Popular Ignorance of Statistics

Mr. Boo · 08-07-2014 02:11PM

shruikan2553 wrote: »

You can use statistics to prove anything you want. There's a graph of the decline of murder rates and decline of internet explorer usage. Correlation does not mean causation.

There's something that irks me about this aphorism. We know that Canadian salmon stocks are not being affected by Somalian piracy, and would be quick to call bs on someone trying to prove otherwise with some dodgy stats. But often when we have a dataset containing correlated variables, there IS something worth inspecting. Rolling off the old "correlation is not causation" bit just sounds more and more like interweb noise.

dxhound2005 wrote: »

I don't agree. Some things are capable of being measured exactly e.g how many people over a certain age died in a particular year. We have a Central Statistics Office and I would imagine they think they are publishing reliable information.

http://www.cso.ie/en/index.html

If we have census data, as is the case with records of death in Ireland, we do not need statistics to describe death rates within the population. Thus, this is sort of irrelevant.

Grayson · 08-07-2014 02:39PM

Mr. Boo wrote: »

If we have census data, as is the case with records of death in Ireland, we do not need statistics to describe death rates within the population. Thus, this is sort of irrelevant.

Statistics and census data are the same thing. Statistics describe historical data. for example if the census says that there were 100 children born in a particular village and 53 were female, that would imply that in the viillage at that point the birthrate was 53% female and 47% male. That's a statistic and historical census data. at the same time

Maybe you're getting confused with probability where someone works out that 53% of the children in the village next year will be female?

folan · 08-07-2014 02:47PM

statistics should not be trusted and some are racist

Mr. Boo · 08-07-2014 02:54PM

Grayson wrote: »

Statistics and census data are the same thing. Statistics describe historical data. for example if the census says that there were 100 children born in a particular village and 53 were female, that would imply that in the viillage at that point the birthrate was 53% female and 47% male. That's a statistic and historical census data. at the same time

Maybe you're getting confused with probability where someone works out that 53% of the children in the village next year will be female?

Maybe if I clarify, OP was talking about the use of probabilistic measures to make inferences about populations. Such inferences are based on some type of survey where we take a representative sample of the underlying population in order to try and describe the characteristic we are measuring at the level of the population (e.g. the average height thing; I don't think anyone has measured every adult male in Ireland, but there's enough data available to make inferences about the population: correct me if I'm wrong).

However, we have a census of deaths; an absolute measure. Because we have a record of every death (give or take a couple I guess...) we don't need to make inferences at the level of the population because we have the actual figure.

So the distinction really is between descriptive statistics taken from an entire population census in order to describe that measured distribution, and statistics derived from representative samples being used to make inferences at the level of the population.

oak5548 · 08-07-2014 02:56PM

Wow op, I wish I was as smart as you :rolleyes:

Pherekydes · 08-07-2014 03:02PM

Mr. Boo wrote: »

Rolling off the old "correlation is not causation" bit just sounds more and more like interweb noise.

Are you claiming it's not true? Then are you claiming that "correlation does imply causation"?

Or just trying to sound Überclever?

Pherekydes · 08-07-2014 03:03PM

oak5548 wrote: »

Wow op, I wish I was as smart as you :rolleyes:

Don't understand the sarcasm in your post.

Karl Stein · 08-07-2014 03:05PM

mickydoomsux wrote: »

Irish people hate stats and figures.

You just attributed an emotion about stats and figures to 100% of people who are Irish.

Facepalm.

Mr. Boo · 08-07-2014 03:06PM

Pherekydes wrote: »

Are you claiming it's not true? Then are you claiming that "correlation does imply causation"?

Or just trying to sound Überclever?

I wouldn't have said it sounded particularly self-satisfied; I'll leave that up to yourself. I'm saying correlation may provide evidence for a causal relationship/effect, warranting further investigation - and - correlations between bee wing beats and seats won by democrats in the US senate are obviously 5hi7e.

Arthur Beesley · 08-07-2014 03:10PM

folan wrote: »

statistics should not be trusted and some are racist

I was going to ask how could a statistic be racist, but I guess I shouldn't.

padraig.od · 08-07-2014 03:11PM

all the tall people left the country due to the recession

Pherekydes · 08-07-2014 03:19PM

Mr. Boo wrote: »

- and - correlations between bee wing beats and seats won by democrats in the US senate are obviously 5hi7e.

Hence the "correlation does not imply causation" qualifier.

What it really means, and nothing else, is "careful now!", and as such is never wrong.

08-07-2014 03:23PM

There's a lot of cod-statisticians strutting their stuff here today

Pherekydes · 08-07-2014 03:25PM

Deleted User wrote: »

There's a lot of cod-statisticians strutting their stuff here today

Cod is an important, but declining, species. Don't disparage cod statisticians!

08-07-2014 03:27PM

Pherekydes wrote: »

Cod is an important, but declining, species.

Stats or GTFO

Mr. Boo · 08-07-2014 03:27PM

Pherekydes wrote: »

Hence the "correlation does not imply causation" qualifier.

