When in doubt, pray

A new University of Florida study indicates that teens who consider themselves very religious are more likely to finish college than those who don't.

But no one can say why.

"For most religious communities represented in our study, there is a strong correlation between religiosity and degree attainment," Ana Puig, research director and affiliate faculty member in counselor education at UF's College of Education, said in a news release. "However, correlation does not mean causality."

The effect was most prominent among Muslim students, but found to be non-existent in religious groups that generally have high education attainment anyway, such as Jewish, Episcopalian and "Eastern religion" students.

The point? "Students and parents are saying that religion is an important part of their academic lives, and we need to listen to that," researcher Mary Ann Clark said.

Comments

[patcon]

Joseph, I think we may not be as far apart as you seem to think.
My only real issue with your post was the focus on "A->B or B->A or B->C->A, or any other combination" seemed to leave out the (very real, and I think likely in this case) situation where A and B are both effects of a common root cause.
So while religiosity and degree attainment may be strongly correlated, I suspect that, rather than one causing the other either directly or indirectly, there is an exogenous factor (or several exogenous factors) which directly and indirectly causes both.
No, I didn't read the press release, as I didn't deem it a study of much importance. It's not as if it could/would be used to guide public policy. It seemed to me to be a rather pointless and frivolous study.

[Joseph]

Dear patcon,

"Even those who "understand" often are incorrect."

I assume that was directed at me.

I am ABD in applied research methods, so I think that I know what I am talking about with respect to statistics.

Did you bother to read the press release? For anyone with even limited knowledge of research methods and statistics would question why this study was even published in the first place.

I just don’t understand why my addendum “correlation implies causation” to the original “correlation is not causation” statement is such a problem for you.

I did NOT say that “correlation is not causation” is a false premise. Read my damn posts.

I DID say that if one finds a significant correlation, then this is an indication THAT a causal relationship MIGHT be present. In other words, this is a call for further research!

I did NOT say that the “indication of a causal relationship” was limited to the two variables in question.

Words like “link,” “connection,” “association,” and so forth IMPLY the presence of a causal relationship, although the specifics are not known.

I don’t understand why the authors of this study would state that they found a significant correlation and then turn around and state that their finding has “no teeth.” They should explain their results and their conclusions, based on theory, even in the press release!

So sure, tell me I missed a thousand endogenous variables in my example. I thought “…and so forth” took care of that, I guess not. Even in your example, a significant correlation between teenage pregnancy and smoking suggests that these two consequences are somehow causally related. Notice how the word “somehow” includes both directions and any chain of endogenous variables that you would like to include!

So please tell me, in your opinion, how in the f@@king world does correlation NOT IMPLY causation?

[patcon]

Funny Arthur, but you've made two mistakes.
1) It's logic class, not statistics.
2) You have fallen prey to the fallacy of the undistributed middle.
"God is love, love is blind, therefore God is blind," is a valid syllogism.
"Men are mammals, dogs are mammals, therefore men are dogs," is not a valid syllogism, even if ...

[Oscar]

A. God is Love
B. Love is blind A=B
C. Ray Charles is blind B=C
Then Ray Charles is God A=C

I love stats class

[patcon]

To all the "understanders" and "misunderstanders" of statistics, "correlation does not mean causation" is important to say in the popular press reporting of such a story because of the misunderstanding/misuse of statistics by the general public.
Even those who "understand" often are incorrect. For example, if a study shows that teenage smoking and teenage pregnancy have a high correlation, it simply means there is a link between the two conditions. It may be that (1) smoking causes pregnancy, or (2) pregnancy causes smoking, or (3) smoking and pregnancy share a common root cause.
So Joseph, you left out another important possibility for the chain of causation when A and B are strongly correlated. It could also be:
E causes C and D.
C causes A, and D causes B.
But A does not cause B and B does not cause A.
Yes, a good study wouldn't settle for simply observing a correlation without looking for a root cause. A report in the popular press isn't a good indicator of what a study actually reported - the press takes what it wants to report and leaves out the "boring stuff". I don't have the desire to examine the original study, so I will neither criticize nor defend this particular study.

[Joseph]

Agreed Drew. But folks, as far as I can tell, don't run correlations to establish causation or because they like SPSS ver. 16! I personally cannot believe that the authors conducted a study of this magnitude and walked away touting a correlational analysis. Who cares? Very weak.

My point and my antagonistic attitude toward the authors is that it makes no sense to restate the over-stated fact that correlation is not causation.

What purpose does it serve? Just shut-up.

Why do an empirical study if you are not looking for cause and effect relationships?

