Johannes C B, Crawford S L, McKinlay J B
New England Research Institutes, Watertown, MA 02172, USA.
Am J Epidemiol. 1997 Sep 1;146(5):429-38. doi: 10.1093/oxfordjournals.aje.a009296.
Although interviewer error is widely recognized as an important source of variation in epidemiologic investigations, scant published information exists documenting the impact of interviewer variation on study findings. Using data from the Massachusetts Women's Health Study, a population-based cohort study of 2,569 middle-aged women (1982-1987), the authors evaluated interviewer variation in responses to different types of questions, and assessed the impact of interviewer variation on inferences derived from study data. Respondent sociodemographic and lifestyle characteristics were similar for the four study interviewers at the first follow-up. No interviewer variation was detected for questions concerning recall of specific events, but responses to questions regarding recall of subjective or personal information or those which required further probing did differ significantly by interviewer. Adjustment for interviewer effects had no impact on the conclusions obtained from one analysis of predictors of depression, despite significant interviewer variation in the outcome and predictor variables, but it did change conclusions from an analysis of the impact of support networks on psychological symptoms, wherein the interviewer variable was strongly related to the outcome after data were controlled for predictor variables. Given these findings, examination of data for interviewer effects is advisable despite incorporation of quality control measures in a study's design.
尽管访谈者误差作为流行病学调查中一个重要的变异来源已得到广泛认可，但关于访谈者差异对研究结果影响的公开资料却很少。作者利用马萨诸塞州妇女健康研究的数据（一项针对2569名中年女性的基于人群的队列研究，时间为1982 - 1987年），评估了访谈者对不同类型问题回答的差异，并评估了访谈者差异对从研究数据得出的推断的影响。在首次随访时，四位研究访谈者所接触的受访者的社会人口统计学和生活方式特征相似。对于关于特定事件回忆的问题，未检测到访谈者差异，但对于关于主观或个人信息回忆的问题或那些需要进一步追问的问题，访谈者的回答存在显著差异。尽管在结果变量和预测变量方面存在显著的访谈者差异，但对访谈者效应进行调整对从一项抑郁症预测因素分析得出的结论没有影响，但它确实改变了一项关于支持网络对心理症状影响分析的结论，在该分析中，在对预测变量进行数据控制后，访谈者变量与结果密切相关。鉴于这些发现，尽管在研究设计中纳入了质量控制措施，但对数据进行访谈者效应检查仍是可取的。
