LEAD STORY-DATELINE: The Wall Street Journal, February 11, 2000.

Education in the home rather than in public schools is becoming more popular in this country. More children are withdrawing from public schools in favor of learning their lessons at home. One of the possible reasons for this movement is the increasing violence in public schools. After the disaster in Littleton, Colorado, the number of registered home-schooled children rose 10%.

Not only are more children staying at home, they seem to be doing very well academically. College-admissions tests indicate that for reading, English and science, home-schooled scores are better than public-school scores. For the ACT, home-schoolers scored 23.4, 24.4 and 21.9 compared to national averages of 20.5, 21.4 and 21.0. Only in math did home-schooled children do worse than the national average (20.4 compared to 20.7). The SAT had similar results with children studying at home scoring 1083 compared to the national average of 1016.

Interestingly, children schooled at home do not come from the highest income levels. The average income for the homes of these children is $40,000 to $50,000 as opposed to the national median income of $50,000 to $60,000. Their parents, however, do have more education than the national average. Home-schooled children also do not fit the racial stereotype, as 8% are nonwhite.

TALKING IT OVER AND THINKING IT THROUGH!

1. Based on this article and your knowledge of college-admissions tests, what variables would you include in a multiple regression model to predict ACT scores? Be certain to include dummy variables (categorical variables).
   

2. If you included family income and family education in the model, is there a potential problem with collinearity? If so, how would you test for this and what could you do to remedy the problem?
   

3. If your model includes the dummy variable home-schooled/public schooled and the variable for family education, could there be an interaction between both of these? In other words, is the level of education more important for home-schooled students in predicting ACT then it is for public-schooled students? How would you represent and test for an interaction between these variables?
   

4. After deciding on the variables to include in the model, explain how stepwise regression, R²_adjusted and C_p could be used to improve upon the model.
   

THINKING ABOUT THE FUTURE!

If college-admissions tests are an indication of the success of students in college, it would appear that higher scores mean more successful students. If true, it would be a logical goal of our educational system to improve the test scores of our high school students. How can we do that? By developing a regression model that includes several important variables, we can explore some of the relationships that act to improve test scores. Once we can identify significant variables, we can then work to understand why they work. For example, family income is usually positively correlated to high test scores. In the case of home-schooled students this relationship may not hold. Why? What is it about students being taught at home that we might be able to bring into the public school classroom to improve performance?

President Clinton is asking for more teachers and smaller classrooms. Is this the answer to improve our test scores? Certainly the individualized attention given students at home may have a positive effect on scores, but is that the entire story or are there other variables at work?

- Sandra Strasser