Thanks to Andrew for pointing me towards this remarkable paper (in his comment re FRC10), I felt it worth putting this on the main page.

The limited information rate between us and the world
In 1954 Paul Fitts reported on a set of three experiments to measure the speed of response in simple human manual operations with various degrees of difficulty. [Fitts, 1954]

There were three different manual tasks:
1/. “Reciprocal tapping”, moving a stylus from a start position horizontally to a target
a) The distance to the target was varied (Amplitude of the movement)
b) The size of the target was varied (Exploring different required accuracy of movement)
c) Two very different stylus weights were tested (exploring dependence on muscle force required)

2/. “Disc Transfer” 16 subjects moving discs from one peg to another
a) With various movement distances
b) With various hole clearances (Exploring different required accuracy of movement)

3/. “Peg Transfer” 20 subjects moving pegs from one hole to another
a) With various movement distances
b) With various hole clearances (Exploring different required accuracy of movement)

He used male college students who were instructed to strive for maximum performance, a challenge to which they probably readily responded!

The results were surprising. Tasks requiring greater precision took longer, but only in proportion to the information implied by that precision. Changing the light one ounce stylus for one weighing a pound had little effect, nor did the overall size of the movements required. There was little variation between the University students used as subjects, and training had little effect either.

His work has had a major impact on the design of interfaces between human and machines. Fitt’s Law has become one of the most successful and well-studied models of human motion. It has provided us with rules for optimising the layout of buttons on computer screens, rules that are still used extensively today, for example to optimise the size and layout of on-screen buttons. His original experiments are still considered to be valid as 16 subjects were used and were tested over 16 levels of difficulty. Each movement duration recorded was the average of more than 600 observations. His experiments have been successfully replicated many times since.

What interests me, but has generally been ignored, is that he approached the problem using the recent information theory of Claude Shannon (which provides an estimate of the fundamental limit to the information capacity of a communication channel from the ratio of its signal to noise) [Shannon Ref]. Fitt reinterpreted Shannon’s equation in terms of the time to complete a manual task with a given degree of difficulty. He considered the required movement amplitude as the Signal, and the dimensional accuracy required to complete the task as the Noise. He configured the experiments to cover various task difficulties over a range of 3 to 10 Bits. He was therefore able to characterise the human subject’s performance in terms of their information capacity. Amazingly the rate of performance expressed as an information rate in Bits per Second fell close to 10 bits per second for all subjects and conditions he explored.

When my attention was first drawn to this work I was staggered by two aspects: the tight distribution of the Bits per second figures for a wide range of conditions, and the closeness to the Bottleneck figures derived from memory competitions. However I initially missed its relevance to my Bottleneck (as my preoccupation has been with the acquisition of novel information, the inward flow). I could not see why a manual task concerned with information flowing outwards should be subject to the same information capacity constraint as in inward learning task. I then realised that these tests required the subjects to use their internal conceptual model of the experiment “out there” in order to complete the task. To know you have hit the target with the stylus and completed the task, we must be able to interpret 2 Dimensional information sensed visually, as a 3D model of the world, through running and updating our internal simulation of the external scene. I believe that the 10 Bits per Second that Fitt measured represents the maximum throughput of our internal simulation.

It has been fascinating to read how many people have struggled with little success to explain Fitt’s observations through low level mechanisms. The facts are counter intuitive: We might expect that using a stylus of 16 times the weight would lower the information capacity as, and similarly with larger displacements. However the experimental reality is that the derived information rate in Bits/second, hardly varies with physical parameters such as length, mass and difficulty. This suggests to me that none of these parameters are involved in the limiting mechanism, and that there is a narrower constraint, the complex mental signal processing required to incorporate what is sensed into our internal model of what is out there. This constraint is determined more by what is in our head than in our arm and hand. (It is worth noting that in fast action computer games, it is the complex three dimensional modelling that limits the ultimate frame rate and screen resolution.)

Note: Fitt observed that the highest bit rates were consistently achieved for displacement amplitudes of 4 to 8 inches. It is interesting to speculate whether this is related to the size of human hands and manual dexterity, or whether this is influenced by the limitations of fast eye movement.

"The Information Capacity of the Human Motor System in Controlling the Amplitude of Movement", by Paul M. Fitts, Ohio State University, journal of Experimental Psychology, 47, 381-391. (There is a link to this in "Links & Blogs" top-right main page)

Richard