near singular matrix error
Posted: Mon May 17, 2010 10:38 pm
hello!
i'm using eviews 6
i'm trying to estimate a regression for the determinants of campus crime using data for the year 2008 with a sample made of different campus.
I have different variables for different type of crime
murder, arson, assault,drugliq, robb, sexoff, weapons, totcrime (total of crime variables)
then I have the following variables related to campus characteristics:
tot_est_enrollment, over_25 retention_rate city suburb town rural parttime women total_faculty total acarank cost residential percent_admitted
over_25, parttime, women are ratio computed over tot_est_enrollment
for the location of the campus i have four dummy variables that are mutually exclusive
acarank is a ratio computed over total_faculty
total is the sum of tot_est_enrollment and total_faculty
i'm trying to estimate the following regression and i get a "near singular matrix"error message:
ls murder c tot_est_enrollment arson assault drugliq robb sexoff weapons totcrime over_25 retention_rate city suburb town rural parttime women total_faculty total acarank cost residential percent_admitted
i see different potential problems creating this error but i'm not sure how to solve them:
1. mutually exclusive dummy variables: i've tried to omit the constant but that didn't solve my problem
2. for the crime variables, a lot of the values are zero (for example, there was no murder this year on the campus) but not all (there was murder on another campus)
3. since i'm doing sum and ratio, the variables may be correlated creating a (perfect) multicollinearity problem
could you help me this?
thank you very much for your help
ps: attached are my data set in excel and eviews format
i'm using eviews 6
i'm trying to estimate a regression for the determinants of campus crime using data for the year 2008 with a sample made of different campus.
I have different variables for different type of crime
murder, arson, assault,drugliq, robb, sexoff, weapons, totcrime (total of crime variables)
then I have the following variables related to campus characteristics:
tot_est_enrollment, over_25 retention_rate city suburb town rural parttime women total_faculty total acarank cost residential percent_admitted
over_25, parttime, women are ratio computed over tot_est_enrollment
for the location of the campus i have four dummy variables that are mutually exclusive
acarank is a ratio computed over total_faculty
total is the sum of tot_est_enrollment and total_faculty
i'm trying to estimate the following regression and i get a "near singular matrix"error message:
ls murder c tot_est_enrollment arson assault drugliq robb sexoff weapons totcrime over_25 retention_rate city suburb town rural parttime women total_faculty total acarank cost residential percent_admitted
i see different potential problems creating this error but i'm not sure how to solve them:
1. mutually exclusive dummy variables: i've tried to omit the constant but that didn't solve my problem
2. for the crime variables, a lot of the values are zero (for example, there was no murder this year on the campus) but not all (there was murder on another campus)
3. since i'm doing sum and ratio, the variables may be correlated creating a (perfect) multicollinearity problem
could you help me this?
thank you very much for your help
ps: attached are my data set in excel and eviews format