study averages

[Guest guest]

If you see from the article Suze posted given in the first study that there

was no dose-response between animal fat and cancer that random results can

throw off results, especially if you decide how to define your groups after you

collect the data (which of course you would do). Interestingly, if the study

had split the " quintiles " up differently, or into different sized groups, the

relationship could have disappeared. If you notice, the groups aren't

equivalent in number of people, span of median fat intake, or span of total fat

percentage intake.

Giving the last " quintile " the folks with the very low incidence of cancer

gives the opportunity to see more folks in the middle, which had a very low

total span of fat intake, who perhaps by chance had more cancer.

If the groups were spread out more, broken into more groups, or arranged in

different ways, folks from the middle could have gotten included with folks or

some folks from the end, and the linear increase that's seen in the middle

would have disappeared.

I have an anecdote about this. Last lab in my biology class we did an

experiment to see if athletes have healthier hearts. Each of us formed a group

of

3-4 people, two who were to be subjects, one who was to be designated an

" athlete " and the other a " non-athlete. " All the subjects did a stepping

exercise

in rhythmic unison for 3 minutes. Pulses were measured before, after, and

every minute after and recorded. The data was organized for starting pulse,

pulse

after exercise, and recovery time (when pulse returned to initial pulse.

I was at first going to designate myself an athlete, becuase I borderline fit

her definition (basically regular exercise or training of some kind, not

necessarily organized.) But the guy I ended up with played organized basketball

and trained regularly, so we put him in the athlete group and me in the

non-athlete group.

There ended up being a girl in each group for the final data (i.e. athlete

and non-athlete), both of whom had tremendously lower pulse rates regardless

athletic ability. In the final data, if you looked at the individual people

(and

there were only six subjects) there was NO correlation whatsoever between the

athleticism of the person and the heart rate or recovery time. If you

couldn't see this from looking at the chalk board you were blind as a bat

(though

only I seemed to notice or care).

However, since my pulse alone was *way* higher than anyones at starting,

finishing, and recovery time (my pulse never actually recovered until after I

stopped measuring it) for reasons talked about in the heart rate thread

(probably

unrelated to heart health), and I was thrown in the non-athlete category, I

threw the averages up tremendously.

In each group, there were people with almost identical starting finishing

pulses and recovery time (I wish I had the data but it's gone), and the only

correlation with anything was sex.

So did the teacher take note? No, because she " knew " that the study was

" supposed " to find a correlation, so she said, " ok, now everyone calculate the

averages. "

I objected. 'But there's one person out of six whose data is off the charts

and will throw the averages off.' But we averaged it anyway, to get a

deliberately less accurate result.

Notice also that if we had three categories instead of two, -athelte-

-nonathlete (or sedentary)- and something in the middle to describe me (regular

personal exercise), then what the study would have found would be no difference

between athletes and sedentary people, but people who get regular exercise but

aren't athletes have much higher pulses.

So how you define your groups and get your averages makes the entire

difference in what your study " finds. " And of course, you always analyze and

organize

the data *after* you collect it.

Chris