Responses to the remaining 12 SMFQ items were used to create three continuous scales (parcels) by summing alternate items. This approach was taken to ease the computational burden (compared with using all 48 individual ordinal responses) and to permit the use of maximum likelihood (ML) estimation, with its more realistic missing data assumptions compared to a least-squares estimator. Furthermore, a CFA-based approach using indicators of a latent construct would permit us to consider the impact of non-constant measurement properties of the parcels (see statistical methods for details). We have chosen a dimensional approach to the SMFQ rather than using a cut-point as there is evidence that depression may be better represented as a continuum [18], and that a dimensional approach should be favoured over a categorical approach for hypothesis testing [19].
