Fig. 2

Directed Acyclic Graph of the hypothesized causal relationships between special educational assistance (SEA) psycho-social difficulties (PSD₅,PSD₈), loss to follow up (L), maternal mental health (MH_m), unobserved environmental and genetic causes (U_E,U_G) and additional confounders (C, i.e. contact with mental health services, maternal education, birth order, birth month, preterm birth). Current and prior psycho-social difficulties PSD₅ are confounders causing bias due to treatment by indication and can be controlled through adjustment. Because maternal mental health (MH_m) predicts loss to follow up (L), which is a collider on a backdoor path between SEA and PSD₈, loss to follow up has to be controlled through inverse probability weighting