[MUSIC] In this lecture, we'll talk about theory construction. In most of the course, we talked about statistical inferences, drawing inferences from data. But of course, what we really want to know is whether your theory is true or not and can predict something. Generating a good theory is difficult, and it depends a little bit on the specific research area in which you're working. But we can give some general recommendations. Now in most of this course we've been talking about statistical hypothesis and drawing statistical inferences. But as a researcher of course what you want to do is draw theoretical inferences. To do this, you of course first have to have a good theory that you can use to create a statistical hypothesis. So we see that statistical inferences and theory generation are basically two sides of the same coin. A scientific theory is a set of testable statements about the relation between observations. And whenever we collect data, we do so with the goal to test some of these statements, to try to make an inference about whether the theory has good predictive power or not. Now we focus a lot on testing these theories but not a lot on theory construction. This is an important topic in itself. Sometimes more difficult to teach. But it's really important to keep in mind that theories are really what's driving the goal of doing research. I'll give some suggestions on how to come up with good ideas to generate theoretical predictions. Some of these are based on this book, Theory Construction and Model-Building, which focuses a lot on how to do this within the social sciences. So if you're interested in this topic, I would highly recommend that you take a look at this book. So which approaches can we use? One used approach in science is to come up with a thought experiment. A very well known thought experiment is done by Einstein when he asks himself, what would you see if you could travel at the speed of light? So by coming up with this thought experiment and thinking about what would happen, you can create theoretical predictions that are testable. Another approach that we use in the social sciences is sometimes to rely on your personal experiences. What happens to you, and can you come up with a theoretical prediction based on something that someone tells you? One example is Stanley Milgram's mother. Over dinner, she complained that nobody in public transport would give up their seats if she entered the public bus. So Stanley Milgram was interested in the social norms in cities and he looked at whether people would actually give up their seats depending on how you asked these questions. He found the 56% of the people actually would give up their seat if you nicely asked. Another approach is to try to empathize with other people. Put yourself in someone else's shoes and think, what drives the behavior of this individual? In psychology, a well known example of this is the question, how is it possible that people who worked in concentration camps did what they did? This inspired a lot of theory development on authority, for example. You can also use observations. Let's say you go out and you buy a car and you look at the car salesman, what do these people do to sell you a car? What kind of ticks do they use and how can this inform you about persuasion and theories about how to make more persuasive arguments? Another approach we often use is to rely on metaphors. Now a metaphor should never be a replacement of an existing theory, but it can be used to inspire theoretical reasoning. In memory research, one of the theoretical models is the model of memory like a storage bin. Whatever you put in last in a storage bin is also most easily retrieved. So one theoretical prediction based on this metaphorical line of reasoning could be, is it also the case in human memory that whatever we store last is most easily retrieved? What's always a good idea as a scientist is to ask why and how like a three-year old. Don't take anything for granted. When you read about theories in the literature, always ask, why would this be the case? What's going on? What's the underlying mechanism? And this can inspire new theoretical questions. Finally, a good idea is to have broad interests as a scientist. We see that many of the most ground breaking ideas come from introducing ideas in other scientific disciplines into your own research line. You don't have to limit yourself to other scientific disciplines. You can also be inspired by movies or books or whatever you see and hear. Now after you have generated some hypothesis and some theory, it's good to try to formalize this theory and this theoretical prediction as well as possible. One idea is to use multiple modalities to do so. So you can use words to describe your theory, figures and numbers. For example, let's say we have the theoretical hypothesis that students are happier than teachers, but especially on hot days. The idea being that on these hot days, students are free to go out and enjoy the weather while teachers are, of course, working inside, preparing your next lecture or grading your assignments. So we can visualize this in the form of a simple graph. We can have a statistical hypothesis saying that this is an ANOVA and we should find an interaction. Or you can even try to specify the numbers that you expect to observe in each of the conditions in this design. When you define your theoretical prediction, it's always good to think about the boundary conditions of the effect that you predict. It's very rare, at least in psychology, to have a universal theory of some sort of effect. There are always boundary conditions. You should not make statement that generalizes over everything that's possible. One thing you can try to do is think, "When would actually the opposite of what I'm predicting be true?" McGuire challenges readers to think about the situation that every proposition is occasionally true, at least in certain contexts viewed from certain perspectives. He calls this perspectivism. His idea is that it always depends. So whatever theory you have, it might be true, but it always depends on certain context. And he even goes as far as to say that the opposite of any theoretical prediction is also true depending on situations. So try to define these boundary conditions and this context as well as possible when you make a theoretical prediction. Fiedler distinguishes between two parts in the scientific cycle. A loosening stage, where we generate theoretical ideas, and a tightening stage where we select and test the best of these ideas. Now when we're in the loosening stage, you have to create as much random variation as possible. Anything goes. Whatever helps you to generate good theoretical ideas goes. So you can play around with data, explore, come up with crazy ideas. And that's all fine. Only later in the tightening stage comes the part of the process where we think okay, which of these ideas that we generated was actually good? Which should we reject and which can we use to keep going? You can think of what a theory would really imply, come up with as many as possible ideas of what the theoretical prediction actually is when you read about something in the literature, for example. You can play around with methods. If you have some sort of measurement device, try to use it in all sorts of different ways. You can also play around with data. You might even simulate data to see what would happen depending on specific assumptions, and then subsequently use the ideas that come from these simulations to test them on people. Or you can explore existing data sets. Now of course we've talked about how exploring data sets can increase the Type I error rate. But here, we're not talking about the testing stage, not the tightening stage of theory testing. We're talking about generating ideas and anything goes. So if you look at data sets and you can come up with a very good theoretical prediction, that's perfectly fine. Of course, after generating this idea, you actually want to move on to the tightening stage and see whether this idea actually holds. But that's a different function. It comes later in the scientific cycle. Now in this lecture, we've talked about generating theoretical ideas. Throughout the course, we mainly focus on drawing valid statistical inferences. But you cannot draw good statistical inferences without having a good theory. So theory construction and generating valid theoretical predictions is just as important as statistical inferences, although it might be more difficult to teach. It's important to keep in mind that a good theory and good statistical inferences are two sides of the same coin. [MUSIC]