U.S. flag

An official website of the United States government

Skip Header


Feasibility of Implementing Mode Switching Rules to Improve Survey Estimates for Language Households

Written by:

Shin and Kominski (2010) document the growth in the number of people speaking a language other than English at home from 23.1 million in 1980 to 55.4 million in 2007. They found that in 2007 about 1-in-5 households spoke a language other than English at home. Of these, about 8 percent reported not speaking English at all. In the 2011 American Community Survey (ACS), 60.6 million people (about 20.8 percent of the population 5 years and over) reported speaking a language other than English at home with nearly 9 percent speaking English less than "very well" (U.S. Census Bureau, 2012).

The ACS is a large national sample survey that produces demographic, social, economic, and housing estimates for a broad set of geographic areas. It uses four modes of data collection -€“ Internet, mail, telephone, and personal visit. After allowing time for Internet and mail responses, the call centers conduct telephone follow up activities using Computer Assisted Telephone Interviewing (CATI) methods. All noninterviews after Internet, mail, and telephone attempts and all sample cases that were not eligible for Internet, mail, or telephone are eligible for selection in the subsample for personal visit follow up. We use Computer Assisted Personal Interviewing (CAPI) for this follow up. We currently apply differential sampling rates at the tract level to select the CATI noninterviews sent to CAPI. Specifically, we select a larger sample of nonrespondents in tracts with the lowest expected levels of response by mail and telephone. U.S. Census Bureau (2010), Section 4.3, includes a complete discussion of this subsampling. We currently do not use any other information (e.g., paradata) when selecting the sample for CAPI follow up.

The motivation for this research is concern (articulated by advisory committee members) about the loss of sample for households that speak languages other than English and have limited English skills. Joshipura (2008) found that "€œlinguistically isolated households" have a significantly lower propensity to respond by mail when compared with households that are not "€œlinguistically isolated." Based on the 2005 ACS she estimated that only about 31 percent of linguistically isolated households (versus 62 percent of households that were not linguistically isolated) responded by mail. This suggests that these households with limited English proficiency are more likely than English-speaking households to be in the universe for CAPI subsampling. It may be possible to improve the quality of survey estimates for the limited English population if the CAPI subsample took advantage of paradata to identify these cases. Specifically, oversampling CATI nonrespondents with limited English proficiency could improve the reliability of survey estimates for these populations.

To assess the potential benefits derived from using paradata on noninterview reason in CAPI subsampling when selecting the CAPI subsample, we must first demonstrate that these paradata successfully identify the cases we wanted to oversample. This project focuses on this aspect of the research. If the results show promise we should consider additional research to determine if oversampling these cases will yield lower overall variances of survey estimates for limited-English proficient households.

Page Last Revised - October 8, 2021
Is this page helpful?
Thumbs Up Image Yes Thumbs Down Image No
NO THANKS
255 characters maximum 255 characters maximum reached
Thank you for your feedback.
Comments or suggestions?

Top

Back to Header