Fig. 1.

Final SEM model. *p < .01, **p < .001. The latent variable ‘Internalising problems’ is made up of total scores for loneliness (β = .64***), depressive symptoms (β = .94***), and anxiety (β = .89***). The path estimates included here are from the full model that allowed the nonsignificant paths to be freely estimated. Their inclusion does not have a substantive impact on the statistical significance, direction, or magnitude of effects compared to the restricted model, where the nonsignificant paths were fixed to zero. The only change was in the magnitude and significance of the path from disease activity to closeness to friends, where the path coefficient changed from − .24 to − .35 and moved from p < .01 to p < .001. Age and gender were not related to any other variables in the earlier multiple regression and so were not included in SEM analysis