You both raise good questions- one the one hand, the more transparency and
explanation the better, but on the other hand there are space constraints
and you hope you can assume some knowledge on the part of your readers.
The most important consideration when you're deciding how much to explain is
to be certain the reader can know what your results mean. Rather than just
reporting a coefficient, explain what that coefficient means in terms of
your substantive question. This is why presenting first differences and
other quantities that are truly of interest is useful. People care about
'results,' so your quantities of interest should truly be of interest and
should be easy to understand. You should explain as much of the model as
you think is necessary for readers to understand your quantities of
interest. No reader wants to be walked through the lines of algebra in any
estimator, so often you have to rely on the reader trusting you that the
model is producing the quantities of interest you say. Often it's enough to
give an introduction to what the model 'does,' a couple of sentences about
what sorts of questions the model is usually used to ask, maybe include a
citation to a more thorough exposition of the model, and then spend the rest
of the explaining time discussing the quantities of interest from the model
rather than the nuts and bolts of the model itself.
Like all parts of the project, though, what works best requires a judgment
call. Without seeing more of the papers you describe, Sean it seems like a
citation to the paper you mention could be used in place of most of
the detailed description of the model, and John, as per above, you probably
don't need to discuss much of the underlying math in those models- just make
sure the reader knows what the output of those sorts of models really says.
You know your own projects better than I do, though, so use your discretion.
Best of luck in the final hours,
Jenn
On Sat, May 3, 2008 at 10:29 PM, John Sheffield <john.sheffield at gmail.com>
wrote:
Similar question: if we're using fairly standard
models like Cox PH or
time series logits, do we need to go into depth on the underlying math?
Thanks,
J
On Sat, May 3, 2008 at 10:27 PM, Sean Li <seanli at fas.harvard.edu> wrote:
My partner and I are writing our paper based on
a statistical method
that is relatively simple to explain conceptually, but whose details are
fairly sophisticated. (Specifically, we're using simultaneous time series
forecasting as developed in Gary's book *Demographic Forecasting* and
implemented in Gary's package YourCast.)
It would take quite a few pages to fully explain all of the math behind
the model we're using (it takes up several chapters in *Demographic
Forecasting*). Since we face space constraints, would it be acceptable
to reference sections of *Demographic Forecasting* instead of deriving
the precise formulas in our paper? Basically, how much of the model should
we try to explain mathematically, and how much should we rely on Gary's book
to provide explanation?
Thanks,
Sean
_______________________________________________
gov2001-l mailing list
gov2001-l at
lists.fas.harvard.edu
http://lists.fas.harvard.edu/mailman/listinfo/gov2001-l
_______________________________________________
gov2001-l mailing list
gov2001-l at
lists.fas.harvard.edu
http://lists.fas.harvard.edu/mailman/listinfo/gov2001-l