a very specialized kind of loneliness
Feb. 10th, 2005 12:33 pm![[personal profile]](https://www.dreamwidth.org/img/silk/identity/user.png){kind=small}
adrian_turtleThis company does not need two analytical chemists because I am so overwhelmed with work that only an analytical chemist can do. I am just tired of being the only one here who understands (without extensive explanation) that:
1. Using more freely varying parameters in the regression analysis will produce an equation that comes closer to fitting the data. It will not necessarily come any closer to describing physical reality. Thinking of these 6 data points as following a 4th-order power law lets us draw a beautiful curve with no noise at all, but it makes no more sense than numerology. Thinking of them as "approximately linear" gives a physical explanation that agrees with what we know of optics and physical chemistry, but it demands an admission that our data is not quite perfect. This is terribly difficult for some people.
2. Not all measurements are equally useful. Consider a situation where we are interested in composition, trying to keep it very consistent between batches. Nobody makes any effort to control temperature, or knows of any way temperature could change the composition of this stuff. Buying a highly accurate thermometer and using it to measure every batch would be a waste of time and money. Yes, bad batches have uncontrolled temperature. Good batches have uncontrolled temperature, too! Controlling temperature is not likely to help anything.
3. The do-overs are important. If we only save enough material for one test, there's no way to know if anomalies came from the material or the equipment. (Or badly timed power breaks.)
1. Using more freely varying parameters in the regression analysis will produce an equation that comes closer to fitting the data. It will not necessarily come any closer to describing physical reality. Thinking of these 6 data points as following a 4th-order power law lets us draw a beautiful curve with no noise at all, but it makes no more sense than numerology. Thinking of them as "approximately linear" gives a physical explanation that agrees with what we know of optics and physical chemistry, but it demands an admission that our data is not quite perfect. This is terribly difficult for some people.
2. Not all measurements are equally useful. Consider a situation where we are interested in composition, trying to keep it very consistent between batches. Nobody makes any effort to control temperature, or knows of any way temperature could change the composition of this stuff. Buying a highly accurate thermometer and using it to measure every batch would be a waste of time and money. Yes, bad batches have uncontrolled temperature. Good batches have uncontrolled temperature, too! Controlling temperature is not likely to help anything.
3. The do-overs are important. If we only save enough material for one test, there's no way to know if anomalies came from the material or the equipment. (Or badly timed power breaks.)
