Background: There are over 17 000 nursing homes in the United States. Within these, special care units (SCUs) provide a separate residential and/or activity locus for residents, and are expected to provide more staff time and more specialized staff assignments. This paper addresses a fundamental issue relating to the nature, quality and quantity of resident care inputs: what impacts of SCUs are associated with added service inputs, and thus with personnel costs, recognizing that personnel account for the majority of costs associated with nursing home care? Aims of the study: The aim of this aspect of the study was to determine the extent to which additions of staff would result in a diminution of deviant behaviors among residents of special care and of traditional care units. Method: The data were collected from a random sample of ten downstate nursing homes located in New York State. Using rigorous sampling procedures, random samples of 40 residents were drawn from each of the facilities, equally divided between special care unit and traditional care unit residents. Thus, the sampling design involved two levels of clustering: subjects were clustered within units and units were clustered within facilities. The observational behaviour measure was taken from the INCARE (institutional version of the Comprehensive Assessment and Referral Evaluation). The behavioral observation measure contains 23 items such as 'disruptive of others', 'picks/pulls clothing', 'repetitive movements', 'repetitive questioning' and 'wandering'. Each item is rated as to frequency of occurrence; ratings are collected on three occasions, and averaged. Outcome and covariate data (e.g., behavior and cognition) were collected by trained research staff who visited each site for three to four weeks of intensive data collection, accomplished through direct resident interviews, staff interviews and questionnaires and chart data abstraction. The clinical staff time data were collected using the InfoAide system, whereby each care provider used a portable barcode scanner to record the type of care given, the recipient and the duration of care.A random effects model using the SAS mixed procedure was applied to the data; adhering to this model, some effects were fixed and some random. The random effects were comprised of the subject (intercept or subject starting point at baseline) and the