Randomisation and masking

SMEs were randomly allocated by an independent statistician using a computer-generated sequence to one of four groups (arms) in a 1:1:1:1 ratio:

There are multiple pairwise and group comparisons that can be estimated from multi-arm trials [16]. Here, we report estimates of the following planned [15] “pairwise” comparisons:

We also report on ‘reactivity’ [17] effects (see ‘control conditions’ below) and temporal change (from baseline to endline).

To allocate SMEs to trial arms, we used covariate-constrained randomisation [18] on the basis of number of employees and SME industry type according to the UK Standard Industrial Classification of Economic Activities 2007 (SIC): 1) manual and secondary sector, 2) service and tertiary sector, 3) social and public sector (see protocol for further details [15]). Within each SME, we randomly selected up to 15 employees from lists of employees provided by the SME. This selection was completely random with no constraint.

Recruitment, enrolment, and delivery of intervention were carried out by the Local Government Implementation Team. Outcome assessors were blinded to allocation [15]. The statistician had no role in delivering the intervention or data collection and was blinded to SME and employee identity.