Changes in the regulatory environment and in resource availability are thought to have had differential impact on home health agencies of different types over the period since Medicare and Medicaid were implemented in 1966. This study uses survival analysis methods including the proportional hazards model to examine an administrative data base of annual survey data for 199 agencies that reported to the Massachusetts Department of Public Health between 1975 and 1986 to examine the following: (1) to see if the major regulatory change imposing the Medicare uniform cost report on all home health agencies resulted in greater hazard rates in the period 1977-1980, (2) to see if organizational mortality rates were higher for public agencies than for free-standing non-profit agencies, and (3) to see if organizational characteristics such as size and scope of services, or locational characteristics such as rurality or poverty level accounted for differences. Results indicate that hazard of failure is inversely proportional to size, and that when size and other variables are controlled, the crude differences in the failure rates of public and non-profit agencies are greatly reduced. Data on proprietary and hospital-based agencies were insufficient for comparative analysis. Hazard ratios indicated that small agencies faced hazard of failure about three times that of larger agencies, controlling for other variables. Failure rates increased in the four year period between passage and actual implementation of the uniform cost report requirement. However, time dependence of the hazard rates appeared insufficient to require inclusion in the modelling process. The period of observation was too short and the population of agencies too small to assess the mortality rate impact of resource constraints in the middle 1980s. Empirical evidence was consistent with the hypotheses that small agencies were at risk during the period when regulatory changes imposed a major administrative requirement and constrained reimbursement options. Rural location, broader scope of services and more specialists had protective effects on agencies. Hypothesized protective effects of variables thought to be associated with community support were found to be of borderline significance, but these possible associations require further study. The proportional hazard regression method provided a useful tool for analyzing an administrative data base with time series data on health care organizations.