Table 3 Key processing steps for implementing AR models in practice when using state-level data

Step 1: Create the needed lag terms for the outcome in the data; we recommend creating between 5 and 10 lags to assist in Step 2.

Step 2: Use common diagnostics for AR models to determine the optimal number of lags for your model; we have found using the partial autocorrelation plot particularly helpful for detecting the optimal number of lags [43].

Step 3: Ensure that the policy variable/indicator is properly coded using change coding (e.g., (A_it − A_{i, t − 1})).

Step 4: Ensure that the right-hand side of the regression model controls for the optimal number of lags, change coding of the policy variable, time fixed effects, and other key controls. Note – state fixed effects are not needed.

Step 5: Check to ensure that you have not overfit your model by confirming that you have at least 10 observations per control variable on the right hand side of the regression model.

Step 6: Do not include any cluster adjustment to the standard errors.