# Combined Effect of Independent Variables

Six Sigma – iSixSigma › Forums › General Forums › Tools & Templates › Combined Effect of Independent Variables

- This topic has 4 replies, 5 voices, and was last updated 4 years, 7 months ago by Robert Butler.

- AuthorPosts
- June 8, 2015 at 4:34 am #55049

prasenjit ghosalParticipant@prasenjit**Include @prasenjit in your post and this person will**

be notified via email.dear members,

i have one dependent variable(heart diseases) and four independent variable(liquor consumption,smoking,high blood pressure and excessive sun exposure).i can find the correlation between one dependent and four independent variable separately, but now i want to establish a relation among them like i want to what is the combined effect of all four independent variable , e.g. if a person having high bp and consume liquor with smoking and having excess sun exposure what is the chance of having heart diseases of such person after considering all four variable together.now my question is i have four data set on 500 people survey having 1.heart diseases-liquor consumption, 2.heart diseases-smoking,3.heart diseases-high bp,4.heart diseases-sun exposure.now using minitab 16 how to find the combined effect all for independent variable on having heart diseases.0June 8, 2015 at 5:38 am #198381

Amit Kumar OjhaParticipant@AmitOjha**Include @AmitOjha in your post and this person will**

be notified via email.Dear Prasenjit,

In order to understand and model the combined effect of multiple independent variables on one dependent variable, you need to perform multiple regression. There are basically two approaches i.e Forward Stepwise and Backward Stepwise.

Both these approached are available in Minitab 16.

You use Forward Stepwise when you start with one most important variable and continue adding others, whereas you use Backward Stepwise Elimination when you start with all the variables and proceed with eliminating one at each step.

Hope it helps…

All the best!!!0June 8, 2015 at 5:40 am #198382Since it sounds like you have a binary response (has heart disease or not), I think you need to start with Binary Logistic Regression. In Minitab that is Stat>>Binary Logistic Regression>>Fit Binary Logistic Model.

Response would be the column containing your Heart Disease data. Then, enter in your Continuous and Categorical predictors.

Minitab Help can guide you on analyzing the results. But, this analysis will show you the combined effect of all the variables. However, I would think you may have interactions among the variables as well. For example, maybe liquor consumption and smoking together have a greater effect then either of them independently? I’d suggesting doing some analysis of the interactions as well.

Hope this helps get you started…. – Good luck!

0June 8, 2015 at 6:54 am #198383

Chris SeiderParticipant@cseider**Include @cseider in your post and this person will**

be notified via email.Good advice @JRBGuy

0June 8, 2015 at 11:35 am #198388

Robert ButlerParticipant@rbutler**Include @rbutler in your post and this person will**

be notified via email.As has been noted – heart disease yes/no is a binary variable so the regression method of choice would be logistic. Before you go there you will have to check to make sure you can actually use the 4 variables of interest in the regression equation. To that end you will need to run co-linearity checks (VIF and condition indices) to make sure the variables of interest are independent enough of one another to allow you to make a statement concerning the odds of heart disease (which is what you will get from a logistic regression) as a function of the independent variables of interest.

You will also have to check for quasi-separation – actually the machine should do this for you and if it exists it will tell you either the model fit is questionable or it will bomb completely and not provide any answer.

I can’t emphasize enough the issue of independence. I deal with medical data on a daily basis and the number of times I’ve seen conflicting claims with respect to significant correlation that were directly attributable to variable confounding are too numerous to mention.

0 - AuthorPosts

You must be logged in to reply to this topic.