Effects of Adjustments on Survey Estimates
Investigator: Richard L. Valliant
Numerous procedures or strategies for adjusting national health survey estimates have been proposed and some have not been studied in sufficient detail to fully understand the consequences of their use. The reasons for the interest in these adjustments include accounting for discrepancies in target population coverage and for unit-nonresponse, calibration of estimates to known values for auxiliary target-population parameters (e.g., raking estimators for totals or means) and constraining estimators of ostensibly the same population characteristics to agree across independent surveys. (The last of these is an important example of where the research is incomplete.) The purpose of this research is to investigate the statistical properties of a broad range of adjustment procedures for a variety of health interview and health care surveys. The investigation will include a study of appropriate variance estimators, such as the well-known linearization and replication methods, which can include extensive use of computer simulation models.
One of the standard techniques used in household and other surveys is raking where estimated counts of units, e.g., persons, are forced to equal known totals from an external data source, like a population census. This is done by computing survey weights in such a way that estimates of population counts by, say, age, race, and sex are exactly equal to census counts (or demographic projections). Rather than controlling to the full age ´ race ´ sex table, only the age, race, and sex margins may be used. In that case, the solution for the weights is found by an iterative procedure. In some cases, estimates from a large survey, such as the National Health Interview Survey, are used as control totals instead of actual census counts. This gives survey designers the flexibility to use variables for which no census data are available.
The general objective of this work is to investigate the properties of raking procedures that are commonly used in survey estimation. In particular, properties, such as bias, variance, and mean square error of estimated totals will be studied.
| Funding: | Health and Human Services, Department of-Centers for Disease Control and Prevention |
Funding Period: 08/23/2004 to 08/31/2006
PSC Research Theme:Analysis and Modeling (Methodology)
PSC blog
Recent resources, events, news
New Publications
Bingenheimer & Geronimus, "Behavior & HIV"
Wildeman, "Imprisonment & Infant Mortality," PSC Research Report
Next Brown Bag
Tues, Dec 1
Arland Thornton & Barb Koremenos
Mobilizing for Human Rights
For live stream
LINK HERE
W A R N I N G
If you are reading this, it may be that you are using rather old web browsing software that does not support modern international Web technology standards. For a better experience of the Web and this site in particular, please upgrade your web browser software today. The following are good choices: Firefox 2; Opera 9; Safari 3.