We do a lot of weld overlay cladding. Is it logical to have " Surface area of cladding" measured in square inches as Opportunities? If not , what are the alternatives?
INITIAL POSTING Once I've calculated my equation for multiple regression, how do I go about validation? My only direction, for now, is to take the residuals of the calculated vs. actual and create a control chart (3sds). Test an appropriate sample size again and calculate those residuals (actual vs. calculated). If they fall within the original control chart, the equation is considered valid. Does this seem correct? Any other suggestions?
RESPONSE: One approach that builds upon the described basic multiple regression model validation approach is:
1. Use historical data to determine a multiple regression equation. A best subsets approach can be very useful to determine the key process input variables (KPIVs) to include within the equation.
2. Create an infrequent subgrouping/sampling plan such that normal variation levels of the KPIVs will impact the XmR response control chart as common cause variability.
3. Select a sample per the sampling plan.
4. For the sample, record a response and the equation’s KPIVs levels when the response was recorded.
5. Use the regression equation to calculate a prediction value for the recorded KPIV levels of the sample.
6. Determine for the sample the difference between the predicted and the measured response. Do not adjust the predicted equation with the newly recorded response value and KPIV level. I will call this “residual†difference a “prediction deltaâ€.
7. Repeat the above procedure over time following the infrequent subgrouping/sampling plan.
8. Compile collected time-series “prediction deltas†in an XmR control.
9. We would expect that a good multiple regression model would have:
• An in-control “prediction delta†XmR chart
• A t-test that would fail to reject the null hypothesis of the equality of the means of the original multiple regression residuals and the “prediction deltas†(should have a mean of zero).
• An F-test that would fail to reject the null hypothesis of the equality of the variance of the original multiple regression residuals and the variance of the “prediction deltasâ€.
Hope this helps. Please contact me directly if you would like to discuss further.
Community metadata Could not be loaded. No pre-configured community forum parent found for the incoming fid: . You need to add this forum into custom_code/community.sql
Send to friend
Comments
forrestbreyfogle 8/31/2004
INITIAL POSTING Once I've calculated my equation for multiple regression, how do I go about validation? My only direction, for now, is to take the residuals of the calculated vs. actual and create a control chart (3sds). Test an appropriate sample size again and calculate those residuals (actual vs. calculated). If they fall within the original control chart, the equation is considered valid. Does this seem correct? Any other suggestions?
RESPONSE: One approach that builds upon the described basic multiple regression model validation approach is:
1. Use historical data to determine a multiple regression equation. A best subsets approach can be very useful to determine the key process input variables (KPIVs) to include within the equation.
2. Create an infrequent subgrouping/sampling plan such that normal variation levels of the KPIVs will impact the XmR response control chart as common cause variability.
3. Select a sample per the sampling plan.
4. For the sample, record a response and the equation’s KPIVs levels when the response was recorded.
5. Use the regression equation to calculate a prediction value for the recorded KPIV levels of the sample.
6. Determine for the sample the difference between the predicted and the measured response. Do not adjust the predicted equation with the newly recorded response value and KPIV level. I will call this “residual†difference a “prediction deltaâ€.
7. Repeat the above procedure over time following the infrequent subgrouping/sampling plan.
8. Compile collected time-series “prediction deltas†in an XmR control.
9. We would expect that a good multiple regression model would have:
• An in-control “prediction delta†XmR chart
• A t-test that would fail to reject the null hypothesis of the equality of the means of the original multiple regression residuals and the “prediction deltas†(should have a mean of zero).
• An F-test that would fail to reject the null hypothesis of the equality of the variance of the original multiple regression residuals and the variance of the “prediction deltasâ€.
Hope this helps. Please contact me directly if you would like to discuss further.
Forrest Breyfogle
forrest@smartersolutions.com
www.smartersolutions.com
512-918-0280