论文部分内容阅读
目的探讨测量误差变量与准确测量变量混合情况下测量误差对联系效应估计的影响。方法利用测量误差大小、准确测量变量与测量误差变量之间的相关性、准确测量变量的个数和联系效应之间的函数,采用R软件做图来讨论分析测量误差对研究真实性的影响。结果当连续变量y和z能准确测量,连续变量X不能准确测量时,无差异性测量误差使所估计的联系效应值总低于实际值,并随X与Z的相关程度的增加,测量误差所致的偏倚会进一步地恶化。在一个错分二分类变量X和一个准确测量连续变量Z混合的情况下,测量误差所致的偏倚不仅跟暴露测量的灵敏度和特异度有关,而且跟X与Z的相关系数以及X的暴露比例有关,并且随着相关系数的增加,AF值逐渐减少。在ρ=0.5时,AF值为1.419,变量X对应变量Y的联系效应估计值大于实际值,但当ρ增至0.9时,AF值为0.474,其联系效应估计值低于实际值,改变了错分偏倚的方向。结论在准确测量变量和测量误差变量混杂的研究中,用线性回归模型来分析估计多个自变量与应变量之间的联系时,对测量误差所致偏倚的识别、控制和评估是十分必要的,对结果的解释要谨慎。
Objective To investigate the influence of measurement error on the estimation of the contact effect when the measurement error variable and the accurate measurement variable are mixed. The method uses the measurement error, accurately measures the correlation between variables and measurement error variables, and accurately measures the function between the number of variables and the contact effect. The software R is used to discuss the influence of measurement error on the authenticity of the study. Results When the continuous variables y and z can be accurately measured and the continuous variable X can not be accurately measured, the non-discrepancy measurement error makes the estimated contact effect value always lower than the actual value, and as the correlation between X and Z increases, the measurement error The resulting bias will worsen further. In the case of a misclassified dichotomous variable X mixed with an accurately measured continuous variable Z, the bias due to the measurement error depends not only on the sensitivity and specificity of the exposure measurement but also on the correlation coefficient between X and Z and the exposure ratio of X , And AF value gradually decreases with the increase of correlation coefficient. At ρ = 0.5, the AF value is 1.419, and the correlation effect between the variable X and the variable Y is greater than the actual value. However, when ρ increases to 0.9, the AF value is 0.474, and the contact effect estimate is lower than the actual value, changing Misdirected the direction of bias. Conclusions In the study of a mixture of accurate measurement variables and measurement error variables, it is necessary to identify, control and evaluate the bias caused by measurement errors when using linear regression models to analyze and estimate the relationship between multiple independent variables and strain variables , The explanation of the results should be cautious.