Measuring the reliability of proxy respondents in behavioural assessments: an open question