PDA

View Full Version : Statistically: How many statistics are useless?



ki4jgt
December 30th, 2010, 05:05 AM
Everyone knows there are anomalies which always seem to defy statistics. In these anomalies statistics are virtually useless. In fact, using statistics in these anomalies, may actually cause the anomalies to be botched to the point that they are no longer an anomaly. (I'm using way too many big words here.)

But I am talking about these kinds of statistics. 100% of divorces are caused by marriage.

handy
December 30th, 2010, 05:46 AM
If you study stat's you'll find out that it is quite easy to make them say anything you want.

That doesn't mean that all statistics are lies, it just means that you have a close look at the stat's for the subject & decide if there would be any reason for them to be perverted.

MisterGaribaldi
December 30th, 2010, 05:56 AM
15 out of 10 statistics are lies or are fundamentally flawed.

I tend to ignore statistics for this reason.

ki4jgt
December 30th, 2010, 06:04 AM
15 out of 10 statistics are lies or are fundamentally flawed.

I tend to ignore statistics for this reason.

My English teacher used to tell us to make up stats. She said that 90% of stats were made up on the spot.

handy
December 30th, 2010, 06:25 AM
My English teacher used to tell us to make up stats. She said that 90% of stats were made up on the spot.

Perhaps in forums & informal conversation. In more formal situations where statistics are involved it can be that the desired results dictate which form of statistical analysis is used. Interestingly the same statistical sample can be analysed in fashions that bring about very different results.

Statistics can be used to create incredibly useful & valid information. It is just that the data from the same statistical sample can also be used for very useful (to a certain party) & totally invalid results.

MisterGaribaldi
December 30th, 2010, 06:44 AM
I deal in facts. Stats are, in my experience, used most of the time to justify someone's point of view, not to present a picture of reality.

handy
December 30th, 2010, 07:09 AM
I deal in facts. Stats are, in my experience, used most of the time to justify someone's point of view, not to present a picture of reality.

You have a limited view of the definition of statistics it would seem, to the statistical formulas that are used on the statistical data.

The statistical data can be extremely accurate, such as how many tickets have been sold each day in this movie theatre for the last 10 years. The answer to that question would be valid statistical data. It is actually a statistic created from many statistics.

As one of the definitions is that statistics are: the numerical facts or data themselves.

So I take it that your meaning is that you don't trust the people that choose which mathematical formula to use on the whatever various collections of statistical data. Like presidential elections in the US for example? :)

If so, as I said in my previous post, I agree with you completely.

MisterGaribaldi
December 30th, 2010, 07:21 AM
handy:

My "limited view" of statistics comes from seeing them used most of the time to prop up something instead of to legitimately "fill in the gaps" where actual data is non-existent or where there exists a legitimate need to simplify the data set being used.

I could give an analogy, but the only suitable ones which come to mind are political in nature, and therefore would be deemed "unsuitable" for UbuntuForums. :(

handy
December 30th, 2010, 07:23 AM
handy:

My "limited view" of statistics comes from seeing them used most of the time to prop up something instead of to legitimately "fill in the gaps" where actual data is non-existent or where there exists a legitimate need to simplify the data set being used.

I could give an analogy, but the only suitable ones which come to mind are political in nature, and therefore would be deemed "unsuitable" for UbuntuForums. :(

It's ok, I gave a perfectly valid political one. :)

MisterGaribaldi
December 30th, 2010, 07:25 AM
Alright...

There's a lot of people out there who think guns are evil and should be regulated or controlled or banned altogether. In a way, this is how I feel about statistics.

cariboo907
December 30th, 2010, 07:36 AM
I worked for the market research division of an advertising firm in the late 80's. The results of all the surveys we did, were always what the clients expected.

radar920
December 30th, 2010, 07:41 AM
150%

handy
December 30th, 2010, 07:49 AM
Alright...

There's a lot of people out there who think guns are evil and should be regulated or controlled or banned altogether. In a way, this is how I feel about statistics.

