Causal Inference with Unmeasured Confounding: an Instrumental Variable Approach