I have to perform some tests on a dataset and I was thinking

I have to perform some tests on a dataset and I was thinking of using linear regression. After I ran the regression test with some of the predictors I looked at the plot of the residuals and it seem like the residual plot line is not normal. So that concerned me so I went back and checked the plot of the independent predictor variables to see if they are linearly related to the dependent variable and it doesn\'t appear they are. So I\'m not sure if I can even use linear regression to look for predictors.

My regression model is looking for the predictors for diabetes based on weight,cholesteral,age, height, BMI. The dependent variable is the glyhb of the subject. Anything above 7 for glyhb is considered diabetic. Figured linear regression would be a perfect fit for this, but it doesn\'t seem so. Am I allowed to use linear regression here? If I can\'t use plain linear regression can I treat the dependent variable as a simple catagorical variable of 1,0(positive or negative for diabetes) and use logisitcal regression here? I just need to be pointed in the right direction. Thanks.

This is the plot of the residuals. It doesn\'t appear to be normal and the Shapiro-Wilk test says it isn\'t normal.

Here is the output from the regression model:

Solution

In the linear regression our assumtion is that the distribution of error is normal. But when we are fitted the data and take residuals (observed - predicted) it may not be the normal variable as it is predicted value.

more over in the R-console display it shows that at 0.05% level of significance mainTable$age and mainTable$stan.glu are rejected that means there is effects of independent variable on dependent variable. Hence ypu can procced with Linear regression.