is there a non parametric alternative to multiple linear regression?

#1
Hi all,

I would like to know is there a non parametric alternative to multiple linear regression?

when I google the quesion I get a lot of info but it doesn't have a very clear answer. So I wondered if there is a test that would be typically used in these circumstances.

Also what should you use if some independent variables are not normally distributed but others are?


Best wishes,
David
 

Dason

Ambassador to the humans
#3
Are you just wondering or do you have a specific reason for asking? I'm just wondering because it might be that looking at a different way of modeling the problem can solve some of the issues.
 
#4
Hi,

Thank you for your replies.

At this point I am not sure if I will need to go down the non-parametric route for my analysis but there is a chance I will need to.

I am finishing entering and cleaning my data, soon I will run my analysis.

My study is looking at the needs of caregivers.

My caregiver participants completed a battery of questionnaires, measuring caregiver burden, mental health, physical depednency of cared for person, emotional and cognitive dependency of person cared for, etc.

Each questionnaire has an overall score. I have around 14 of these questionnaires, so 14 continuous variables.

I also have a number of categorical variables (carer sex, care recipient sex, age category, type of illness, etc).

My main concern is identifying predictors of whether or not caregivers use respite care services.

But I might also look at the relationship between the variables and the type of condition the person cared for has (we have collapsed the illness conditions into three categories).

I wil be analysing the data using SPSS / PASW 18. I am not sure if I will be able to use more specialist software if the analysis requires it.

Best wishes,
David
 
#5
If your outcome is whether or not people used care services (i.e., binary Yes/No variable) then it sounds like you should be using logistic regression.
 
#6
Hi,

Yes I have been reading more, and now I think logistic regression for the use or non-use of respite care, and perhaps a linear regression for predicting carer mental health.

But do I need the independent variables to be normally distributed, etc, to use them in a logistic regression?

If so is there a non-parametric equivalent of logistic regression?


Best wishes,
David
 
#7
There is no non-parametric form of any regression.

Regression means you are assuming that a particular parameterized model generated your data, and trying to find the parameters. Non-parametric tests are test that make no assumptions about the model that generated your data. Those two assumptions are incompatible.

It's certainly good for you to think about the possibility that the real process that generated your data is not linear or not logistic, but there is no statistical procedure that will discover the parameters of other possible models without any assumptions about what the model space is.
 
Last edited:

Dason

Ambassador to the humans
#8
There is no non-parametric form of any regression.

Regression means you are assuming that a particular parameterized model generated your data, and trying to find the parameters. Non-parametric tests are test that make no assumptions about the model that generated your data. Those two assumptions are incompatible.

It's certainly good for you to think about the possibility that the real process that generated your data is not linear or not logistic, but there is no statistical procedure that will discover the parameters of other possible models without any assumptions about what the model space is.
One could consider a smoother a nonparametric regression. It gives an idea about the mean with respect to the covariate but you don't need to make distributional assumptions for most.