Best way to describe changes in performance through time. I'm wondering how to explain changes in production/performance metrics quarter to quarter. In the example below, one can see that the average processing time for "widgets" has decreased substantially over two years.  However, the last two quarters show a processing increase of 17%.   Given the long term trend, this really isn't significant.  What statistics would be best used in describing this? 

 A: The statistic/characterization/summary that you are looking for is "there has been a downward trend for the last 7 periods and the first 3 period were inconsistent with that trend' The TSOUTLIERS packages may be useful to generate this statement along with other commercial offerings that might even be more useful for short series like this one..
It's a poor Doctor that doest try his own prescriptions. I gave you 10 values to my toy of choice and lo and behold it concluded that there two distinctly different means and one anomaly.
  . The identified equation (statistic) is  . Your statistic (%change) was descriptive and not inferential.
As @DJOHNSON pointed out you could build a causal model relating the # of gidgets to possible predictors. One would not be advised to simply use simple regression as you have time series data thus potentially (nearly always !) auto-correlated observations. The preferred method for dealing with causal time series data is called Transfer Function or simply ARMAX . 
