Top Page | Upper Page | Contents | About This Site | JAPANESE

Making Hypothesis by Correlation

"Correlation relationship and cause-effect rerationship are diiffrent" is basic konwledge of data analysis. But correlation is a strong hint to find cause-effect.

Dot Graph and Correlation , Correlation of Category Data and Asymmetry of Quantity Data are used.

To Find the Relationship between Cause and Effect by Correlation

If you find the correlation, there are two cases. One is
(1) X and Y are cause-effect.
The other is
(2) X and Y have the same cause.

If you are not a specialist of the phenomena, it is difficult to do the next stage of the analysis. Communication and cooperation with specialists is important.

In my experience, it is difficult to tell the probability of the case (2) for the specialists.


Statistical models are used as the method of screening variables that we should analyze deeply.

In this page, "statistical models" are that in multi-variable analysis and data mining. For example, in multi-regression analysis,
Y = A1 * X1 + A2 * X2 + EEE + An * Xn + B
is the model.

Y is used as the "effect" data. X is used as "maybe it is a cause" or "maybe effect of another cause".

It is not good to pick-up only one variable because we do not know why the function of selection of variance selects the variables.

If you use some models and analyze strong correlation variables, you may find some characters of the data. It is the hint to solve the problem.

Analysis by Graphs

After screening, it is easier to analyze the relationship among variables by graphs.

If you analyze the data visually, you also find some characters of the data.

For Analysis of Mixed Cause-Effect

For simple structure of cause-effect, Multi-Regression Analysis and Pattern Recognition are the useful tools.

For complicated structure, Correlation Analysis for Multi-Variable , Bayesian Network , Associations Analysis and Principal Component Analysis are the useful

Some tools make arrow graphs. We need to know that the arrow mean the structure of the data, and does not mean cause-effect relationship. arrows

To Find the Relationship between Cause and Effect by No Correlation

Basic way to use correlation is that we use the correlation as the proof of cause-effect. We need to find storng correlation for thie approach.

By the way, if we know the fact of strong correlation by the mechanism of the phenomena, but the data tells us there is not strong correlation, the data means the abnormal situation. And it could be a hint to study the cause-effect relationship.

NEXT Search for hidden variables