Missing data problems are commonly encountered in biomedical studies. There have been longstanding interests on the mechanism governing missingness of data. In particular, the notions of missing completely at random, missing at random, and missing not at random are the commonly used classes of the mechanism for data missingness and are crucially relevant for developing appropriate statistical methods. In this talk, we present a novel approach for choosing an appropriate mechanism for the data missingness from the two possibilities: missing at random or missing not at random. By leveraging the power of instrumental variables, the testing procedure does not require specification of a model for missingness. We demonstrate the validity and feasibility of the new test by simulation studies and case studies.