There is a lot of debate about whether questions should be presented on a grid or in a single item per screen. Operationally, grids take less time for respondents to complete. Use

of grids should decrease response burden, although new research shows that respondents seem to prefer a single item per screen. From a measurement point of view, grids pose numerous issues: higher item non-response, higher item non-differentiation, and sometimes higher measurement error.

In this experiment, we are testing the Vitality (4 items) and Mental Health (5 items) scales of the SF-36v2® Health Survey. The SF-36v2 asks 36 questions to measure functional

health and well-being from the patient's point of view. It is called a generic health survey, because it can be used across age (18 and older), disease, and treatment groups, as opposed to a disease-specific health survey which focuses on a particular condition or disease. Two of the four items of the vitality scale and two out of five items of the mental health scale are reversed in scoring.

A sample of 2,500 KnowledgePanel® respondents was randomly assigned to one of five experimental conditions: Group 1: Standard grid; Group 2: Shaded grid; Group 3: One item per screen with horizontal response options; Group 4: One item per screen with vertical response options; Group 5: One item per screen with vertical shaded response options. Approximately 360 respondents completed the survey per condition for a completion rate of 73.4%. The survey was optimized to be seen on a screen with minimum resolution of 800 by 600 pixels. During the study we collected the browser type for each respondent. This allowed us to exclude cases in which the survey was taken either on a MSNTV or on an iPhone/PDA because they could not properly see the grid items. The final sample used for the analysis, after exclusions, was of 1,419 cases for an average group size of about 280.

We hypothesized that items presented on a grid would lead to more measurement error as indicated by a higher rate of “inconsistencies” in the self-reports to grid questions and a lower rate of inconsistencies in the self-reports to the single-item questions. We speculated that presenting items on a single screen allows the respondent to bring more cognitive focus to each question and therefore be more consistent in their answers to questions. In contrast, when items are on a grid, it is easier for the respondent to get confused, especially when the meaning of some of the items is reversed. We computed an index of consistency by correlating the total sum of scores for the reversed items with the total sum of scores for the non-reversed items. If respondents are consistent in their answers the correlation between reversed and non-reversed should be higher. We calculated Cronbach's alpha scores to measure consistency in answers for each of the five experimental conditions.

The direction of the study findings were consistent with our hypotheses -- lower alpha level for the grid presentation and higher correlation for the single-item presentation -- although the differences among groups do not reach statistical significance.