Box 2.

Recommendations for best practice.

• Ecological or organisation-level correlation studies often provide only weak research evidence, and only those with adequate reliability, power and validity should substantially influence policy.

• Where patient-level associations are of interest, organisation-level analyses may not be the right approach. Where organisation-level analysis is performed, it is important that this is stated or identified directly.

• Regarding unit-level (Spearman–Brown) reliability, there need to be enough patients included at each organisation, and enough variation between organisations, for a reliable measure.

• The total number of organisations rather than individual patients is the most relevant sample size for the analysis of organisation-level associations. The data illustrated in Figures 1 and 2 can be used as ‘ready reckoners’ by researchers and policy makers when designing or interpreting the findings of such studies.

• The way that the organisation characteristics or performance are measured is also important. The construct validity and unit-level reliability of what is being measured are important to consider and can also influence the power of the study.

• It is important to report any multiple testing and exploratory analyses as well as significant findings, particularly when using publicly available data for which multiple possible correlations could be considered.