My professor and teacher Jan Smedslund recently passed away, almost 97 years old. 
As a tribute to his research and influence, I have created an online app that anyone can use to consult him. Upload a research article on this webpage, and it will tell you if the research findings could have been known before collecting the data. Jan called this pseudo-empirical research, and you will find the explanation below: https://jansmedslundmachine.streamlit.app, and here is the link to a launch report that explains it all.
In 2022, he suddenly gave me a call, asking me to co-author an article with him. He was 92 at the time.
He and his son Geir had discovered that psychological research displays a strange phenomenon: It always seems to explain 42.8% of what it sets out to explore. Every year since 1945, psychological research will, on average, explain 42.8%, not more, not less. Why this number, and why is there no development? Jan and Geir asked me if my work with semantic algorithms can shed light on this.
It turned out we could explain this. It is linked to conventions in psychological research methodology. When psychologists do research on how phenomena are related, they will frequently check how their data behave in something called a “factor analysis”. It is like throwing all the variables in a bucket, shaking it and see which clusters appear when you pour out the contents. If two variables come out of the analysis with factor loadings about 0.70, we say that they probably are the same. Only if a variable loads with less than 0.30 on a factor can we say that it measures something else. We treat the two variables as independent. But check this: If we divide 0.30 by 0.70, we get 0.43. This implies that variables linked in statistics are not independent at all – they simply share more or less meaning,
Jan Smedslund has talked about this since the 1970s. He called this phenomenon “psycho-logics” and said that the relationships in psychology are pre-determined by the way we talk about them. Because of this, the connections between the variables are not “empirical”. They are not something we discover. They were there all the time, in the way we describe the variables. In December 2022, we published the article in Frontiers in Psychology. I am deeply grateful to Jan for him and Geir taking me with Jan on his last publication.
Now, with semantic algorithms, we can model the relationships before asking anyone. We can model and display how questions in questionnaires are related, how they cluster in 3-dimensional space, and why the clusters – the psychological “contsructs” do not cause reach other. They are instead “reasons” for each other. They are mutual information.
Here is the first plot, showing how a long range of single questions about leadership and motivation cluster in high-dimensional space: survey_items_3D
Here is the second plot, showing how these single questions cluster on the level of “constructs”: survey_constructs_3D
And here is a really interesting plot showing that transformational leadership does not “cause” motivation and work outcomes. Instead, you can see that what we say about these phenomena cluster in semantic space. In this plot, you see the single questions as small dots, how they cluster as constructs, and how the constructs look in 3D. You can rotate and look at the model from many angles: Molecular_leadership_3d
Finally, if you wonder what this would look like in at published scientific journal, here is the same plot again in 2D with psychometric statistic numbers on them. It is similar to the one Kai Larsen and I published in 2014: Sem_diagram_voyage_ai