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.

I have data that map one real (float) and several nominal arguments to one real value:

y = f(r, n1, n2, n3, n4)

If I check dependency of y on r (r is my real-valued variable), I see that y depends linearly on r with a lot of noise on top of this dependency.

Now I want to check if there is any dependency of y on nominal variables: n1, n2, n3, n4. I also would like to know what nominal variable has the largest influence on y. So, my question is: What methods can I used to do that?

share|improve this question
    
Can you show us the data (using graphs)? –  Roman Luštrik Feb 27 '13 at 8:52
    
@Roman Luštrik, I am not sure what graph you need. How should I build it? The only graph that I have is y as function of x (it ignores values of other variables). It looks like y linearly depends on x with a lot of noise. The second problem, I think I am not allowed to post the data on line, since it is property of the company that I am working on. –  Roman Feb 27 '13 at 8:59

1 Answer 1

Regression with dummy variables for the levels of the nominal variables (factors) would probably be an obvious first thought, but as @RomanLuštrik points out the first thing to do is plot the data.

An example plot with one linear variable and one factor:

enter image description here

(That data is from this page)

There are various other plots that can be done.

Here's another example:

enter image description here

(Same data, somewhat different kind of plot)

Here is a document that discusses some other plots for this kind of data (like boxplots).

share|improve this answer
    
thank you for the answer. Unfortunately, I do not understand the data. There are four columns which of them is function and which are arguments. If I assume that the last (or fist) columns is the function than I have one real-valued argument and two (not one as you say) nominal-valued variables (or factors). I also do not understand the first plot. It looks like there are four plots. Why four? And what is x and what is y. What is "priorlmp" and what is "Improv"? –  Roman Feb 27 '13 at 10:23
    
Improv is the $y$, PriorImp is the $x$. RDExp is a factor (nominal). The other one is actually a replicate number but I have treated it kind of like it was a factor in the second plot. As for what the data measure, I have no clue. The fourth plot (lower right) combines the two plots on the left. Don't worry about it - just try to comprehend the general idea of doing a separate y vs x for each level of a factor (that's called a coplot). The second plot is called an interaction plot. If you gave some sample data, someone might be able to generate one for you. –  Glen_b Feb 27 '13 at 10:36

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.