See the final report

H1: There will be a higher rate of compliance at follow up among businesses which have previously been audited than businesses which have not.
H2: There will be a higher rate of compliance at follow up among businesses which received the alternative audit process than businesses which did not.
H3: There will be a higher rate of compliance at follow up among businesses which received the additional audit activities (reminder and My account sign-up) than businesses which did not.

We conducted a randomised controlled trial (RCT) involving 1,860 small businesses. Businesses were randomly assigned to one of four conditions:
1. Control: no audit and a compliance check.
2. Treatment 1: standard audit, survey and compliance check.
3. Treatment 2: alternative audit (simple language, salience, checklists, planning prompts, social norms) survey and compliance check.
4. Treatment 3: alternative+ audit (alternative audit plus timely reminder and actively engaged to sign up to a 'My Account tool)' survey and compliance check.

The primary outcome of this trial was business compliance with payment of wage entitlements to employees. The primary outcome was operationalised as a binary variable: business is compliant/ not compliant.

We will test all three hypotheses in a single overall model. This model will be a logistic regression with three dummy variables, corresponding to the three hypotheses, coded 1 for businesses receiving the component specified in the hypothesis and 0 for those that did not.
Given that the additive effects of the components across the four experimental conditions may not be statistically significant at each step, but could be statistically significant when combined, we will also conduct pairwise tests to test for difference in compliance rates between each of the treatment conditions and the no-audit control condition. If the pairwise comparisons yield positive trends but not statistically significant differences, we will consider pooling the conditions to test the effect of the behaviourally informed audits against control and any audit against control.
The baseline rate of compliance among the three treatment groups may add additional statistical power to our analysis as a covariate, but cannot be included in the models specified above. Thus, we will also conduct our analyses among the three baseline audited treatment groups. These will be similar to the two models described above, with the difference being that there will be only two dummy variables for the overall model instead of three, and that the standardised audit will comprise the control group for the pairwise comparisons.
It is our intention to break the matched quadruplets (formed for the randomisation procedure) in our analysis. In order to account for correlations induced by the matched random allocation , we may use number of employees, age of the business and industry as covariates in our analyses. The inclusion of these covariates will be subject to tests of model fit and balance checks across the treatment groups.

1,860 small businesses.