[EMAIL PROTECTED] (Robert J. MacG. Dawson) wrote in
news:[EMAIL PROTECTED]: 

>      No, it isn't!   The mean of the daily proportions is not in
>      general the 
> same thing as the overall mean. This is closely related to Simpson's
> paradox.
> 
>      Drive      Table     Proprortion
>      through          DT
> 
> Tue     30     10     .75
> Wed     30     10     .75
> Thu     30     10     .75
> Fri     30     90      .25
> Sat     30     90     .25
> Sun     30     90     .25
> ___________________________
>               #   0.5
>                     =======  
> Total     180     300 | 0.375

It's interesting how most "ecological fallacies" really come down to 
forgetting that you can't add fractions by taking the sum of the numerators 
and dividing it by the sum of the denominators.

Another factor is the tendency to reify averages; if, say, a very short kid 
transfers into a class, the average of the heights of the students will go 
down, for sure, but we tend to use sloppy language and say "the height of 
the average student has gone down."  In this case, nobody would seriously 
think that the transfer made any of the kids shorter, but when dealing with 
less familiar examples we often do the equivalent.

As another example, let's say that the population of Slobovia is divided 
into two ethnic groups, Danierians and Pirenians.  We observe that the 
average height, at the present time, of adult Pirenians is lower than that 
of adult Danierians by a statistically significant amount.  We also know 
that adult height changes little if at all over the course of an 
individual's adulthood; if Jack is shorter than Joe at age 25, he will 
still be shorter than Joe at age 55.

But we *cannot* magically slide from this to the conclusion that the 
difference in average heights for the two groups is therefore "immutable," 
because the "average Danierian" and the "average Pirenian" are *not* 
individuals whose height is under the control of individual biological 
processes that maintain stability of height over time; they are 
abstractions, namely aggregate measures taken over a population whose 
composition will change over time.  Jack at 55 is the *same* individual as 
Jack at 25, but the "average Pirenian" of 2034 is *not* the same as the 
"average Pirenian" of 2004; the latter two are aggregate measures of 
*different* populations.

In fact, we can say that there's a source of variation in the average 
height of either group that is *not* present in the measured height of any 
individual from either group, namely the effect of variation in group 
composition, something that simply doesn't exist at the individual level.  
Therefore, we will often make wrong inferences if we take group averages 
(or other composite measures) and plug them into models that are valid for 
individual measurements.

As you've probably guessed by now, the above example is simply a 
translation of some aspects of the race/IQ debate.  However, a few months 
ago I actually ran into a confusion between average heights and individual 
heights.  It was in a debate over the causes of the "obesity epidemic" and 
I observed that the average weight of any developed country's population 
has been going up steadily for decades, well before there was much increase 
in obesity, and, at least in the US and the UK, there were no sudden 
"jumps" in average weight that could be accounted for by an Event.  

However, for much of that period the average *height* of the populations in 
question was also increasing, so the increased weight didn't translate into 
a higher average Body Mass Index (BMI), since the BMI is an increasing 
(linear) function of weight and a decreasing (quadratic) function of 
height.  However, the increase in average adult height has levelled off in 
most developed countries, so now any increase in average weight translates 
into an increase in average BMI.  And in fact the levelling-off of the 
increase in average height occurred earlier in the US than in the UK, and, 
surprise surprise, so did the increase in obesity.

Anyway, someone else in the debate, who was strongly attached to the notion 
that there *was* an Event that caused a jump in weights (I think he was 
asserting that it was the introduction of high-fructose corn syrup), said I 
had to be incredibly stupid to believe that in the past adults kept growing 
taller past the age of 20 or so.  He took an assertion about averages and 
treated it as an assertion about individuals, with absurd results.

My point, BTW, was not that the increase in obesity wasn't real, but that 
everyone seemed to be asking "what did people do differently to cause it?" 
and not considering the possibility that the cause was something they 
*failed* to do differently.  Of course nutritional issues are subject to 
passions that make the race/IQ debate look very tame by comparison, and 
there seems to be a strong preference for exotic explanations over mundane 
ones.

One last statistical issue about the increase in obesity; obesity is 
defined as a BMI of 30 or over, which is to say that it's defined by 
dichotomizing a continuous variable, the cutpoint being out in the right 
tail of the distribution.  That means that P(BMI>=30) is *not* a linear 
function of average BMI; if, say, BMI is distributed approximately normal, 
then as the curve slides to the right, the area to the right of the 
cutpoint will increase slowly at first (since what's sliding under the 
cutpoint is all tail) and then much more rapidly as the "body" of the curve 
starts bumping into the cutpoint.  Thus small increases in average weight 
can lead to large increases in obesity, and again a steady trend starts to 
look like an Event.

.
.
=================================================================
Instructions for joining and leaving this list, remarks about the
problem of INAPPROPRIATE MESSAGES, and archives are available at:
.                  http://jse.stat.ncsu.edu/                    .
=================================================================

Reply via email to