Regressing LogLog Equations

[quiveria]

So I have actual observations I am regressing for a double-log equation, such that the original equation is Y = x1*x2, so I would be regressing ln(Y) = B0 + B1*ln(x1) + B2*ln(x2)

Can I treat this like a linear regression, where I take the natural log of my observations and just plug them in and run it through Excel? Or is there some sort of special regression form I have to use?

 

[Deleted member 4993]

So I have actual observations I am regressing for a double-log equation, such that the original equation is Y = x1*x2, so I would be regressing ln(Y) = B0 + B1*ln(x1) + B2*ln(x2)

Can I treat this like a linear regression, where I take the natural log of my observations and just plug them in and run it through Excel? Or is there some sort of special regression form I have to use?

That equation has three variables - I don't think excel can handle that.

To do that properly, you'll need to run ANOVA in some statistical package.

 

[galactus]

SK is most likely correct, but you may be able to do this regression using Excel Solver.

See here:

http://archives.math.utk.edu/ICTCM/VOL1 ... paper.html

or here:

http://homepage.mac.com/samchops/B73317 ... index.html


I did not look at the sites at length, so perhaps it is of no use.

The Solver is very versatile. I have used it a lot.

 

[quiveria]

Excel can handle multiple variables. You can have something like 16, I think. So that's not the problem. Is how I'm treating the variables by doing their natural logs and then running them linearly the correct way to do it?

 

[Deleted member 4993]

Yes - I found out that excel can do multiple regression - but it needs extra service pack to be added to the regular set of program.

So if you have that service pack - your approach is legitimate.

However, for most of of the other stat-pack that I have used - you would not need to take logs. It can create trouble if in reality your variables are shifted by an additive constant.

 

[quiveria]

Yes - I found out that excel can do multiple regression - but it needs extra service pack to be added to the regular set of program.

So if you have that service pack - your approach is legitimate.

However, for most of of the other stat-pack that I have used - you would not need to take logs. It can create trouble if in reality your variables are shifted by an additive constant.

Er, even though the theory says that it should be the natural log?

 

[galactus]

Yes, there are numerous sites that offer free excel add ins for more advanced math and stats.

I even saw one for engineering.