Section 25 Maybe the nodes are not variables, maybe they are … propositions, events, schemas?

Mostly in this guide I assume that the component parts of mini-maps are variables: things which could be different, take different values. On balance, variables are the best choice for this job.

But there are other solutions which come close and which are worth a quick mention.

25.1 Events

When we are coding retrospective narratives (“this happened, then that happened”): couldn’t we treat these as events rather than variables? And what would the difference be? (The QuIP approach of eliciting causal information in the context of questions about “changes” invites the respondent to provide this kind of narrative-style information.)

In event-based maps we would have nothing like the same kind of predictability which we claim for variable-based maps. At best we could claim that a causal chain was plausible in retrospect.

The thing about event-based historical scenarios is that the cause does not just create a causal pulse on the next variable but also co-creates that next variable.

[add more here!]

25.2 Propositions

Our nodes could be propositions in the sense of propositional logic. One big difference to a variable-based approach concerns negative statements. If we have a Boolean, false/true variable like “Project is launched, false/true”, this suggests that we are supposed to be equally good at recognising each value of the two values of the variable. Whereas propositions are arguably asymmetric in terms of evidence: the evidence for NOT-C is just the absence of the evidence for C, which can be problematic if the search is unbounded, because we don’t know if really NOT-C or just that the evidence hasn’t arrived yet.

Some event-based narratives, especially non-scientific ones, can be better understood as links between propositions than about links between variables.

25.3 Schemas / Schemata

Most human decision-making is “fast” Kahneman (2011) and associative – the kind of thinking which is better understood in terms of cognitive schemas. Causal maps with variables or propositions for their nodes are better at dealing with “slow”, deliberate reasoning. We could in principle use something like causal maps to model the links between schemata - for example, an arrow could show that schema E is more likely to be activated when schema B is activated. The trouble is, it is really difficult to accurately elicit this network from respondents.

What is great about schema theory is that it provides a model of how schemas can evolve. This kind of training has also been proposed for FCMs Koulouriotis, Diakoulakis, and Emiris (2001).