Is SEM Causal Modeling?

You will often hear SEM referred to as a causal modeling approach. SEM does not determine causation between two variables. This is a misnomer that is used quite often with SEM. As stated earlier, SEM uses a covariance matrix as its input, so you are essentially looking at cor- relations between variables to determine how one influences the other, but it will not deter- mine causation. Just because two things are highly correlated does not mean one causes the other. For instance, let’s look at an example about Divorces in the U.S. and how it relates to Organic food sales. If you look at the data from 2010 to 2017, the correlation between those two concepts is −.98, which means that when consumers started eating more organic foods, the divorce rate in the U.S. went down at almost the same rate. So, does this mean that the key to a good marriage is eating fewer preservatives in your food? Of course not; just because a correlation is highly significant does not mean that one thing is “causing” the other. In this case, we have what is called a spurious relationship.The constructs appear to be interconnected, but in reality, they have no association with each other.There are numerous examples of spurious relationships, one of the most popular being that the length of women’s skirts can determine the direction of the stock market. While those two things may have a strong correlation his- torically, it does not show evidence that one is causing the other.

Figure 1.3 Example of Spurious Correlation

SEM is a great technique to determine how variables influence one another, but to deter- mine causation you would really need to use an experimental design.While the majority of SEM research has been performed with non-experimental data, SEM is more than capable of analyz- ing experimental data. SEM can examine the differences of a control and a manipulation group while exploring their influence on multiple dependent variables, which can overcome many of the limitations of other statistical techniques.Though SEM requires a large sample, this statistical technique can provide you with flexibility in the conceptualization and analysis of an experimental design. So, while a SEM analysis itself will not determine causation, you can use SEM to analyze an experimental design study to aid in understanding causation between constructs. Later in this book (Chapter 8), I will talk more about how to use SEM to analyze experimental data and pro- vide a detailed example of how to analyze data with numerous experimental groups.

Source: Thakkar, J.J. (2020). “Procedural Steps in Structural Equation Modelling”. In: Structural Equation Modelling. Studies in Systems, Decision and Control, vol 285. Springer, Singapore.

Leave a Reply

Your email address will not be published. Required fields are marked *