Yang Mei-Xia, Zhou Yi-Biao, Jiang Qing-Wu
Xuhui District Center for Disease Control and Prevention, Shanghai 200031, China.
Zhonghua Liu Xing Bing Xue Za Zhi. 2007 Aug;28(8):810-3.
To explore the impact of measurement error on the associated effects under the incorrectly-measured variables when mixed with precisely measured variables.
Based on the functions of measurement error, correlation of incorrectly-measured predictors and precisely measured explanatory variables, number of precisely measured explanatory variables and associated effect, the 'R Project for Statistical Computing' method is used to analyze the impact of measurement on the validity of a study.
Under the scenario that the continuous response Y and the continuous explanatory Z are precisely measured but the continuous predictor X is incorrectly-measured, when focusing on inference about the effect of X on Y, the non-differential measurement error always makes the value of estimated effect less than the actual value, and the attenuation effect of measurement error more closely worsens the correlation of X and Z. Under a misclassification dichotomous predictor X with an additional precisely measured explanatory variable Z and focusing on inference about the effect of X on Y, the misclassification bias is not only related to the sensitivity and specificity of exposure measurement, but also to the correlation between X and Z and exposure proportion of X. The attenuation factor (AF) decreases gradually with the increasing correlation between X and Z. For instance, in the p = 0.5 scenario, AF is 1.419, and the estimated effect of dichotomous predictor X on continuous response Y is more than the actual effect. When it increases to 0.9, AF is 0.474, the estimated effect becomes less than the true effect.
In the studies of the impact of measurement error in linear regression with additional precisely measured explanatory variables, the impact of measurement error on the associated effect is relatively complex, suggesting that it is necessary to control and to assess the measurement error bias in order to correctly interpret the results of a study.
探讨当错误测量变量与精确测量变量混合时，测量误差对相关效应的影响。
基于测量误差的函数、错误测量预测变量与精确测量解释变量的相关性、精确测量解释变量的数量以及相关效应，使用“R统计计算项目”方法分析测量对研究效度的影响。
在连续反应Y和连续解释变量Z被精确测量但连续预测变量X被错误测量的情况下，当关注X对Y的效应推断时，非差异测量误差总是使估计效应值小于实际值，且测量误差的衰减效应更紧密地恶化X与Z的相关性。在具有额外精确测量解释变量Z的二分预测变量X错误分类且关注X对Y的效应推断时，错误分类偏差不仅与暴露测量的灵敏度和特异度有关，还与X和Z之间的相关性以及X的暴露比例有关。衰减因子(AF)随着X和Z之间相关性的增加而逐渐减小。例如，在p = 0.5的情况下，AF为1.419，二分预测变量X对连续反应Y的估计效应大于实际效应。当它增加到0.9时，AF为0.474，估计效应变得小于真实效应。
在具有额外精确测量解释变量的线性回归中测量误差影响的研究中，测量误差对相关效应的影响相对复杂，这表明有必要控制和评估测量误差偏差，以便正确解释研究结果。
