A randomized experiment to increase response rates to a healthcare survey among individuals with a high predicted probability of preferring Spanish
*Marc N. Elliott, RAND Corporation 

Keywords: Indirect Estimation, Survey Mode, Spanish Language, Randomized Experiment

The Consumer Assessments of Healthcare Providers and Systems (CAHPS) surveys of Medicare beneficiaries are conducted initially by mailout-mailback of a paper instrument, with telephone followup of nonrespondents. Mail surveys are only in English, with a rarely-used option to request a Spanish language instrument, so response rates by Spanish-preferring beneficiaries are low and predominantly by more-costly telephone followup. We used CMS administrative data, including residential address and surname information, to predict the probability of Spanish-phone response in a previous year with 98% concordance. In the subsequent year, 5,000 of the 10,000 beneficiaries with the highest predicted probability of preferring Spanish were randomized to a bilingual mailing. This intervention resulted in substantially increased mail and total response rates but no change in phone response rates among the target population, with a majority of mail response in Spanish in the intervention condition.