Suppose, in a regression, you know you have an omitted variable, W_i. In your model, you estimate the coefficient on D_i (i.e. the effect of interest) the to be positive. Under what conditions on and Cov(W_i, D_i|X_i) is this information still useful? Be specific.