Let me tell you about my data:
My data comes from the North Pacific and was collected on a cruise that I was on last August. We collected water samples at all ranges of depths – from very very deep (~ 3km) to very very shallow (~10m). We collected hundreds of little bottles of water which we keep here, in a cold room, so that we can reanalyze that water for nutrients like silica, nitrate, and phosphate. All those bottles have lived in that cold room for about a year and will probably live there for another 40 years or so. Not that that’ll do anybody and good, it’s just kinda the way things go around here. I’ve tripped over mud cores collected in the late seventies in that cold room. Seriously, scientists don’t know when to let go.
I say reanalyze because all these samples were run on the ship last summer. They were run by a dude from a lab hired out by the University of Washington. This guy had an instrument called an auto analyzer, which does exactly what it sounds like it does. It analyzes things automatically, hooray! We like auto analyzers because they’re fast. No messy pippetting, no headaches of tinkering with your spectrophotometer, just load your samples in and the numbers come roaring right out. The problem is that with speed, you loose precision.
Auto Analyzers suck/rule depending on your situation.
Precision is the ability to repeat the same measurement and get the same result. So say you run one sample three times and you get 180, 182, and 179, you’re in good shape. We like precision more than speed because precision means we can make statements about what we see and have the wherewithal to back it up. You don’t want to go out on a limb and make some big dramatic statement when you don’t know how good your data is.
I’m currently having data problems because of this business with the auto analyzer and it’s crappiness in the world of precision. You see, I’m looking for differences in the realm of 4% so we’re talkin’ small changes in nutrient content seen over hundereds of water samples taken from hundreds of depths at dozens of stations. If this auto analyzer can make only make measurements precise to 2-4%, we’re in trubs. The error inherent in the measurement overshadows any real signal I could hope to find. Now I’ve got to go back through all this data with a fine tooth comb and make sure the differences that are going to lead to BROADER GLOBAL IMPLICATIONS are really real and not just a figment of analyzation. Doesn’t that sound like fun, kids*???!!!
*Answer = No.