A Bell-Type Inequality for Real Experiments and the Relevance of Fair Sampling Assumption