Timing pressure conflates speed with ability, so let's get rid of it

One mainstay of the most seminal cognitive assessments is the inclusion of timing constraints, either on a per-item or per-section basis. Why so?

The defenders' case for timing is two-fold. First, it further standardizes performance conditions, and, second, it tends to increase g-loading across the broader population. If both of these things are true, shouldn't timing constraints measure ability?

To rigorously examine that question, a bottom-up approach seems best. What is a timing constraint, and, fundamentally, what does it demand of the tester? As noted above, they are a time limit for either individual items, like in the WAIS series, or for sections, such as in the Old SAT and Old GRE. For cases like the former, the goal is to see if the tester's "horsepower" can propel them to a solution in time. In the latter, this partially applies too, but it also prompts the tester to "manage their time" or, in other words, to strategize.

All of these components are g-loaded, but, ultimately, both constraint types are built to reward testing speed. This leads us to the question at the center of this paradigm: what is mental speed's relationship with mental ability?

Mental speed is simply the quickness of mental execution, formalized in the CHC taxonomy as processing speed, Gs, one of the broad abilities sitting beneath g (McGrew, 2009). The mainstream view treats speed as a basic cognitive resource, even a mediator of g's biological substrate: Salthouse (1996) built an entire theory of cognitive aging on it, Jensen spent decades clocking reaction times against IQ (Jensen, 2006), and Penke et al. (2012) found that white matter integrity relates to g largely through its effect on processing speed. I think that view mistakes the plumbing for the water. My read is that speed is a byproduct of mental efficiency, the "filter" rather than the "engine" of the brain, and a filter's quality tells you how cleanly a brain executes, not how far it can reason.

By contrast, mental ability in its purest form, referencing the CHC hierarchy, is reasoning, or Gf, which loads on g at or near unity in Gustafsson's (1984) model. The field remains split on whether the two are identical, but nobody disputes that Gf is the closest thing to g we can measure.

The reason that a statistical relationship exists between g and speed is, I suspect, simple: a brain with a better engine is more likely to have a better filter. This is particularly precarious for individuals with spiky profiles, such as those with ADHD, where delayed frontal maturation (Shaw et al., 2007) means executive functioning is a notoriously poor indicator of true reasoning ability.

In practice, timing constraints place an artificial cap on the expression of reasoning. Take an example from the VPS Beta, where one tester spent 88 minutes to correctly solve an item (without much, if any, break time based on the metadata). Any timed test would have boxed the tester out, preventing this individual from putting their full ability on display. This example alone illustrates how timing binds the measurement of reasoning to speed. It may require more time to complete, but that is a sacrifice worth making for the most rigorous measurement.

To be clear, there are elements of cognitive proficiency that display genuine g, such as applied working memory. This differentiates rote tests (like digit span tasks) from the application of reasoning (like mental arithmetic), where application is a significantly more credible manifestation of g. At the latent level, working memory capacity and reasoning are nearly the same thing (Kyllonen and Christal, 1990), and complex span tasks predict reasoning far better than simple span tasks do (Conway et al., 2005).

The testing establishment already concedes most of this; it just won't say so out loud. Every extended-time accommodation ever granted is a signed admission that the clock measures something other than ability, or there would be nothing to accommodate. The LSAT retired its logic games section in 2024 after litigation over accessibility rather than any question of validity. The most authoritative development, though, is the upcoming SB6's structure, which places working memory and processing speed in a separate General Cognitive Processing composite that no longer feeds the FSIQ derivation (Fun Facts about the Stanford-Binet 6). The flagship instrument of American intelligence testing has decided that speed is not intelligence.

Removing the clock has some downsides, though. Some neurodivergent takers fixate, for example. An untimed item can become a rabbit hole, and I've watched it happen in the data. Unlimited time is also unlimited opportunity to consult a model, which is why the instrument's process-level checks exist. Continuing on, a long, untimed battery fatigues people, as seen in the dropout numbers from my last post. Finally, speed is a legitimate clinical signal in contexts I'm not measuring; a neuropsychologist tracking decline obviously needs it. However, a reasoning instrument does not.

Decidedly, speed is not ability. Time pressure injects speed variance, which correlates with g without actually being g. In other words, correlation is not causation. Perhaps the most critical implication of this stance is this: the positive manifold is real, but the habit of treating every correlate of g as a component of g is an artifact of the method rather than a finding. That is a piece for another day, though.

As always, I welcome your thoughts, rebuttals, and questions!