quadratic model ?

[caremiriam]

I'm using regression analysis.
I plotted variable A on Y axis, variable B on X axis. After trying a linear model, then i tried a quadratic model, and then a cubic model. Quadratic model works best. It's like a U shape.
However, out of some funny reasons, I'm told to do the regression in the opposite way in order to predict B from A.
so i did the same procedures, while the data looks like a long C shape, this time, linear model works best.
Can anyone tell me why this is the case?
Thanks.

[Dr N S Gandhi Prasad]

Hello
This is something undesireable.Such attempts lead to a saying " Statistics can prove any thing". Before undertaking any regresion analysis, one should be aware of cause and effect relations. In the regression analysis, one makes a hypotheis that a Vraiable ( Dependent variable say Y Crop Productivity )is influnced by an independent variable ( say X Fertilizer ) apart from random compnent. Such hypotheses can not be reversed.

If such cause and effect relationship between X and Y can not be established, it is better to go for simple correlation to assess the degree of mutual association between variables.
Good Luck
Dr N S Gandhi Prasad
Professor of Statistics ( Retd. )
Hyderabad India