.
As this age group is open-ended it is possible that the 8 vegetarians
were all aged in their 90s and that they were matched with meat-eaters
in their 80s. Although this is unlikely to have happened, the authors
should certainly have reported the mean ages (and measures of SES) in
the 4 diet groups and run an analysis of variance to check that there
were no significant differences between the values. If the vegetarians
were, in fact, 1 or 2 years older on average than the meat-eating groups
(entirely possible given the rather loose matching on age) this might
have been enough to account for their apparently poorer health. Given
that only 2.2% of participants in the full survey were classified as 'vegetarians'
(even though more than half of them reported consuming fish) the authors could
surely have matched much more closely on age (e.g. in 2-year or even
1-year age groups).

No competing interests declared.

RE: Did the authors check the effectiveness of their matching?

burkert replied to astatistician on 26 Mar 2014 at 15:25 GMT

We have verified the effectiveness of our matching. Since we matched the subjects one-by-one, no differences in age, sex, or SES exist between the groups. The socio-demographic characteristics of the subjects are shown in Table 1. Age was only measured in age-groups spanning 5 years (e.g. 20- to 24-year-olds) in the AT-His and the oldest age group covered persons aged 80 years or older. The percentage of subjects having a low, middle, or high SES is reported in Table 1. The mean SES in all dietary habit groups was 8,14 (SD=2,72).

No competing interests declared.

RE: RE: Did the authors check the effectiveness of their matching?

astatistician replied to burkert on 01 Apr 2014 at 09:56 GMT

Given that the authors only had age group and not exact age available to them it is impossible to verify their assertion that "no differences in age ... exist between the groups" because they cannot calculate and compare mean age in each of the 4 diet groups. At least there appears to be no difference in mean SES between the 4 diet groups.

Given that the vegetarians were numerically by far the smallest of the 4 diet groups I wonder why the authors did not increase the power of the study by matching each vegetarian with 2 or more participants from each of the other 3 diet groups. This would have reduced the possibility of chance findings. Better still, why not include all of the survey participants in the analysis and adjust for sex, age and SES (or subdivide the data by sex and adjust for age and SES in each subgroup)?

I am also surprised to learn that 13 vegetarians were excluded from the analysis "since not all of them corresponded to a subject of the same sex, age, and SES from a different dietary habit group". This is most surprising given that vegetarians only constituted 2.2% of the survey population. Which sex, age and SES categories did the excluded vegetarians belong to?

No competing interests declared.