in what way does high within-groups variance obscure between-groups variance?