FAQ
It is currently Sun Oct 22, 2017 11:50 am


Author Message
marnixR
Post  Post subject: trends and correlation  |  Posted: Fri Jun 21, 2013 6:20 am
User avatar

Joined: Thu Aug 04, 2011 8:35 pm
Posts: 4883
Location: Cardiff, Wales

Offline
this was triggered by the peak oil thread, but i don't want this thread to devolve into another peak oil derivative

it's basically that i've noticed how some people use correlated trends to imply that we are able future behaviour on the basis that if there is a correlation, then there must be a fixed link between the 2

that's fine if the correlation really implies causation and we understand how causation forces the 2 trends to be correlated - if not then we end up with a rule of thumb, and the trouble with rules of thumb is that they only apply within a limited field of experience and break down outside it

hence it's no good showing that a correlation has held true for the last 100 years to imply that it always will : you first need to understand what is the basis of the correlation in order to determine under what circumstances it will hold true in the future

after all, the only thing the economic models gave the financial markets was a feeling rather than the reality of control, and it did not prepare them for the crash of 2008 - so much for any correlations and trends that may have been observed if a crucial factor such as the sub-prime market failed to be included in the analysis

_________________
"Reality is that which, when you stop believing in it, doesn't go away." (Philip K. Dick)
"Someone is WRONG on the internet" (xkcd)


Top
tridimity
Post  Post subject: Re: trends and correlation  |  Posted: Fri Jun 21, 2013 12:31 pm

Joined: Fri Aug 12, 2011 5:55 pm
Posts: 1117

Offline
Quote:
marnixR wrote:
it's basically that i've noticed how some people use correlated trends to imply that we are able future behaviour on the basis that if there is a correlation, then there must be a fixed link between the 2


Absolutely, you can test association using Spearman or Pearson statistical tests until you are blue in the face - but if you do not prove direct causation by way of eliminating and restoring the suspected causative factor and witnessing a rescue of the original situation - then it is not possible to credibly claim causation.

Quote:
marnixR wrote:
that's fine if the correlation really implies causation and we understand how causation forces the 2 trends to be correlated - if not then we end up with a rule of thumb, and the trouble with rules of thumb is that they only apply within a limited field of experience and break down outside it


Mm, and what complicates matters is the shear complexity of Economics. There are so many factors to consider that even if you do have a knowledge of direct causation between two factors, there are so many interacting variables that making accurate economic predictions becomes overwhelmingly difficult.

Quote:
marnixR wrote:
hence it's no good showing that a correlation has held true for the last 100 years to imply that it always will : you first need to understand what is the basis of the correlation in order to determine under what circumstances it will hold true in the future


I am reminded here of the black swan theory - just because all of the swans you have ever seen have been white, does not mean to say that all swans are white, or that we should not consider the possibility of observing swans of a different colour in the future. I guess this is basically about staying awake and not becoming too complacent in our assumptions and paradigms.

_________________
gone also


Top
seagypsy
Post  Post subject: Re: trends and correlation  |  Posted: Tue Jun 25, 2013 4:54 pm
User avatar

Joined: Wed Jan 30, 2013 4:44 am
Posts: 176
Location: DFW, TX

Offline
I agree. I often see statistics used in political and social debates but in reality statistics (in these areas) mean very little. IMO, statistics are the least accurate form of data collection. They depend on accuracy and honesty of reporting, taking into account ALL possible factors (which they never do) and there is never a control situation. They are not double blind studies and even those are blighted with interpretation bias.

And depending on who collected the stats, what the source of reporting was or where the source of reports was, you may get entirely different data which shows that you will get conflicting data all over the place. Which should rule out a positive correlation and proof of causation. But people who use these as argument points do so because they have nothing else to go on other than their own personal beliefs and/or fears, especially in political debates. They want their fears or beliefs to be justified by science so they through numbers out there, number that seem to support their point of view, but the other side has the same arsenal of numbers and both sides are able to interpret the numbers to their own whims. Both will claim the other person's numbers are wrong or lies when it may be that they are both being technically honest. but they both fail to realize that the statistics simply do not support either side of the debate.

_________________
It takes a certain amount of obvious stupidity to keep the science geeks from falling asleep. I am happy to do my part in keeping them awake.


Top
tridimity
Post  Post subject: Re: trends and correlation  |  Posted: Wed Jun 26, 2013 9:39 am

Joined: Fri Aug 12, 2011 5:55 pm
Posts: 1117

Offline
Quote:
seagypsy wrote:
IMO, statistics are the least accurate form of data collection. They depend on accuracy and honesty of reporting, taking into account ALL possible factors (which they never do) and there is never a control situation. They are not double blind studies and even those are blighted with interpretation bias.


