I sometimes get poor defect curve fits when running a forecast.  What can I do to improve this?

Getting a good curve fit is important - the poorer the fit, the less confidence you will have in the projected new time for that set of data. The goodness of fit is particularly important during the latter stages of Phase 3 because defect data is weighted more heavily at that time. There are two things you can do to improve your defect curve fits:

1.  Assess the reliability of early data points carefully. If you have little confidence in the earliest defect data, you should throw it out. Our theoretical defect data curve algorithm assumes defects are being captured from the start of Phase 3 in a consistent and systematic manner. In practice, however, systematic logging of defects is often not performed until around Systems Integration Test (71% of phase 3 duration). If your early defect data is very erratic, chances are those data points are unreliable. If you know (or suspect) this to be the case, exclude the unreliable data points from curve fitting and forecasting. Unreliable data introduces random noise into SLIM-Control's calculations and creates unreliable forecasts.

2.  Try to determine the best tuning factor for your defect data. This is fairly easy to do once any early noisy data points have been discarded. There are two ways to get a handle on the defect tuning factor.  You can use either one, or combine the two methods:

• The first method uses past projects to generate a historic defect tuning factor. You can use any similar projects captured in SLIM-DataManager. Individual defect tuning factors for each project can be displayed on DataManager’s Project List View by selecting Tools | Customize View Layout from the menu. To get the average Defect Tuning Factor for one or more completed projects, import them into SLIM-Control using the History | Load Projects menu item and then run the Defect Tuning Calculator.  If you plan on using an average DTF from your history, you may wish to inspect the data and eliminate outliers to prevent distortion of the average.
• Once you have entered at least three actual total defect data points into SLIM-Control the Defect Tuning Calculator can also calculate a defect tuning factor from your actual defect data.

Once the expected defect total has been tuned to your history or actual data, you should see an improvement in your defect curve fits. If you’re tracking Defects Found by Category, here's one final step that will help ensure optimal curve fits:

• Tune the % of total defects for each category using your actual data. For more information on how to do this, see the FAQ entitled, "How Do I Tune My Defect Category Percentages?".

If you are still getting poor data fits after performing the steps outlined above, either try to determine the reasons or simply exclude defect data from the forecast curve fit.