A modest proposal about uncertainty

Joan Roughgarden in Beyond Belief made a very astute observation of a problem, and then proposed a lousy solution to it. The problem she was addressing had to do with the public perception of evolution as something quite uncertain scientifically (“theory vs fact”). She observed that the public sees science changing its stance on many things, especially in medicine. One day, you should eat bran. The next, bran is bad for you. The next, bran is good for you again. As a result the public observes that some sciences are uncertain, and can’t distinguish one field from another or one type of claim from another, so they apply doubt to all of science even when it is not warranted by the science. Her solution involves using religious analogies, interpreting phrases in the Bible to explain things like natural selection and mutations, in order to communicate it to a group of people who share and value that vocabulary. Dawkins rightly chews her out for this approach, pointing out how far she is stretching the meaning of the phrases just to fit her philosophy.

The problem she is stating, however, is quite real. How can we expect the public to make decisions about medicine, global warming, evolution, the big bang, etc… when they (somewhat rightly, somewhat wrongly) observe that the scientists themselves are arguing about it? The Intelligent Design folks are currently using this observation to sow doubt with the public in their efforts to “teach the controversy” of evolution to inject creationism into the schools. It is a failure of the scientists, and the media that covers them, to communicate with the public. Can we do better?

I have a proposal, which I’ll sketch out in a toy example. The problem is not the communication of facts, or even of the procedures of science. The problem is with the communication of uncertainties. In day to day life, we easily handle claims with different levels of uncertainties. The sun rises in the east each morning has low uncertainty. The claims of the auto salesman or the politician have higher uncertainty. Quantifying it is, of course, more challenging but the qualitative features of uncertainty are known to nearly everyone. So scientists and journalists really need to take efforts to communicate the uncertainty of every claim, not just the fact of the claim or how the new observations differ from the old observations. How could this be done? I think, at least roughly, one should include a plot of the probability distribution with any claim. One doesn’t need to know advanced math to see the picture. If every claim is accompanied by a plot of the uncertainties, the public will get used to reading them. Let me demonstrate with a toy example.

Say, I am trying to determine the origin year of homo sapiens. I realize there isn’t just one year, and there is a process, but it is not much harder to include those in this simple analysis. I have several homo sapiens fossils where I’ve measured the age, which allows me to calculate my best guess of the age, and the distribution of my uncertainty shown here.

blah2-2011-01-29-11-35.png

I’ve used a normal, Gaussian, distribution here although in fact it probably should be something skewed left and probably a lot flatter to reflect our greater uncertainty with age, and that we have other observations that put confident lower limits on the origin of homo sapiens. Again, the details aren’t important because all attempts at clarifying the distribution only further help with communicating the uncertainty to the public. A few observations are in order here:

  1. there are many possible values for the origin that lie well outside of our data yet have non-zero probability
  2. Our “best guess” is around the middle of this distribution, but it really can’t be interpreted as “homo sapiens originated 250,000 years ago” as it might read in a newspaper

Now, we have a new paper that adds another fossil much older than than the previous ones, around 340000 years ago. Newspapers may claim “origin of homo sapiens 150% older than originally thought”, or “estimates of the origin of humans overthrown by new data”. How might it look with the uncertainties plotted?

blah3-2011-01-29-11-35.png

There are a number of lessons that can be read from this.

  1. the new data updates our “best estimate” by only a little – the old data, combined with the new data, are used for the estimate
  2. our uncertainties have widened – by having a larger range of data, our uncertainties may have increased with new data.

In reality, estimating an origin (first event) will update a bit differently than this example shows. For example, the uncertainties in the right-half of the distribution may not be affected at all by an older observation. If this data were in medicine, however, and we were estimating the effect of some new treatment, then the update would be very similar. A single result of a strong effect may not increase our best estimate for that effect by a huge amount. The uncertainties in many medical treatments, or dietary recommendations, straddle the origin: there is significant probability for no effect. It would be fruitful to see the plot of uncertainties, pushed a little this way and that, updated in perhaps a wiki style by scientists as new data come in. There would be many lessons, all of which would help the public understanding of science.

  1. observations rarely overturn well-supported scientific understanding
  2. not all topics have equal uncertainties – doubting everything the same amount is not rational
  3. certainty is never an option, but sometimes the uncertainty is so low that there is a practical certainty
  4. nature itself, not authority, determines our best guess and some of our uncertainty
  5. if the thing you are measuring has a small effect, then you should expect a series of measurements of the effect to change sign: bran is good, bran is bad, bran is good, etc…. This doesn’t mean that the scientists are waffling, it only means that the effect is small and difficult to detect – and probably meaningless.

I think the public could learn to, at least qualitatively, understand and use plots like these. Perhaps there is a better way to display it that does not do violence to the truth, and I’d be open to that. I think getting in the habit of making plots like this would be good for the scientist as well, forcing them to address and communicate the actual uncertainties in their claims.

Advertisements

About brianblais

I am a professor of Science and Technology at Bryant University in Smithfield, RI, and a research professor in the Institute for Brain and Neural Systems, Brown University. My research is in computational neuroscience and statistics. I teach physics, meteorology, astonomy, theoretical neuroscience, systems dynamics, artificial intelligence and robotics. My book, "Theory of Cortical Plasticity" (World Scientific, 2004), details a theory of learning and memory in the cortex, and presents the consequences and predictions of the theory. I am an avid python enthusiast, and a Bayesian (a la E. T. Jaynes), and love music.
This entry was posted in Uncategorized. Bookmark the permalink.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s