Let me use your example to illustrate my point. If the sex of people
is included in the regression, say ln(wage)=a+b*age+c*sex, my question
is in what sense is sex controlled? You may say that: at the same age,
a male will on average get c dollars more in wage than an average
female. Now I might add: a person's education level and experience
might contribute to his/her wage too, so we might add these two
variables as well.
Now let's look back at the coefficient of sex, if c is smaller, then
the controlled sex effect is largely a manifestation of education and
experience, because in general women get lower education level and
less experience because of child bearing and raising burden; if c is
unchanged, then it shows male does have a higher compensation rate
than a female, which might come from discrimination, demand
difference, etc, something we don't know.
So my interpretation is that sex as a variable is largely a
manifestation of something else. Things don't seem so easy if one
variable, say, education, is correlated with other variable,
experience. Higher education often implies fewer years of experience,
given lifetime is largely little varied across people. Samples often
include those with higher education+ fewer experience, and those with
lower education + more experience, and they are pooled together, the
coefficient we get is neither for the former nor for the latter type,
but something average between them. And the coefficient for education
can only be interpreted when we maintain the presumption of average
experience, but not be able to be interpreted for ANY particular
individual observation with his/her particular individual experience,
alas, we can't say, "other things equal (including his/her
experience), one more year of education would bring his/her wage up ??
On 10/15/05, Chris Kopp <email@example.com> wrote:
> Let's keep this off-list.
> "Control for" in the social science meaning refers to account for the
> effect of a variable by including it into the model. For example, if
> you have the equation ln(wage)=a+b*age, people might say that the sex
> of the person should be included too. So you might add the dummy
> regressor "woman"; and say that you "control for" the sex of the
> person. It's just social science slang and has nothing to do with the
> meaning of holding something fixed in physics.
> Hope that helps clear things up
> On Oct 15, 2005, at 19:52, Sean wrote:
> > What's the relationship between statistical model and the real world?
> > Aren't social scientists supposed to explain the real world based upon
> > their statistical models? I'm still confused.
* For searches and help try: