This means that the approximation result f(x) should be between
a) ρOPT <= f(x) <= OPT, when ρ < 1
b) OPT <= f(x) <= ρOPT, when ρ > 1
So, a) occurs when it's a maximization problem, and b) occurs when it's a minimization problem.
For example of a) when ρ=0.5, and you want to find a 0.5-approximation algorithm for a NP maximization problem of a graph G=(V,E)
when, OPT = |E|
you need to find a approximation algorithm f(x) which finds at least
0.5|E| <= f(x) <= |E|
(for 0.5-approximation in a graph, randomized algorithm almost fits to the approximation with factor 0.5)
so, if ρ becomes closer to 1, it becomes a tighter approximation algorithm.