First
of all, I would like to thank those of you who haven't bothered to
cancel your subscription to this blog. Secondly, I'd like to note that
if I'm only going to post every 2.5 years, I completely don't blame you
for canceling your subscription to this blog.
Welcome
back! Feel free to read the last post again, just in case you have
forgotten all the minute details of what I wrote back in 2010. (Sigh.)
Now, shall we continue?
I
will remain true to my last post, where I said, "In fairness, maybe the
study is just fine, and the results are true and accurate. How can you
tell that? I’ll get into more details of that next time." I'd like to
cover just a little bit of experimental theory, because if you're like
me, it's been a long time since that high school class where you had to
keep a composition lab notebook that got turned in once a week. In real
life, I actually do still have a lab notebook. However, instead of the
black and white cover with wide ruled pages, it's got a brown hardback
cover, a control number, and I have to make scanned copies once every
six months to go into a quality assurance archive.
When
was the last time you heard a friend, colleague, or tv reporter say,
"studies show that..." or something similar? Probably very, very
recently. But how often do you hear that person cite the actual study
("a March 2011 study in the Journal of Agriculture and Food Chemistry
led by Dr. Knows Something at the Institute of Human Health...")?
Probably less often. And even less often than that, you go pull up that
academic journal, read the article itself, and follow up on the
references. Right?
So...
how do you know whether the study was even real? Or done correctly? Or
had valid results? Let’s be honest: academic journals are typically
written in ways that make the authors sound smart. I’ve peer-reviewed
and edited some of these articles, and I can tell you from an insider
perspective that the bigger the words and the longer the equations you
use, the less people will question you because they don’t want to look
stupid by not knowing your terminology. These things aren’t written for
your average Reader’s Digest subscriber. They’re written to be
convincing. But even the RD subscriber can pick out a few things that
will help you determine how seriously to take a study. We’ve already
talked about the basis for research itself- now let’s look at
experimental aspects.
When
you're looking to conduct scientific research, how you go about setting
up, conducting, and interpreting your experiment makes all the
difference in the world. While this isn't a comprehensive list, a few
things to keep in mind are sample size, sample population and controls,
scenario realism, and result framing.
First
of all, let’s consider sample size and population. Whether you’re
looking at rats, humans, lima bean plants, or air conditioners, the
number of items in your experiment makes a difference. Conducting an
experiment on a single person isn’t likely to yield results worth
anything. Just because I give a person a heart medication and they die
of pancreatic cancer doesn’t mean that the medicine caused that. Maybe
it did. Or maybe not. Just because a mother who drank coffee in
pregnancy gives birth to a child with allergies doesn’t mean that all
pregnant moms who drink coffee will then have allergy-prone offspring.
On the other hand, if you track 1,000 patients with heart disease being
given a heart medication, and 200 of them die of pancreatic cancer,
there’s a much higher chance that you’ll be able to link the two. In
short, if you’re reading up on a study and they only used two people,
take the results with a grain of salt. If they used 70,000, that’s a
point in the “good research” column.
But
just having an absurdly large population size won’t get you good
results. Sample population and controls are also critical. Remember how
in math you had to keep solving for x? The “x” was the variable. If you
got into more complicated math, you had more variables- x, y, z. In
super-duper math, we ran out of regular letters and had to switch to
Greek letters. At some point, it just becomes too much to keep track of
and control. Well, same thing happens in experiments. The more variables
you have, the less control you have over your experiment. Anything you
can keep entirely constant is your control. Picture this: you want to
find out whether watering lima bean plants with dilute acid kills them
or makes them grow stronger. (Don’t ask me why- maybe you bought stock
in a vinegar production technology and like limas?) You already know
that testing 2 plants probably isn’t a large enough sample size. What if
one was already diseased? So you get 100 lima bean plants. You know
that they are all about the same age and from the same seed producer.
You put them in identical container sizes, and fill the pots with soil
from the same source. You put them all in a location with the same
amount of light exposure, and where the temperature is kept at a
constant 80 degrees. What you’ve just done is controlled a bunch of
variables. Light, nutrients, temperature, container size, soil
permeability, etc. are all factors that *could* have affected
these plants. But by controlling your environment, and controlling your
population to plants with similar genetic and age factors, you can be
more sure that what you’re watering them with is what would be causing
growth differences.
And
how do you water them? Ideally, you’d want your “watering” scenario to
be realistic. You’d want to know how often lima bean plants should be
watered, and how much water they need. It’s not realistic to water a
tiny plant with a gallon of water (or acid) a day. It’s also not
realistic to never water them. In real life, maybe 0.5 cups every other
day is just about right. So you set up your experiment where you water
25 of the plants with pure water, 25 plants with half water, half dilute
acid, 25 plants with three-fourths dilute acid to one-fourth water, and
25 with the dilute acid. For control, you water all of them using the
same batch of water and acid, and you measure out each 0.5 cup “serving”
to each plant. Now you’ve got a reasonable population size, a
controlled environment and population sample, and a realistic scenario.
You should be able to feel fairly confident that differences you see
among the groups of 25 are a result of what you’re watering them with.
By the by, feel free to steal this lima bean thing as your kid’s science fair project this year. You’re welcome.
Last
but not least, how you frame your results makes a big difference in the
“take-away” message, or conclusions, from your experiment. There’s a
book called “How to Lie with Statistics” that was written in the 50’s
(still completely valid and recommended reading!) whose title is
self-explanatory. Let’s say- and I’m completely making this up- that the
25 plants that got pure water grew 5 inches each, the 50 plants with
mixed water grew 10 inches each, and the plants with just acid died
completely. If your experimental conclusion is that “watering plants
with dilute acid kills them”… yes. That’s true in this case. BUT, you’ve
left out the important point that the 50 plants that received slightly
acidic water grew better than the water-alone ones! This might not seem
like a huge deal, but stuff like this gets used as a scare tactic. For
example, we hear all.the.time. about needing to reduce the sodium in our
diets, right? Because a high-sodium diet can kill you! Well, yes, but a
no-sodium diet can kill you too. A lot of life- and this should not
surprise you- is about middle ground. The truth is often somewhere in
between extremes. Make sure that the way results are framed makes sense,
considers context, and isn’t just serving an agenda.
This is getting really long, so, if you’re still reading, bless you. But let me wrap up.
Have
you ever heard of "anecdotal evidence?" Anecdotal evidence is someone's
story, basically. Let's say that I know three people who ate the
cafeteria's pasta salad. All three got sick. Anecdotally, evidence
suggests that pasta salad from the cafeteria makes you sick. However,
does that really prove that the pasta salad makes them sick? (Well… *I*
wouldn’t eat it, but that’s just me…) As you know by now, that’s not a
real experiment. You don’t know that those three people didn’t also
drink contaminated water. Or get simultaneous exposure last week to the
flu. The best solution here would probably be to send a sample of that
pasta salad off to a lab and have it tested for bacteria. But in the
absence of that, maybe just have the chicken instead, and don’t go write
a book about how all pasta salad causes illness.
So
next time you hear “studies show,” question it. See if sources are
actually cited. See if the study meets the criteria of being independent
and unbiased. Then look at how the experiment was run. Think about the
sample size, how variables were controlled (and which weren’t
controlled), and how the results are presented. You’ll be in a much
better position to decide whether or not that study really DOES show
what it claims to.
Class dismissed.
-Schientist