It is often said that neuroimaging studies (e.g. linking BOLD activity to task type in an fMRI experiment) can only provide correlational evidence, and that only actively manipulating parts of the brain (e.g. their excitability, with TMS) can provide causal evidence for the observed effect.
I am not sure I understand this, in light of the even-older adage “correlation does not imply causation”. As I understand that adage, for the observation “A correlates with B” to also imply “(variation in) A causes (variation in) B”, two things need to happen in addition to simply measuring A and B:
- A always needs to happen before B (temporal precedence, as opposed to simultaneity of measurements)
- levels of A are (actively) manipulated, and the corresponding level of B is (passively) measured, as opposed to both being passively measured
However, (assuming I am not wrong on the statistical point above!), both of these conditions are, seems to me, met in the case of the fMRI event-related designs, where levels of the IV are generated at non-regular intervals, and in each case the DV (BOLD signal) is measured. Even block designs would meet these conditions, if the temporal irregularity of the conditions is not essential.
Why is it then that these fMRI designs - a case of manipulations (various stimulus conditions) organised in an event-related or blocked design, which directly determine the BOLD signal - do NOT constitute causal evidence? Isn’t this equivalent to turning the TMS on and off as a manipulation, and measuring the effect on behavior/the brain in each case?
Is it perhaps that the manipulation needs to refer not to the stimulus but to the brain itself, but if so, why is that a condition (one that favors TMS as causal evidence, but not fMRI)?
I hope the question makes sense. Thank you for any thoughts!