ProduceRegressionEstimate

Linear regression is a very simple technique to perform estimates as long as there exists a linear dependency between some size factor and the duration of the task. Calculating a linear regression requires a minimum of 3 data items. However, more will produce more reliable results. The calculation behind the technique will draw a line in your dataset that is as close as possible to all points, making the average distance from entry to line as small as possible.

For this page, it will be assumed that your dataset contains an FP value for the function points and an Actual value for the time the task actually took. Feel free to use your own values. For example, comparing the line of code count to the amount of bugs faults reported may reveal interesting information.

Set-up

Access the estimation interface for the task that needs estimation.

Procedure

1. Calculate the total amount of function points identified in your task by dragging the FP column in the task breakdown to the Data box of the Distribution tool. The calculated fields will be populated.
2. Double-click on the sum to copy it to the short term memory and drag it to the task's FP metric for future use.
3. Switch the tool section to the Regression tab and the data panel to Data Pool.
4. Filter the relevant dataset and drag the FP column to the Independent box and the Actual column to the Dependent box. Base facts for the linear regression will be populated.
5. Verify that the correlation (R²) is sufficient. The value will be between 0 and 1. The closer to 1, the better. Values under 0.75 should be considered irrelevant. Repeat previous step as needed.
6. Drag your function point count to the Predictor field. The Forecast field will be updated.
7. Drag the Forecast field to your task's estimation metric.

Close-out

Verify that the data used as historical data is relevant to your project.

Consider the Confidence and Range values. The confidence is a parameter you can adjust to calculate the statistical likelihood of your estimate. The default value of 0.90 provides a high fidelity value, but the initial range it provides may be larger than you would like.

Also see

• Analyze and Measure Tasks
• Produce an Estimate With Performance Average

---