RE: Trendlines

[Steven_S_Prevette@rl.gov]

Is there anytime that a 'trendline' provides value to a time series
chart?  NO (IMHO, as with all responses below)
 
Or more specifically a control chart?  DEFINITELY NO

How helpful is the line in determining 'trends'?  IT WILL ONLY INCREASE
FALSE ALARMS

Does a trendline help identify the cause? DEFINITELY NO

What is the operational definition of 'trend'?  SEE BELOW

Does a rising or declining trend (in the sense of trendline) actually
mean there is a problem?  NOT NECESSARILY

----

The major problem with these "trendlines" (more properly, regressions) is
that the user generally never checks to see if the "trend" is significant.
That is, if I plot a set of random numbers, and put a "best fit", least
squares, linear regression, trend line through the numbers, the slope of the
line will not be exactly zero.  So, the user will say "Aha!" there is a
trend.  Excel provides a little bit of a hint in that you can display the
"r-squared" value, which can be used to assess the goodness of the fit, but
this is very difficult and most people don't do it.

I should also comment that simply putting +/- 3 standard deviation control
limits about a regression line is incorrect.  The proper term is "prediction
interval", and the predication interval increases as you move away from the
center of the data (and especially as you extrapolate away from the plotted
data).  On a linear regression, the prediction interval lines are curved
arcs.  The upper arc is concave, and the lower arc is convex.

To me, the operational definition of a "trend" (and this is proceduralized
at Hanford at http://www.hanford.gov/safety/vpp/spc.htm) is the detection of
a out-of-control point (or points) on a control chart.

Steve Prevette
QA Engineer, ESH Planning and Performance
Fluor Hanford, A Fluor Global Services Company
ASQ Certified Quality Engineer
steven_s_prevette@rl.gov
509-373-9371
===================================================================