Take the 2-minute tour ×
Cross Validated is a question and answer site for people interested in statistics, machine learning, data analysis, data mining, and data visualization. It's 100% free, no registration required.

Background Context:

I am amazed at the complex statistical questions people ask on this site and I have learnt a lot from them. Thank you all!

However, I have focused on correlation because it is the easist method to understand for a person like me who has no formal training in statistics but is required to do market research as part of my job. A lot of my projects seek to test the relationship between two variables, so Pearson's r is the one I use.

I know each statistical method has a specific purpose and requires certain basis assumptions to be met to function well.

Questions:

  • Can correlation (in particular, Pearson's r) be an end in itself?
  • Given all the advancements in statistical testing, is it still a robust method for data analysis?
share|improve this question

4 Answers 4

up vote 5 down vote accepted

It is true, as @lejohn said, that if all you have is a hammer, everything looks like a nail.

It is also true, though, that if all you have is a nail, then you might only need a hammer!

The thing to do is to define your substantive question, whether it be from market research, psychology, physics or whatever. Then investigate methods, probably not on your own. The method to solve your problem MIGHT be correlation. It might be something else that is very simple. But it might not.

share|improve this answer

For me the answer to your question is no.

I do not think that a given method or technique can be an end in itself. If you have some data at hand or even before you start collecting data, you should ask yourself what are the problems you want to tackle, or what are the questions you would like to answer. When you have the data and the question you can start thinking about techniques.

If you are focussing on one particular technique you are likely to hide a lot from yourself. In that case, the questions you will answer or would like to answer are very much conditioned by the technique you intend to use. Think a lot about the following proverb:

If all you have is a hammer, everything looks like a nail

In addition I would like to note the following. There are indeed a lot of fancy techniques out there. However, the mere fact of using a fancy and sophisticated technique does not validate a statistical analysis by itself. Or to phrase it differently, a fancy technique is no subsitute for a convincing empirical strategy.

share|improve this answer
    
Thanks lejohn. It makes sense! Let's see what others have to say... –  Adhesh Josh Sep 30 '11 at 8:26
3  
+1 Every bit of this excellent advice applies to the sequence of questions you have been asking on this site, @Adhesh. I would prefer, though, that one remark had been made with slightly different emphasis: the time to start thinking about analysis techniques is before you collect data, not afterwards (which for many people is too late). –  whuber Sep 30 '11 at 14:41
3  
"Hiring a statistician after the data have been collected is like hiring a physician when your patient is in the morgue. He may be able to say what went wrong, but he is unlikely to be able to fix it" = George Box (I believe) –  Peter Flom Sep 30 '11 at 19:38

Plotting you data should never be overlooked. In this example, the correlation between X and Y is 0, but surely the two variable are related.

 |  x                         x
 |   x                      x 
 |    x                    x   
 |     x                  x 
Y|      x               x
 |        x           x
 |          x       x
 |            x x x
 |_______________________________
               X

share|improve this answer
    
Indeed. See also the other parts of the Anscombe quartet en.wikipedia.org/wiki/Anscombe%27s_quartet –  Peter Flom Sep 30 '11 at 19:41

I agree with the others who have posted in this thread but have one point to add: Simple methods -- like correlation -- are more justifiable when you can be more sure that there aren't more complicated things going on.

In experimental work where you have used randomization to take into account potentially complicating observed and unobserved variables, you're get much more mileage out of a simple correlation than you will out of observational data where you've got all kind of complicated interplay between the variables you've measured and the variables you haven't. I'm not sure what kind of market research data you have. If it's observational, I'd be even more worried about relying only on correlations.

That said, simple methods like correlation play a very important role. What I try to do in most of my work is start out with simple relationships like correlations, t-tests, and chi-squared tests that establish the relationship in terms that people are more likely to understand intuitively. Only after I've built a substantive understanding, I'll present more complicated models. At that point, it's about addressing threats and creating better estimates than it is about making the strong substantive point.

share|improve this answer
    
Welcome to our site, Benjamin! Thanks for weighing in with your thoughts. –  whuber Sep 30 '11 at 16:11

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.