 Instrumental variables are commonly used for dealing with endogeneity in empirical research. When you apply instrumental variable analysis, there are certain practices that you should follow. The first thing that you need to do when working with instrumental variables is to understand the endogeneity problem in the first place, because sometimes instrumental variables can actually make the situation worse. So it's a good idea to start by looking at what is the cause of variation of the independent variables in the analysis. So if you suspect that one of your predictor variables is endogenous, then you should have an idea on what is the cause of the variation of that variable. If you have omitted a variable from the model, is there a good reason to believe that there is a specific variable that you should have included ideally but which you didn't? Or is there a specific reason why you think that simultaneity between one of the predictors or more than one of the predictors and the dependent variable is a concern? Then you explain that problem to your reader and then you make a decision on whether to apply instrumental variables or not. There are quite a few different workflows for how to apply this model and relating to different kinds of tests that you can do when you apply this instrumental variable model. The key problem here, the endogeneity problem is that X and U are correlated. So that is the problem that you want to deal with instrumental variables and you include the instrument Z to estimate this X, Y relationship consistently. The problem is that we don't really observe U so we don't know whether X and U are correlated. That is something that we have to infer based on theory. If we include this instrumental variable here, what we are actually doing is not necessarily solving the problem which is trading one untestable assumption for another. So instead of assuming that U and X are uncorrelated, we are assuming that Z and U are uncorrelated. This is called the exclusion criterion of instrumental variables and it is fundamentally untestable. So when you work with instrumental variables, these two assumptions, the relevance criterion and the exclusion criterion must be justified. Because otherwise if these can be justified, then the estimation of this coefficient cannot be assumed to be consistent. So how do you justify the relevance criterion and the exclusion criterion in empirical analysis? There are some workflows available like this one from general operations management and I like this kind of flow charts. They really tell you what are the options that you can use and what kind of tests you do and in which order. And this is my take on instrumental variable workflow. The first thing that you need to do is to decide whether you actually need an instrumental variable or not. And that starts by explaining the endogeneity problem. If the endogeneity problem is considered to be just minor, then it is possible that using the instrumental variable actually makes things worse. Even if the assumptions hold, the reason for that is that for example two states least squares, it's biased in small samples and it is always less efficient than applying normal recursion analysis. So if the endogeneity problem is not severe, then it may be better to just run the recursion analysis and interpret the results with caution than to go with the instrumental variables. So you need to answer the question, does the problem warrant an instrumental variable source? If it does, then you need to go and look for instruments. And when you look for instruments, it is important to justify the exclusion criterion. So why do you think that the instrument that you have is uncorrelated with all the omitted causes of the dependent variable? This is something that is fundamentally untestable empirically, so it has to be argued based on theory. And so the first thing when you start writing about instrumental variables is not how you apply two states least squares to your data. Instead, you need to justify why you think that the instruments don't correlate with the unobserved causes of the dependent variable. And that can be hard to do and if you don't understand the phenomena that you're studying, well, it is nearly impossible to do. So this is very much a theoretical exercise instead of an empirical exercise. The next step in the workflow is performing tests on the instruments. You need to assess the weak instrument's problem if the instruments are not correlated strongly with the instrumented variable. That increases the bias of instrumented variable estimators and makes those estimators less efficient. Also, it is possible to test the exclusion criterion in a way and those tests assume that you have more instruments that you need, in which case the model is over-identified and then you can test some of the instruments assuming that the others are valid in the exclusion criteria. And then the final step, you do an endogeneity test using the instrumental variables and if the test indicates that there is endogeneity, then you should apply instrumental variables in all the main analysis of your paper. If the test shows that there is no endogeneity problem or you cannot conclude that there is an endogeneity problem, then it may be okay by just excluding those instruments from further analysis. So that's a simple workflow on how to do an instrumental variable analysis and I have separate videos of each of those steps.