Join 3,555 readers in helping fund MetaFilter (Hide)


By answering this question, you will have proved yourself a genius.
November 9, 2011 9:12 PM   Subscribe

The presence of A is a good sign B exists, but as people learn about the A trick they take pains to display it so it isn't as reliable a sign of B anymore. What's the term for this, is its rate of adoption reliably based on other variables, and who writes about it?

The popularity of strong action verbs in resumes is the best example I can think of at the moment. At one point it might have indicated a good candidate but now everybody knows to do it. What is that?

The behavior of spammers trying to pass a Bayesian filter is another example and I'd imagine a spam filter has an enormous body from which to draw conclusions. If only I knew what they were.

P.S. The suggested duplicate list for this question seems to be saying something about me. I'm impressed and possibly insulted!

Meta-Median from other medians? February 23, 2009
Human biases in judging threats and costs December 29, 2008
I need examples of really bad research. September 21, 2008
How do we establish causation? January 8, 2008
This Dummy needs a guide to learn Bayesian... May 8, 2007
posted by michaelh to Science & Nature (20 answers total) 28 users marked this as a favorite
 
When talking about spam filtering, you're referring to Bayesian poisoning.
posted by sbutler at 9:27 PM on November 9, 2011 [1 favorite]


Not sure if this is true or merely anecdotal, but supposedly customers at restaurants habitually skip the cheapest wine on the list and will often buy the second-cheapest... to avoid appearing cheap, or because they think they're avoiding the "worst" bottle. So restaurants have responded by swapping their actual cheapest with their second cheapest.

I've also heard about a similar phenomenon in the funeral industry, where people supposedly will routinely skip over the two cheapest caskets offered to them. So funeral homes offer as the third-cheapest the coffin which will actually give them the biggest profit margin. This the kind of thing you are looking for?
posted by Conrad Cornelius o'Donald o'Dell at 9:30 PM on November 9, 2011


This the kind of thing you are looking for?

Yes. And, thank you for the term Bayesian Poisoning, sbutler. The article doesn't really speak to the rate at which poisoning attempts happen, just effectiveness. Do you know where rate is discussed?
posted by michaelh at 9:38 PM on November 9, 2011


I've heard it summed up as 'observing a system changes it', or 'we change what we measure'. Search engine optimization is a good example, as are all forms of teaching to the test.

It doesn't have to be intentional 'gaming of the system', either. Health example: high cholesterol predicts heart attack. But some things can reduce cholesterol without lowering your risk of heart attack (one medicine was recently rejected by the FDA for this reason). So now that people are trying to lower their cholesterol, it's not a 'pure' signifier of their underlying risk and the correlation between heart attack and cholesterol isn't as good as it was.

The "law of unintended consequences" is often used to refer to similar phenomena as well. Without an incentive to X, only people who are Y will do it - so X is a great way to find people who are Y. But once you start rewarding X, you're not only rewarding people who are Y - you're also rewarding people who do X without being Y. (e.g., people who know tricks to do well on a given test, even if they don't actually know the material that well).
posted by Lady Li at 9:45 PM on November 9, 2011 [1 favorite]


It sounds like a form of social or intraspecies mimicry.
posted by Rhaomi at 9:45 PM on November 9, 2011 [1 favorite]


More specifically, Bowerian mimicry:
Browerian mimicry, named after Lincoln P. Brower and Jane Van Zandt Brower, is a form of automimicry; where the model belongs to the same species as the mimic. This is the analogue of Batesian mimicry within a single species, and occurs when there is a palatability spectrum within a population. One example is Monarch Butterflies (Danaus plexippus), which feed on milkweed species of varying toxicity. This species stores toxins from its host plant, which are maintained even in the adult (imago) form. As the levels of toxin will vary depending on diet during the larval stage, some individuals will be more toxic than others. The less palatable organisms will therefore be mimics of the more dangerous individuals, with their likeness already perfected. This need not be the case however; in sexually dimorphic species one sex may be more of a threat than the other, which could mimic the protected sex. Evidence for this possibility is provided by the behavior of a monkey from Gabon, which regularly ate male moths of the genus Anaphe, but promptly stopped after it tasted a noxious female.
posted by Rhaomi at 9:47 PM on November 9, 2011


In sociology a good concept to look into is "status contamination."

This is very easily observed in gendered naming conventions, but also in consumption patterns and diction. Interestingly, class differences in consumption and speech may move in either direction.
posted by bilabial at 10:16 PM on November 9, 2011 [1 favorite]


Gaming the system might be the closest term, though not exactly right.

