Next message: Fred Hebard: "Re: Nautical Almanac Usage"
On Oct 24, 2004, at 6:30 AM, Jim Thompson wrote:
> Bill, After again reading David's article, I ammended my summary to
> this:
>
> 3. Inspect the set of sights for consistency and discard obvious
> outliers,
> or sights that fail to increase and decrease in time or altitude
> reasonably
> smoothly. It helps to plot the sights on a graph of altitude over
> time; or
> to build a table, and then use arithmetic to calculate the change in
> time
> and altitude between sights, and then inspect the table for outliers.
> Beware
> discarding an apparent outlier that might in fact be a good sight:
> David
> Burch explained a method of using the slope or rate of ascent/descent
> to
> optimize a run of sights in an article posted at
>
Jim,
I believe if you compute means or slopes with and without outliers, you
often will find little difference in the results for five observations,
barring _gross_ errors such as reading the wrong degree from the arc or
minute from the watch. With fewer observations, changes in summary
statistics will be more evident, but then it becomes more difficult to
distinguish the outlier from the good data point. In my opinion,
obvious gross errors should always be discarded but not slight errors;
I think part of the value of this exercise is to detect blunders so you
can refine your technique to minimize their occurrence.
Fred
