4. Which is a reason we might prefer weighted least squares linear regression (WLS) over propensity score matching (PSM) for estimating an ATE? a. WLS allows extrapolation outside the support of the data using its propensity model. b. WLS uses all of the data, while PSM uses only the matched data. c. PSM is easy to make doubly robust, while WLS is not. d. We don't have to estimate propensity scores for WLS estimates. 5. Which is an advantage of using an ML algorithm with an inverse propensity weighted (IPWML) loss function over mean-squared error loss? a. IPWML lets us make unbiased predictions, even in the context of regularization b. IPWML is a more flexible way to estimate E[Y|do(X=x), Z=z] c. IPWML lets us use more flexible model specifications than MSE loss minimization. d. IPWML lets us avoid estimating P(X|Z). 6. In which context might bias amplification be a concern? a. Adjusting for a cause of the outcome which we think is otherwise unrelated to the causal state. b. Adjusting for a cause of the causal state when we're sure we've adjusted for all other confounding variables c. Adjusting for a cause of both the causal state and the outcome