B

Randomization (allocation of participants to study arms) and matching are standard design strategies that equalise or balance potential confounders between comparison groups, thereby minimising confounding [Pourhoseingholi et al., 2012, PMID 24834204].  
Random selection (random sampling from the source population) only improves representativeness and reduces selection bias; it does not ensure that exposure groups are comparable with respect to confounding variables. Blinding, meanwhile, mainly prevents information/observer bias rather than confounding [Fosgate, 2021, PMID 33135243].