The analysis method is non-linear regression, in most cases using the logit model. The
standard errors are estimated with the Huber–White robust estimator which takes
account of sample stratification and weighting, and also allows for the probable
presence of heteroscedasticity in surveys of this type.
All analyses have been conducted using both employment weights and establishment
weights, and these will be reported side by side. Employment weights, which are
relevant to a ‘coverage’ concept, are more informative when the aim is to assess
employees’ exposure to employer policies and practices: descriptive results using these
weights estimate the population’s proportion of employees exposed, while modelled
results estimate population differences in exposure for the average employee.