Proxy respondents are widely used in population health surveys to maximize response rates. When surveys target frail elderly, the measurement error is expected to be smaller than selection or participation biases. However, in the literature on elderly needs for care, proxy use is most often considered with a dummy variable in which endogeneity with subjects' health status is rarely scrutinised in a robust way. Pitfalls of this choice extend beyond methodological issues. Indeed, the mismeasurement of needs for care with daily activities might lead to irrelevant social policies or to private initiatives that try to address those needs. This paper proposes a comprehensive and tractable strategy supported by various robustness checks to cope with the suspected endogeneity of proxy use to the unobserved health status of subjects in reports of needs for care with activities of daily living. Proxy respondents' subjectivity is found to inflate the needs of the elderly who are replaced or assisted in answering the questionnaire and to deflate the probability of unmet or undermet needs.