A methodology question

#1
Hi,

I have a multiple regression model e.g. y=x1+x2+e
Can I estimate the model by using the model in two ways by using same values of y and x1, but two different sets of values of x2 obtained from two different sources? I want to do it as both sources for x2 are not highly reliable, but that's I what could get. I want to include this methodology in my prospectus. If you know some peer reviewed paper doing it, let me know. Thanks a lot!

tal
 

hlsmith

Omega Contributor
#2
What are you planning on doing with them? Using them as two different predictors, merging them via picking and chosing, or merging them via a calculation? Why don't you describe the general study description and what these variables are in particular, plus whether they are continuous or categorical. This information may help us understand the scenario.
 
#3
Here is an example:
In the year 2011, for all public schools in a state -
Y=average math scores in SAT in the school
x1=average income of parents in the school
x2=whether they have a math club in the school.

I can get y and x1 from a government organization
I can get x2 from two sources - two different surveys conducted by two different private organizations.

Can I estimate the model by using the model in two ways by using same values of y and x1, but two different sets of values of x2 obtained from two different sources? I want to do it as both sources for x2 are not highly reliable, but that's I what could get. I want to include this methodology in my prospectus. If you know some peer reviewed paper doing it, let me know.

Thank you for your response.
 

hlsmith

Omega Contributor
#4
Thanks for the description. I still don't know how you are using the two sources of X2? It still seems from your wording you may enter them as x2 and x3, is that what you want to do? If they both say the same thing and if they were both continuous and incorporated in your model it would inflate your variance, I am sure there is some comparable penalty for including both as independent variables when they are categorical. It seems like you need to test how much they are the same or different. If they are different, it seems flawed to incorportate both. Is there any way to merge them. How do you know they are unreliable (because they are not the same)? If this is your prospectus, can you then propose to contact them all in the future?