2. What to do?
Improve the sensor or the conditions.
Discard data that you “know” is wrong.
Take “average” over multiple readings.
Add more information.
Editor's Notes
Some causes of measurement error:device malfunctionfaulty assumptionsnot measuring what you think you’re measuring
improve conditions: while measuring blood pressure, I learned to stand still, otherwise machine would keep trying.discard data: the GPS picture on the previous page is a classic example. Notice that the error is at the very beginning, so using the past as a guide doesn’t always work.averaging: a moving average is a great way to filter out “noise”, but we’re usually making the hidden assumption that our noise has a normal distribution. Hardest problem was pulse oximeter timings, which showed two distinct intervals (mixed with noise!).more info: pulse oximeter very sensitive to movement, so we added an accelerometer. Another instance is the Withings scale, which asks for age and sex to use better model.Taking a step back, these are all variations of “improve your model”. (Using raw data from a sensor has a model that the sensor provides accurate data, so improving the sensor is improving the model.)The world is a messy place, so can we learn to embrace it? At a very small scale, Google or Wikipedia can be wildly inaccurate – but they both apply these same ideas in very different domains.