Hi,

I am building a regressional model with 13predictors:
A,B,C,D,E,F,G,H,I,J,K,L,M

The Response, Y, is obtained from the ratio A/B.

I would like to know if it would be safe to exclude A and B from my
predictors in my regressional analysis.

In article <1166318377.673423.293830@16g2000cwy.googlegroups.com>,
brakwabena <omanekodie@gmail.com> wrote:
Hi,

I am building a regressional model with 13predictors:
A,B,C,D,E,F,G,H,I,J,K,L,M

The Response, Y, is obtained from the ratio A/B.

I would like to know if it would be safe to exclude A and B from my
predictors in my regressional analysis.



If the response Y is obtained from the ratio A/B, that
should be included. If it is obtained only from A/B,
the rest should be discarded.

Do not build linear regression models just because there
are canned programs for doing them. Build your model from
YOUR probabilistic considerations, in total disregard of
what you know how to calculate. THEN talk to someone who
does know how to calculate, and how to approximate.

If you can wear both hats, fine. But your question is
strong evidence that you cannot.


--
This address is for information only. I do not claim that these views
are those of the Statistics Department or of Purdue University.
Herman Rubin, Department of Statistics, Purdue University
hrubin@stat.purdue.edu Phone: (765)494-6054 FAX: (765)494-0558

Dear Herman,

Thank you for responding to my post.
I did not quite get what you said.
I would appreciate it if you could kindly put it in other words.
To seek further clarification from you, I am rephrasing my earlier
enquiry.

" 11 characteristics (C,D,E,,,,,M) of 120 houses were collected in a
survey after which characteristics A and B were estimated from the data
collected. finally, the ratio of A and B were created.

From the description above, it appears your A may be the asking or
selling

I am building a
regressional model to know how well this RATIO can be predicted given a
few of the characteristics of houses.

Considering that the response is directly calculated from the ratio of
A and B, can I discared A and B......... while I regress the Response
on C,E,D,,,, M in my analysis?"

Thanx for your time; it is deeply appreciated.

Brakwabena.


Herman Rubin wrote:
In article <1166318377.673423.293830@16g2000cwy.googlegroups.com>,
brakwabena <omanekodie@gmail.com> wrote:
Hi,

I am building a regressional model with 13predictors:
A,B,C,D,E,F,G,H,I,J,K,L,M

The Response, Y, is obtained from the ratio A/B.

I would like to know if it would be safe to exclude A and B from my
predictors in my regressional analysis.



If the response Y is obtained from the ratio A/B, that
should be included. If it is obtained only from A/B,
the rest should be discarded.

Do not build linear regression models just because there
are canned programs for doing them. Build your model from
YOUR probabilistic considerations, in total disregard of
what you know how to calculate. THEN talk to someone who
does know how to calculate, and how to approximate.

If you can wear both hats, fine. But your question is
strong evidence that you cannot.


--
This address is for information only. I do not claim that these views
are those of the Statistics Department or of Purdue University.
Herman Rubin, Department of Statistics, Purdue University
hrubin@stat.purdue.edu Phone: (765)494-6054 FAX: (765)494-0558

brakwabena wrote (in response to Herman Rubin):
Dear Herman,

Thank you for responding to my post.
I did not quite get what you said.
I would appreciate it if you could kindly put it in other words.
To seek further clarification from you, I am rephrasing my earlier
enquiry.

" 11 characteristics (C,D,E,,,,,M) of 120 houses were collected in a
survey after which characteristics A and B were estimated from the data
collected. finally, the ratio of A and B were created.

From the description above, it appears your A may be the asking or
selling
price of 120 houses and B the number of square ft. of the house
perhaps,
because those are the most readily avaiable DATA from all real estate
listings of houses.

Your C,D,E, ..., M are other characteristics of the houses, and you are
interested to see whether the Cost/sq.foot is related to the OTHER
characteristics, and how well it is related.


I am building a
regressional model to know how well this RATIO can be predicted given a
few of the characteristics of houses.

Several of my students in the Data Analysis courses had considered
models of this kind. If yours is NOT what I described, it probably is

a variant of that kind of empirical modeling.


Considering that the response is directly calculated from the ratio of
A and B, can I discared A and B......... while I regress the Response
on C,E,D,,,, M in my analysis?"

Certainly! But you may also consider (independently) a model of
A being the Y and include B with the independent variables, because
of the RATIO aspects of the A/B would not generally fit well to the
non-ratio characteristics of the other variables.

Regression A on (B,C,....) is essentially equivalent to regressing
A/B on C/B, D/B, ... which alters the constant and the error
distribution of your model.

-- Reef Fish Bob.

Thanx for your time; it is deeply appreciated.

Brakwabena.


Herman Rubin wrote:
In article <1166318377.673423.293830@16g2000cwy.googlegroups.com>,
brakwabena <omanekodie@gmail.com> wrote:
Hi,

I am building a regressional model with 13predictors:
A,B,C,D,E,F,G,H,I,J,K,L,M

The Response, Y, is obtained from the ratio A/B.

I would like to know if it would be safe to exclude A and B from my
predictors in my regressional analysis.



If the response Y is obtained from the ratio A/B, that
should be included. If it is obtained only from A/B,
the rest should be discarded.

Do not build linear regression models just because there
are canned programs for doing them. Build your model from
YOUR probabilistic considerations, in total disregard of
what you know how to calculate. THEN talk to someone who
does know how to calculate, and how to approximate.

If you can wear both hats, fine. But your question is
strong evidence that you cannot.


--
This address is for information only. I do not claim that these views
are those of the Statistics Department or of Purdue University.
Herman Rubin, Department of Statistics, Purdue University
hrubin@stat.purdue.edu Phone: (765)494-6054 FAX: (765)494-0558