.
Our motivation in examining API was to better understand the factors that best predict .
this score, and separate out any predictive variables that are strongly correlated to the outcome. .
While we may not be able to offer any policy recommendations, our hope is that the legitimacy .
of this Index can be tested through our analyses and that schools can better understand the factors .
that help to influence their scores so that they may better serve their students. .
We conducted our analyses on data published during 2012, which is relevant as the .
makeup of the API metric and data collection methods have been slightly altered over the years .
(i.e. consecutive years are not necessarily comparable). In our preliminary data exploration, we .
made the decision to focus exclusively on high schools because of the observation that there is .
significantly less variability in these cases. We hypothesize this may be a result of more focused .
and directed curriculum. Additionally, we noticed that some schools had a large percentage of .
disabled students. We concluded that these were likely special education schools, and felt it was .
inappropriate to include them in our analyses. In examining the summary statistics, we noticed a .
natural break in the data at about 40%, and thus we only considered schools with a student body .
consisting of less than 40% disabled students. .
Model Building .
We selected potential variables using a combination of personal experience and published .
theories from education professionals. Both of these sources focused on a few main areas: .
parental education, socioeconomic status, the number of disabled students, and the English .
language skills of students. We encountered trouble, specifically when searching for an .
interaction effect, because of the interrelated nature of these components. Parental education .
appeared to be the most strongly correlated of all our variables under consideration to API, .
