Frequently Asked Questions - Mental Chronometry
Also check out JitterFaq for more info on variable-ISI experiments...
1. What is mental chronometry?
As important (or moreso) than finding out where your activation happened is finding out when it happened. Where did the information flow during the processing of your stimuli? Which structures were active before other structures? Which structures fed output into other structures, and which structures processed end results? Such questions suggest a fairly crude schematic of brain processing, but still a useful one: if you could answer all of those accurately about a given task, tracking the millisecond-by-millisecond flow of information through the brain, you'd have a much fuller picture of the information processing than from a static activation picture. So mental chronometry experiments attempt to attack those questions. First done with reaction time data, using experiments that would add or remove certain stages of a task and find out whether reaction time was sped or stayed the same or what, chronometric experiments have moved into fMRI, with moderate success. Formisano & Goebel (below) review recent developments in fMRI chronometry, with a thorough overview of potential pitfalls. They examine several studies in which flow of information is pinned down with a couple hundred milliseconds - not quite the temporal resolution of EEG, but much better than previously thought could be achieved with fMRI.
2. But how do I get better temporal resolution than my TR?
Simple: don't always sample the sample points of your response. If you always sample the BOLD response 2 seconds and 4 seconds and 6 seconds after your stimuli are presented, for your whole experiment, you'll have a very impoverished picture of the shape of your HRF. But if, for example, you sampled 2 sec. and 4 sec. and 6 sec. post-stimulus for half the experiment, then cut one second between trials and sampled 1 sec. and 3 sec. and 5 sec. for the rest of your experiment - why, then, you'd have a better picture. The cost, of course, is reduced power and expanded confidence intervals at the points you've sampled.
With a good picture of the shape of your HRF, though, you could then compare HRFs from two different regions and see which one had started first, or which one had reached its peak first. If HRF timing is connected in some reliable way to neuronal activation, you then don't need to sample the whole experiment at a super-fast rate - you could infer from only a limited-sample picture of one part of the HRF where neuronal activity had started first and where it had started second, which allows you to rule out certain flows of information.
3. How does variable ISI relate to mental chronometry?
In order to do a mental chronometry experiment, you need to absolutely maximize your statistical efficiency - your ability to pin down the shape of your HRFs. You're going to be comparing HRFs from several regions, and you need to look for the differences between them, which means you can't assume much (if anything) about what shape they're going to be, or else you're going to bias your results. Assuming nothing about your HRF and still getting a good idea of its shape, as Liu describes, means you need an experiment with very high efficiency - and, as Dale demonstrates, those are precisely those designs with variable ISIs (both on MentalChronometryPapers). Only by randomizing (or pseudo-randomizing) your ISIs can you pack enough trials into an experiment for sufficient power and still have enough statistical flexibility to get a good look at the shape of your HRFs.
4. What are some pitfalls in mental chronometry with fMRI, then?
The big one is a crucial assumption mentioned above: that HRF timing is connected in some reliable way to neuronal activation. We assume you're not interested so much in the HRF for its own sake, but rather as an indicator of neuronal activity. So let's say you get two HRFs, one from visual cortex and one from motor cortex, and the motor HRF starts half a second later than the visual. Is that because neuronal activity started half a second later in motor cortex? Or is it because the coupling between neuronal activity and BOLD response is just slower in motor cortex in general? Or is it because the coupling in that particular subject just happens to be looser for motor than for visual - and maybe it'll be different for your other subjects! Clearly, no matter how good a look you get at your HRF, questions like these will dog your chronometric experiment unless you're careful about validating your assumptions. Several excellent studies have examined the issue of variability of HRF between regions, subjects, and times, and those studies are crucial to check out before drawing conclusions from this sort of data (check out HrfPapers for some).
Of course, other more mundane issues may well torpedo chronometric conclusions: if you don't have high enough efficiency in your experiment, you won't be able to distinguish one HRF's shape from another with high accuracy, and you'll have a hard time telling which one started first or second anyways.
5. If I want to do a mental chronometry experiment, how should I design it?
Chronometric experiments depend crucially on determining HRF shape. So start with maximizing that - you need a design that will absolutely maximize your efficiency, no matter what the power cost. M-sequence or permuted block designs are good ways to start (see Liu, MentalChronometryPapers). It should be obvious that with designs like that, you need experimental tasks that will generate fairly reliable activations; your experiment will suffer in terms of power from its focus on finding HRF shape and not using shape assumptions. Choosing a task with a reasonably long latency is also important - even with the best possible design, fMRI noise is such that resolution below a couple hundred milliseconds is simply not possible for now. So if your task only lasts half a seconds, you may not be able to get much information about the chronometric aspects with fMRI. As well, in an experiment like this, having more samples is always better - so you want to have the shortest possible TR. If you can focus your experiment to a smaller segment of the brain than the whole thing, you can get a good number of slices and still have very fast TRs.
One thing to try and avoid in doing chronometry is to toss it in as a fishing-expedition analysis: if your experiment isn't designed with doing chronometric analysis in mind, you'll almost certainly have trouble finding reliable latency differences in your subjects. Unless you've got an eye on this from the start, it's probably not worth doing. But if you do, you can get some pretty sweet looks at the temporal flow of activation and information around your subjects' heads.