This weeks ask a Science blogger question is:
"If you could shake the public and make them understand one scientific idea, what would it be?"
Random sampling. If I want to know how many crimes there were in the country last year, you get a more accurate answer if you take a random sample of people and ask them than if you add up all the crimes recorded by the police. Now the crime rate in your sample might not be exactly the same as the population, so you have introduced an error by sampling, but we can mathematically estimate the size of the error, while using police records introduces an error of unknown size -- we miss all the crime that aren't recorded. Alas, too many people don't understand this. They can't tell the difference between dodgy Internet surveys where the sample isn't random, and surveys that use proper random sampling. My first post on this theme was in October 2002 which was before I started blogging. And in eighty odd posts on the Lancet study, the most common errors made by critics were because they just didn't get random sampling
The round up of answers to last week's ask a Science blogger is here.
In general, "shaking" (numbered ping-pong balls in a box, names in a hat, M&M's in a jar, the public, etc) IS a very good way to randomize the sample.
Non-linearity - in some circumstances, a small chance in one input to a system can have a large impact on the overall system.
Or, if we're counting "the dismal science" as a science - opportunity cost: the cost of doing anything includes not being able to do other things with the resources you use to it.
I'd say one thing the public really needs to understand is the concept of "scientific uncertainty".
As it now stands, most people hear the word "uncertainty" (eg, associated with global warming) and conclude that "those scientists don't know ANYTHING. Why should we take their policy advice?"
The Bush administration and its think (ie, gas) tanks have been especially adept at exploiting this.
Gravity. Not just a good idea. **It's the law**.
Random sampling is an important idea, but what is especially difficult to understand about it at first is that if your sample is truly random, the accuracy of your results depends *only* on the size of the sample, *not* on the size of the population -- a random sample of 1000 is just as good at determining aggregate characteristics of a population of a billion as it is at determining those characteristics for a population of 10,000.
Whenever I hear someone comment on a sample by noting what a small percentage of the population it is, it is immediately clear that they haven't understood this fact.
Rather than explain a statistical method, how about teaching the public about science being probabilistic? This way, when the lab coat wearers talk about likelihoods without using the language of certitude, the public understands the concept of uncertainty.
Best,
D
That mathematical equations are not amenable to reason, pleading, threats, prayer, offers of compromise, etc. etc. etc.
Globalization.
You know, that thing that nature did for us a few billion years ago. The fact that we all live together on the thin surface of a sphere flying through empty space, that no matter which direction you travel you run into other folks and then come back. That we're all in this together.
Random sampling. If I want to know what the global average temperature last year, you get a more accurate answer if you take a random sample of the earth's temperature than if you use a non-random sample as we now do. Now the temperature in your sample might not be exactly the same as the true average temperature, so you have introduced an error by sampling, but we can mathematically estimate the size of the error, while using a non-random sample introduces an error of unknown size -- we miss all the temperature data that aren't recorded. Alas, too many people don't understand this. They can't tell the difference between dodgy temperature surveys where the sample isn't random, and surveys that use proper random sampling.
Err Paul, the temperature is quite similar in places near to the ones where you take the samples, so we can estimate the size of the error by not measuring it everywhere. Thanks for trying.
As usual, you completely missed the point. In case you forgot, we're talking about random sampling. My point isn't that we need to measure the temperature everywhere, but that we need a random sample, which we do not have. What we do have is a convenience sample, which is not adequate.
If you apply statistical tools to non-random data as if it is random the answer will be wrong.
That's nice, but no-one is treating weather stations as if they were a random sample. For instance, if they were a random sample you could just average them to estimate global average temperature, but that's not how it is done.
"My point isn't that we need to measure the temperature everywhere, but that we need a random sample, which we do not have. What we do have is a convenience sample, which is not adequate."
Of course, taking a look at the datapoints one has in relation to the potential universe of datapoints is a first step in determining the possible degree of bias in a nonrandom sample, and what adjustments may be advantageous.
I looked at Fig. 3 of http://www.ipcc.ch/pub/wg1TARtechsum.pdf and am of the opinion that what sample we have probably gives a decent representation of the Bigger Picture. The entire matrix of factors, for instance altitude*latitude*urban, does not need to have a value for every intersection to be able to account for the effect of each dimension in the final model; that specific fact represents a major cost savings in industrial sampling/QC processes and is therefore the subject of periodically repeated articles in niche industrial journals on applied lab measurements, etc.
"Theory" (of the scientific kind) would be another good subject to shake John and Jane Q. Public over.
As it now stands, most people equate "theory" with "Aw, it's just a theory" (You know what I mean, Vern?)
Critical to the explantion of "theory", of course, would be the concept of "falsifiable".
Laurence: "As it now stands, most people equate "theory" with "Aw, it's just a theory" (You know what I mean, Vern?)"
My standard response to "Evolution is only a theory" is "yes, just like gravity and atoms."
I once had an interesting e-mail exchange with a fellow on the Dover, PA school board about the subject of "scientific theories" (about a year before the ruling in the case).
Prior to my "conversation", I had no inkling of how hopelessly dense some of these people can actually be. I use quotes because it was not a "conversation" in the usual sense of word, (unless talking to oneself can be considered a conversation)
I used Einstein's "General Relativity" as an example to illustrate to the fellow that scientific theories are more than just "wild unfounded guesses".
Here's the response I received to my example and I quote, (including the caps -- for emphasis, I presume):
"THE LAW OF GRAVITY IS A LAW, ITS NOT A THEORY. THINGS DO NOT FALL TO THE GROUND BECAUSE OF A THEORY."
I attempted to explain further, but, alas, failed miserably -- and, as they say, "the rest is history".