Mixed-Mode

Reasons for Use

Researchers often employ mixed-mode survey designs to address problems associated with undercoverage of key groups of interest or to improve participation rates. Some mixed-mode designs can also be conducted at lower cost than single mode surveys, particularly when survey modes costing less (e.g. mail) are employed first to obtain a large percentage of the completed interviews and more costly methods (e.g. in person) are used later to attempt to interview initial nonrespondents.

The decision to use a mixed-mode approach and the particular design to employ is based on several considerations. First is the need to identify the survey design that best meets the study's objectives, in terms [Page 473]of the research question(s) posed, population of interest, and amount of data to be collected. Next is the desire to reduce the total survey error in a project, which is the error from all potential sources, including coverage, sampling, nonresponse, and measurement error. The decision is also affected by the time frame available for data collection. Some modes (such as mail surveys) require considerably longer field periods than other modes (such as telephone surveys). Finally, cost is an important consideration, given that researchers typically need to operate within a fixed budget.

Bias and Cost Differences

From a research perspective, an optimal mixed-mode design is one that for a fixed budget will reduce the total amount of error in the survey to the greatest extent possible, thereby reducing the potential for bias in the survey estimates. This typically involves consideration of survey coverage, sampling error, nonresponse, and measurement error. The frame used for sampling potential respondents must include all units in the population of interest. If particular types of sample units (persons, households, businesses, etc.) are not included in the sampling frame, then coverage bias may result. Modes differ in their degree of coverage, with face-to-face surveys (based on residential addresses) often having the highest levels of coverage and Internet the lowest.

Coverage rates by mode can and do change over time. For instance, in the late 1990s landline telephone surveys had household coverage rates of 95%-98%. Since the advent and rapid adoption of cellular telephones, however, landline telephone frame coverage of households has dropped significantly (e.g. to less than 80% by 2008). Coverage of the general population for mail survey has improved steadily over time with the development of computer technology and large address databases. Internet surveys of the general population based on sampling email addresses tend to have very low coverage because there is no centrally maintained or comprehensive listing of email addresses. However, Internet surveying via email addresses of known lists of individuals (e.g. members of a professional organization or students at a university) can have extremely high coverage.

...