 Bucks on structural ecosystem modelling typically differentiate between three types of structural ecosystem models. The path analysis model, the confronted factor analysis model and the structural ecosystem model. Let's take a look at what these models are and also cover some other basic terminology in structural ecosystem model. The three main analysis books on structural ecosystem modelling typically present a typology of three different models. These three different kinds of models are the path analysis model, the confronted factor analysis model and the structural regression model. The idea of a path analysis model is that these are models that don't have any latent variables except the error terms. So these are the kind of models that econometricians would refer to as simultaneous equation models and you would use for example GMM or seemingly unrelated recursions for their estimates. Confrontal factor analysis models on the other hand are models where we have latent variables and observed variables so that the latent variables are modeled as causes of the observed variables and the relationships between the latent variables are left unexplained so they are freely correlated. So we are simply interested in knowing what the observed variables have in common and that's the tool that confronted factor analysis answers. Then the structural regression model is a combination of a path analytical model and a confronted factor analysis model. So the idea is that we take a confronted factor analysis model and instead of saying that all the factors are freely correlated, we don't care about their correlations, we specify that the factors are actually related to each other following a path analytical model. So we can for example run a recursion model or run a mediation model using the latent variables as variables in those models. So because the variable, a latent variable is simply a variable for which we don't have data, there is no other special thing about latent variables. All these three modeling techniques or approaches are basically the same. We just don't have data for some of the variables when we do latent variable modeling and that needs to be taken into account when we estimate model. The variables can be in a couple of different configurations in a structural equation model. So this model is not identified, it's just an example, but we can basically have this typology of latent variables versus opposite variables, endogenous variables and exogenous variables. A variable in an SCM model is endogenous if it has incoming recursion paths. If it does not have incoming recursion paths, then it is exogenous. So endogenous if we have incoming paths, exogenous if we don't. This is of course a bit confusing because the term endogeneity and endogenous variable is used for a slightly different meaning in econometrics. So the fact that you have an endogenous variable in a structural equation model, for example Y here, does not mean that you have an endogeneity problem. You have an endogeneity problem if you have unmodeled correlations between error terms or any of the predictors of the variable to which that error term belongs to. How we typically model these is that the latent variables, those are basically the variables that represent constructs in our theory. And quite often in a factor analytical model, we model these what we call reflective indicators. So A1 is an indicator, it depends on A and some E which represents measurement there or here. There is of course a lot more to this, but this is just a simplified case. Then we have also causal composite or formative indicators. These are often used problematically. And even if you have something that is measured without any error, which we assume when we use these kind of variables shown as Y1 here, is that there is no measurement error. So that's a bit difficult to justify. But there's also another alternative for modeling these kind of no measurement error scenarios. And it is to use a latent variable A for those indicators and just fix the error variance to B0. I will address these techniques more on the videos that talk about measurement. But this is the basic configurations. We have offset variables, we have latent variables, we have endogenous variables, we have exogenous variables. And typically you should avoid having absurd variables as exogenous predictors, unless you really know what you're doing. Then the variables can have different configurations. So we have paths from exogenous to endogenous variables, paths between endogenous variables, and then we have correlations and variances between exogenous variables. We cannot have a path between exogenous variable and another exogenous variable, because when we draw a path from variable to another, then the target variable will become endogenous. So by definition exogenous variables cannot have incoming rigors and paths. So that's a relationship that is not allowed. One thing that beginners of SCM wonder why cannot we have correlations between endogenous variables or correlations between exogenous variable and endogenous variable in the model as model parameters. The reason for this is that all endogenous variables are actually weighted combinations in the linear model, weighted combinations of exogenous variables. So whatever the variance is, is simply a function of the other model estimates and their variance itself is not estimated. So for example if we know that the variance of X is 1 and the variance of Z is 1, we know that X and Z are uncorrelated. We take a sum of X and Z, then we know that the variance of the sum will be 2 and it does not need to be estimated, because it's a function of two of those exogenous sources of variation that we estimate. So this is the basic terminology and the basic kinds of models that we deal with in structural ecosystem modeling.