Hardware · for the thesis protocol · calibrated
Force-matching device
The force-matching task asks someone to reproduce a force applied to their own finger, once by pressing directly and once by moving a slider that sets the force the motor applies for them. People reproduce it accurately through the slider and overshoot when they press themselves. The size of that overshoot is the measure, and it is what the schizophrenia thesis uses to ask whether self-generated sensation is attenuated the way it should be.
A trial runs in three phases on one finger. The motor presents the target force for three seconds, the force is removed for two, and then the participant reproduces it over four while the same transducer records. The measurement is the mean over the window from 2.5 to 3.0 seconds into that last phase, which is McNaughton’s window taken unchanged.
So the rig needs one force channel that can both deliver a commanded force and measure a produced one, and it needs to do that identically in both conditions, because the comparison between them is the entire experiment. It was designed and machined in-house with my brother Anıl Bekdemir, a mechanical engineer, and the acquisition software is mine.
It was built because there is nothing to buy. No manufacturer sells a force-matching rig: the paradigm lives in a few labs, each of which made its own. So the choice was not between building and buying. It was between building and not running the experiment.
The design is not original and is not meant to be. The instrument descends from Shergill and colleagues at UCL, through the Cambridge group who fixed the Direct and Slider terminology along with the target forces, to McNaughton at Macquarie, who published a device and a timing standard openly. This rig follows McNaughton for the timing and the analysis window, and the older UCL arrangement for how the indirect condition delivers its force. The motor is a Maxon RE50 driven by an EPOS4 controller. The build notes will carry the full citations, since knowing which variant a rig implements is the difference between a replication and an unrelated experiment with the same name.
Calibration is done. The force channel was traced against dead weights from 0 to 4.066 N, which covers every target the task uses, so no experimental reading has to be extrapolated. The linear fit gives R² = 0.9998, an RMSE of 0.021 N and a worst single error of 0.038 N.
Rising and falling loads were measured separately instead of averaged, because averaging is what hides direction-dependent error. Hysteresis came to 0.0498 N against a 0.05 N limit, which passes by 0.24 mN and is by far the tightest margin in the set. Seven acceptance criteria were measured and seven passed. The eighth is drift, which cannot be computed from one session at all; this record is the reference the next one will be compared against.
The calibration data and the build notes will be published together with the device, since a device nobody can rebuild is not much better than no device. How the rig got here, including the lineage I had to choose between and the description of it I got wrong, is in the development log.