Partial Sum of Squares

[d_hlambert]

Hi,

Does EViews allow us to report the 'partial sum of squares' values for each coeffient in an ordinary least squares regression?

I'm wanting to use this value to estimate the influence of each variable.

I know that in SAS the command is 'ss2'.

I'm quite new to EViews and programming in general, so any help on the actual commands I need to use to get the partial sum of squares would be greatly appreciated.

Thank you in advance,
David

[EViews Glenn]

It's not a built-in feature, but can be calculated with only a bit of work. Can you tell me a bit more about your base equation (how many variables, etc.) and what you want the final output to look like? And lastly, how exactly do you want to use the results? As a table of output for presentation, or just to find which variables are (partially) significant?

[d_hlambert]

Hi Glenn,

Thank you for your reply, it's greatly appreciated!

The base equation is usually something like: "room_nights c au_gdp_capita hh_disp_y_capita unemploy_r nz_gdp us_gdp jp_gdp ext_shocks" (7-8 variables).

The final output that I have extracted from SAS before is usually something like this:

Parameter Estimates

Parameter Standard
Variable DF Estimate Error t Value Pr > |t| Type II SS

Intercept 1 -2747.28424 445.49104 -6.17 <.0001 71926
aust_gdp_capita 1 0.03843 0.10173 0.38 0.7079 269.88862
hh_disp_y_capita 1 0.03810 0.04777 0.80 0.4305 1203.29661
unemploy_r 1 69.35739 358.95772 0.19 0.8479 70.60809
nz_gdp_capita 1 0.10315 0.11097 0.93 0.3590 1634.17784
euro_gdp_capita 1 0.54247 0.11587 4.68 <.0001 41452
us_gdp_capita 1 0.02187 0.08567 0.26 0.8000 123.20403
jp_gdp_capita 1 0.00019043 0.00081434 0.23 0.8165 103.42364
ext_shocks 1 -88.38970 35.80797 -2.47 0.0186 11524

Then I usually use the 'Type II SS' values for all variables (except the intercept) and weight them against each other to get a view on the relative influence of each variable (eg. 2%, 21%, 43% etc).

So really I don't mind whether it prints a table for presentation, or just an output where I can work out the influence of each variable.

Also, is this an okay way to try and work out the influence? Should I be using something like Cook's distance statistic?

Sorry I'm a bit new at all this :), appreciate the help.

Thank you again,
David

[EViews Glenn]

Can you post your workfile with the data here, or send it to support@eviews.com with a flag to let me know that it's there...

[PCA]

Do you have any additional information about how to do this?

I'd be looking to get the partial sum of squares for each variable placed into a matrix. That would be optimal, but more importantly I just need these values.

[PCA]

Any additional insight as to how to get partial (or sequential based on largest t-stat) sum of squares for each explanatory variable.

I believe these are referred to as Type I (sequential) and Type II (partial) in SAS. Not entirely sure on that however.

[EViews Glenn]

All of these are easily obtained by running individual regression models based on the original specification with subsets of the original variables. We've never added them to EViews since they are either order dependent (Type I) or don't have adding up properties unless you have an orthogonal design (Type II and Type III). Our view has always been that these are used more in analysis of designed experiments than in the econometric analysis that is the EViews core. But that may not be accurate...

There are also some issues about properly detecting interaction terms in Type II calculations since EViews doesn't have an interaction syntax (except for @expand).

If there is a lot of interest, we might get around to an add-in sometime, but it's not high on our list (though we wlll take note of the request). Some enterprising person out there might want to take a crack at it and not wait for us...