Ong Ai Rene, Hu Mengyao, West Brady T, Kirlin John A
Institute for Social Research, University of Michigan, 426 Thompson Street, Ann Arbor, MI 48104, USA.
US Department of Agriculture, Economic Research Service, Washington, DC, USA.
Public Health Nutr. 2018 Jul;21(10):1781-1793. doi: 10.1017/S1368980018000137. Epub 2018 Feb 22.
To understand the effects of interviewers on the responses they collect for measures of food security, income and selected survey quality measures (i.e. discrepancy between reported Supplemental Nutrition Assistance Program (SNAP) status and administrative data, length of time between initial and final interview, and missing income data) in the US Department of Agriculture's National Household Food Acquisition and Purchase Survey (FoodAPS).
Using data from FoodAPS, multilevel models with random interviewer effects were fitted to estimate the variance in each outcome measure arising from effects of the interviewers. Covariates describing each household's socio-economic status, demographics and experience in taking the survey, and interviewer-level experience were included as fixed effects. The variance components in the outcomes due to interviewers were estimated. Outlier interviewers were profiled.
Non-institutionalized households in the continental USA (April 2012-January 2013).
Individuals (n 14 317) in 4826 households who responded to FoodAPS.
There was a substantial amount of variability in the distributions of the outcomes examined (i.e. time between initial and final interview, reported values for food security, individual income, missing income) among the FoodAPS interviewers, even after accounting for the fixed effects of the household- and interviewer-level covariates and removing extreme outlier interviewers.
Interviewers may introduce error in food acquisition survey data when they are asked to interact with the respondents. Managers of future surveys with similarly complex data collection procedures could consider using multilevel models to adaptively identify and retrain interviewers who have extreme effects on data collection outcomes.
了解在美国农业部的全国家庭食品获取与购买调查（FoodAPS）中，访员对他们收集的粮食安全、收入及选定调查质量指标（即报告的补充营养援助计划（SNAP）状态与行政数据之间的差异、初次访谈与最终访谈之间的时间长度以及缺失的收入数据）的回答所产生的影响。
利用FoodAPS的数据，拟合具有随机访员效应的多层次模型，以估计每个结果指标中由访员效应引起的方差。描述每个家庭社会经济地位、人口统计学特征和参与调查经历以及访员层面经验的协变量作为固定效应纳入。估计了访员导致的结果中的方差成分。对异常访员进行了剖析。
美国大陆的非机构化家庭（2012年4月 - 2013年1月）。
4826户家庭中对FoodAPS做出回应的个体（n = 14317）。
即使在考虑了家庭层面和访员层面协变量的固定效应并剔除极端异常访员之后，FoodAPS访员之间所考察结果（即初次访谈与最终访谈之间的时间、粮食安全报告值、个人收入、缺失收入）的分布仍存在大量变异性。
当要求访员与受访者互动时，访员可能会在食品获取调查数据中引入误差。未来具有类似复杂数据收集程序的调查管理者可考虑使用多层次模型来适应性地识别和重新培训对数据收集结果有极端影响的访员。
