 Cross-sectional datasets allow us to study differences between cases. When we add more observations over time for each case, then we can also study change over time. Whether we should be interested in the difference or in change depends fundamentally on our research question. Let's take a look at the next example. So we have firm 1 and firm 2 and we have their development over time. Now the question is which one of these firms is doing better? That depends on the context of the question. For example, if the question was which of these firms would have been a better investment in the last five years, then firm 1 would have been a lot better because they are more profitable on average. If on the other hand we are a manager of a company and we want to imitate either firm 1 or firm 2, then we need to consider do we imitate the company that is declining or do we imitate the company that is increasing its performance. So this company is clearly doing something that causes its performance or profitability to go up and typically a manager would want their company to be more profitable. Therefore, imitating firm 2 is probably a better idea. So depending on the context of the question, firm 1 can be better or firm 2 can be better depends on whether we are interested in difference or in change. Difference and change are supported differently by various models. This is the latent growth model. So interest here gives us the initial difference and slope here gives us the direction of change for each company. We might be interested in explaining these two variables. We can add a third variable x here. So are we now interested more in how x explains initial difference or whether how x explains whether some companies are interested, are improving more than others. Typically we are more interested in change because a process that unfolds over time is more related to causality. The same thing can be seen in the mixed model. So we have difference and change in the level 2 firms or initial difference and change in the level 2 effects and which one we are interested in again depends on the context. In first differencing and fixed effects estimation we only get the change. So when we take the unobserved differences out from the data then we can't say anything about differences because we take the differences out. We only get estimates of change. So which one should you be interested in difference in change or change depends on the research question. Difference is modeled by the between effect. Differences are more difficult to interpret causally because causality is a process that unfolds over time and if you don't observe the time limits and you only observe averages over time then it's difficult to say if there's a causal process or not. Then there's also the contextual effect which can be used to estimate whether the differences are simply due to the change process that unfolds over time or whether there's some more fundamental difference between the observed units. Then we have the change. So if we're interested in change over time why some companies develop to be better and some companies decline then we need to take a look at the within effect and it is easier to interpret causally. However the within effect is rather short term so sometimes you need to take a look at these longer term effects so looking at difference when you want to analyze longer term causality might be a good idea. And the difference between difference and change is fundamentally related to research question and you need to first think through which of these quantities is of interest for you before you construct and interpret your model.