Cannot proceed with Chow test

[Jacob Gardiner]

Hello guys, I am very new to Stata and as part as my thesis I need to run a chow test on two regressions. However, when trying to add my sum of residuals to find the restricted and unrestricted RSS I find I cannot perform the test as they are decimals. Hence whenever I try and find my critical F value it is always much higher than the F statistic I make.

When I have watched tutorials, often peoples Residuals are always positive numbers higher than one.

Please find attached the three regressions, one complete and two, 2000-2007 and 2007-2015 respectively.

Any help would be greatly appreciated

Many Thanks!
Jake

[Richard Williams]

How is your dependent variable scaled? It seems to have a very small variance, but that reflects how it is scaled. Also is there some reason you need to do this by hand? Stata can do this for you. If by residuals you mean residuals sum of scares there is nothing inherently problematic in the numbers you get -- if y were scaled differently the rss would get rescaled too.

[Richard Williams]

Somebody pointed out to me that I said residual sum of scares, not squares. But if somebody is new to Stata or statistics, maybe scares is more appropriate. ;-)

[Jacob Gardiner]

Hey Richard,

Many thanks for the help! My dependent variable is Vehicle production per capita; per country.. hence the data is scaled so small. I'm not too sure why I am working it out by hand to be honest, perhaps because I am so unfamiliar with stata! hahaha.

I'll attempt to run the test using this guide tomorrow: http://www.stata.com/support/faqs/st...how-statistic/

Sum of scares is a very apt description

Cheers,
Jake

[Richard Williams]

There are several ways to do this, but the Gould FAQ is a good one. I would probably go straight to the approach that uses contrast.

Also, both your analysis and the FAQ include the constraint that the constant is the same for both groups. Sometimes Chow tests are done where you relax that constraint. That is, you allow for a dummy for group membership, but constrain the slopes to be the same for the variables in both groups, i.e. no interactions with group membership. In the FAQs example, you can do this with the command

contrast g2#c.x1 g2#c.x2,overall

If you want a more extended discussion of different approaches and how to interpret them, you can see

http://www3.nd.edu/~rwilliam/stats2/l51.pdf

http://www3.nd.edu/~rwilliam/stats2/l52.pdf

http://www3.nd.edu/~rwilliam/stats2/l53.pdf