We thank Dr Sashidharan and colleagues (Sashidharan et al. Reference Sashidharan, Bhui and Duffy2013) for their interest in our paper (Singh et al. Reference Singh, Burns, Tyrer, Islam, Parsons and Crawford2013). They state several well-rehearsed opinions and also raise important questions about facts and science. The latter include: (a) that the study was not blind and practitioners were aware of the purpose of the study; (b) that denominator should have been different since ethnic bias might be operating prior to the decision to assess someone under the Mental Health Act (MHA); (c) that since ethnic bias might operate in risk assessment, risk variable should not be included in the regression model; (d) ethnicity–site interaction is a major weakness in our study; and (e) upstream factors should be better explored in future research. We will restrict our response to these questions.

The AMEND study specifically set out to assess the impact of two changes made to the MHA (2007 amendment to the 1983 Act) – the single definition of mental disorder and inclusion of the Appropriate Treatment Test. Prospective data were collected over 4 years. Practitioners were contacted at the start of the study to explain its purpose and were not told that we were exploring ethnic bias in clinical practice. We therefore do not accept the claim that explaining the study once would change practitioners' behaviour over the next 4 years.

We used the assessed population as a denominator since we wanted to explore why some assessed patients get detained and others (almost a third in our study) do not. We found that individuals with serious mental disorders who were at risk and had poorer social support were most likely to be detained. Our findings could be interpreted simply as this: populations at higher risk of serious mental illness will also have higher rates of detention. This is unsurprising. Consequences of an illness are more frequent in groups with higher rates of that illness, analogous to the well known finding that resistant hypertension is more common in ethnic minority groups who are also at greater risk of its cardiovascular complications (Sarafidis et al. Reference Sarafidis, Georgianos and Bakris2013).

We agree that we cannot rule out ethnic bias in who gets assessed or in the assessment of risk, but we cannot claim that such bias therefore exists. A basic tenet of science is that the burden of proving a hypothesis rests on those who state the hypothesis. To make an assertion and then demand that others find evidence to refute it is not science but ideological positioning. Interestingly, in our study we found no evidence of ethnic differences in clinicians' assessment of the presence of risk. To address the point about modelling risk in the analysis, we have re-analysed the data using the logistic regression methods described in the paper, but without entering the risk variable as a possible predictor variable. The results show that ethnicity, even when forced to remain in the model, still has a non-significant effect on the odds of detention (p = 0.582) and the odds ratios for other predictor variables remain similar.

Ours is the largest, but by no means the only study to have found that controlling for confounders eliminates or significantly diminishes the effect of ethnicity on detention; others have reported the same (Cole et al. Reference Cole, Leavey, King, Johnson-Sabine and Hoar1995; Lawlor et al. Reference Lawlor, Johnson, Cole and Howard2012). Previous studies, including meta-analyses (Bhui et al. Reference Bhui, Stansfeld, Hull, Priebe, Mole and Feder2003; Singh et al. Reference Singh, Greenwood, White and Churchill2007) suffer from precisely the weaknesses that concern Sashidharan et al. – differing denominator populations, lumping of diverse ethnic groups into single categories, failure to account for controlling factors, etc. We believe that our study deals with many of these inadequacies in a large and robust dataset.

We have no evidence to suggest that our site differences show ethnic bias operating in psychiatric practice in London and not in Birmingham and Oxford. We agree that site–ethnicity interaction may well be explained by variables suggested by Sashidharan et al.: poorer access to alternatives, more co-morbid conditions, substance misuse, homelessness, etc. These are all inadequacies in service provision to meet the needs of the local population. There may well be differences in clinical practice across sites and different clinical thresholds for assessment and detention, dependent upon resource availability.

We have no intention to ‘rehabilitate’ psychiatry, as stated by Sashidharan et al. British psychiatry has a strong tradition of attempting to understand ethnic differences in mental health care, challenge psychiatric practice where needed, and alter service provision to meet the needs of ethnic minorities. We agree that we must do more to understand upstream factors and reduce ethnic difference in mental health care. We quite understand that when powerful opinions on an alleged fact are promulgated by any individual or group, it is not easy to alter these through argument alone. But when the level of evidence improves to the point at which the facts tend to support an alternative explanation, at some point the opinions have to change. We do not presume that the results of the AMEND study are sufficient to effect this change, but they speak quietly for themselves and lessen the need for dogma in this particular debate.