How AP analyzed test scores under No Child Left Behind law

The Associated Press used school enrollment figures and state compliance plans from the U.S. Education Department to estimate the number of students nationwide whose test scores aren't being counted in required racial categories under the No Child Left Behind law.

The law signed by President Bush in 2002 requires public schools to test more than 25 million students periodically in reading and math, reporting the schools' overall measure of success. No scores are excluded from the overall measure.

But the schools also must report scores by categories, such as race, poverty, migrant status, English proficiency and special education. Failure in any category means the whole school fails.

States are helping schools get around that second requirement by using a loophole in the law that allows schools to ignore scores in mandatory categories that are too small to be statistically significant.

State educators decide when a group is too small to count. And nearly two dozen states have successfully petitioned the government to exclude larger numbers of students' test scores in the required categories.

AP's analysis focused only on scores in the law's five major racial categories: white, black, Asian, Native American and Hispanic. It found schools aren't reporting to the government 1.9 million students' scores in those categories.

The actual total is probably larger since AP did not examine students in other mandatory reporting categories - such as poor, migrant, limited English and special education children - to avoid double-counting students who fall into multiple categories.

AP analyzed student enrollment for each school in grades 3 through 8 and the 10th grade. The last is the most common grade for testing in high school, federal officials said.

AP used 2003-04 enrollment figures for 48 states and the District of Columbia provided by the Education Department's National Center for Education Statistics, the most recent numbers available.

North Dakota and Montana calculate group sizes based on confidence intervals rather than fixed numbers.