So you've got some statistics that show a correlation between being Muslim and certain kinds of crimes. Repeat after me, very slowly:

CORRELATION IS NOT CAUSATION

The trouble with this analysis isn't the math or the statistical data. It is the implicit proposition that one's religion is the only relevant factor when it comes to one's propensity to commit a crime. Heaven knows, there are a myriad things that probably have bearing, such as social background, economic status, just to name two off the top of my head. In that context, it doesn't seem like a stretch to claim that the analysis is both simplistic and incomplete. Anyone who claims that this statistical analysis is rigorous enough to draw any sort of firm conclusion on is guilty of sloppy statistical methods.

Quote:

---

What I haven't done is download the source data and run the analysis myself.

---

That is misdirection.

Quote:

---

CORRELATION IS NOT CAUSATION

The trouble with this analysis isn't the math or the statistical data. It is the implicit proposition that one's religion is the only relevant factor when it comes to one's propensity to commit a crime. Heaven knows, there are a myriad things that probably have bearing, such as social background, economic status, just to name two off the top of my head. In that context, it doesn't seem like a stretch to claim that the analysis is both simplistic and incomplete.

---

The guy admits that it's incomplete himself. He just posted it because he found the correlations interesting.

I wouldn't be surprised if people native to countries, legal systems, cultures, or, yes, religious dialects that are, to put it simply, nasty, or tend to put extreme burden of proof on the rape victim, are more likely to commit sex crimes. Bush called islam a religion of peace, not me, and I have no particular respect for it (in fact, less so than the median religion).

But this is an embodiment of "lies, damn lies, and statistics."

First, the sample sizes. Each of his graphs by percentage of muslim population is based on 1 or 2 cities.

Second, he's classifying the cities by muslim population.

He's ignoring any other demographics, such as:

income (he dismisses this as "insulting to the poor," but gimme a break, it tends to be associated with a wide range of cultural and social demographics)
whether the muslims are immigrants or not
culture of city
culture of muslim immigrants (they may come from different countries, not just the middle east)
age
gender
education
local laws
local police
city layout that may make sex crimes easier
population density in various parts of the cities
the demographics of the criminals who are actually committing the crimes and the victims, instead of the not quite so tangible association of any group and total population of a city

And somebody responded to him (no idea if this statistic is real): "You must answetr the following question for me please: what percentage of convicted sex offenders in the UK are Moslem? According to the 2005 NSY figures, UNDER ONE PERCENT!"

That seems like a much stronger number, even if it goes against my own expectations, than a rather convoluted correlation of muslim percentages of populations.

To be fair, he doesn't have access to so many demographics, and he admits that it is correlation and not causation (even though he goes on to say that he has drawn conclusions from it). But it's also fair to say it's not particularly convincing evidence (one way or the other; I'm not disputing that religion may play a role). It might be something to make a person say, hey, maybe a real in depth study should be done on this subject, by people who don't have an agenda?

Rico is disappointed but this veiled attempt at spiritual mudslinging. *Dons the sad panda hat*

~Rico

I'm not looking for why the work might be invalid, I'm looking for a mathematical explanation that contradicts his findings.

Correlation is not causation is a fine statement, and it is a true statement, except when it isn't. Absence of evidence is not evidence of absence, nor is absence of evidence evidence of presence. In the medical sciences we use correlation all the time, doing our very best to minimize confounding factors to judge the efficacy of drugs, diets, or treatments.

The entire list of confounding factors Veckums listed can't be run through for correlation trends, I can't even begin to see how "culture of the city" can be numerically expressed. But things like age, median income, etc CAN be numerically scored and manipulated to find correlation.

So take off the sad panda hat and put on the thinking cap.

Jimro

Nope hat stays on. Anyone thats been in a stat class knows you can work numbers to say what you want.

~Rico

Hey, you can correlate whatever you like. The question is not whether you numbers are correct - it's whether or not your statistical results are applicable.

Quote:

---

it is a true statement, except when it isn't

---

What on earth is this supposed to mean? Correlation is not causation.

Quote:

---

Absence of evidence is not evidence of absence, nor is absence of evidence evidence of presence.

---

You know, if you're trying to show that Muslims are intrinsically predisposed to certain kinds of crimes, I'd say that the burden of proof belongs to you, and you'll have to do a lot better than that.

Quote:

---

put on the thinking cap

---

No, you put on *your* thinking cap and listen good, because that little analysis you posted has got absolutely no merit on which to draw any sort of statistical inference. To say it *can't* be done does not excuse the author of the "analysis" from sloppy thinking.

Quote:

---

To say it *can't* be done does not excuse the author of the "analysis" from sloppy thinking.

---

I guess we'll just have to wait until the author finishes his incomplete findings (if he does complete it).

Quote:

---

The entire list of confounding factors Veckums listed can't be run through for correlation trends, I can't even begin to see how "culture of the city" can be numerically expressed.

---

That's part of my point (the difficulty of supporting any claims from such an analysis) to a degree. I did say it wasn't his fault he couldn't measure so many things.

Measuring a huge number of demographics to eliminate confounding factors to run in parallel with his analysis would be a big job, that would be limited to the demographics available on UK online sources, which is why I suggested that a dedicated study with more access might be better suited to it.

Any such correlations of population and area would be a lot more suspect as a source of conclusions than simple crime statistics, even with a huge number of demographics calculated, however.

Quote:

---

I'm not looking for why the work might be invalid, I'm looking for a mathematical explanation that contradicts his findings.

