turn on suggestions

Auto-suggest helps you quickly narrow down your search results by suggesting possible matches as you type.

Showing results for

Find a Community

- Home
- /
- Analytics
- /
- Stat Procs
- /
- Regression analysis suggestion

Topic Options

- Subscribe to RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Printer Friendly Page

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Email to a Friend
- Report Inappropriate Content

09-11-2010 04:48 AM

Hi all,

I am currently running regression analysis to find out Impact of Each indep variable on dep variable

say,

y = b1x1 + b2x2 + b3x3 + b4x4 ....

The output that I seek is var x1 accounts for 20% , var x2 accounts for 30% impact. to calculate this I am using standardized coefficients.

However, I see that there are a lot of variables that have**correlation amongst them.**

I know factor analysis and principle analysis are methods of dealing with the problem of correlation .. however, in the end I want to see the impact in terms of percentage .. Like var x1 has 20% var x2 has 12% and so on ..

** what correlation would be considered as high .. like above 0.7 or above 0.8 or some number ... **

It would be helpful if you guys could suggest me some method to this ..

Thank you,

Rockerd.

I am currently running regression analysis to find out Impact of Each indep variable on dep variable

say,

y = b1x1 + b2x2 + b3x3 + b4x4 ....

The output that I seek is var x1 accounts for 20% , var x2 accounts for 30% impact. to calculate this I am using standardized coefficients.

However, I see that there are a lot of variables that have

I know factor analysis and principle analysis are methods of dealing with the problem of correlation .. however, in the end I want to see the impact in terms of percentage .. Like var x1 has 20% var x2 has 12% and so on ..

It would be helpful if you guys could suggest me some method to this ..

Thank you,

Rockerd.

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Email to a Friend
- Report Inappropriate Content

09-13-2010 09:08 AM

The only time statements like "var x1 has 20% var x2 has 12%" have meaning is when the variables are not correlated with one another.

As soon as you have correlations, statements of the sort "var x1 has 20% var x2 has 12%" are meaningless.

I suggest you adopt a different mindset. You can still determine which*combinations *of variables are important in the prediction equation; you cannot determine which individual variable is causing the impact via statistics alone, nor can you quantify the percent impact of an individual variable.

One technique that allows you to determine the combinations of variables that are important in the predicting a response(s) is Partial Least Squares regression.

As soon as you have correlations, statements of the sort "var x1 has 20% var x2 has 12%" are meaningless.

I suggest you adopt a different mindset. You can still determine which

One technique that allows you to determine the combinations of variables that are important in the predicting a response(s) is Partial Least Squares regression.