If I’m understanding your situation correctly, T is not the mechanism being tested, but rather it’s the experimental manipulation that randomly assigns people to the genuinely hypothesized mechanism R.  To give a concrete example, there is research about the link between mood (e.g., happy, sad) and memory.  In your example, mood (R) causes memory performance (B).  Of course, we can’t just tell somebody, “be happy” so the intervention might be watching either a very sad or very happy video clip (T).  Your concern was if we show a happy or sad video clip (T), how do we know if it really created the moods (R) that could lead to memory performance (B).  What researchers in Psychology often do is a, “manipulation check,” to see if the experimental manipulation really worked.  In this example, we might give a mood measure immediately following the video to see if the happy group really is happier than the sad group.  Given random assignment, we treat a significant difference (and hopefully a very large effect size) as evidence our manipulation worked.  Then we test the actual hypothesis with the dependent measure.  Probably, if we have a scientific theory or model, we have ideas about mechanisms connecting variables, but I don’t feel we need to always test those mechanisms to say we’re doing science.  Afterall, any correlational design avoids (explicit) causal claims and I would consider these studies quite important to scientific progress.  Hope I understood.  What brings up these questions, Hendrik? Best wishes, Kevin

Hey Kevin, thanks for your reply. I understand your point, and although this is not exactly what I meant with my initial question, it is a topic I also often dealt with.

I came up with these questions because I sometimes think I am "doing it wrong". A lot of my ideas that I turn into experiment have to goal to "see if something affects behavior/ or attitudes". Given that I understand science to have the goal to explain WHY something happens (not just to show THAT something happens), I was under the impression that I should rather focus on experiments that take something which I "know" influences behavior or attitudes in real life and then design an experiment that tests potential causal mechanisms underlying this relationship.

Maybe I am making it too complicated.

Oh, I see, you mean on a grander scale!  In my view, it's legitimate - and probably better - that some science begins more theoretically and that motivates empirical research, and some science begins as atheoretical empirical research that motivates forming theories.  In my field, developmental psychology, there is an ongoing debate about exactly what, "development," is (e.g., when child learn multiplication facts, would you say that's development?).  Piaget started one major perspective that begins with a theoretical notion of how development is fundamentally distinct from learning, and this motivated the creation of numerous experimental tasks and brought meaning to many disparate patterns in children's behavior.  On the other hand, another major perspective is Information Processing and these scientists have historically began by cataloging in minute detail however children behave and how it changes over time.  They're building a theory from the "bottom up" and essentially concluding whatever changes over age, must be development.  There's a strong case to be made for either, and they have both contributed to progress in the field.  The same could be said for other sciences.  For example, Einstein's Theory of General Relativity motivated the empirical work during an eclipse to test if objects with considerable gravity (e.g., stars) could actually bend light - something probably nobody would have guessed without Einstein's theoretical work.  On the other hand, Kepler empirically determined his laws of planetary motion (e.g., elliptical orbits) but he had no idea why that's how planets move.  It wasn't until about 80 years later that Newton explained it with gravity.  So I'd say it's okay to be motivated empirically, or theoretical, or both.  Best wishes with your quest, Hendrik! ~ Kevin

Dear Kevin,

even though its been some time, thanks for your answer. I just read something which relates to what I was asking here. It is from a recent paper by Maniadis et al. in The Economic Journal 127 (605): "let us start from the hypothetical case where for a given scientific field, ‘surprise’ discoveries (rather than theory-driven ones) are necessary in order to achieve publication in a major academic journal. For simplicity, assume that the research design has maximum power (b = 0) and the usual a = 0.05. Further, assume that ‘surprise’ associations are those findings that are ‘1 in 100 results.’ That is, of 100 potentially surprising associations studied, one is known to be true (p = 0.01). In this case, the probability of both a true association and a rejection of the null hypothesis is 0.01 9 1 = 0.01. And, the probability that there is both no association and rejection of the null hypothesis is 0.99 9 0.05 = 0.0495. Thus, the total probability of rejection is 0.01 + 0.0495 = 0.0595. Thus, when a researcher rejects the null hypothesis in this case, there is an 84% (0.04995/0.0595) chance that there is no association (a false positive). Accordingly, there is only a 16% chance that the statistically significant finding represents a true association." (p. F212)

What relates to my question is the relation between "surprise discoveries" and "theory driven ones". The rest of the quote shows what a downside of this "hunt for surprise discoveries" are relative to theory driven research.

Dear, you find useful information in links below:

journals.openedition.org/educationdidactique/497

https://educationdidactique.revues.org/497

https://journal.alt.ac.uk/index.php/rlt/article/download/1143/1393