Linear, step wise regression vs binary logistic regression - which method to choose

#1
Dear friends
I have a dilemma about choosing a right logistical method.

i want to find predictor factors of life satisfaction and have 5 covariats.
My dependent variable - life satisfaction is on nominal level. 0 - satisfied, 1 not satisfied
Excoriates vary from ordinal to nominal levels.

So which method should i choose? I had a binary logistic regression but i don't think now it is an appropriate method.

Thank you for the advice
 

hlsmith

Omega Contributor
#2
Re: Linear, step wise regression vs binary logistic regression - which method to choo

Why don't you think it is appropriate?
 
#3
Re: Linear, step wise regression vs binary logistic regression - which method to choo

because when i change the direction of the dependent viable 0 not satisfied, 1 satisfied (i have now 0 - satisfied, 1 not satisfied ) i get very different OR and 95 % CI.
 
#4
Re: Linear, step wise regression vs binary logistic regression - which method to choo

Of course it's going to be different, as you are now working out the OR of being 'not satisfied' rather than 'satisfied'.
 
#5
Re: Linear, step wise regression vs binary logistic regression - which method to choo

Of course it's going to be different, as you are now working out the OR of being 'not satisfied' rather than 'satisfied'.
it is very different. for example, if for not satisfied OR for gender is 0.06, on satisfied version i get OR 4,58 on THE SAME variable.
 

maartenbuis

TS Contributor
#6
Re: Linear, step wise regression vs binary logistic regression - which method to choo

They should be very different. Think of it this way: if the odds of being happy for women is twice the odds of being happy for men, then the odds of being unhappy for women should be half the odds of being unhappy for men. So the relationship should that the OR in the happy version is 1/OR in the unhappy version, and vice versa. The fact that the ORs on the same variable are very different depending on how you code your dependent variable is normal.

However, there is a big problem with your data: [MATH]\frac{1}{.06} \approx 16.7 \neq 4.58[/MATH]. I suspect that when reverse coding your sattisfaction variable missing values were mistakenly turned into observed values, or some similar mistake.
 
#7
Re: Linear, step wise regression vs binary logistic regression - which method to choo

They should be very different. Think of it this way: if the odds of being happy for women is twice the odds of being happy for men, then the odds of being unhappy for women should be half the odds of being unhappy for men. So the relationship should that the OR in the happy version is 1/OR in the unhappy version, and vice versa. The fact that the ORs on the same variable are very different depending on how you code your dependent variable is normal.

However, there is a big problem with your data: [MATH]\frac{1}{.06} \approx 16.7 \neq 4.58[/MATH]. I suspect that when reverse coding your sattisfaction variable missing values were mistakenly turned into observed values, or some similar mistake.

Thank you sooo much!! I will see it now
 
Last edited:
#8
Re: Linear, step wise regression vs binary logistic regression - which method to choo

They should be very different. Think of it this way: if the odds of being happy for women is twice the odds of being happy for men, then the odds of being unhappy for women should be half the odds of being unhappy for men. So the relationship should that the OR in the happy version is 1/OR in the unhappy version, and vice versa. The fact that the ORs on the same variable are very different depending on how you code your dependent variable is normal.

However, there is a big problem with your data: [MATH]\frac{1}{.06} \approx 16.7 \neq 4.58[/MATH]. I suspect that when reverse coding your sattisfaction variable missing values were mistakenly turned into observed values, or some similar mistake.
just a question
Would it be the same way if i have an ordinal variable (scale 1-5)