Fig. 1

Combined Confirmatory Factor Analysis and cascaded path regression model implemented in SEM. The graphical convention for expressing structural equation models has observed variables as rectangles and underlying latent variables (essentially, factors) as ellipses with loading onto marker variables (such as COMQ-12 question items here) as an outwards arrow. For additional simplicity and clarity here: a loading arrows are made thinner than those for substantive regressions, b observable variables determining construct validity are italicised, but observable question items are not, and c the error terms for the observable variables usually shown as a circle plus arrow into the for each are omitted. The juxtaposed numerals are standardised regression weights (SRWs) a universal metric and one type of effect size, similar in concept to the factor loadings in Additional file 2, but on a different scale. No ordinary item loading has standardised regression weight SRW less than 0.45, except the two for the cross-loading item 5 (discomfort/pain) at 0.41 and 0.26, so all items are doing useful work ‘marking’ the underlying factors (latent variables -- ellipses). One marginal link had remained in the first two CFA models, the inter-factor correlation between hearing symptoms and generic impact. However, in the third CFA, and in the full SEM this is omitted as not necessary (CFA) and not improving fit (SEM). Major contrasts in link strength are seen within the construct validity links. The large SRW for disease activity influence on ear symptoms score factor (0.65; p < <0.001) and the moderate one for SF-36 on activities and uptake (−0.29; p = 0.013) contrast with the small and marginal one for hearing level on hearing factor score (0.234; p = 0.066)