Corelation and regression
Six Sigma – iSixSigma › Forums › Old Forums › General › Corelation and regression
 This topic has 4 replies, 5 voices, and was last updated 16 years, 7 months ago by Bill Craig.

AuthorPosts

October 4, 2005 at 9:14 am #40908
I am working on six sigma project on reduction of chemical consumption (Say Z)
There exist strong correlation between Chemical Z and other three variable.
Should I go for regression analysis when there exist correlaton
The values are r = 0.294 and p value = 0.000
dd0October 4, 2005 at 11:49 am #127818
BuridanoParticipant@Buridano Include @Buridano in your post and this person will
be notified via email.Not done yet ?
if you have three Xs ( predictors ) than you should have more than one correlation coefficient.
what about scatter plots ?
0October 4, 2005 at 12:31 pm #127824
Robert ButlerParticipant@rbutler Include @rbutler in your post and this person will
be notified via email.As written, your missive gives the impression that all you have done is dump a few numbers in a machine and pressed a button. If true, then you need to realize you are only doing regression – what you want to do is regression analysis. To do this you need to take a step back and first really look at your data. The quickest way to do this is plotting – lots of plotting.
1. What do the plots of Z vs. the three variables look like? Linear, curvilinear, matzo ball and pencil dot, shotgun blast, etc.?
2. What do the plots of the three variables against each other look like? Shotgun blast, linear trend, etc.
3. Do the plots of the 3 X’s against one another suggest that the model search might be able to support an examination of interactions? If so, based on your knowledge of the process, would it make physical sense to include them.
4. Do you have time ordered data? If so, plot the Z against time. Do you know where in that time interval changes were made in one or more of the X’s? If so, mark the locations and values on the plot? What do you see?
5. Plot the data in any way that makes sense to you and see what you can see.
Once you have really looked at your data and seen what, if any, trends are present you will be in a position to start thinking about possible terms for a regression analysis and when you ask the machine to run a regression for you it won’t be a blind regression – it will be a regression based on reason and thought.
0October 15, 2005 at 2:41 pm #128372
Pawan VermaParticipant@PawanVerma Include @PawanVerma in your post and this person will
be notified via email.Dear sir
I dont think so that u should go for the regression model because ur r shows that u have negative correlation bt r^2 is only 8.6% that shows that only approx 9 % causes are identifiable in variation in ur Chemical consumption…..Do following things
1) Choose some other factors which r causing variation in Consumption
2) Remove outliers
3) Check for the correlation between the independent variables ( Multicollinearity)
Get back 2 me for further queries.
Pawan Verma
Doing Post Graduation in Industrial Engg.0October 15, 2005 at 5:08 pm #128373
Bill CraigParticipant@BillCraig Include @BillCraig in your post and this person will
be notified via email.DD,
Looks like the rvalue doesn’t support your claim of a strong correlation.
If your X factors show correlation with each other, those with High VIFs should be evaluated and some should be dropped from the model (Multicollinearity). Kind of like using relative humidity and evaporation rate as two of your X factors. (one is dependent on the other).
If I am not mistaken, you should not have several correlation coeficients (Only one for your model), but you should have a pvalue for each term in your model.
I am pretty familiar with multiple regression using JMP and Minitab. I can offer more advice if you let me know which you are using. (If either!)
Good luck!0 
AuthorPosts
The forum ‘General’ is closed to new topics and replies.