Statistical differences between models

[startz]

Code: Select all

ls y DUK DUS DEU DUK*X DUS*X DEU*X

Thanks for your suggestion. When I enter equation: y c DUK DUS DEU DUK*X1 DUS*X1 DEU*X1 and so on for al five variables I get the error near singular matrix, also without entering the constant. This is probably because all dummies are mutually exclusive.
Starting to get a bit desperate here as I do not seem to be able to solve this problem that shouldn't be so difficult.

When you include the constant, you should expect "near singular matrix." This is the dummy variable trap. I don't see why you get the error when the constant is omitted though.

[Movid]

That's more than what I could have asked for.
Basically what I've done is running 4 separate regressions for 4 datasets (entire dataset + its three subsets based on region), all consisting of the same variables. Equation: y c interest quantum low high under over

I'm trying to figure out whether there are significant differences between the coefficients of the variables when comparing each subset (America, UK, Cont. Europe) to one another. As Startz suggested I created dummy variables for each variable to indicate its region. He suggested equation
y (c) DUK DUSA DEUR DUK*X DUSA*X DEUR*X which in my case would lead to the following:

y ie ia iu ic interest*ie interest*ia interest*iu interest*ic qe qa qu qc quantum*qe quantum*qa quantum*qu quantum*qc le la lu lc low*le low*la low*lu low*lc he ha hu hc high*he high*ha high*hu high*hc ue ua uu uc under*ue under*ua under*uu under*uc oe oa ou oc over*oe over*oa over*ou over*oc Where -e is the Entire dataset, -a is America, -u is UK, and -c is Continental Europe.

This, however, leads to near singular matrix error, regardless of whether the constant is included. Any help on solving this issue or different approaches on my statistical differences problem is greatly appreciated! Dataset is attached.

[startz]

That's more than what I could have asked for.
Basically what I've done is running 4 separate regressions for 4 datasets (entire dataset + its three subsets based on region), all consisting of the same variables. Equation: y c interest quantum low high under over

I'm trying to figure out whether there are significant differences between the coefficients of the variables when comparing each subset (America, UK, Cont. Europe) to one another. As Startz suggested I created dummy variables for each variable to indicate its region. He suggested equation
y (c) DUK DUSA DEUR DUK*X DUSA*X DEUR*X which in my case would lead to the following:

y ie ia iu ic interest*ie interest*ia interest*iu interest*ic qe qa qu qc quantum*qe quantum*qa quantum*qu quantum*qc le la lu lc low*le low*la low*lu low*lc he ha hu hc high*he high*ha high*hu high*hc ue ua uu uc under*ue under*ua under*uu under*uc oe oa ou oc over*oe over*oa over*ou over*oc Where -e is the Entire dataset, -a is America, -u is UK, and -c is Continental Europe.

This, however, leads to near singular matrix error, regardless of whether the constant is included. Any help on solving this issue or different approaches on my statistical differences problem is greatly appreciated! Dataset is attached.

You have a better shot of getting someone to look at this if you attach your EViews dataset.

[Movid]

You have a better shot of getting someone to look at this if you attach your EViews dataset.

That makes sense. I ran the regressions for the subsets on separate workfiles. The file attached combines all data.

[startz]

You have a better shot of getting someone to look at this if you attach your EViews dataset.

That makes sense. I ran the regressions for the subsets on separate workfiles. The file attached combines all data.

That clears it up very quickly. You have included two complete sets of dummies. That's another form of the dummy variable trap.

[Movid]

That clears it up very quickly. You have included two complete sets of dummies. That's another form of the dummy variable trap.

Two? I have four dummy variables (under/over/low/high) but they do not always have opposing values (e.g. under/over, low/high can both be 0).
What would you suggest me to do?

[startz]

That clears it up very quickly. You have included two complete sets of dummies. That's another form of the dummy variable trap.

Two? I have four dummy variables (under/over/low/high) but they do not always have opposing values (e.g. under/over, low/high can both be 0).
What would you suggest me to do?

I may have gotten it wrong, but it appears that "ie ia iu ic" is one complete set of dummues and that "le la lu lc" is another.

[Movid]

I may have gotten it wrong, but it appears that "ie ia iu ic" is one complete set of dummues and that "le la lu lc" is another.

True, but they're not the only ones. I created those dummies to match the values of variable with a region as was suggested earlier. Originally I did not have those dummies, but ran equation y c interest quantum low high under over on 4 separate datasets. Now I merged all data into one file. To my knowledge creating just one set of dummies would only indicate differences in influence on the dependant variable, whereas I want to compare the differences of the coefficients each variable for each region. Hence, I created a set of region dummies for each variable, which leads to the following dataset:
Variable (Dummies)
Interest (ie, ia, iu, ic ) (e=Entire dataset, a=America, u=UK, c=Continental Europe)
Quantum (qe, qa, qu, qc)
Low (le, la, lu, lc)
High (he, ha, hu, hc)
Under (ue, ua, uu, uc)
Over (oe, oa, ou, oc)
Where Low, High, Under, and Over are dummies.

Running a regression on all these variables/dummies, with or without constant, leads to a near singular matrix error. Also when I multiply the variables with the dummies, as was proposed earlier.

[startz]

I may have gotten it wrong, but it appears that "ie ia iu ic" is one complete set of dummues and that "le la lu lc" is another.

True, but they're not the only ones. I created those dummies to match the values of variable with a region as was suggested earlier. Originally I did not have those dummies, but ran equation y c interest quantum low high under over on 4 separate datasets. Now I merged all data into one file. To my knowledge creating just one set of dummies would only indicate differences in influence on the dependant variable, whereas I want to compare the differences of the coefficients each variable for each region. Hence, I created a set of region dummies for each variable, which leads to the following dataset:
Variable (Dummies)
Interest (ie, ia, iu, ic ) (e=Entire dataset, a=America, u=UK, c=Continental Europe)
Quantum (qe, qa, qu, qc)
Low (le, la, lu, lc)
High (he, ha, hu, hc)
Under (ue, ua, uu, uc)
Over (oe, oa, ou, oc)
Where Low, High, Under, and Over are dummies.

Running a regression on all these variables/dummies, with or without constant, leads to a near singular matrix error. Also when I multiply the variables with the dummies, as was proposed earlier.

This isn't really an EViews question. Perhaps if you posted in the Econometric Discussions section someone might be able to be more helpful.