»Statistical tricks and statistics lies«

Here I will post some statements that try to make clear how you can be fooled by some combination of data:

Wrong conclusion because of failure to see a COMMON reason

•	Never sleep in a bed, because most people die while in a bed.

•	Do storks bring children? The birth rate is lower in cities (= where only little storks exist)

•	The more calcium you have in your bones, the younger is your aunt

•	The shorter your number on your licence plate, the less accidents you have.

All the statements ARE correct but they imply that one thing leads to another (= storks -> children) while there is another reason that leads to BOTH statments. Example:
"The lower (A), the lower (B)"
That statement may be correct but DOES NOT necessarily mean that (A) leads to be (B) (although for humans it sounds so).
In fact there may be (C) which leads to (A) and (B).
In the "bed example" it's the typical resting behaviour of humans.
In the "stork example" it's the coincidence that big cities have a lower birth rate AND less places for storks.
In the "calcium example" it's the aging.
In the licence plate it's the fact that governmental cars get the best drivers and the shortest numbers.

Wrong conclusion because you behaviour does not change someone else's behaviour

•	Always take a bomb with you when you travel on a plane, because it is highly unlikely that 2 guys (= you and someone else) will have a bomb on the same plane.

Wrong conclusion because you confuse cause and effect

•	When a waiter takes your plate just when you finish eating, you were really lucky that you haven't been eating slower.

•	Forbid Hashish. 90% of heroine junkies started with hashish

•	Forbid brutal PC games. All persons who ran amok liked brutal PC games.

The "hashish example" is typical for political discussions. But replace "hashish" with anything usual, like "eating pizza" or "taking a shower" and you see, that you don't have to ask how many junkies smoked hashish, but how many hashish smokers turned junkie. In the same way you have to ask how many gamers ran amok, and not how many amokers played games.

Wrong conclusion because you fail to see the alternative

•	Don't drink Cocal Cola Light. Sweeteners cause cancer.

•	The company fired 5000 employees although they made 5 billion profit

•	Don't become religious. 0.5% of religious fanatics will commit suicide.

Let's analyze these statements:
Although the Diet Coke statement might be correct, you would have to compare it to drinking "Coke with Sugar". It could be that sugar causes much more cancer than sweeteners.

The company example implies that the company is bad, to be comparable you would have to ask "How many more would they have to fire when they only made 1 billion profit". Such statistical confusion is common in politics: "The other party outsourced X and now we have Y more unemployment".

In the suicide example you need to compare the suicide rate of religious people to that of non-religious people. Otherwise this above statement doesn't give much valuable information.

posted by knn

ebooks-download.com

in-my-opinion.org -> Technology, Computers, Science, Internet -> Language, Math and Names -> Statistical tricks and statistics lies

You should look at the book "A Rulebook For Arguments" by Anthony Weston ♣. Especially the 'fallacies' section. It says some really cool stuff about just the sort of errors you've mentioned above; explaining exactly how they are wrong.

posted by fatpie42

"The beauty of the Superman came to me as a shadow. What are the gods to me now!"

What about

"Who do you think will make the best U.S. president?"

a) George W. Bush
b) The Other Guy

White laugh

posted by Marl64

Marl64:

"Who do you think will make the best U.S. president?"

a) George W. Bush
b) The Other Guy

Yeah, there is a related statistical trick. It goes like this: You make 2 groups of people.

You ask the 1st group:
1) Is the population of the country Myanmar less than 2 million or more than 2 million?
2) What do you think the population of the country Myanmar is?

You ask the 2nd group:
1) Is the population of the country Myanmar less than 100 million or more than 100 million?
2) What do you think the population of the country Myanmar is?

Now, although these are basically the same neutral questions, the results will be different, because people think that these 2 questions ("more than x million" and "what is the population") are related, although they aren't.

Another example (although this time not neutrally worded) how wording can change the outcome is the following:

1st group: Do you think we should put hostile alien spies into prison when we catch them?

2nd group: Do you think we should put foreign agents into prison only because we can?

posted by knn

Fooling others with 'Mathematical Averages'

One of the most abused things in statistics is the 'Arithmetic Average', short: 'Average'

When a village has 1000 farmers, but only 1 of them has pigs (say 1000 pigs), then it's correct to say: "Every farmer of this village has 1 pig on the average" but it is wrong (in a human sense). The correct way to say it is "No farmer has pigs, except the pig famer"

That is the reason why "the average salary in the US is XXX $" is bogus because some very very rich guys are always pulling the arverage to the top.

Another bad thing is, that sometimes there are extremes to one side (the rich guy who earns 500 million per year) but no extremes to the other side (there simply noone who spends 500 million more than he earns per year).

Another bad thing about the 'Aritmetic average' is that it can feature strange numbers, like "Couples have 3.7 kids". THERE ARE NO 3.7 KIDS!

Thus, wherever you can don't use the arithmetic average, instead use "the Median".

Using the median goes like this: To find out
You would make a long line of employers' earnings. You sort from left to right, where the left employer earns the least and the right one earns the most. The average would be the employer in the middle of this line.

That gives you a much better average than the arithmetic average.

posted by knn

The time now is 2 December 2008, 01:06
php B.B.