We have all have actually done that in Oz (re. guns that is), (with very few exceptions) because a lot of people thought the same way as you stated.

So I totally agree with your thoughts on that subject (guns not statistics) too. :)

We are an agreeable pair. lol

handy
December 30th, 2010, 08:03 AM
I worked for the market research division of an advertising firm in the late 80's. The results of all the surveys we did, were always what the clients expected.

It is very easy to see a firm that dealt in statistical analysis, reckoning its way into staying in business & keeping its clients. Most especially the ones that paid me the most for my services. (I'm not saying that your company did that cariboo, I'm just stating what so many of us think about both advertising & statistics.)

It is the nature of business to behave like that, & that is the reason why so many people distrust statistics. Which is most unfortunate as statistics can & so often do give us incredibly valid & useful information.

Unfortunately so many people have been trained not to take any notice of statistics, due to the inherent ability of them to be twisted 180 (or any other number of) degrees into total falsehood.

This is unfortunately all to humanities detriment as there are so many valid & crucially important statistics that are released to the public that are ignored & disbelieved.

When you only believe in facts, then what do you use to warn you to take preventative action if not projections that come from statistical analysis. (I know there is this & that, but most of our friends aren't research scientists or investigative journalists that actually tell the truth.)

JDShu
December 30th, 2010, 08:33 AM
People die on average once.

But really, statistics don't lie. They CAN'T lie, they're numbers. Its the people who use them in order to give false impressions, or the people who misinterpret them that is the problem. It takes critical thinking to figure out if a given statistic is relevant to the topic at hand.

handy
December 30th, 2010, 10:00 AM
People die on average once.

But really, statistics don't lie. They CAN'T lie, they're numbers. Its the people who use them in order to give false impressions, or the people who misinterpret them that is the problem. It takes critical thinking to figure out if a given statistic is relevant to the topic at hand.

I agree re. the statistical data.

I also agree that the anomalies arise from the statisticians who are usually tasked with the job of giving their employer the results that their employer desires.

Under those circumstances the statistician has a vast array of choices of formulas from which to choose in an effort to give their employer (or government department) the desired result.

We who are interested in a topic are left with the job of doing research in an effort to understand the topic better & to then decide whether the statistical analysis we have been presented with is valid or questionable.

How many people bother to do this?

How many people have the time to bother to do this?

I do, most don't.

Paqman
December 30th, 2010, 10:11 AM
But I am talking about these kinds of statistics. 100% of divorces are caused by marriage.

That's just plain bad statistics, although it's a pretty common mistake: Correlation does not imply causation (http://en.wikipedia.org/wiki/Correlation_does_not_imply_causation).

bouncingwilf
December 30th, 2010, 10:18 AM
"Lies, Damned lies and statistics" is a quote ascribed to Winston Churchill I believe.

However, they're not all bad, I used statistical methods extensively in my work as a problem solver. They can be very powerful.



Bouncingwilf

spupy
December 30th, 2010, 10:34 AM
My English teacher used to tell us to make up stats. She said that 90% of stats were made up on the spot.


74% of all statistics are made up. The other 36% have calculation errors. :)

handy
December 30th, 2010, 10:40 AM
74% if all statistics are made up. The other 36% have calculation errors.

:lolflag: So in your case 100%!

I'm glad you're not a statistician spupy. :)

koenn
December 30th, 2010, 11:09 AM
The problem is not so much the statistics, but
a- the people that make them
b- the people that interpret them

There's an obvious correlation between a person's IQ and their attitude towards statistics :

http://sporkforge.com/prmimg/04d1/fg261393420029c5i0003000057fb1020.png

It remains to be seen whether the underlying cause is that people with higher IQ
- are more capable of doing statistics correctly
- are more capable of understanding what information is or isn't represented by the data
- tend to have more knowledge of the problem domain so they can see facts and stats in the right context and perspective

spupy
December 30th, 2010, 11:18 AM
The problem is not so much the statistics, but
a- the people that make them
b- the people that interpret them

