Link to Exercise #3
Link to MBA 510 Main Page
 

Q.: Does a 4 point increase mean 4 percentage points (0.04), or decimal
points (0.4), or 4 units?
A.: I meant 4 units.  Such as a change in pressure from 28"Hg  to 32"Hg.

Q: Won't that give me a very large number for the odds ratio, R?
A.: Well, yes.  Even though I asked for the effect of a 4-point change
in Pressure on the probability, just give me the effect on the odds
ratio, R.  Our calculators probably don't show enough digits to see the
difference in the probabilities.

Q: What do you mean by "the effect of a 4 point increase in Pressure?"
Am I supposed to go back and add 4.0 to all of the values in the Pressure
column in the data table?
A.: No.  I want you to predict the effect of a hypothetical change in
Pressure, assuming that Temp and Humidity remain
constant.  In the logistic regression you do this by using the
odds-ratio equation.  Note that the effect is a multiplicative effect;
that is, you should give me a factor by which the odds ratio will be
multiplied as a result of a 4 point increase in Pressure.  In the linear
regression, you should give me the predicted numerical change in Y
(which is supposed to be 0 or 1, but the predicted values probably won't
be) as a result of a 4 point change in Pressure.  You should use the
coefficients in the Paremeter Estimates section to form the regression
equation to help you answer this.

Q.:  Why would we want to use a level of significance (alpha, a) of 8%
in the linear regression?
A.:  My mistake.  I should have inserted that sentence in the logistic
regression section.  I was looking at the p values from the Wald
chi-square statistics in the Parameter Estimates section.  The
maximum-likelihood tests are usually better, and they don't require an
8% alpha.

Q.:  But I didn't find a difference in the significant variables between
the two regressions.  Should I still answer question 7?
A.:  Yes, but if you found that the same two variables were significant
in each model, you should still tell me which model is better and why it
is better.  If you used the p values from the maximum-likelihood tests
for the logistic regression, you may find that the same two variables
are significant in each model.  I was thinking of the Wald tests given
in the Parameters Estimates section when I wrote that question.