Home › Forums › General Forums › Tools & Templates › R Square Issue
This topic contains 5 replies, has 4 voices, and was last updated by Jim Frost 1 week, 6 days ago.
Sir, I run regression. I face a problem that the value of R^2 is very low. How Can I improve it.
Hello @bali,
R^2 is an indicator for how strong a relationship could be. A low R^2 value could mean that the relationship in question may not be very strong.
However, I really need more information from you in order to be of any help. To start with;
1. Do you have the data from your response and predictors?
2. If not, do you have a graphical output to show your result? (I need to see either data or results)
3. What is your R^2 value?
4. What is your P value?
5. What type of industry do you work in?
Regards,
James
Reduce spread of the data is the short answer.
This can be accomplished by things such as….
A. Reducing measurement system error
B. Finding more true causes (X’s) impacting your Y with cause and effect
sir I collect the data from farmers to find the relationship of inputs and output in term of energy. My topic is the energy use and rice productivity. I used Cobb-Douglas production function for my study.
The R^2 value was 0.14925.
So, you took some data plugged it into a regression program and hit run and got a low R-square. I’d recommend backing up and starting over.
First – plot the data. Plot it in any way that makes sense and look at the result. R-square can be viewed as a measure of the strength of a relationship but it is easily fooled and you should NEVER assess a regression analysis on the basis of that one statistic.
Second – plot the data – what makes you think you are going to have a linear relationship between the inputs and outputs? If there is curvature in the data when plotted against one or more of your predictors and all you have are linear terms then, of course R-square is going to be low.
Third – plot the data – If you have influential data points then they could easily influence and even destroy any trending pattern that might exist.
Then there is the issue of predictor co-linearity – but we’ll wait to talk about that after you have really taken the time to plot your data and look at the results.
R-squared seems like a simple measure that is intuitive to understand. However, I’d argue that it is neither. A low R-squared isn’t necessarily bad. It might reflect a large amount of variability that is inherently unexplainable. Or, you might be able to increase it. There’s a number of things you need to check.
Jim
© Copyright iSixSigma 2000-2017. User Agreement. Any reproduction or other use of content without the express written consent of iSixSigma is prohibited. More »