Regression analysis with holding a certain group of variables fixed

#1
Hi,

I am currently analysing peer-to-peer lending data from Lending Club, but hitting a wall. The dataset consists of data on borrowers and their loans (personal income, dti, amount they wish to borrow, interest rate, where they're from, etc). What I have done is the following:

- Based on where the borrower lives (state), I have created groups based on demographic variables. So if a borrower lives in a state with less than desirable demographics, he/she would belong to group 1 for example, while someone from a good state would belong to group 3.

I want to research if someone from a 'bad' state receives worse loan terms than someone from a good state, having the same personal characteristics (like income, fico score, dti, work experience).

So, my question is:

How do I perform a regression where the dependent variable is for example interest rate, while holding the borrower characteristics fixed across the different groups?

Any help would be greatly appreciated!
 

Karabiner

TS Contributor
#2
If you perform a multiple regression or an analysis of covariance with both personal characteristics
and state as predictors, then you analyse the effect of state while adjusting for personal characteristics.
You can even include interactions between "state" and characteristics ("is the effect of state different
for males versus females", for instance). Adjusting is not "holding fixed", but for me it looks like that it
is what you are essentially looking for.

With kind regards

#Karabiner
 
#3
If you perform a multiple regression or an analysis of covariance with both personal characteristics
and state as predictors, then you analyse the effect of state while adjusting for personal characteristics.
You can even include interactions between "state" and characteristics ("is the effect of state different
for males versus females", for instance). Adjusting is not "holding fixed", but for me it looks like that it
is what you are essentially looking for.

With kind regards

#Karabiner
Thank you so much for your contributions, this was indeed what I was looking for!
 

noetsi

Fortran must die
#4
I think it might be better to do a multilevel model with individuals nested inside state.

Regression will always control for all the variables in the model in theory.