The implausibly-named G. B. Blows wrote:
> > The table below represents the mean vacancy rates of eight office
> > submarkets of Atlanta, Georgia for a period of nine years. Quarterly
> > vacancy rates were used in calculating the means.
> >
> > SUBMARKET Mean Vacancy Rate (%)
> > Buckhead 16.85
> > Downtown 20.73
> > Midtown 19.75
> > North Central 16.73
> > Northeast 16.95
> > Northwest 16.81
> > North Lake 20.38
> > South 28.26
Assuming that Mr. Blows is not another of the tirekickers who think
we'll steal their data set if they show us the whole thing, but who
expect a (free!) analysis nonetheless (spiritual kin to the patient at
the doctor's office "whose friend thinks he might have VD"), I am
puzzled.
My first reaction is that these data are a hopelessly oversimplified
summary of a very complicated data set. The full data set is a set of
proportional time-series, probably highly autocorrelated and
cross-correlated. Without knowing how much vacancy rates fluctuate over
time, you cannot say which of these differences are significant. You
would also need to know market size (as a time series) for each region,
etc.
Even making simplifying assumptions (eg, that each submarket's vacancy
rate is stationary and that there is no autocorrelation) I do not see
*any* valid tests that can be done to compare these data, as there is no
indication of within-submarket spread. [An awful thought: does somebody
think it's sqrt(npq) where $n$ is 36, the number of quarters? Please,
please, somebody, tell me that the fact that it's a perfect square is a
coincidence...]
I would be curious to know (a) who set this assignment and (b) what
they conceive the answer to be.
-Robert Dawson
.
.
=================================================================
Instructions for joining and leaving this list, remarks about the
problem of INAPPROPRIATE MESSAGES, and archives are available at:
. http://jse.stat.ncsu.edu/ .
=================================================================
