the approach to causal inference that we've talked about so far is often called the reduced form approach to causal inference built on the logic of the randomized controlled experiment and it uses control groups to create the counter factual and thereby confront the fundamental problem of causal inference but it turns out that we can't always use this reduced form approach in order to use it we have to identify a treatment group and a control group we can do that by running an experiment or by finding a natural experiment in the data and if they are confounders using the tools of econometrics to adjust for the impact of those confounders as we derive the average treatment effect that would have occurred had we been able to run the randomized controlled experiment but we have to be able to identify a treatment group and the control group what if the treatment is a unique event in history when we first talked about the fundamental problem of causal inference we motivated it with the assassination of John F Kennedy and the hypothesis that some historians have made that this assassination of Kennedy caused our involvement in the Vietnam War now how do historians go about making that argument they can't use the reduced form approach to causal inference there is no treatment or control group so instead historians go back into the historical record they read everything that Kennedy wrote they listen to all of his speeches they analyzed the way Kennedy made decisions they talked to his advisors or read what they wrote and they form in their minds a model of how Kennedy made decisions they then ask what decisions would Kennedy have made that would be different from those that Lyndon Johnson made had Kennedy not been assassinated and they use that model to try to ascertain whether those decisions that Kennedy would have made would have caused us to be involved in the war in Vietnam what the historians are using is what we call a structural approach to causal inference we're using what we call a structural model to derive the counter factual so the structural approach uses a model we call a structural model to derive the counter Factor the historians will typically use verbal reasoning to create that counter factual scientists and economists will typically use mathematical models to create the counter factual think for example about climate change the important question in climate change aside from just the description of how the climate is changing is how much is human activity contributing to climate change so when climate scientists confront that question they can't use a reduced form approach there is no treatment group and control group we don't have a set of planets where on some planets as human activity and on other planets that look similar there isn't human activity we have a single planet that's impacted by the treatment of human activity so what climate scientists do is they create a model of the climate they use all the information they have available to them they construct all the processes that climate goes through as it's changing how particulates in the air in the atmosphere affect climate in different ways and they try to arrive at a model that accurately predicts what's actually been happening with human activity once we're comfortable that we have a good model that predicts well we can take that mathematical model put it on the computer and dial it back in time go back to 1950 go back to 1800. and rerun the model but this time exclude human activity so by re-running the model without human activity climate scientists create the counter factual what would the climate look like today had there not been human activity contributing to climate change we can then compare the prediction of that model without human activity to what the climate actually looks like today to determine how much of climate change is due to human activity or think about macroeconomics in macroeconomics we often look at unique events in macroeconomic history macro economists deal with the economy as a whole and they ask questions about what different policies might have done to an economy economies like the U.S economy are relatively unique they change over time and there aren't many other economies that are roughly similar to the US economy so suppose that we want to ascertain how much of an impact a stimulus program following say the Great Recession of 2008 had on the economy on outcomes like unemployment or GDP so can we use a reduced form approach well we don't have a treatment group and the control group the last time any event like the Great Recession happened in U.S history was the Great Depression when the U.S economy was very different so with macro economists do is they create a model of the U.S economy they try to use all the information we have all the best thinking within economics to try to structure that model and try to find a model that accurately predicts macroeconomic changes then we can look back at 2008 and rerun the model just like the climate scientists did but take the stimulus program that the government passed after the recession started and see what would have happened to unemployment or to GDP had the stimulus not been put in place so again we're creating a counter factual to see what the causal impact of the stimulus program on unemployment and GDP was at least within the context of this model the covet recession was a similar unique event and again stimulus programs were passed by government and we might look back and ask what was the impact of those stimulus programs on important outcomes like unemployment and GDP again we can use a structural model to do that to create the counterfactual or we can even use those models to make predictions about future stimulus programs what would the impact be if we hit another recession of passing a stimulus program of a certain size so macroeconomists just like climate scientists use models to create the counter factuals when they think about the impact of big macroeconomic policies that are relatively unique events in economic history or you could think of policies that have never been tried before suppose that we think of a large Healthcare reform a healthcare reform that hasn't been tried before we can't create a treatment and a control group to think about what the impact the causal impact of that reform would be on important outcomes like changes in the healthcare Market Health outcomes of individuals the number of uninsured people and so forth so how could we go about it well again economists might construct a model of the healthcare Market that includes all the best information that we have that predicts well what's been happening so far and then run the model with the new healthcare reform to see what the model predicts would be the outcome of the healthcare reform so that's the underlying logic of the structural model of the structural approach we use a model to construct the counter factual just as in the reduced form approach we use control groups to construct the counter factual now there's a place for each of those approaches depending on what it is that we're trying to analyze and oftentimes the two approaches are used together evidence from reduced form approaches to causal inference often informs how we construct structural models that we use for a structural approach to causal inference each approach has drawbacks or advantages or disadvantages and in each approach there is a critical question that we would have to ask to think about whether we believe the causal inference analysis in the reduced form approach the important question to ask is whether there were any confounders that haven't been taken account into account in the analysis in the structural approach the important question to ask is whether the model the structural model is used to create the counter factual is Rich enough to actually create a believable counter factual that we can use for causal inference