Replication data: Challenges and benefits of including institutionalized, cognitively impaired, and unable-to-respond individuals in a representative survey of the very old

Institutionalization, cognitive impairment, and the inability to conduct an interview due to health impairment are among the top exclusion criteria for most large-scale social and aging surveys. Reservations about targeting vulnerable groups result from economic or legal restrictions of recruitment and concerns regarding research ethics or the validity of the data obtained. However, failure to include these individuals may lead to substantial bias. Metadata showed that privileged data access and checks against nursing home repositories prevented the undercoverage of institutionalized individuals. Measures to include difficult-to-survey groups led to a marked increase in response rates. Individuals with health impairments substantially contributed to the representativity of the sample. Nonresponse bias was cut in half when compared with a less inclusive study protocol. From a Total Survey Error perspective, reductions in nonresponse bias, low item-nonresponse, and evidence of measurement invariance across self-reports and proxy reports for key outcome variables show significant benefits of including difficult-to-survey groups in estimating characteristics of this population. Institutionalization, cognitive impairment, and the inability to conduct an interview due to health impairment are among the top exclusion criteria for most large-scale social and aging surveys. Reservations about targeting vulnerable groups result from economic or legal restrictions of recruitment and concerns regarding research ethics or the validity of the data obtained. However, failure to include these individuals may lead to substantial bias. Metadata showed that privileged data access and checks against nursing home repositories prevented the undercoverage of institutionalized individuals. Measures to include difficult-to-survey groups led to a marked increase in response rates. Individuals with health impairments substantially contributed to the representativity of the sample. Nonresponse bias was cut in half when compared with a less inclusive study protocol. From a Total Survey Error perspective, reductions in nonresponse bias, low item-nonresponse, and evidence of measurement invariance across self-reports and proxy reports for key outcome variables show significant benefits of including difficult-to-survey groups in estimating characteristics of this population.