Quality Digest

We are trending data from various manuf processes with the objective to see shifts/consistencies and OOCs. Some of the data being graphed are:
1. purity of batches
2. % Yield of production batches
3. time to completion
4. defects/rejects

The approach employed has been to graph the data from each batch over time (cumulative over several years) using M+/- 3SD as the UCL and LCL. If a point is outside the CLs, they are then considered OOC. In the graph, a mean of all data (several years), the mean+/-3sd (sds from all data), and the target line and the individual data point are plotted.

Questions:
1. Is this the right approach? In one of Dr. Wheeler's articles, this approach using the M+/-3sd is the incorrect way of computing limits. If this is the case, should we use XmR charts for the 1st 3 sets of data (purity, yield, time) and use the u, p, c or np charts for the defects? Then use the OOC guideline to monitor OOCs for the purpose of trending.

2. If so, could you clarify this for me also? In an SPC class note: "Based on empirical data, Shewhart found that control limits set at 3 sds from the mean provide the most economical balance between the risks of false signals and unrecognized signals." Is this incorrect?

3. Also, if we are interested in this year's performance, do we use data from previous years (+ the current year) to calculate the X and the R? This is a huge amount of data. Is there a way of reducing this effort?

Thank you.