What it really means, and nothing else, is "careful now!", and as such is never wrong.

Just as well nobody said it was wrong.

Pherekydes · 08-07-2014 03:29PM

Mr. Boo wrote: »

Just as well nobody said it was wrong.

Maybe not wrong, but you did say it just sounds more and more like interweb noise.

Mr. Boo · 08-07-2014 03:30PM

Deleted User wrote: »

Stats or GTFO

http://www.sciencedirect.com/science/article/pii/S0960982213007070

Mr. Boo · 08-07-2014 03:31PM

Pherekydes wrote: »

Maybe not wrong, but you did say it just sounds more and more like interweb noise.

The use of the phrase, yes. If I drop into a conversation about relativity and just post "E=mc^2" with no real context I'd sound equally goofy.

Amirani · 08-07-2014 04:01PM

shruikan2553 wrote: »

You can use statistics to prove anything you want. There's a graph of the decline of murder rates and decline of internet explorer usage. Correlation does not mean causation.

Showing a correlation can hardly be considered "proving something".

Correlation doesn't automatically imply causation, but it can mean that the correlated variables do indeed have a causal effect. Blind use of small statistical sets aren't generally seen as "proofs".

Arthur Beesley · 08-07-2014 05:48PM

Pherekydes wrote: »

Cod is an important, but declining, species. Don't disparage cod statisticians!

I made he switch to scrod a while back.

bnt · 08-07-2014 05:54PM

dxhound2005 wrote: »

I don't agree. Some things are capable of being measured exactly e.g how many people over a certain age died in a particular year. We have a Central Statistics Office and I would imagine they think they are publishing reliable information.

http://www.cso.ie/en/index.html

A couple of things:
- please don't quote a whole post just to reply with a line or two. A good guideline is to quote only the specific bit to which you're responding, and write at least as much.
- in this case you should have quoted nothing, since you're responding to something I didn't say. Yes, some things can be and are measured accurately, and I never claimed otherwise.

Though it's all commonly lumped under the heading of "statistics", your example (how many people over a certain age died in a particular year) is better described as data or information. Statistical methods can then be used to analyse the data to give you e.g. average life expectancy. I used the example of height: the statistical bit only kicks in after the heights are measured, and the more accurately they're measured, the more useful the statistics that can be derived from them.

The CSO publishes both information and statistics derived from the informations. From their "About Us" page:

The mandate of the CSO, as set out in that Act, is "The collection, compilation, extraction and dissemination for statistical purposes of information relating to economic, social and general activities and conditions in the State". The CSO is also responsible for coordinating the official statistics of other public authorities and for developing the statistical potential of administrative records.

Arthur Beesley · 08-07-2014 06:10PM

bnt wrote: »

A couple of things:
- please don't quote a whole post just to reply with a line or two. A good guideline is to quote only the specific bit to which you're responding, and write at least as much.
- in this case you should have quoted nothing, since you're responding to something I didn't say. Yes, some things can be and are measured accurately, and I never claimed otherwise.

Though it's all commonly lumped under the heading of "statistics", your example (how many people over a certain age died in a particular year) is better described as data or information. Statistical methods can then be used to analyse the data to give you e.g. average life expectancy. I used the example of height: the statistical bit only kicks in after the heights are measured, and the more accurately they're measured, the more useful the statistics that can be derived from them.

The CSO publishes both information and statistics derived from the informations. From their "About Us" page:

Thanks for the very informative information on quoting bnt.

In other news, obvious is obvious.

Turtwig · 08-07-2014 06:14PM

I wish how to lie with statistics was mandatory reading for everyone.

Arthur Beesley · 08-07-2014 06:29PM

Turtwig wrote: »

I wish how to lie with statistics was mandatory reading for everyone.

So we could all do it? Nice.

Amirani · 08-07-2014 06:33PM

Arthur Beesley wrote: »

So we could all do it? Nice.

So we could all know when it's being done.

Arthur Beesley · 08-07-2014 06:38PM

Diego Simeone wrote: »

So we could all know when it's being done.

I was (clearly) being facetious.

Peregrine · 08-07-2014 07:43PM

bnt wrote: »

For example: the average adult male height in Ireland is 177.5 cm*. What do we learn from that?
...
- "Is there an "average man" who is that height?" The media seems to like this one, but it's irrelevant.

If by average you mean the median, the population of males in Ireland was an odd number and the entire population was surveyed, then yes there is an 'average man' who is that height.

UCDVet · 08-07-2014 08:13PM

This isn't limited to statistics but it's often used in conjunction with it. People, so many people, don't understand the difference between CORRELATION and CAUSATION. And it drives me crazy.

First_October · 08-07-2014 08:34PM

To be fair even highly trained statisticians have poor intuitive notions of statistics. The popular ignorance of statistics that the OP refers to is sadly an inevitability of our psychology.

I'd highly recommend anyone to read `Thinking, Fast and Slow' by Daniel Kahneman for a brilliant account of this.

Popular Ignorance of Statistics

Comments