Whether it is:

A--->B or

B--->A or

B--->C--->A

Or any other combination, a significant correlation coefficient IMPLIES causation in some direction and some strength.

In other words, correlation directs our attention to causation, but it does not specify which variable is exogenous or endogenous.

The "academic" authors should have had better sense.

[howie]

Learn how to pray for blessings in the form of serious cash from whatever you call your Supreme Being because college, student loans, the interest on those loans, the current economic downturn, and some SERIOUS inflation are gonna make times very tough. And if you aren't a believer during college, you may well see the light once you graduate and are faced with loan payments so high that you won't be able to start a family until you are forty.

[Drew Smith]

When we say that correlation does not imply causality, we're referring to the idea that correlation *between* two variables (A and B) does not automatically allow us to conclude that "A causes B". It may *instead* be the case that B causes A, or it may *instead* be that C causes A and C causes B.

[Chris White]

Most of the very religious people I knew in college rarely spent late nights partying and were often much more disciplined than their secular classmates, myself included.

[Joseph]

Pete,

Go back to statistics class.

Whether or not a relationship between two variables may or may not be considered spurious does not change the fact that a significant correlation implies causation.

For example, if I have a r between race dummied as white and non-white and education of .80, and then, when I control for mother’s education, the r drops to .3, this in no way indicates that there is not a causal relationship between the two original variables.

Thus, race is still an implied “cause” of education; but I have only further specified the relationship by suggesting that it may operate through mother’s education.

The case that you speak of, when a control variable completely reverses the direction of the original hypothesized relationship is rare and it should not be used as an example to critique correlation. When this situation does occur, one should first check his/her data looking at the operationalization of variables.

Having said all of this, this study is junk, and it is not surprising to me that a couple of education professors are at the root of it. Something like this would never fly in a reputable psychology department.

Furthermore, there are many tools that can be used to establish casual relationships between variables, such as multiple linear regression, quadric regression, hierarchal linear modeling, and, my favorite, structural equation modeling (SEM).

This is nothing more than a descriptive study, and a waste of our time. Besides that, with 11,000+ cases, SEM would have been terrific! Moreover, with a sample of this size, one is bound to find significant results for basic statistics, like correlation and t-tests, and the researchers, if they were worth their salt, would have presented theory to account for which ones they considered legitimate.

So, in my final analysis, in the Stone Age, this would have been great research; but today, it is nothing more than crap on paper.

[Joseph]

Pete,

Go back to statistics class.

Whether or not a relationship between two variables may or may not be considered spurious does not change the fact that a significant correlation implies causation.

For example, if I have a r between race dummied as white and non-white and education of .80, and then, when I control for mother’s education, the r drops to .3, this in no way indicates that there is not a causal relationship between the two original variables.

Thus, race is still an implied “cause” of education; but I have only further specified the relationship by suggesting that it may operate through mother’s education.

The case that you speak of, when a control variable completely reverses the direction of the original hypothesized relationship is rare and it should not be used as an example to critique correlation. When this situation does occur, one should first check his/her data looking at the operationalization of variables.

Having said all of this, this study is junk, and it is not surprising to me that a couple of education professors are at the root of it. Something like this would never fly in a reputable psychology department.

Furthermore, there are many tools that can be used to establish casual relationships between variables, such as multiple linear regression, quadric regression, hierarchal linear modeling, and, my favorite, structural equation modeling (SEM).

This is nothing more than a descriptive study, and a waste of our time. Besides that, with 11,000+ cases, SEM would have been terrific! Moreover, with a sample of this size, one is bound to find significant results for basic statistics, like correlation and t-tests, and the researchers, if they were worth their salt, would have presented theory to account for which ones they considered legitimate.

So, in my final analysis, in the Stone Age, this would have been great research; but today, it is nothing more than crap on paper.

[mlm]

As a recent UF grad, this actually makes sense, even if it's not just UF kids polled. Nearly every church/temple/mosque in Gainesville is within walking distance of campus and all have support systems for students. As for "party habits", everyone parties in college, to some degree.

[Pete Smith]

In a significant amount of cases correlation does not imply causality-often a third variable is involved or the causation works in an opposite direction from what is assumed.

Perhaps the "very religious" are less likely to fall victim to the party habits of many.

[Joseph]

In order to set the record straight, yes, it is true that "correlation doesn’t mean causality"; but it sure as hell implies causality.

"A" and "B" are (somehow) linked. Correlation between "A" and "B" tells us that if "A" occurs, then in a significant amount of cases "B" is soon to follow.

That is the definition of cause and effect!