There's an obvious correlation between a person's IQ and their attitude towards statistics :

http://sporkforge.com/prmimg/04d1/fg261393420029c5i0003000057fb1020.png

It remains to be seen whether the underlying cause is that people with higher IQ
- are more capable of doing statistics correctly
- are more capable of understanding what information is or isn't represented by the data
- tend to have more knowledge of the problem domain so they can see facts and stats in the right context and perspective

Well, guess what, the chart you got there is a perfect example of bad statistics. ("whoosh" for me if you posted it as a joke.)
What's the Y axis? What are the values on both axes? How do you measure "attitude"?

handy
December 30th, 2010, 11:30 AM
Attitude is the "Z" axis. :)

koenn
December 30th, 2010, 11:35 AM
("whoosh" for me if you posted it as a joke.)

The graph is a joke, obviously (http://sporkforge.com/misc/fake_graph.php) (but having to admit that kinda ruins it :( )

otoh, I'm quite serious about the gist of that post, that stats are often misused, purposely or unknowingly, but that this doesn't diminish the value of statistics as such.

handy
December 30th, 2010, 11:41 AM
I think that it certainly leaves plenty of room for companies that supply statistical data to clients to make themselves very comfortable.

Of course it is easier for such companies to manipulate their clients during certain economic & political conditions than others. Just like it is easier to manipulate the mindset & emotions of populations of people under certain sets of circumstances than others. This has been shown time & time again historically.

ki4jgt
December 30th, 2010, 02:23 PM
The problem is not so much the statistics, but
a- the people that make them
b- the people that interpret them

There's an obvious correlation between a person's IQ and their attitude towards statistics :

http://sporkforge.com/prmimg/04d1/fg261393420029c5i0003000057fb1020.png

It remains to be seen whether the underlying cause is that people with higher IQ
- are more capable of doing statistics correctly
- are more capable of understanding what information is or isn't represented by the data
- tend to have more knowledge of the problem domain so they can see facts and stats in the right context and perspective

My IQ is 133 and I hate statistics :-) I've always been an anomaly :'-(

DZ*
December 30th, 2010, 04:58 PM
People have to make decisions in situations where they don't know with 100% certainty what reality is.

If we only deal in facts, we'll be hit by a killer asteroid, eventually, or hell ... err ... I mean Gulf Stream will freeze over :)

Statistics is then used to estimate relative likelihoods for various possible states of reality given imperfect evidence and prior knowledge.

Ideally, statistics should also be used to guide behavior given that a particular state of reality is correct, while taking into account consequences or cost of that behavior.

"How likely is that this man raped the victim given the DNA found on her bed? Based on that and other knowledge, should he be convicted?"

"How likely is it I am pregnant?" (most often asked question ever that starts with "how likely")

"Stats" are aggregate summaries of data. "Statistics" as a scientific discipline utilizes stats toward its goal of making decisions under uncertainty.

MisterGaribaldi
December 30th, 2010, 05:30 PM
The problem with your examples, DZ*, is probabilities aren't substitutes for facts.

The man either raped the woman, or he didn't. You can't be guilty by proxy.

A woman is either pregnant, or she isn't. There's no such thing as being "slightly pregnant".

Statistics have their place, but the most they're good for is approximations.

DZ*
December 30th, 2010, 06:20 PM
The problem with your examples, DZ*, is probabilities aren't substitutes for facts. The man either raped the woman, or he didn't. You can't be guilty by proxy

All real world problems are like this. We can only disprove theories (e.g. crime scene DNA didn't match the suspect). Proving theories is impossible and instead we have to act as if a theory is true based on the probability odds.

We will never know if in fact that man raped the woman unless he did it in the court room in front of the jury. "DNA match" is a probability statement because two unrelated people have non-zero probability of sharing the DNA profile. Does that mean the DNA evidence should be ignored? No, but you cannot interpret it without statistics.

ki4jgt
December 30th, 2010, 07:12 PM
All real world problems are like this. We can only disprove theories (e.g. crime scene DNA didn't match the suspect). Proving theories is impossible and instead we have to act as if a theory is true based on the probability odds.