Another example: being nice to waitstaff. Everyone has heard that this is how to show that you're a good person. It's still fine, but it's no indicator of who's a good person. And when someone posted a comment in the Steve Jobs thread about how he was so great because he once held open a door for some people, I had to roll my eyes. Again, it's a fine thing to do, but people do it to appear good, not because they're saints.
posted by John Cohen at 10:17 PM on November 9, 2011 [2 favorites]


http://en.wikipedia.org/wiki/Goodhart's_law is what you're looking for
posted by drethelin at 10:19 PM on November 9, 2011 [6 favorites]


I would refer to this as an example of spoiling the commons. Generically, that refers to any kind of strategy which is productive as long as it doesn't become widespread. As soon as it's overused it switches from being a boon to being a burden.

Advertisers do this all the time. Someone finds a new advertising trick, and it's really effective for them. Soon everyone is doing it, and it no longer works. (Pop behinds, anyone?)
posted by Chocolate Pickle at 10:43 PM on November 9, 2011


Sounds like signalling theory to me.
posted by unknowncommand at 11:20 PM on November 9, 2011 [2 favorites]


In economics, this is known as signalling, for which a Nobel prize was given. Wikipedia mentions a corresponding theory in biology.

I don't know that there is any specific name for the damage done to a signal. As a computer scientist, I lean heavily on electrical engineering terms and so I'd just refer to the noise:signal ratio.
posted by pwnguin at 5:52 AM on November 10, 2011 [1 favorite]


I don't know that there is any specific name for the damage done to a signal.

A pooling equilibrium.
posted by ROU_Xenophobe at 5:57 AM on November 10, 2011


A related phenomenon you might be interested in is the Hawthorne effect.
posted by adamrice at 7:31 AM on November 10, 2011


Lots of what in here, good stuff. Does anybody know about the when -- the speed at which these things happen?
posted by michaelh at 7:50 AM on November 10, 2011


I imagine the term and speed would differ depending on the field. If you think about it in a "biology of sexual attraction" sense, your phrase "The presence of A is a good sign B exists, but as people learn about the A trick they take pains to display it so it isn't as reliable a sign of B anymore", could be about make-up (or any number of animal strategies to falsely give the impression of genetic health).

The presence of full, red lips, big eyes, cheek bones and lustrous hair used to be a good sign that the individual is genetically healthy, but as more people learned about it they took pains to display these characteristics (using make-up and conditioner) so it is not a reliable sign of genetic health anymore.

I guess Richard Dawkins would have written about this in his "the selfish gene" book.

It's the ratio between effectiveness of a strategy and general awareness of its existence.
posted by guy72277 at 8:52 AM on November 10, 2011


As to how fast they happen, it varies enormously. Once pop-behinds were effective for the first advertiser, it was only a few weeks before lots of other advertisers started using them.

On the other hand, things like protective mimicry can take thousands of generations to evolve. (See Batesian Mimicry.)

A good example of that is the coral snake (venomous) and the scarlet kingsnake (which isn't).
posted by Chocolate Pickle at 11:24 AM on November 10, 2011


I was wondering whether the answer to you question was actually "game theory" (about which a lot has been written.)

If you think abut the chess games where IBM's deep blue was pitched against Kasparov, Kasparov managed to beat the machine using a specific type of strategy, but once that strategy became known to deep blue's programmers, it lost its effectiveness and went on to become the grand machine master.

From your initial phrase:
A = Strategy
B = Win

So if A = B at the outset, over time A (to the n) no longer = B

I've never read about game theory, so this could be a bunch of codswallop! I'm sure it is explained far more elegantly in print.
posted by guy72277 at 11:52 AM on November 10, 2011


Since others have touched on Bayes - which is the mathematical basis for these things - I'll skip that.
Lots of what in here, good stuff. Does anybody know about the when -- the speed at which these things happen?
As another poster mentioned, in game theory this is known as signaling - a method by which an informed party sends an observable indicator of his or her hidden characteristics to an uninformed party.

For a signal to be effective, it has to be something that is hard to fake.

Thus your question about "how fast [mimicry] occurs" depends on the system in question.
Evolutionary changes will be slow, for instance an animal's coloration shifting towards that of a different and poisonous creature - mimicking the "I am poisonous!" signal.

Wearing an Armani suit might be a useful signal of wealth or personal ability. Which is great until someone creates a convincing fake. If the fake suits are relatively indistinguishable, and easily acquired, then wearing an Armani suit is no longer a useful signal.
posted by handle_unknown at 7:18 PM on November 10, 2011 [1 favorite]


Thus your question about "how fast [mimicry] occurs" depends on the system in question.

Yes, and I'm wondering if the rate might have found to be some function of variables measurable in the systems. For example, if the speed of communication was twice as fast, mimicry might spread twice as fast or four times as fast. Has anybody tried to measure it in enough situations totals a stab at it?
posted by michaelh at 8:08 PM on November 10, 2011


« Older How do you emotionally deal wi...   |  Advice and resources for deali... Newer »
This thread is closed to new comments.