One of the first lab exercises I taught this semester dealt with simple use of the metric scale, but I found the opportunity to teach a much more powerful lesson in that part of our lab experience. When scientists plan experiments, we hope we take into account all the things that need to be controlled so that only one variable remains- the thing we intend to test. Due to ignorance, willful or innocent, however, sometimes things surprise us for which we do not account.
I asked the students to record the values on the board so that we could analyze the variance of data from person to person. I found another phenomenon that was easy to account for but not necessarily apparent. In the jar of pennies to weigh, there were pennies of different coinage. In 1981, the US Mint stopped coining pennies in pure copper and started wrapping copper around a zinc filler, changing the weight. Another student, unable to read apparently, weighed a 1000ml beaker instead of a 500ml one. This resulted in some widely varied numbers.
By and large, the students obtained weights where the only value that varied was in the last significant figure. That is common in science- it's the figure we're not EXACTLY certain of. For the different pennies (1997 and 1978), there was a 25% difference in weight (2.23 v. 3.41g). If you didn't know about the change in mintage and assumed that a penny was a penny, you might factor in both weights and have skewed data. Or you might throw out the one aberrant design because it was skew, but then you'd have to say "using pennies minted in the 1990s" in your description of the objects weighed. As for the beakers, it was easier to throw out the one value because he knew what he'd done wrong and could easily explain why it was okay to throw it out.
Outliers if not properly identified cannot be removed. That doesn't seem to stop many of my colleagues from deleting, losing or omitting data that counters the conclusion at which they wish to arrive. I have actually been TOLD by PhDs to omit data for various reasons. I must thank Genevieve Pont-Kingdon at ARUP for NEVER having given me the impression that so doing was acceptable.
In my last post, I mentioned briefly variegations across a species. Our lab studied 18 different cultivars of the species Vitis vinifera (wine grape), which isn't all of the cultivars available. There are at least seven members of the Vitis genus, and there are many other members of the fruiting vine family. To assume that Vitis vinifera cabernet sauvignon's behavior explains that of Vitis riparious (Norton- a North American native vine) or that of Watermelon would be silly, yet that's exactly what scientists try to tell us sometimes.
When I made conclusions, I said things like this:
I did not try to say that water deficit stress will affect other wine grapes grown in Africa or Iceland in the same way or that resveratrol was affected the same way systemically. You must restrict your conclusions to the limits you define or else you start running into other variables. Even then, sometimes they show up when you least expect it, even in something as simple as a penny.
I asked the students to record the values on the board so that we could analyze the variance of data from person to person. I found another phenomenon that was easy to account for but not necessarily apparent. In the jar of pennies to weigh, there were pennies of different coinage. In 1981, the US Mint stopped coining pennies in pure copper and started wrapping copper around a zinc filler, changing the weight. Another student, unable to read apparently, weighed a 1000ml beaker instead of a 500ml one. This resulted in some widely varied numbers.
By and large, the students obtained weights where the only value that varied was in the last significant figure. That is common in science- it's the figure we're not EXACTLY certain of. For the different pennies (1997 and 1978), there was a 25% difference in weight (2.23 v. 3.41g). If you didn't know about the change in mintage and assumed that a penny was a penny, you might factor in both weights and have skewed data. Or you might throw out the one aberrant design because it was skew, but then you'd have to say "using pennies minted in the 1990s" in your description of the objects weighed. As for the beakers, it was easier to throw out the one value because he knew what he'd done wrong and could easily explain why it was okay to throw it out.
Outliers if not properly identified cannot be removed. That doesn't seem to stop many of my colleagues from deleting, losing or omitting data that counters the conclusion at which they wish to arrive. I have actually been TOLD by PhDs to omit data for various reasons. I must thank Genevieve Pont-Kingdon at ARUP for NEVER having given me the impression that so doing was acceptable.
In my last post, I mentioned briefly variegations across a species. Our lab studied 18 different cultivars of the species Vitis vinifera (wine grape), which isn't all of the cultivars available. There are at least seven members of the Vitis genus, and there are many other members of the fruiting vine family. To assume that Vitis vinifera cabernet sauvignon's behavior explains that of Vitis riparious (Norton- a North American native vine) or that of Watermelon would be silly, yet that's exactly what scientists try to tell us sometimes.
When I made conclusions, I said things like this:
For Vitis vinifera cultivar gewuertztraminer under water deficit stress in greenhouse conditions, we observed a 10-fold reduction of resveratrol in the leaves and a 2-fold reduction in the berries. A total of three biological samples were tested on three separate occasions to arrive at this figure.
I did not try to say that water deficit stress will affect other wine grapes grown in Africa or Iceland in the same way or that resveratrol was affected the same way systemically. You must restrict your conclusions to the limits you define or else you start running into other variables. Even then, sometimes they show up when you least expect it, even in something as simple as a penny.
No comments:
Post a Comment