Sunday, March 22, 2009

Making better use of statistics

Too often, we've all relied on anecdotal evidence or personal perceptions to guide our arguments and form our thoughts. The following video, which was highlighted on The Online Citizen caught my attention.

The information about the developing world is nice, but I think it is more important for us to realize that we need tools (to view trends) as well as hard data (statistics) to help us reason about the truth (statistics sometimes do lie, however).



I encourage readers to check out the professor's tool website at http://www.gapminder.org/

3 comments:

skeptic said...

I don't think statistics lie. It is the humans who lie because their assumptions are wrong.

You don't use a spanner to hit a nail.That is not the fault of the spanner as it was designed to work for a particular task.

But you see this problem especially in the social sciences where they misuse the tools available. People should know when the statistics tools are applicable.

Chee Wai Lee said...

Hey Skeptic,

Actually statistics do sometimes lie, because like it or not, they are simply mathematical profiles of the real world.

For example, when you do a summation of data over a year, you ignore the dimension of time. And in the case of statistical sampling, one has to pay attention to the confidence intervals.

It is important to understand what the statistics mean and how they can be flawed. In this respect, I think we can both agree that human assumptions and interpretation often take advantage of a lack of understanding about how statistics are gathered.

skeptic said...

Hi Chee Wai,

"For example, when you do a summation of data over a year, you ignore the dimension of time."

Actually, this is assuming that the points have no time correlation. Sometimes this is true. But when it is false, it is the human who applied the wrong assumptions and hence the wrong tools. And there are statistical tools to study time series if one assumes the right conditions correctly.

You have to look at the right tools before applying; for example, I don't apply quantum mechanics to describe planetary orbits because I know what the underlying assumptions are. That doesn't mean quantum mechanics lies. It just says that this thing will predict so and so under certain assumptions(conditions). It is up to the human to check whether they are valid.

BTW: Even confidence intervals rely on certain assumptions like the frequentist interpretation of probability rather than bayesian or the fact that you must have sufficient number of points for it to work.

But human beings lie to themselves by ignoring the regime of suitability which is why I say statistics never lie but humans do.