Analysis, Lack Thereof

I realized one thing while reading news these days: 

• Gold rises because of inflation worries. 
• Dollar falls because of inflation worries
• Invest in real estate now because stocks are not doing well
• Invest in gold now because stocks are doing well
• Euro falls against yen because of political unrest in northern Africa
• Mid size IT companies in India will go through a consolidation phase because Apple's IPAD has little competition. (Well I am exaggerating a little here)

I mean what kind of headlines are these? Mostly they seem to report data. And some opinions which hardly prove the assertions. Articles which probably shouldn't even find a place in a blog. And we go on reading these? Why don't we get an in depth analysis to justify the statement being made?

When we make an assertion the first thing that is expected is: do we have data to support it? When we say A (some result) happens because of B (some cause), A and B should have a high correlation. 

What does a high correlation between A and B mean?

• When B rises, A rises and vice versa
• The relationship between a change in B and the resultant change in A is stronger than the relationship between C (any another possible cause) and A.

I am reminded of a survey conducted in a women's college and in a men's college. It was found that on an average alumni of the the women's college had 2.7 children each while those from the men's college had 2.1 children each. And a conclusion hence was drawn that women had more children then men.

Now number of children that a person has (let's say, result Z) was compared with with gender of a person (X = women, Y = men). And it was found that Z for X was 2.7 and Z for Y was 2.1. Now what's wrong with this analysis?

The problem is that Z (number of children) does not depend on gender. If the survey had taken a wider sample it would have obviously found that Z is the same for X and Y. (Common sense says that it cannot be different for X and Y). In this case, it was easy to identify the flaw.

But what do you do if the issue is more complex? When common sense may not be able to help?

• Look at larger/wider samples and see if the conclusion reached initially is still valid.
• See how Z correlates with other possible factors P, Q, R etc and see whether Z correlates better with any one or more of them.
• A good correlation (Cause and effect) has to satisfy the rules mentioned in bold and italics above.

I wish the analyses in the articles would do this before reaching a hasty conclusion.