Estimating recall bias without gold standards: job tenure in Colombia