Eva Young wrote:

> I've never had a statistics class, but I have had calculus, through
> differential equations, and I'm not sure what Atherton is refering to with
> his reference to "regression models" and "variances."  It kind of reminds
> me of how computer techies talk -- and when you ask questions -- rather
> than being afraid of sounding "stupid" -- a few follow up questions reveal
> the person doesn't know what they are talking about.

I've had calculus, though not through differential equations, and when
I became an education major I was shocked at the complexity of the
statistics courses (even after having had statistics courses in psychology
and computer science).  Regression models and variance components
are conceptually complex, but absolutely necessary for understanding
educational research and making educational policy decisions, but it
is not surprising that people outside of the field are unfamiliar with
the terminology.  I'll try to give a quick overview so that we can all be
on the same page and we can continue to discuss these issues intelligently.

"Models" in mathematics and statistics (generally) refer to equations that
describe something about the real world.  For example the equation,
distance = rate X time, tells you how far you'll travel if you drive
at a specific speed for a given amount of time.  It allows an
accurate prediction of the distance from your starting point.  In statistics
we also create models to describe situations and make predictions.  The
idea of a regression model is to derive a formula to describe the relationship
between one factor and given a number of other factors that one collects
data about.  For example, if we were interested in knowing what factors
contributed to heart attacks, we could go collect data and then use
a computer program to create a formula that would describe
the influence of each factor on heart attacks.  This formula would actually
allow us to predict how likely an individual would be to have a heart
attack, given say their age, weight, sex, amount of exercise, etc.

Now let's consider the specific model that Ms. Johnson cited.  This
model was designed to describe and predict student achievement based
on parent involvement, teacher quality, and class size.  First point:
It is the researcher who gets to pick the factors.  In this case the
researcher picked parent involvement, teacher quality, and class size;
they could have just as easily picked student intelligence, socioeconomic
status, ethic background, or any other set of factors.
Usually, a research has some notion of what might be the most
important factors and picks those.  Second point:  The factors that you
select never completely describe or fully predict the factor that you
are interested in.  SO, in this student achievement model, parent
involvement, teacher quality, and class size cannot fully account
for everything that contributes to achievement.  There is always
some, "portion of the variance" unaccounted for by the model.

That is why the numbers didn't make any didn't make any sense.
When Ms. Johnson originally reported them. She said, research shows
that "...student achievement can be accurately measured as follows:
49% attributed to parent involvement, about 42% teacher quality,
and about 8% to class size."  Well if you sum up these percentages
the result is 99%, leaving only 1% is account for by any other factors
that might reasonably contribute to achievement.  Well then, when I
check the article it turned out that the numbers were really 49%,
43%, and 8% which sums to 100%, and that makes even less
sense.  So here is the problem.  Ms. Johnson most likely got
her numbers from a pie chart on page 9, which divides the pie
into only three pieces, the three factors mentioned above.  The
pie chart is titled, "Influence of Teacher Qualifications on
Student Achievement." Then, in really small print it says
"Proportion of Explained Variance in Math Test Scores Gains."
Well, it's true that if you only included these three factors in
your model, they will account for 100% of the EXPLAINED
variance, because they are the only factors that you included
in your model. What is not mentioned is how much of the
variance is not accounted for by the model.  In other words,
the three variables (parent involvement, teacher quality, and class size)
may account for say 20% of the total variance, leaving 80% of the
variance unaccounted for by this model, but we don't know
because this article never reports what amount of the influence
on student MATH achievement was not accounted for by their
model.  This is really important because without knowing the
influence of the unknown factors you can't truly calculate the
influence of the known factors.

The upshot of this is that this report is very misleading
(be it intentional or otherwise),  thus it is reasonable that
Ms. Johnson could have misinterpreted the data. And, it's
true when Ms. Young says, "figures don't lie, but liars
figure."  The upshot of this is that we can't expect to improve the
schools and help students using lies and false data.  This is
why it's so important for school administrators be able to
evaluate the research.  I don't fault anyone for not knowing
statistics when they are elected to the school board, but
they better get up to speed before they start making decisions
that are going to impact our children.

Michael Atherton
Prospect Park



_______________________________________
Minneapolis Issues Forum - A Civil City Civic Discussion - Mn E-Democracy
Post messages to: [EMAIL PROTECTED]
Subscribe, Unsubscribe, Digest option, and more:
http://e-democracy.org/mpls

Reply via email to