 The ordinal and nominal models are related in an interesting way, and understanding the relationship can help you in choosing which modeling approach to use. This article by Fullerton in Sociological Methods and Research presents the following categories. So they have 12 different categories organized along two dimensions, approach to comparisons and parallel odds, or proportional odds or parallel lines assumption here. And we have the most constrained model is the traditional ordered logistic regression analysis that I cover in another video. And then we have the traditional multinomial logistic regression analysis that I cover in another video. And then we have these 10 other modeling approaches that can be understood as lying somewhere between the multinomial model and the normal ordered logistic regression model. Instead of memorizing all these models and what they do, let's take a look at the dimensions. And the first dimension is the approach to comparison. And the idea of approach to comparison is that we are predicting a slightly different thing in these different models. So the cumulative we are comparing, we are predicting the log of one outcome, log of odds. So we have also one outcome less than one or one or more, for example. So we are comparing two outcomes if your score is at one or more than one, for example, at two or more than two. So that's the normal ordered approach. Then you have the stage approach. And the stage approach asks or answers slightly different questions. It answers the question that given you have reached a particular value, what is the probability of going beyond? So for example, given that we know that the response is at least two, what are the odds or what's the probability of the response being more than two? And these kind of models are useful when you want to test theories that predict that, for example, people or companies follow through different stages. For example, if you want to explain technology at absent, first the person hears about technology, then they make a decision of whether they want to try the technology, when they try the technology, whether they start using the technology, when they use the technology, whether they abandon the technology. And this is kind of like a model that tries to explain how you continue following through stages. So that's the stage model. And then the adjacent comparison is comparing one category against another one exactly. There's no like one category and more than one. It's an exact comparison between two categories. And we had the other here, the other approach, the proportional odds assumption. The first assumption is that the effects of the explanatory variables are the same for all values of y. So this is the normal parallel lines for proportional odds assumption, so that you only estimate one regression model that explains all values of the dependent variable or all categories of the dependent variable at the same time. Then we have the least constraining. So effect of the explanatory variables are freely, very freely between values of y. So the choice between living in Finland to living in Sweden is explained by different variables in a different way than the choice between living in Finland versus living in Norway, for example. And then we have two different categories that allow some effects to vary and some effects are constant. For example, we could have that the running in a running race, you could say that your stamina is a long run race that predicts how well you do. But then also being able to run really fast the last 100 meters, that predicts whether you're going to be the first or second or something like that. But it doesn't really predict much whether you're the tenth or eleventh. These two come in different two variants. The common factor means that we constrain the effects of some variables to be proportional to each other, but we allow the strength of those variables to differ between the different categories. So how do you choose your modeling approach then? You need to consider first what is the nature of the phenomena that you're explaining. Is it a rank? Is it something that you continue along like a stage model? Or is it just comparisons between two adjacent positions? How do you present the variable? How do you explain the transitions? Is the effects going to be the same for all transitions? Or are the effects going to be different within each transition? So that's the question of how your theory explains the phenomena. And this is a very useful framework because it allows you to take like a bigger picture of how do you want to model your categorical and possibly older variable.