Make unbiased predictions.

Start with an estimate of the prediction you need to make.
This estimate should typically be around the average value of the subject you are predicting. One example is starting with an average GPA score if you’re trying to predict someone’s GPA.

Determine the prediction that matches your impression of the evidence.
For example, if you have evidence that a girl could read fluently by the age of four, you may increase your GPA prediction for her, as her high reading level probably means a high level of intelligence.

Estimate the correlation between your evidence and the subject of prediction.
This correlation can be obtained from the number of shared factors. Shared factors are any factors that are shared by both your evidence and your subject of prediction—for example, any factors that cause children who are good readers to become academically successful. One shared factor is genetically determined aptitude. In this case, the correlation between a high reading level and GPA scores would be about 30%.

Move the correlation value of the distance from your original prediction estimate to the adjusted prediction estimate. Using the example above, move 30% of the distance from the original estimate of the GPA to its estimate after adjusting the girl’s reading level.