The 2019 ASR design is similar to the 2018, 2017 and 2016 ASR designs2, which used a full cross-sectional national sample of refugees entering within the past five years. The population of interest – the study population – for the 2019 ASR is defined as refugees entering the U.S. between FY 2014 and FY 2018, inclusive, who are at ages 16 and over at the time of the 2019 ASR interview. Because the interviews were conducted in early 2020, the population includes a small number of refugee respondents younger than 16 at the time of arrival to the U.S.
The 2019 ASR targeted 1,500 completed interviews from refugee households entering the U.S. between FY 2014-2018. The sample was designed to allow for separate estimates and analyses from each of the three designated cohorts. Moreover, the design needed to accommodate both household- and person-level analyses. The sample was drawn as fresh cross sections by cohort; there was no longitudinal component. The survey objectives required that – in addition to primary stratification by cohort – the sample of households (i.e., PAs) be stratified at least by year of entry and geographic region of origin.
The 2019 ASR employed a stratified probability sample design of refugees. The first stage of selection was the household (PA) and the second stage was the selection of persons within households. Within each of the three cohort strata, the following factors were used for stratification:
cohort, year of arrival (for cohorts 1 and 2 only); geographic region, native language, age group, gender, and family size at arrival (1, 2, 3+ persons). Proportionate stratified samples were drawn independently within cohort.
A replicated sample design was used for the 2019 ASR. A larger sample of 10,800 was drawn and randomly split into equal sized replicates to produce 30 independent sample replicates per cohort. Under this strategy, 20 replicates comprised the 7,200 that were expected to be needed based on the assumptions of tracing and nonparticipation used above. Upon commencement of field operations, 10 of the 30 replicates for each cohort were released into the field and the results monitored by cohort, thus releasing about 3,600 sampled PAs
