A Causal Approach to Nonrandom Measurement Errors