Current Projects

Functionality and Usability of Data from the
American Community Survey

_____________________________________________________________________________

Background

This study will examine the functionality and usability of data from the American Community Survey (ACS), the ongoing continuous survey that the U.S. Census Bureau intends to implement as replacement for the decennial census long form (thus permitting the 2010 decennial census to collect only those basic data items currently on the census short form). The main charge of the panel would be to study the effects of introducing estimate based on multi-year measurements (including moving averages) into applications where static long-form estimates are currently applied. The panel may also identify possible opportunities for combining ACS results with data from other sources (e.g., census household surveys and administrative records). The major goal is to provide a base of information to ease the transition from the long form to the ACS for a wide variety of organizations and individual data users.

Many of the questions related to the use of ACS data are based on a fundamental statistical trade-off between bias and variance. In simplest terms, the census offers users estimates that may have large statistical biases (due to time lags) but, relatively, very low variances. On the other extreme, the estimates that would arise from looking at a single year of ACS data: a small sample, relative to the census long-form sample? would have low statistical bias but potentially substantial increases in variance. The goal for optimal use of ACS data is to strike a compromise between these positions. The proposed moving average methodology strikes one possible middle position; it remains to be explored by this study whether that position is optimal for meeting data users’ needs. Another important task would be to assist users of the data in under- standing how to take advantage of more frequent, more current estimates, not just
understanding how to continue their current uses. A major potential concern over using ACS estimates from a statistical perspective is the impact on estimates of change: e.g., change in per capita income from one year to the next:: when said estimates are based on moving averages. In particular, the properties of the variance of the difference are considerably more complicated and may easily be misunderstood. A more general version of the same basic problem would occur in applications if moving average estimates were incorporated in statistical analyses such as regression and time series models. For that reason, the panel would also assess using the estimates for individual years in such applications.

The basic structure of the ACS suggests other areas of statistical concern and interest, among them the ACS methodology for over sampling small areas and its implications. From a modeling perspective, the implicit hierarchical structure of ACS estimates merits extensive consideration. For instance, small areas (e.g., a small school district) are intended to be estimated using five-year moving averages. However, that small area may be nested within a larger population construct (e.g., a county) for which three-year moving averages may be constructed, and that construct may itself lie within a large population unit (e.g., a state) for which one-year ACS estimates may be valid. The potential of exploiting this hierarchical structure has not been fully examined, and the panel would also explore this area. In a more general sense, the panel would consider the advantages and possible difficulties moving averages or other estimates pose in a wide variety of public and private planning contexts, including transportation planning, disaster response, and assessment of local conditions.

_____________________________________________________________________________

Roster

_____________________________________________________________________________

Staff

_____________________________________________________________________________

Publications