TO PROVIDE SOME ASSESSMENT OF THE EXTENT TO WHICH INDIVIDUALS INITIALLY IDENTIFIED AND INVOLVED IN RESEARCH BECOME UNAVAILABLE IN LATER STAGES OF THE PROJECTS, THIS STUDY INCLUDED A REVIEW OF THE LITERATURE ON RESPONDENT LOSS. ACCORDING TO A DOZEN TYPICAL STUDIES, LOSS RATES CAN VARY FROM LESS THAN 1 PERCENT TO ONE-HALF OF THE ORIGINAL SAMPLE. MANY FACTORS CONTRIBUTE TO EITHER A HIGH OR LOW LOSS RATE, INCLUDING LENGTH OF FOLLOWUP, WITH THE HIGHEST LOSS RATES FOUND IN THE LONGER FOLLOWUP STUDIES. HOWEVER, WHERE THE INVESTIGATOR GIVES ATTENTION TO THE PROBLEM OF LOSS, WHERE RESOURCES ARE AVAILABLE, AND WHERE PROPER TECHNIQUES ARE USED, RATES CAN BE CONTROLLED EVEN WHEN THE STUDY COVERS SEVERAL YEARS. TAKEN BY ITSELF, A RANDOM LOSS OF RESPONDENTS MIGHT NOT POSE PROBLEMS. UNFORTUNATELY, THE LOST CASES CANNOT BE ASSUMED TO BE RANDOMLY DISTRIBUTED. EVIDENCE SUGGESTS THAT LOSSES CAN BE ANTICIPATED AMONG PERSONS OF GREATER SOCIAL VULNERABILITY, I.E., THOSE WHO ARE AT SOME FORM OF SOCIAL OR INSTITUTIONAL DISADVANTAGE. A PERSISTENT OBSERVATION IN EDUCATIONAL COHORT STUDIES IS THAT LOSSES ARE MORE LIKELY TO INDICATE SUBJECTS' LOW LEVEL OF EDUCATIONAL AND OCCUPATIONAL ASPIRATION, AND ARE LIKELY TO BE CORRELATED TO DELINQUENCY. DATA FROM THE MARION COUNTY YOUTH STUDY (OREGON) REVEAL THAT LOSSES CAN BE REDUCED BY STIMULATING SUBJECTS' INTEREST IN THE RESEARCH, ASSURING ANONYMITY, AND PAYING SUFFICIENT ATTENTION TO THE IMPORTANCE OF RESPONDENT LOSS AND TO PATTERNS IN LOSS REGARDING CORRELATION TO ATTRIBUTES SUCH AS DELINQUENCY. TABULAR DATA AND REFERENCES ARE PROVIDED. (TWK)