0

I am building a logistic model with about 20 variables. I have used the following code:

`fullmod = glm(cancer ~ B_SEX+BLINE_AGE_AT_BASELINE+B_BMI+B_chro+B_fdrc+B_hrt+B_LMET+ B_MET+B_FV+B_EDU+B_INC+B_MAR+B_EMP+B_sm_status+B_sm_y+CDHQ1_ALCOHOL_MYP+CDHQ1_HEI_2005_DERV+red.meat.w.g+red.meat_cat+location,family=binomial,data=all)

Summary(fullmod) ` I got the following output:

enter image description here

I have investigated the correlation structure/linear dependencies of independent variables , found this: enter image description here

So, Sex is correlated with the intercept, I found some explanation regarding this situation here enter link description here

But what is the solution, how I can remove this singularity

Fahmida Yeasmin
  • 17
  • 2
  • 1
    The link you included is unrelated. That's about the coefficients being correlated, not the predictors. Are you positive `B_SEX` has more than one value in it? This would occur if all values were the same. Are there any missing values in the predictors? – Noah Nov 28 '20 at 19:30
  • Yes, B_Sex has two values 1=Female with 34436 entry, 0=Male with 19493. and Yes, Other predictors have missing values, but not in the B_SEX. – Fahmida Yeasmin Nov 28 '20 at 20:47
  • 1
    If you run `with(na.omit(all), table(B_SEX))`, are there still two unique values? When missing data are present, all rows with missing data are removed by `glm()`, so the dataset that `glm()` fits the model on may not have the same levels as the data you submitted. – Noah Nov 28 '20 at 23:54
  • Thanks a lot. I got it now, after running your code, I got only one value in B_SEX, which is 1=Female. Then I have tried na.action=na.pass code, but it gives me the following error: "Error in glm.fit(x = c(1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, : NA/NaN/Inf in 'x' ". apperantly "glm" will not use rows containing NAs. do you have any suggestion? – Fahmida Yeasmin Nov 29 '20 at 02:11
  • 1
    `glm()` cannot include missing values in the predictors or outcome. Search on this site for how to deal with missing values in regression. It's a big topic and the best answer depends on your purpose. – Noah Nov 29 '20 at 22:37

0 Answers0