東華大學圖書館 |

Language: English

Help

回圖書館首頁

手機版館藏查詢

Back

Switch To: Labeled | MARC Mode | ISBD

Bias in ecologic and semi-individual...

Webster, Thomas Fredrik.

Linked to FindBook

Google Book

Amazon

博客來

Bias in ecologic and semi-individual studies.

Record Type:	Electronic resources : Monograph/item
Title/Author:	Bias in ecologic and semi-individual studies./
Author:	Webster, Thomas Fredrik.
Description:	323 p.
Notes:	Source: Dissertation Abstracts International, Volume: 64-01, Section: B, page: 0167.
Contained By:	Dissertation Abstracts International64-01B.
Subject:	Health Sciences, Public Health. -
Online resource:	http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=3077482
ISBN:	0493980415

Bias in ecologic and semi-individual studies.
Webster, Thomas Fredrik.

Bias in ecologic and semi-individual studies. - 323 p.

Source: Dissertation Abstracts International, Volume: 64-01, Section: B, page: 0167.

Thesis (Sc.D.)--Boston University, 2003.

Semi-individual studies measure exposure on the group level but outcome and covariates for each individual. Such studies are common in environmental and occupational epidemiology. For example, air pollution studies typically assign exposure based on measurements at central air monitors. Are semi-individual studies subject to ecologic bias or are they just individual-level studies with measurement error? I qualitatively and quantitatively examined bias in simple ecologic and semi-individual studies constructed from 2 x 2 tables (or linear risk functions) and analyzed using ordinary or weighted least squares. Since only exposure is measured at the group level, we can view semi-individual studies as individual with measurement error. Indeed, assigning each individual the average exposure of the group introduces no bias in an ordinary least squares analysis. Such studies are, however, still subject to ecologic bias. How can we resolve this apparent paradox? The use of group averages introduces a special kind of measurement error that reduces exposure variance, causing magnification of bias due to group-level confounding and effect modification of the risk difference. Non-differential misclassification of binary exposure within groups biases results away from the null just as in fully ecologic studies. When semi-individual studies include an individual-level binary covariate, the amount of bias caused by residual confounding by group is intermediate between that of fully individual and fully ecologic studies. In sum, we can think of semi-individual studies as individual with exposure measurement error, but they also share certain characteristics of ecologic studies. Since the magnification factor is maximal when exposure is binary, textbook examples will tend to overestimate the amount of ecologic bias. When designing studies, investigators should try to maximize differences in exposure between groups while making exposure within groups as homogeneous as possible. Epidemiologists often partially aggregate data; certain aspects of ecologic bias may be more common than is usually supposed.

ISBN: 0493980415Subjects--Topical Terms:

1017659
Health Sciences, Public Health.

Bias in ecologic and semi-individual studies.
LDR:02978nmm 2200277 4500 001 1858937
005 20041020095006.5
008 130614s2003 eng d
020 $a 0493980415
035 $a (UnM)AAI3077482
035 $a AAI3077482
040 $a UnM $c UnM
100 1 $a Webster, Thomas Fredrik. $3 1946607
245 1 0 $a Bias in ecologic and semi-individual studies.
300 $a 323 p.
500 $a Source: Dissertation Abstracts International, Volume: 64-01, Section: B, page: 0167.
500 $a Major Professor: David M. Ozonoff.
502 $a Thesis (Sc.D.)--Boston University, 2003.
520 $a Semi-individual studies measure exposure on the group level but outcome and covariates for each individual. Such studies are common in environmental and occupational epidemiology. For example, air pollution studies typically assign exposure based on measurements at central air monitors. Are semi-individual studies subject to ecologic bias or are they just individual-level studies with measurement error? I qualitatively and quantitatively examined bias in simple ecologic and semi-individual studies constructed from 2 x 2 tables (or linear risk functions) and analyzed using ordinary or weighted least squares. Since only exposure is measured at the group level, we can view semi-individual studies as individual with measurement error. Indeed, assigning each individual the average exposure of the group introduces no bias in an ordinary least squares analysis. Such studies are, however, still subject to ecologic bias. How can we resolve this apparent paradox? The use of group averages introduces a special kind of measurement error that reduces exposure variance, causing magnification of bias due to group-level confounding and effect modification of the risk difference. Non-differential misclassification of binary exposure within groups biases results away from the null just as in fully ecologic studies. When semi-individual studies include an individual-level binary covariate, the amount of bias caused by residual confounding by group is intermediate between that of fully individual and fully ecologic studies. In sum, we can think of semi-individual studies as individual with exposure measurement error, but they also share certain characteristics of ecologic studies. Since the magnification factor is maximal when exposure is binary, textbook examples will tend to overestimate the amount of ecologic bias. When designing studies, investigators should try to maximize differences in exposure between groups while making exposure within groups as homogeneous as possible. Epidemiologists often partially aggregate data; certain aspects of ecologic bias may be more common than is usually supposed.
590 $a School code: 0017.
650 4 $a Health Sciences, Public Health. $3 1017659
650 4 $a Biology, Biostatistics. $3 1018416
690 $a 0573
690 $a 0308
710 2 0 $a Boston University. $3 1017454
773 0 $t Dissertation Abstracts International $g 64-01B.
790 1 0 $a Ozonoff, David M., $e advisor
790 $a 0017
791 $a Sc.D.
792 $a 2003
856 4 0 $u http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=3077482