All true but statistics can at least point us in the direction or asking the informative questions about potential cause-effect relations.

Quote:
seagypsy wrote:
They want their fears or beliefs to be justified by science so they through numbers out there, number that seem to support their point of view, but the other side has the same arsenal of numbers and both sides are able to interpret the numbers to their own whims.


Yeah, if you ignore the human influence and human motives for distorting or misreporting statistical data then statistics would be one of the best approaches to understanding humanitarian issues. The alternative is to rely purely on anecdotal evidence, which as you say is not a desirable approach in these situations. Which is why it is important to think independently and to take into consideration all of the available information - both statistical and anecdotal - keeping in mind one's own subjective prejudices. It's not easy but this is the optimal approach to discovering truth.

_________________
gone also


Top
seagypsy
Post  Post subject: Re: trends and correlation  |  Posted: Wed Jun 26, 2013 9:58 am
User avatar

Joined: Wed Jan 30, 2013 4:44 am
Posts: 176
Location: DFW, TX

Offline
tridimity wrote:
Quote:
seagypsy wrote:
IMO, statistics are the least accurate form of data collection. They depend on accuracy and honesty of reporting, taking into account ALL possible factors (which they never do) and there is never a control situation. They are not double blind studies and even those are blighted with interpretation bias.


All true but statistics can at least point us in the direction or asking the informative questions about potential cause-effect relations.

Quote:
seagypsy wrote:
They want their fears or beliefs to be justified by science so they through numbers out there, number that seem to support their point of view, but the other side has the same arsenal of numbers and both sides are able to interpret the numbers to their own whims.


Yeah, if you ignore the human influence and human motives for distorting or misreporting statistical data then statistics would be one of the best approaches to understanding humanitarian issues. The alternative is to rely purely on anecdotal evidence, which as you say is not a desirable approach in these situations. Which is why it is important to think independently and to take into consideration all of the available information - both statistical and anecdotal - keeping in mind one's own subjective prejudices. It's not easy but this is the optimal approach to discovering truth.


Potential for great usefulness is certainly there. It's just sad that human imperfection makes it extremely difficult to get a bias free interpretation of the data. It also makes some of terrible spellers. Some day, I will write T-H-R-O-W when I intend to state that people are symbolically launching information at one another rather than t-h-r-o-u-g-h. As well as using plurals and singular nouns in proper places. I must stop posting when I am tired. But I am always tired.

_________________
It takes a certain amount of obvious stupidity to keep the science geeks from falling asleep. I am happy to do my part in keeping them awake.


Top
kojax
Post  Post subject: Re: trends and correlation  |  Posted: Sun Jul 21, 2013 10:03 pm
Original Member
Original Member

Joined: Mon Aug 15, 2011 11:43 am
Posts: 582

Offline
seagypsy wrote:
I agree. I often see statistics used in political and social debates but in reality statistics (in these areas) mean very little. IMO, statistics are the least accurate form of data collection. They depend on accuracy and honesty of reporting, taking into account ALL possible factors (which they never do) and there is never a control situation. They are not double blind studies and even those are blighted with interpretation bias.

And depending on who collected the stats, what the source of reporting was or where the source of reports was, you may get entirely different data which shows that you will get conflicting data all over the place. Which should rule out a positive correlation and proof of causation. But people who use these as argument points do so because they have nothing else to go on other than their own personal beliefs and/or fears, especially in political debates. They want their fears or beliefs to be justified by science so they through numbers out there, number that seem to support their point of view, but the other side has the same arsenal of numbers and both sides are able to interpret the numbers to their own whims. Both will claim the other person's numbers are wrong or lies when it may be that they are both being technically honest. but they both fail to realize that the statistics simply do not support either side of the debate.


The problem is, when statistics are used in a political debate they are asking you to trust both:

1) - The raw data
an
2) - The presenter's interpretation of that data.

#2 is always suspect in a political debate.

Statistics are easy to distort because the required brevity of a statement made in political debate requires most of the raw information to be omitted. You're only hearing conclusions, with a smattering of the data behind them, and are being asked to assume the rest of the data exists without knowing whether it exists or not.

tridimity wrote:

Quote:
marnixR wrote:
that's fine if the correlation really implies causation and we understand how causation forces the 2 trends to be correlated - if not then we end up with a rule of thumb, and the trouble with rules of thumb is that they only apply within a limited field of experience and break down outside it


Mm, and what complicates matters is the shear complexity of Economics. There are so many factors to consider that even if you do have a knowledge of direct causation between two factors, there are so many interacting variables that making accurate economic predictions becomes overwhelmingly difficult.


