tcarter2 at elp.rr.com
Fri Aug 15 20:30:13 EST 2003
Ian Goddard <igoddard at erols.mon> wrote in message news:<fkfljvcdcvj3csi0v12p6tp8foutmst2r2 at 4ax.com>...
> tcarter2 at elp.rr.com (Thomas Carter) wrote:
> >Hi Ian,
> > Foremost allow me to apologize if I upset you. I know how
> >easily we humans can be irritated by opposing opinions, and I try to
> >avoid it. But I guess I'll have to try harder. I'd like to continue
> >our discussion in a dispassionate, scientific manner. If you can't,
> >feel free to cuss away. I have a thick skin, and a short memory for
> >such things. :>)
> IAN: Tom, it's important to accurately represent another's response.
> You represent my response as "cussing away." "Cuss" means to curse,
> and curse means "a profane word or phrase." A review of my reply
> to you will not find the profanity that you imply exists therein:
It never occured to me that the remark could be taken that way.
Even so the remark was unappropeiate and I apoligize, and hope that
you too, have a short memory for such things.
> As I said there: "you could broach your comments as hypotheses and
> questions, rather than, as you have, accusations of 'data doctoring'
> and firm conclusions of 'fatal mistakes' and such by the authors that
> you use to caste sweeping aspersion on peer-review per se." Yet once
> again in this response you're back declaring that your quick-glance
> conclusions are inherently accurate prior to consultation with the
> authors regarding what are really questions, not clear conclusions.
Yes I could have, but they are my conclusions, not questions nor
> IAN: The study (http://www.jacn.org/cgi/content/full/19/2/242) says:
> "To investigate alternative explanations for the association of high
> mid-life tofu intake with poorer cognition in late life, analyses were
> conducted to identify possible confounding factors [...] Thirty-one
> factors were examined [...] birth in Japan, years of childhood lived
> in Japan [...] When the 933 subjects who had lived one or more years
> of their childhood in Japan were excluded from the analysis, the
> corresponding values were beta=-1.69, p=0.01 for 1965?67 tofu intake;
> beta=-0.65, p=0.01 for 1971?74 tofu intake, and beta=-0.72, p=0.015
> for the composite tofu intake index."
> So even when the subgroup you claim biased the results was excluded,
> a statistically significant relationship to tofu intake remained.
A statistically significant relationship between tofu and what?
I'm glad you brought that up. It's another example of data doctoring
that I had missed.
Thruout the article the authors based the cognitive differences on
the FAILURE rate in the CASI score. Which was about 1000 out of 7000.
After adjusting for the confounding factors they wished to, (leaving
out grade school in Japan among others) they managed to get an Odds
Ratio for failure rate of 1.62 for the hi-hi group against the low-low
group with a significant p value of .03. I will now show that had they
left out the 933 men born in Japan the p value would have most likely
been over .12, a mathematically insignificant value and would IMO have
likey made the report unpublishable for lack of interest.
Below I post the full paragraph of the partial quote Ian posted
just above. It shows that when the 933 men born in Japan were left out
for the average CASI scores the p value rose by over a factor of four,
from .0025 to .015, showing that the scores of the men born in Japan
were dragging down the the scores of the hi-hi group. The obvious
conclusion is that the p value for the falure rate in the CASI test
would have risen by about the same, or from .03 to .12. The average
CASI scores for the two groups are never given in the report. Only the
p values are given which are significant because they are the averages
of 7000 men, not the falures of only 1000. My estimate is that the
difference in the CASI average scores would have been less than three
points, a very uninteresting figure and that's why the authors used
the failure rate instead carefully concealing the fact that the
average scores were quite similar.
In a scientific paper reporting on the differenc in intelligence
of a group of men twice placed in a low tofu group with that of
another group twice placed in a hi tofu group the actual test scores
of the groups are concealed. Furthermore the data that are spoonfed
to us are selected to exagerate the difference in test performance.
When they wish to titalate us with large differences, they use failure
rates. When they wish to assure us with low p values, they use average
scores. When they wish to persuade us with doctored statistics, they
leave out key factors.
Here is the paragraph from which I based my estimate.
"Initial multivariate analyses were done using linear regression
modeling with the participant's CASI score as the dependent variable.
Controlling covariates included all of the candidate confounders and
predictors of CASI score mentioned above, together with midlife FEV-1
(a pulmonary function indicator previously reported to predict CASI
scores) and midlife BMI (a predictor of brain weight in subsequent
analyses). In a model using the 19651967 tofu intake question as the
primary predictor variable, while controlling for all other
independent variables, there was a significant inverse association
with CASI score (beta=-1.4, p < .006). Similar associations were seen
with the 19711974 tofu intake variable (beta=-0.74, p=.0005), and
with the composite variable (beta=-0.86, p=.0025). When the 933
subjects who had lived one or more years of their childhood in Japan
were excluded from the analysis, the corresponding values were
beta=-1.69, p=0.01 for 196567 tofu intake; beta=-0.65, p=0.01 for
197174 tofu intake, and beta=-0.72, p=0.015 for the composite tofu
intake index. There was no evidence of interaction between
apolipoprotein E genotype and tofu in their associations with
cognitive test performance."
More information about the Neur-sci