A set of numeric weights attached to answers on a loan application or items in a credit report. The sum of the weights for a particular application produces a credit score, which is a number that indicates how likely it is that a person will pay back a loan.

Credit scorecards are created using statistics. The first step is to collect past loan applications and put them into two groups: the "goods", which are the people who repaid their loans, and the "bads", which are the people who defaulted on their loans. Next, the distributions of each variable on the loan application or credit report are examined and the distributions of the goods are compared to the distributions of the bads. For example, suppose we look at the variable age:

Age Range  Goods   Bad   Ratio  round(10xlog(Ratio))
under 21    10     40     1/4       -6
21-35       20     30     2/3       -2
36-55       30     20     3/2        2
over 55     40     10     4/1        6

Taking the logarithm of the ratio puts it in a more convenient form: items with a larger number of goods are positive, and items with a larger number of bads are negative. Scaling and rounding the result makes the weights easy to understand and manipulate.

Of course, it's not this easy. How did we know to create exactly four age groupings? In statistics, this process is known as binning. Another problem is which variables to use in the scorecard. Variables are often correlated--for example, age, salary, and home ownership--and if multiple highly correlated variables are used, the net effect is to count the same information more than once.

An alternative to scorecards is neural networks. The advantage of neural networks is that can be better predictors. The major disadvantage is that they reveal nothing about the factors that influence the prediction. A scorecard shows immediately which factors are important, whereas a neural network usually gives no insights whatsoever.