[EMAIL PROTECTED] (Andrey) wrote in [EMAIL PROTECTED]:">news:[EMAIL PROTECTED]:
> Hello everybody, > > I need help determining abnormalities in a set of measurements. We > measure the distance between two consecutive welds (in a mathematical > car model) to determine missing / duplicate welds. The entire > population can be between 5 - 20 welds. Let's say a typical weld > spacing is between 40-60mm. Thus, a 0mm spacing indicates a duplicate > weld, while a 100mm spacing probably means a weld is missing. The > first thing that came to my mind was to flag all measurements that > were outside of mean +/- 2 standard deviations. The problem is, I > cannot start with a mean because 1) it is specific to each weld set, > 2) the distribution might be skewed (if only missing or only duplicate > welds are present). So I decided to use the median as a measure of > what "good" weld spacing each. My question is, is there a "standard > deviation about median" or some other function that shows how far a > sample is away from the median? The interquartile range (difference between the 75th and 25th percentiles) is commonly used as a measure of spread when medians are reported. However, I wouldn't use it or any other summary measure before trying to more thoroughly characterize the distribution of weld distances and trying to understand the nature of the process that produces the welds. This really looks to me more like a statistical process control problem than an outlier-detection problem, so I'd consult the SPC literature. . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