We will never know if in fact that man raped the woman unless he did it in the court room in front of the jury. "DNA match" is a probability statement because two unrelated people have non-zero probability of sharing the DNA profile. Does that mean the DNA evidence should be ignored? No, but you cannot interpret it without statistics.

Kind of like DNA, and pregnancy tests always being %100 acurrate if they're wrong, but then you find out that someone accidentally switched the results with someone else, or that kids can be born with the DNA of only one of their parents and not both. Point being, even though we know the statistics, and yes, we can use them (In cases like rape), but they're still only statistics and present very little perspective on what is reality. Which is what I meant by statistics forcing the anomaly to fall in with them. Inaccurate statistics or (incomplete - as all statistics are) tend to leave the anomalies standing by themselves and even in some cases forces it to adapt to something which is not in anyway related to it. Though all statistics are based on 1:1 ratio, they soon become irrelevant after 1:2, 2:2 except in matters where math is concerned. Every statistic must first be approached on a 1:1 basis before it is ever passed to 1:2 or 2:2 or even 2:1.

koenn
December 30th, 2010, 08:05 PM
"Stats" are aggregate summaries of data. "Statistics" as a scientific discipline utilizes stats toward its goal of making decisions under uncertainty.
Not really. Stats are a way of organizing data, and a tool to detect and describe relations between them. In that sense, they actually are "aggregate summaries of data." but it's the transition from 'data' to 'infortmation' that is more important.
(Statistics as a science describes the methodology for making that transition a valid one)


... statistics forcing the anomaly to fall in with them. Inaccurate statistics or (incomplete - as all statistics are) tend to leave the anomalies standing by themselves and even in some cases forces it to adapt to something which is not in anyway related to it.
...
statistics don't force anything on anyone. But sometimes people misunderstand them, and sometimes they are misused.
Take "IQ" - it's all stats : a number that represents your score on a standardized test, compared to other people's score, on a Gauss curve, with a given standard deviation.
Now, if you scored 175; you'd be somewhat of an anomaly, but how would that number be "incomplete" or "force you" ?

wannabegeek
December 30th, 2010, 08:08 PM
this a great thread...so many witticisms I have a tear in my eye...
One of my favs so far is "People die on average once"......

I use stats to figure out how much heat is turning into kinetic energy and things like that because the number of particle collisions is immense....
The average value is often the only usable or knowable value...

Probability is one of the basis for quantum mechanics...nuclear decay and light transmission are described as a probability

Electrons do not orbit the nucleous, they exist in a probability cloud...and this fact was confirmed using experiment and statistical analysis.... the social sciences abuse the hell out of it...IMHO

del_diablo
December 30th, 2010, 08:17 PM
Sturgons law should be a good answer.

DZ*
December 30th, 2010, 08:31 PM
Though all statistics are based on 1:1 ratio, they soon become irrelevant after 1:2, 2:2 except in matters where math is concerned. Every statistic must first be approached on a 1:1 basis before it is ever passed to 1:2 or 2:2 or even 2:1.

I have no idea what that means. But it reminds me of a joke
Q: What are the chances of seeing a dinosaur on Manhattan?
A: 50%. It's either there or not.

DZ*
December 30th, 2010, 08:39 PM
kids can be born with the DNA of only one of their parents and not both

Last time that happened was 2000 years ago with a woman called Mary.

It defied scientific explanation though, because having all of her genes was supposed to result in a daughter.

ki4jgt
December 30th, 2010, 09:06 PM
Not really. Stats are a way of organizing data, and a tool to detect and describe relations between them. In that sense, they actually are "aggregate summaries of data." but it's the transition from 'data' to 'infortmation' that is more important.
(Statistics as a science describes the methodology for making that transition a valid one)


statistics don't force anything on anyone. But sometimes people misunderstand them, and sometimes they are misused.
Take "IQ" - it's all stats : a number that represents your score on a standardized test, compared to other people's score, on a Gauss curve, with a given standard deviation.
Now, if you scored 175; you'd be somewhat of an anomaly, but how would that number be "incomplete" or "force you" ?

