Prompted by an item on Friday’s “More or Less” on BBC Radio 4, I quickly knocked together a causal diagram in an attempt to see possible causal linkages between two apparently correlated parameters, in this case the number of off-licences and number of teenage drinkers within a given area, suggested by a study correlation.
It’s fairly rough and ready, but the process of developing a causal diagrams prompts you to think a bit deeper than the simple “one cause then one effect” rationale. Have a look at this one. I can think of load more stuff to put in, but you soon run out of space.
Even though this is a fairly uninformed analysis it becomes apparent that in this case the two parameters in question are probably being affected by a common factor – I would say here, the general size of demand for alcoholic drinks in the given area. It would seem that the number of underage drinkers is also more directly connected to the ability for them to acquire alcohol either through older 3rd parties or directly from less discerning outlets, of which off-licences would be a sub-set.
For me, it is the process of putting the causal (loop) diagrams together that forces us to ask more guided questions, in turn gaining direction and insight for further study. One can also go looking for “causal loops”, which can result in negative or positive feedback behaviour within the “system” under investigation. The process does however often result in a complex diagram (at first sight), which becomes more difficult to communicate unless you show the sequence of reasoning behind it or simplify the diagram (sometimes this can go too far). You might use the diagram for the basis of a “rich picture”, which can be easier to discuss.
The parameters need not be measurable or quantifiable i.e. they can be a bit woolly, to get a feel for how the system in question fits together and works. You can fit ‘hard’ and ‘soft’ systems together.
The arrows effectively represent some form of temporal relationship between one parameter and another. On closer analysis this may turn out to be a linear function, a decision, another sub-system with another set of causal behaviours, or something yet woolier (gut feel, conjecture etc.). The signs at the end of each arrow show the type of relationship one might expect:
‘+’ : one variable increases with respect to another
‘-‘ : one variable decreases with respect to another
If you find a “loop” you can link multiply the signs together to establish if it’s likely to be a negative (tending to be self-limiting) or positive feedback loop
Once happy with the system you’ve represented, possibly having qualified relationships with expert knowledge, you may be able to start modelling the relationships mathematically. You begin by turning it into a classical block diagram – the arrows become blocks (functions or sub-systems) and the parameters become the inputs and outputs. In a linear system the blocks can often be represented by Laplace Transforms, which can be resolved into differential equations. In a computer model the blocks can be linear, non-linear, deterministic, random etc. and “run” to see what happens. Haven’t done this for a while (in MSc Thesis 1992) so won’t say any more.
Clearly I’m a bit out of practice so will probably blog some more causal diagrams in the future to get my “eye-in”.
Here’s a link to some guidelines that I found to remind me a couple of years ago : http://www.thesystemsthinker.com/tstgdlines.html
That’s it for now. Soup’s ready.