==============================================================================
Mediation sensitivity: ρ* at which each indirect effect (ACME) = 0
==============================================================================
(|ρ*| = unobserved mediator–outcome residual correlation needed to nullify)

Access-effect channels (treatment = internet):
  Information acquisition            a=+1.0616  b(γ)=+0.0236  ACME=+0.0251  ρ*=+0.031
  Employment formalization           a=+0.0685  b(γ)=+0.2844  ACME=+0.0195  ρ*=+0.165

Use-type channel (within users; treatment = online learning):
  Learning → information             a=+0.1633  b(γ)=+0.0387  ACME=+0.0063  ρ*=+0.045

------------------------------------------------------------------------------
Interpretation: ρ* is the partial mediator–outcome correlation; an unobserved
confounder would have to induce a residual correlation of at least this size
(with the SAME sign) to drive the indirect effect to zero. Values near the
conventional 0.1–0.3 'moderate' band indicate the channels are suggestive,
not bullet-proof — consistent with the paper's cautious mediation language.
------------------------------------------------------------------------------
