Survey researchers desire both a good response rate and quality data within the limits of available resources. We report on data collection results from a mixed-mode representative sample survey of 800 continuing care retirement communities (CCRCs) for seniors, where the targeted respondent was the CCRC director. The CCRCs belong to the American Association of Homes and Services for the Aging (AAHSA)-a national membership organization. To achieve a better response rate than in previous establishment surveys of AAHSA members (averaging 20% for mail surveys), we drew on Dillman's recommendation to be flexible. We provided sample members multiple modes to respond to a 45-item survey about resident participation in physical activity. Our data collection design included: awareness messages about the upcoming survey using AAHSA's established channels for communicating with members; an advance letter; U.S.- mailed questionnaire; U.S.-mailed reminder post card; e-mail reminders; phone call reminders; and, the option to complete the survey by web. By adding the web option, we increased survey completions by 47% (adding 128 web completions to 270 mail completions, for an overall response rate of 50%). However, web respondents had four times higher average item non-response than mail respondents (12% to 3%, p < .01). A similar inverse relationship between unit- and item-level non-response occurred in a randomized experiment with a subset of the sample (n = 157) for which we had no e-mail addresses (for whom we only offered the mail option). The reminder postcard treatment group (no advance letter) had a higher response rate than the advance letter treatment group (no reminder post card) (35% to 23%, p = .10). Yet, the reminder postcard treatment group had higher average item non-response than the advance letter treatment group (4% to 3%, p < .01). We will discuss this inverse relationship and its implications for establishment survey design.