Say someone had studied the test before hand. . . After taking so many of those tests you'd be able to get them right pretty easy. Or say someone grew up learning a certain way whatever ways an IQ test measures. It would be an anomaly to that individual.


this a great thread...so many witticisms I have a tear in my eye...
One of my favs so far is "People die on average once"......

I use stats to figure out how much heat is turning into kinetic energy and things like that because the number of particle collisions is immense....
The average value is often the only usable or knowable value...

Probability is one of the basis for quantum mechanics...nuclear decay and light transmission are described as a probability

Electrons do not orbit the nucleous, they exist in a probability cloud...and this fact was confirmed using experiment and statistical analysis.... the social sciences abuse the hell out of it...IMHO

But all those responses depend on 1:1 values before they ever get to 1:2 or 1:( x ) . They depend on the fact that the only thing which moves the mercury is another particles collission. So we know that it is only particles and not for example photons (Unless the thermometer is exposed to light.) But out of Science and Math, other social areas are not good to force statistical data into and when you hear crazy people every day trying to force statistics into situations like (Personality) for instance then you're bound to have trouble. As far as IQs go, statistically people with higher IQs exhibit so many psycological behaviors, it's not funny, now to force the idea that just because you have a high IQ, you're going to have some kind of psychological issues, is a very bad statement. Even though Statistics back the idea.

ki4jgt
December 30th, 2010, 09:14 PM
I have no idea what that means. But it reminds me of a joke
Q: What are the chances of seeing a dinosaur on Manhattan?
A: 50%. It's either there or not.

You just answers your own question. It's either there or it's not, then statistics should be brought in to determine what we need to do about it if it is


Last time that happened was 2000 years ago with a woman called Mary.

It defied scientific explanation though, because having all of her genes was supposed to result in a daughter.

You need to look into more modern cases of this. Children can be born without their mothers DNA. Look into the statistics (It's very small)

DZ*
December 30th, 2010, 09:27 PM
You need to look into more modern cases of this. Children can be born without their mothers DNA. Look into the statistics (It's very small)

I don't believe that it is possible. You do realize that the sperm is haploid, right? To produce a baby like that would require a mysterious duplication of father's genome.

OTOH, Parthenogenesis (mothers clone) is theoretically possible but I believe it never have been documented in humans.

DZ*
December 30th, 2010, 09:30 PM
re: '"Stats" are aggregate summaries of data. "Statistics" as a scientific discipline utilizes stats toward its goal of making decisions under uncertainty.'


Not really. Stats are a way of organizing data, and a tool to detect and describe relations between them. In that sense, they actually are "aggregate summaries of data." but it's the transition from 'data' to 'infortmation' that is more important. (Statistics as a science describes the methodology for making that transition a valid one)

My view is that statistics is not merely a methodology for obtaining information from data. Information is contained in the likelihood of data under an assumed state of reality (in the sense of the amount of information that observations carry about an unknown parameter).

What we often want to do next is to reverse that and estimate how relatively likely this particular state of reality is, given the information, and make decisions based on that.

koenn
December 30th, 2010, 10:03 PM
What we often want to do next is to reverse that and estimate how relatively likely this particular state of reality is, given the information, and make decisions based on that.
Hm, yes. It wouldn't occur to me as the essence of statistics, but I see
how statistics can be used to that effect as well.

koenn
December 30th, 2010, 10:25 PM
Say someone had studied the test before hand. . . After taking so many of those tests you'd be able to get them right pretty easy.
Well, I explained IQ as a score on a test, I didn't imply anything about what that score would mean (what does an IQ test measure, intelligence ? the "ability to do IQ tests" ? ...)

otoh, lots of those tests require that you can see relations between abstract drawings. In my opinion, intelligence is about being capable of dealing with abstractions, and about seeing relations, so I think those tests do measure "intelligence".
If doing lots of those tests actually trained you in dealing with abstractions and seeing relations, it's only right that you're score goes up : it should reflect your improved capabilities.

in any case, defining what it is your measuring, and making sure your tests measures that what you want to measure, is a crucial step in designing tests. That's not specific to statistics.



