Monday, March 29, 2010

Wright's law of empirical illustrations

There comes a time in a person's life when they want to create a law, and for me that time is now. I have a law, actually it's probably more like a general principle, regarding how people use empirical illustrations. It's fairly common in a conversation for someone to make a statement like this: assertion/ generalization + supporting empirical evidence.

Here's my law: On average, the further away the empirical evidence, either in time or distance, the less valid the assertion/ generalization.

A hypothetical example: If someone says that it's dangerous to swim in the ocean, and they talk about a shark attack that happened that day at the nearby beach, well, that's probably stronger evidence than if they cite an attack in Australia last month.

A real example: I was talking with a friend about spiritual & church things, and he was wondering if the Christian Church was a positive force in the world. We talked for awhile, and he suddenly he bust out: "What about the Inquisition?" I'm not sure how I answered him, but in thinking about it later, I realized that the very fact that he had to go to a different continent and go back in time about 500 years answers his question.

Can you think of other examples?


Jay Livingston said...

Not in these enlightened times, of course, but several decades ago, evidence of the perfidy, treachery, and general untrustworthiness of Jews was that they killed Christ -- not exactly a recent event. Come to think of it, this same "fact," even though it was 14-15 centuries old at the time, might have been cited as evidence to justify that Inquisition you mentioned.

Knumb said...


But I am thankful Nancy doesn't read your blog.

I'd have even less of a chance in arguments.

/just sayin'

Brad Wright said...

Great example, Jay.

Drek said...

Oddly, the only thing I can think about in response to your proposed law is that it seems to be making a lot of implicit assumptions about the process generating said empirical evidence. So, for example, if it's a Poisson process then it doesn't really matter how long it's been since the last event, we should see occurrences at the same average rate regardless of the specific interval. Or, put differently, just because it's been over 300 days since the last serious industrial accident doesn't mean we can conclude that previous accidents don't tell us anything about the likelihood of future accidents. Likewise, whether or not geographic dispersion matters depends on the process. In other words, the effects of a severe earthquake in Haiti may be valid empirical evidence for funding good preparedness in San Francisco, despite the large geographic separation, because the nature of the mechanism is unaffected by distance. So, basically, I think your law needs some scope conditions.

But, ignoring all that for a moment: Claiming that Haiti was struck with a massive earthquake because the ancestors of the current population signed a pact with the devil seems to be somewhat in line with what you ask for.

Brad Wright said...

I'm sure there are exceptions, but most of the social processes that I'm familiar with have some sort of time or place patterning. I can't think of many that are random.

With industrial accidents, for example, I personally would feel safer in a factor that hasn't had an accident in 10 years vs. one that's had three in the past year.

That's not to say this principle is determinist, e.g., the factory with three accidents is guaranteed more likely, but it seems a useful heuristic.

Jay Livingston said...

On the 10-year accident thing, maybe a choice that's more in line with what Drek was saying is this:
One factory had one accident in 2003, one in 2006 and one in 2009; the other had one in 2001 and two in 2009. They both have had 3 major accidents in 10 years. Is there any reason to feel safer in one than in the other. (Maybe the example would be better if it involved more accidents and perhaps more years.)

Brad Wright said...

Okay... I don't think that I intended this principle to be that fine-tuned.