Audiomaxxer
Sign in
Measurement quality

Test-retest reliability

Take the same screening twice in an hour and the two results will not match exactly. That is not a bug — it is what measurement looks like. The useful question is how big the ordinary spread is, so you know when a change means something.

Precision is not accuracy

Precision (reliability)
How close repeated measurements of the same thing land to each other. You can check this yourself by repeating a screening under identical conditions.
Accuracy (validity)
How close a measurement is to the true value from a calibrated clinical setup. This cannot be checked from inside the app — it needs a comparison against clinical results.

Where the spread comes from

  • You: attention, tiredness, how willing you are to respond to a tone you are only half sure about.
  • The room: background noise raises the level at which a quiet tone becomes detectable.
  • The hardware: headphone fit and position shift, and different models deliver different levels for the same setting.
  • The software stack: browser and operating-system volume paths are not calibrated to a physical reference.

Catch trials and the quality score

Some trials in an Audiomaxxer screening are silent. If you report hearing a tone that was never played, that is a false alarm, and enough of them mean the result reflects guessing rather than detection.

Those catch trials feed a screening-quality score alongside response consistency and estimate stability. A low-quality run is worth repeating rather than interpreting — and it is far better to know that than to build a trend out of unreliable points.

Estimate stability vs test conditions

Estimate stability describes how tightly the adaptive algorithm has narrowed in on your threshold given the answers you gave. It is a statement about the maths, not about the room.

Test conditions are reported separately: headphones, volume setting, and how quiet your surroundings were. A very stable estimate collected in a noisy kitchen is still a poor measurement, and keeping the two figures apart stops one from disguising the other.

Reading a change over time

  • Compare runs with matching conditions: same headphones, same volume setting, similarly quiet room.
  • Treat a shift smaller than your own typical run-to-run spread as noise.
  • Prefer trends over three or more sittings to any two-point comparison.
  • Research mode locks a protocol so repeat runs are genuinely comparable — that is what makes the spread meaningful.