Or say someone grew up learning a certain way whatever ways an IQ test measures. It would be an anomaly to that individual.
IQ tests are to be conducted by trained psychologists - they're supposed to know about such cases.

ki4jgt
December 30th, 2010, 10:56 PM
Well, I explained IQ as a score on a test, I didn't imply anything about what that score would mean (what does an IQ test measure, intelligence ? the "ability to do IQ tests" ? ...)

otoh, lots of those tests require that you can see relations between abstract drawings. In my opinion, intelligence is about being capable of dealing with abstractions, and about seeing relations, so I think those tests do measure "intelligence".
If doing lots of those tests actually trained you in dealing with abstractions and seeing relations, it's only right that you're score goes up : it should reflect your improved capabilities.

in any case, defining what it is your measuring, and making sure your tests measures that what you want to measure, is a crucial step in designing tests. That's not specific to statistics.



IQ tests are to be conducted by trained psychologists - they're supposed to know about such cases.

I've dealt with my share of those. I deal with people (Sometimes :-) - not a lot) But from what I've seen (As with any human) Psychologists only know what you tell them and have to guess the rest (From statistical data :-)). They're no different than we are. I had to learn that the hard way, when I was getting my psychological examination/Evaluation <Job related> :-). I didn't tell mine everything about my life. He diagnosed me with so many things it wasn't even funny. I didn't believe him at first, but then I started to (He turned me into a nervous wreck). Later someone else saw me and said he had made so many mistakes. Had statistics been involved, I would've been in the nut house by now, thankfully they hadn't because after the way I felt. I was about ready to rip someone's head off. If someone relies simply on statistics, they there is no way in which extra statistics may be added to the equation. I understand we need them, but some of them are just plain useless I'm sorry, but I will always believe that.

***EDIT: Do I blame the guy?? A little. . . He shouldn't be working at a job where he has to deal with other's pain, if he can't sense it himself. Humans are the biggest anomalies I've ever seen. But all in all, it was my fault. I took the guy seriously.

radar920
December 30th, 2010, 11:15 PM
Statistics are only relevent if they fit my personal beliefs and and feelings, everything else is useless.

handy
December 30th, 2010, 11:32 PM
Last time that happened was 2000 years ago with a woman called Mary.

It defied scientific explanation though, because having all of her genes was supposed to result in a daughter.

:lolflag:

That's what miracles have to do... ;)

koenn
December 30th, 2010, 11:33 PM
@ ki4jgt

Well, of course, the underlying assumption was that those psychologists know their job and do it right.

Sounds like you've had your share of bad luck with those.

Also, line DZ said, statistics are aggregate summaries of data. Using them for conclusions on one specific individual is simply a wrong application.

MisterGaribaldi
December 31st, 2010, 12:06 AM
I don't mind IQ tests being conducted by trained psychologists. Heck, I've studied for all my drug tests and always passed them so who am I to criticize?

ki4jgt
December 31st, 2010, 02:20 AM
@ ki4jgt

Well, of course, the underlying assumption was that those psychologists know their job and do it right.

Sounds like you've had your share of bad luck with those.

Also, line DZ said, statistics are aggregate summaries of data. Using them for conclusions on one specific individual is simply a wrong application.

Not just me, I've only had one intense psychology session for my Job. But some of my siblings and friends have also had bad expiriences. Things like tying to make them feel every little thing in their lives, when they weren't even having problems to begin with. School psychologists who pulled them aside to ask them if anything was up in their lives and then went into every single detail. Worst part is, I used to want to be one. Now I don't know. They're just too pushy! :-(