THE PROBLEM INVOLVED IN ELICITING AND REANALYZING DATA STEMS FROM THE RELUCTANCE OF RESEARCHERS TO DISCLOSE DATA FOR DOCUMENTATION AND DATA CONTROL SYSTEMS, AND VAGUE AND NONEXISTENT POLICY FOR ASSURING ACCESS. OBTAINING ACCESS TO EVALUATIVE DATA ON SOCIAL PROGRAMS CREATES PROBLEMS BEYOND THE SCOPE OF ATTENDING LABORATORY DATA. THE PRINCIPLE OF OPEN ACCESS TO EVALUATIVE DATA IS WARRANTED ON SCIENTIFIC GROUNDS. DATA MUST BE SUBJECTED TO CROSS-EXAMINATION TO ADVANCE THE STATE OF THE ART IN PUBLIC PROGRAM DEVELOPMENT; ON PUBLIC INTEREST GROUNDS DATA COLLECTED UNDER GOVERNMENT SUPPORT MUST REMAIN PUBLICLY AVAILABLE. IT IS ALSO JUSTIFIED ON THE GROUNDS OF PROTECTING THE PUBLIC FROM FRAUD. PROVISIONS ARE RECOMMENDED TO GOVERN THE ACCESS TO EVALUATE DATA ON SOCIAL PROGRAMS. PROPRIETARY RIGHTS TO DATA OUGHT TO BE MADE CLEAR PRIOR TO FUNDING A PROJECT, AUTHORITY FOR RELEASE OF DATA SHOULD BE ESTABLISHED BEFOREHAND. RELEASE OF DATA SHOULD BE FEASIBLE UPON COMPLETION OF THE PRELIMINARY ANALYSIS, AND RELEASE OF DATA AFTER PRELIMINARY ANALYSIS SHOULD INCLUDE ALL STATISTICAL RECORDS. IN SOME INSTANCES, IT IS NECESSARY FOR SECONDARY ANALYSTS TO HAVE ACCESS TO IDENTIFIABLE RECORDS RATHER THAN RECORDS CONTAINING ONLY STATISTICAL INFORMATION. PLANS FOR DOCUMENTATION OF THE DATA SHOULD BE SUBMITTED AS PART OF THE RESEARCH DESIGN PLAN IN AN EVALUATION. THOSE PLANS MUST INCLUDE ESTIMATES OF COSTS OF SUCH DOCUMENTATION AND IDENTIFICATION OF SPECIFIC STANDARDS USED. CONCERNING DATA MAINTENANCE FOR SECONDARY ANALYSTS AND CUSTODIANS, IN SOME CASES IT IS REASONABLE TO EXPECT THE ORIGINAL PRINCIPAL INVESTIGATOR TO ARCHIVE THE EVALUATION AT HIS HOST AGENCY, TO MAINTAIN AND CORRECT ERRORS IN THE FILE, AND TO PERFORM SECONDARY ANALYSIS ON DEMAND BY OUTSIDE ANALYSTS. THE MAIN ARGUMENT FOR THIS OPTION IS THAT THE PROCESS OF SECONDARY ANALYSIS CAN BE SIMPLIFIED GREATLY. ANOTHER OPTION FOR DATA MAINTENANCE IS CENTRALIZED ARCHIVE OR SET OF ARCHIVE. THE ADVANTAGE OF THIS OPTION IS THE EASE WITH WHICH EVALUATIVE DATA COULD BE LOCATED AND ACCESSED. THE OPTIMAL SINGLE FORM OF DATA MAINTENANCE IS CLEAR. THE CURRENT HAPHAZARD RESPONSES ARE NOT CONDUCIVE TO MAINTAINING THE INTEGRITY OF THE SCIENTIFIC PROCESS. REFERENCES ARE INCLUDED. (STB)