The Effect of Complementing Current Population Survey Data with American Community Survey Data 
Jaingjaing Yang
Department of Statistics
Colorado State University
Monday, May 10, 2004
1pm
Room 006, Statistics
The M.S. oral examination of Jiangjiang Yang will take place Monday, May 10,
2004. The examination will begin with JJ’s seminar, “The ffect of
complementing current population survey data with American community survey
data,” at 1:00 p.m. in Room 006 of the Statistics Building. The oral
examination, which is open to all interested faculty, will immediately
follow the seminar presentation.
ABSTRACT
Traditionally the Bureau of the Census has relied mostly on the Current
Population Survey (CPS) to help estimate the labor force statistics.
However, the data from the CPS are not large enough in quantity. As a result
BLS can only develop reliable estimates with the CPS at the national level
or the state level. At the substate levels, BLS can rely little on the CPS
due to its lack of size.
The American Community Survey (ACS) is a recently developed ongoing
largescale survey that the Census Bureau uses to collect census type data
on some basic population characteristics. After the introduction of the ACS
we now see an opportunity to combine the new source of data with what we
have from the CPS in estimating the labor force statistics. We would need to
incorporate the ACS data into the CPS framework to get the desired results.
In this paper we study and compare the Current Population Survey (CPS) and
the American Community Survey (ACS). We construct an ACS model compatible
with the structural CPS model that has already been established by the US
Bureau of Labor Statistics (BLS). We then combine both the ACS and the CPS
into one statespace model and evaluate its performance in estimation in
comparison to the CPS model alone. More specifically, we want to compare the
mean square errors of the estimates (predicted by the Kalman Filter) of the
CPSonly model and the CPSACS combined model. We look at the ratio of the
two mean square errors; and we use a factorial analysis to explore the
relationships between the ratio and a number of error factors.