---

Well I don't think his math is the issue. He showed a correlation of Muslim population in particular cities and sex crimes, and I could check his math, but it would take a long time and I have no reason to suspect it is wrong. The only contradiction to the correlation might be taking a larger sample to see if it stands up, or measuring other demographics to see other correlations.

A mathematical contradiction of his conclusions would be the statistics of muslims convicted of sex crimes, but I have no idea if that is accurate.

Quote:

---

What on earth is this supposed to mean? Correlation is not causation.

---

Ok, so you are saying that tobacco products do not cause cancer.

Think about that.

Statistical analysis allows us to justify millions of dollars fined from the innocent tobacco companies.

The ENTIRE justification for "tobacco smokes you" is statistical analysis.

Of course correlation does not equal causation, except when it does. Do seatbelts save lives? Statistically they do. You can try to argue the math, and you can give plenty of arguments like "ice cream causes burglaries" (statistically higher ice cream sales correlates to higher rates of burglary).

Stop arguing "correlation isn't causation". If data correlates then it is either related or random. If it is random then it will NOT show up again in other studies of the same question. If it is related, like ice cream and burglaries, then there may one or more steps removed (rising heat causes people to buy ice cream AND sleep with their windows open, which leads to greater numbers of burglaries of opportunity).

Yes we all know that there can be confounding factors, but the fact remains that the correlation is there, and it is either random or evidence of a true relationship.

No one has said that this is completely random, an easy way to show that it is random is to analyze a different city in the same way and come up with the opposite result.

My guess is that a truer relationship is that the many of the Muslims in the burroughs of London are immigrants, and all immigrant populations have historically had a higher crime rate than the "natives". By analyzing the data based on immigration data instead of religious data might prove my hunch right, or wrong.

You can say something is repugnant, or not politically correct. You can even put on the sad panda hat. None of that deals with the subject, only your inability to deal with the subject.

Jimro

You're trying to argue for a direct link between being Muslim and committing certain kinds of crimes. So far, you have shown a correlation, but have been incapable of dealing with the "confounding factors", as you put it. Your argument falls on its face like a legless cheetah.

Quote:

---

No one has said that this is completely random

---

You're beginning to get mixed up in your mind. You are arguing about randomness? Causality? Or have you somehow managed to conflate the concept of independence with randomness?

Quote:

---

an easy way to show that it is random is to analyze a different city in the same way and come up with the opposite result

---

Go take a statistics class. Run; do not walk. Your brain demands it.

Quote:

---

You can say something is repugnant, or not politically correct.

---

Who says?

I could take an hour and run you through the basic definitions for statistical independence and correlation using some simple point mass functions. Unfortunately, I'm hardly going to get paid for it, and I don't see your education as a particular priority. That and I really should be going. So I'm going to ask you to take a long hard look at what we've already written concerning the inability of the posted article to say anything about the causal relationship you posit. Nobody has shown any inability to deal with the subject here except you; please come back when you are ready to display a firmer grasp of the material.

Executive summary: Stop hand-waving so we can have a real discussion.

*sad panda face*

Pundit, are you married? Are you rich? Will you marry me? Answer the second one first. :3

~Rico

Pundit,

Quote:

---

Go take a statistics class. Run; do not walk. Your brain demands it.

---

I'll try to use small words.

Randomness shows up everywhere. In the medical sciences doctors and researchers consider 5% a baseline for randomness. The better designed a study reduces the R factor by various methods, but that is another matter.

This means that statistical anomalies will show up 5% of the time, that out of any 20 given studies using statistical analysis 1 study will not be consistant with the other 19.

If a study does not have an R factor of 5 or less then meaningful correlation has not been shown.

Have I lost you yet?

By doing another analysis in a different city and population and getting the opposite result that puts the random factor at 50%, both studies are "invalid" until more studies are done to put one of the results below the 5% margin of randomness, and the other answer 95% consistant.

Still with me?

Just like it was "proved" that smoking causes cancer. There are studies that show that smoking is related with a decreased risk of cancer because they are statistical anomalies.

Did I use too many big words?

The guy showed a correlation between Muslims and sex crimes. The "confounding factors" can be ignored if more studies invalidate his results as randomness.

Follow me yet?

You can run yourself, don't walk, to a bookstore and buy "Studying a study, testing a test: How to read the Medical Evidence" by Riegelman.

Jimro

RAmen.

So you show a very weak correlation, a 1.5 degree C rise in 400 years with a decline in pirate population. If you adjust the scales so that they are more relative to each other then the trend line approaches zero rise. Also you might want to adjust the X line so that it doesn't go 35000 to 45000 and then decrease, for the first two points of the graph indicates more pirates equals more temperature.

Altho I do like it as an explanation for "global warming", it makes as much sense as Al Gore.

Jimro

I think the purpose is to show with the right adjustments you can make number show anything you want.

~Rico

Yus.

By implying that you can manipulate numbers this argument is invalid?

Is the tobacco industry a victim of number manipulation?

I know that correlation is either random or related. Pirate population to Global Mean Temperature is a random relationship. I really don't know much about historical pirates but I do have a firm grasp on meteorology.

If you want to disprove a statistical result you do so by showing that the results are invalid, not by saying "I don't agree with the result therefore it must be invalid".

Attacking the method, the same method used to determing efficacy in drugs, is a juvenile argument.

Jimro

Pot. Kettle. Black.

And with at that I take my leave. I have more important matters with which I must attend

~Rico

Why is that turtle looking at you? o.o

It's the beginning of a beautiful friendship!

(teehee)