Below is our story: a university has two systems, 1 series and 2 lines. The school leadership requires increased the number of girls in school. The following figure is the situation at the time:
Figure 1: 1980, 1990 Schools 1 and Series 2 boys and girls
In 1980, there were 50 boys and 50 girls, so there was a 50% of girls. By 1990, 10 boys and 15 girls and 15 girls increased to 60%. Therefore, the system performs the decision of the leadership.
In 1980, 2 had 25 boys and 5 girls, so about 17% of girls. By 1990, 2 have 75 boys and 20 girls. The ratio of 2 girls is therefore increased to approximately 21%, so that the series 2 also fulfilled the decision of the school leaders. Figure 2: The number of men and girls across the school and the proportion.
The proportion of the above statement indicates that the two systems have added the number of girls from 1980 to 1990. Everything looks very normal. However, the data added by two series of men and women represents a completely different ratio. The school leaders are calculated. He first adds the number of 1980 boys and girls, and then adds 1990. He then calculates the ratio. The proportion of girls calculated by school leadership is 29% from 1980 to 29% in 1990! However, two systems are called the proportion of girls from 1980 to 1990. They can prove. The problem is: what is wrong. Some people say that the above example will not appear in reality. They said it is an example of human manufacturing. We only explain that some of the most useful software metrics in the industry is similar to the properties of the Simpson's paradox. This is the defect-density metric (MEASURE DEFECT-DENSITY). Please read the speeches 2. Second Lecture.)
