Get access

Role of Cognitive Testing in the Development of the CAHPS® Hospital Survey


  • Roger E. Levine,

    Search for more papers by this author
    • Address correspondence to Roger E. Levine, Ph.D., American Institutes for Research, 1070 Arastradero Road, Palo Alto, CA 94304. Floyd J. Fowler Jr., is with The Center for Survey Research, University of Massachusetts, Boston, MA. Julie A. Brown, B.A., is with the RAND Corporation, Santa Monica, CA.

  • Floyd J. Fowler Jr.,

  • Julie A. Brown


Objective. To describe how cognitive testing results were used to inform the modification and selection of items for the Consumer Assessment of Health Providers and Systems (CAHPS®) Hospital Survey pilot test instrument.

Data Sources. Cognitive interviews were conducted on 31 subjects in two rounds of testing: in December 2002–January 2003 and in February 2003. In both rounds, interviews were conducted in northern California, southern California, Massachusetts, and North Carolina.

Study Design. A common protocol served as the basis for cognitive testing activities in each round. This protocol was modified to enable testing of the items as interviewer-administered and self-administered items and to allow members of each of three research teams to use their preferred cognitive research tools.

Data Collection/Extraction Methods. Each research team independently summarized, documented, and reported their findings. Item-specific and general issues were noted. The results were reviewed and discussed by senior staff from each research team after each round of testing, to inform the acceptance, modification, or elimination of candidate items.

Principal Findings. Many candidate items required modification because respondents lacked the information required to answer them, respondents failed to understand them consistently, the items were not measuring the constructs they were intended to measure, the items were based on erroneous assumptions about what respondents wanted or experienced during their hospitalization, or the items were asking respondents to make distinctions that were too fine for them to make. Cognitive interviewing enabled the detection of these problems; an understanding of the etiology of the problem informed item revisions. However, for some constructs, the revisions proved to be inadequate. Accordingly, items could not be developed to provide acceptable measures of certain constructs such as shared decision making, coordination of care, and delays in the admissions process.

Conclusions. Cognitive testing is the most direct way of finding out whether respondents understand questions consistently, have the information needed to answer the questions, and can use the response alternatives provided to describe their experiences or their opinions accurately. Many of the candidate questions failed to meet these standards. Cognitive testing only evaluates the way in which respondents understand and answer questions. Although it does not directly assess the validity of the answers, it is a reasonable premise that cognitive problems will seriously compromise validity and reliability.