# How to find the association/relationship between two continuous variables?

Hello,

The attached image is the scatter plot of income and age. I understand linear and polynomial relations, but in the above case what is the right way or approach to find how age and income are related. I found the correlation between the two but it is pretty low (0.2).

Also, how to fit a curve in this case? Any help is much appreciated.

Regards

• It's pretty clear those dots at 50,000 aren't the real monthly income: presumably, your data are censored at this value. There are also a visible preponderance of incomes at 40,000, 35,000, and 30,000, suggesting that some appreciable proportion of the incomes have been rounded heavily. These issues can influence the choice of regression methods. The dot with an age of 0 and the relatively large number in the 90s and 100s suggest there may be issues with how the ages are recorded, too. What is the purpose of exploring the age-income relationship? – whuber Feb 29 '16 at 21:00
• Yes, you are right regarding the 50,000 part. I capped the data @ 3-sigma since the there were observations in the range of 1-10 million in monthly income which were heavily skewing the results. That being said, the rest of the data you see is what i got. I am working on a logistic model and age & income are predictors. I am trying to find if both the variables should be included or not. – Raj Feb 29 '16 at 21:28
• That's a completely different question! You should be focusing first on the relationships between your predictors and the response rather than on modeling any relationships among the predictors. – whuber Feb 29 '16 at 21:31
• That's true, but the modeling is just a practice exercise. So i am trying to learn everything i can while doing it. – Raj Feb 29 '16 at 22:18
• That makes your question vague, then, because it appears to have no definite purpose. As an example of why this matters, you have supplied no reason to model income as a function of age: you could just as well model age as a function of income or, more generally, to model the joint variation of both. That requires three distinct statistical techniques and will produce three very different results. Exercises are fine things to do, but they must have a point and they need to be as clearly framed as possible. Perhaps you could edit your post to clarify these issues. – whuber Mar 1 '16 at 13:37