Yeah. The problem is that the system is dynamic. There are chain reactions where one thing causes another, which causes another, ... etc.

In he health insurance industry, on the other hand, they system an insurance person analyzes is less dynamic. They can use statistical analysis to determine who will or will not fall off a roof to a higher degree of reliability, because one person falling off a roof usually doesn't cause another person to fall off a roof also. They're isolated events. If anything causes them it's an external factor, and so that factor may be detectable because it is clearly isolated from its effects.

But when effects are causes, and causes are effects, that's just a mess.

tridimity wrote:

Quote:
marnixR wrote:
hence it's no good showing that a correlation has held true for the last 100 years to imply that it always will : you first need to understand what is the basis of the correlation in order to determine under what circumstances it will hold true in the future


I am reminded here of the black swan theory - just because all of the swans you have ever seen have been white, does not mean to say that all swans are white, or that we should not consider the possibility of observing swans of a different colour in the future. I guess this is basically about staying awake and not becoming too complacent in our assumptions and paradigms.


That's a problem of using large scale observations to make small scale predictions. However you can still predict that the next swan you see has X% probability of being white. You just can't say for sure it will be white.

You can also predict that out of a large group of swans, most of them will be white.

However, just like insurance, you have the advantage here that the cause of swans' being one color or another is (mostly) an external cause.


Top
kojax
Post  Post subject: Re: trends and correlation  |  Posted: Mon Jul 22, 2013 12:43 am
Original Member
Original Member

Joined: Mon Aug 15, 2011 11:43 am
Posts: 582

Offline
So the short version is that statistics are great for detecting an external cause, but almost useless for detecting an internal cause.


And people like to claim statistical certainty without actually giving you their data.


Top
tridimity
Post  Post subject: Re: trends and correlation  |  Posted: Mon Jul 22, 2013 10:53 am

Joined: Fri Aug 12, 2011 5:55 pm
Posts: 1117

Offline
Quote:
kojax wrote:
(2) - The presenter's interpretation of that data.

#2 is always suspect in a political debate.


True.

Also (3) - The way in which the presenter portrays the data - may be misleading:

Quote:
'Scientists say 1 in 2 people are of above average weight.' Sounds like we're a nation of fatties. It's a scientific analysis of the facts, and almost certainly true... the above headline could just as accurately say, 'Scientists say 1 in 2 people are of below average weight.'

[It's also possible to] phrase things in a leading way: 90% fat free! Would you buy it if it were 90% cyanide free?

[All 3 describe the same data]:

i. The amount of energy wasted is increasing
ii. Energy wasted per person is decreasing
iii. The rate at which energy waste is increasing is slowing down

Two sound positive and one negative.


Quote:
kojax wrote:
So the short version is that statistics are great for detecting an external cause, but almost useless for detecting an internal cause
.

What do you mean by internal and external cause? Do these have meaning outside the context of Psychology?

_________________
gone also


Top
kojax
Post  Post subject: Re: trends and correlation  |  Posted: Mon Jul 29, 2013 11:11 am
Original Member
Original Member

Joined: Mon Aug 15, 2011 11:43 am
Posts: 582

Offline
tridimity wrote:

Quote:
kojax wrote:
So the short version is that statistics are great for detecting an external cause, but almost useless for detecting an internal cause
.

What do you mean by internal and external cause? Do these have meaning outside the context of Psychology?


I just mean that the cause is not being indirectly observed, only the effect.

Suppose a given population group has a history of health problems, but also wasn't previously insured and didn't have ready access to prevention. It's virtually impossible to use statistics to reliably determine what the likelihood of them continuing to have health problems will be once they're insured and getting prevention. (Not saying an insurance statistician wouldn't venture a guess.)

However, if they had always had access to prevention, so by insuring them we're not altering the odds of anything, then their health history can be used to predict the likelihoods pretty well. If the cause remains unchanged, we can predict that it will continue to do what it has done before. We don't need to actually understand the cause.

The trouble with econ is everything depends on everything else, so the cause always does change whenever we alter any of its effects downstream. The only way it doesn't is if we're making a prediction about an element that is so small it doesn't affect the larger whole, only gets affected by it. (That's why it's rarely a good idea to mix micro and macro econ, unless you're being careful to remember what is what.)


Top
Display posts from previous:  Sort by  
Print view

Who is online
Users browsing this forum: No registered users and 0 guests
You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
Jump to:   
cron

Delete all board cookies | The team | All times are UTC


This free forum is proudly hosted by ProphpBB | phpBB software | Report Abuse | Privacy