Causal explanations
I am reading “Navigating the Neural Space in Search of the Neural Code” by Jazayer and Afraz from MIT, Neuron, 2017. It’s a paper on experimental design. Simple principle: if the independent variable is randomized, you can assess causality. If it is not, your analysis is correlational. I have a certificate from a course on causality from Harvard from edX, but for the love of me, I cannot remember anything. I remember there was talk about statistical bias. I remember a simple graphical model where an intermediate variable that affects both A and B (no interaction between A and B), can lead to a statistical association that is not causal. I remember the ‘explaining away’ effect in the flow of causality in a graphical model, where the causal influnce flows through a v-structure, i.e. if A and B have a causal impact on C and you observe C, there will be a statistical association between A and B. This is illustrated by the following example: imagine A is a robber, B is an earthquake and C is the alarm going off. If we know that there was an earthquake and the alarm rang, the prospect of a robbery becomes less likely, because the earthquake ‘explains away’ the alarm. But yeah, that’s basically all I remember. Just know, you need to randomize a variable to determine causality. This is because then other factors that might cause an apparent association and bias are statistically evenly distributed over your samples and this destroys the association.
I find this quote from the paper inspiring: “‘‘Can Molecules Explain Long-Term Potentiation?’’, Lichtman and Sanes argued that long-term potentiation (LTP) may not be straightforwardly explained in terms of its underlying molecular causes.” Some molecules modulate LTP and do not mediate it. Off-target pathways can lead to spurious causal effects. There is a discrepancy in scales-- it is difficult to understand phenomena at the cellular level in terms of interactions at the molecular level. Again, because of the mismatch between lower and higher levels, it would be difficult to draw causal connections between small groups of neurons and the ability to perceive, move, or perform a cognitive task.
Kommentaarid
Postita kommentaar