i'd think its becouse of the ortho-effect.
i've seen this as a correct argument analyzing the substitutent effect on the acidity of benzoic acids.
when a group is placed ortho to the carboxilic acid, regardless of it being an electron withdrawing or donor group, the main effect observed is a steric hindrance of the carboxilic site due to the proximiti with the substitutent, leading always to a reduced acidity, even in o-nitrobenzoic acid.
so i'd think this situation is analogous, and the expected increased basicity will be observed with m- and p-metylaniline. go see if this is true, and if it isnt, then we'll just have to find another explanation