How are folks doing their multiple rate analysis: defects per unit analysis by groups
One example we do here in our company is particles per disc surface by product, or scratches per surface by production machine.
I'm wondering if there is a better way to analyze this data in JMP
Today, I use = Analyze > Quality and Process > Pareto Plot, then do a "Per Unit Analysis".
This gives me a "Per Unit Rates" table and "Test Rate Across Groups" output which is useful for supporting the analysis and conclusions.
I was thinking that perhaps there is a better way to do this within Fit Y by X or other platform. I'd like to see other graphical analysis options and have some more flexibility in the way that I can explore the data, more than the Pareto Plot platform provides.
First comment is to be careful in the analysis of Pareto charts. I have found many folks interpret these charts incorrectly. You are not only looking at the ordering of the height of the bars, but you are looking for "jumps" between the bars. Follow the special cause/common cause model from Deming. (or the assignable cause/random variation model from Shewhart). These charts also lose a critical component to understanding variation...the time series.
1. The response variable of "defects" is less than optimum for understanding causal relationships. There are too many potential failure mechanisms in creating a defect. My first thought is to create continuous scales for your measurements (Y's). Can the defects be described better? (e.g., Length, depth, location of scratches). Can you increase the scale from discrete (defects) to ordinal (with a minimum of 5 points on the scale...e.g., 1-5)? The more continuous like, the more efficient the study.
2. Develop a list of hypotheses, possible explanations as to why the defects occur. You have listed by Product and Production Machine. Why would those contribute to defects? Your hypotheses should be explanations as to why the phenomena occurs, the more specific, the easier to to think about how and what data to get to provide insight to your hypotheses.
3. With your multiple hypotheses (and multiple x's), you can open up many opportunities to look for relationships between input variables and response variables. You can:
design sampling plans to help separate assign leverage and to assess consistency.
design experiments to reduce to identify causal relationships between factors and response variable.
look for multivariate relationships between dependent variables (this can help identify continuous variables that may be associated with the defects).
use variability plots to look for patterns in the data and associations between inputs and outputs (you may learn to love graph builder...)
use control charts to look for consistency.
run fit model (and fit Y by X) to determine relationships between inputs and outputs.