When planning, it’s useful to know how well your plans predict your actual performance. For example, in PSP, correlation is measured between estimated and actual size and between estimated size and actual effort. We can also correlate between planned and actual effort.
We use a mathematical tool called linear regression” and correlation using “Pearson’s Correlation”, or the correlation coefficient. . The correlation coefficient, “R^2” measures the degree to which the predictor variable predicts the response variable. These are sometimes called the independent, and dependent variables. .
“R” varies from -1 to 1.0, we usually take the square of the value. R=1.0 implies perfect correlation, the two values move together with no variation. R=-1.0 implies perfect anti-correlation, that is they move in opposite directions. There may or may not be a direct causal relationship, all the math tells us is that they have tended to move together. Causality is another discussion, as is Significance.
For planning, the following guidelines are useful.
- R^2 >0.9 the relationship is highly predictive, you can use it with confidence
- 0.7 <R^2 < 0.9 , the correlation is still strong, it’s good enough for planning
- 0.5 <R^2 <0.7 , may be good enough for many purposes, but use the relationship with caution
- R^2 < 0.5, the relationship is not strong enough to be reliable for planning.
In the following example, we use a scatterplot. Scatters are useful for visualizing relationships between two variables. The trendline was calculated by Microsoft Excel. The correlation between planned values (x axis) and actual (y axis) is fairly weak. This might result from estimation, or there might be process issues. This is a pretty typical starting point, but some work must be done before the planned values can useful for making a reliable work plan.
References and Bibliorgraphy
The PSP Body of Knowledge, V2.0 www.sei.cmu.edu/reports/09sr018.pdf