The National Center for Education Statistics (NCES) at the US Department of Education publishes reports on a variety of surveys. In addition to these reports, NCES also provides a Data Analysis System (DAS) that can be used to create custom analyses of the data in these surveys.

The data analysis system (DAS) is an easy-to-use web-based tool for writing and executing data queries. The trick is in crafting the right query and in correctly interpreting the results.

Be careful about hidden assumptions in the variables, especially when those assumptions are relevant to the nature of the analysis:

• Some variables restrict the data set to just dependent students. For example, the DEPINC variable measures the income of parents of dependent students, excluding independent students, while CINCOME is a continuous variable that measures the income of independent students in addition to the income of parents of dependent students. This can lead to inaccurate results at colleges with a disproportionate share of independent students, such as for-profit and community colleges.

• Other variables restrict the data set to just Bachelor's degree students. For example, the COLLGRAD variable in the NPSAS measures whether a student was a graduating senior or not. It assumes that the undergraduate degree program (UGDEG) is a Bachelor's degree and that the student completed the degree program (PROGSTAT). Using the COLLGRAD variable will omit Associate's degree and Certificate recipients.

• The NPSAS includes two data sets, one for undergraduate students (UG) and one for graduate and professional students (G1P). Often researchers use just the undergraduate data set and are mystified as to why their data does not match national data for all colleges.

• Both the CENTILE and CENTILE>0 directives generate percentile distributions for a variable, but the latter omits zeros from the distribution. This can yield a big difference, for example, if one is measuring debt at graduation. Using CENTILE lets one find the median debt for all students, including those who did not borrow, while using CENTILE>0 lets one find the median debt among just the students who borrowed to pay for their education. Sometimes researchers report a median debt figure without clarifying whether the median is for all students or just the students with debt.

• Institution level (2-year, 4-year) is not the same as degree program (Associate's degree, Bachelor's degree). A 4-year institution, for example, can graduate students with Associate's degrees and Certificates in addition to Bachelor's degrees. Sometimes researchers using institution level as a proxy for degree program, not realizing that results can differ significantly.

• Some data sources, such as the Integrated Postsecondary Education Data System (IPEDS), are limited to first-time first-year undergraduate students. This can lead to inaccurate results at colleges with a high transfer-in rate.

Read the variable descriptions carefully, especially if they reference another variable for further information. For example, the net cost, effort and need variables in the NPSAS exclude negative values and round up small values to $10. This prevents one from demonstrating that financial aid to higher income students may exceed financial need.

Long-time users of the DAS add or subtract half a point (0.5) to numeric thresholds to test for equality. For example, "< 1000.5" is the equivalent of "<= 1000". The modern DAS includes an "=" checkbox that should have the same effect, but there have occasionally been errors in its implementation.

The weighted sample sizes presented in the DAS reports can themselves be treated as data for analysis, not just the estimates in the report. For example, one can use them to calculate the distribution of recipients as opposed to dollars.

Become a power user of the DAS to save copies of your table parameter files (TPF), making it easier to reproduce results. It's also a good idea to save the data in a spreadsheet, giving it and the TPF similar names.