Appendix E**Composite Estimation**

This appendix briefly presents the simplest composite estimator for a rotating panel design. There is an extensive literature on composite estimation to which the reader can turn for descriptions of more elaborate approaches.

This discussion assumes an annual data collection. Let _{} denote a parameter to be estimated in year *t*, and let _{} denote the “usual unbiased estimate” of _{} based only on data from time *t*. Note that it is also possible to compute an estimate _{} based only on the overlapping (or “matched”) sample, i.e., using only the sample that is being interviewed for the second time (or more). Since values exist from the previous year for the matched sample, it is possible to compute _{} and so _{} is an estimate of the change between year *t* and year *t*–1 based on the same sample units. By adding this estimate of change to last year’s estimate of _{} a second estimate of _{}, namely, _{} is obtained. (In words, the estimate for this year is equal to the estimate for last year plus the estimated year-to-year change.) Since there are now two estimates of the same quantity _{}, it is natural to combine them to obtain an improved estimate. Thus the so-called composite estimate of _{} is _{} a weighted average of two estimates (assuming α is between 0 and 1). Because is a better estimate than _{} then in practice, the following composite estimate is used: _{}.

The value of α is chosen to minimize the variance of _{}.

Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.

Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter.
Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.

OCR for page 105

Appendix E
Composite Estimation
This appendix briefly presents the simplest composite estimator for
a rotating panel design. There is an extensive literature on composite es-
timation to which the reader can turn for descriptions of more elaborate
approaches.
This discussion assumes an annual data collection. Let θt denote a
ˆ
parameter to be estimated in year t, and let θ t denote the “usual unbiased
estimate” of θt based only on data from time t. Note that it is also possible
ˆ
to compute an estimate θ t ,m based only on the overlapping (or “matched”)
sample, i.e., using only the sample that is being interviewed for the second
time (or more). Since values exist from the previous year for the matched
ˆ ˆ
ˆ
sample, it is possible to compute θ t –1,m , and so θ t ,m – θ t −1,m is an estimate
of the change between year t and year t –1 based on the same sample units.
By adding this estimate of change to last year’s estimate of θt–1, a second
ˆ ˆ ˆ
estimate of θt, namely, θ t −1 + θ t ,m – θ t −1,m is obtained. (In words, the estimate
for this year is equal to the estimate for last year plus the estimated year-to-
year change.) Since there are now two estimates of the same quantity θt, it is
natural to combine them to obtain an improved estimate. Thus the so-called
ˆ ˆ ˆ ˆ ˆ
composite estimate of θt is θ tc = αθ t + (1 – α )(θ t −1 + θ t ,m – θ t −1,m ), a weighted
ˆ
average of two estimates (assuming a is between 0 and 1). Because θ tc−1 is a
ˆ , then in practice, the following composite estimate
better estimate than θ t –1
ˆ ˆ ˆ ˆ ˆ
is used: θtc = αθ t + (1 – α )(θ tc−1 + θ t ,m – θ t −1,m ).
ˆ
The value of a is chosen to minimize the variance of θ tc .
105

OCR for page 105