Dominance analysis / relative importance / R2 decomposition

[Klaus Pforr]

Dear listers,

I am looking for available Stata ados in the area of dominance analysis, relative importance, and R^2 decomposition. So far I found <domin> by J. Luchman and <rego> by F. Huettner. Does anybody know any other related ados.

best regards

Klaus Pforr

[Rich Goldstein]

sorry, but I don't know of any others; in fact, I didn't even know about -rego- and -findit- does not find it; where can it be found please?

[Klaus Pforr]

-rego- can be found here: http://uni-leipzig.de/~rego/
Franz Huettner and Marco Sunder presented it at the UK Stata user conferencen 2012 (http://www.stata.com/meeting/uk12/ab...k12_sunder.pdf)
There is also a publication with this ado referenced at the uni-leipzig page.

Klaus

[Rich Goldstein]

thank you

[Joseph Luchman]

Hi Klaus,

Besides - domin -, any of the Shapley Decomposition methods (- shapley - (SSC), - shapley2 - (SSC), - shapleyx - (http://www.homepages.ucl.ac.uk/~ucakjpr/stata)) similarly decompose the R2. The only method I believe that differs in terms of implementation is the "epsilon" option offered by - domin - which uses the method of "relative weights" (Johnson, 2000).

It's worth noting that domin's "Standardized Weight" results match - rego -'s % Shapley R2 and as well as the results from the "- shapley -" programs on SSC in terms of the general dominance results when not using the "epsilon" option.

- joe

Johnson, J. W. (2000). A heuristic method for estimating the relative weight of predictor variables in multiple regression. Multivariate Behavioral Research,35(1), 1-19.

[Klaus Pforr]

Thank you, Joe

[Daniel Weimar]

Hi everyone,

I got an additional question about this Topic. If Johnson`s Epsilon method (Domin) delivers the same decomposition values as a shapley procedure, why is this method referred as a special procedure anyway? To what Approach should I refer in a paper: to Johnson`s paper or to shapley`s?

[Joseph Luchman]

Hi Daniel,

They (epsilon and dominance/Shapley) are not equivalent procedures.

See:

Thomas, D. R., Zumbo, B. D., Kwan, E., & Schweitzer, L. (2014). On Johnson's (2000) relative weights method for assessing variable importance: a reanalysis. Multivariate behavioral research, 49(4), 329-338.

Epsilon is useful in some cases as its computation/run time is much smaller for large regression models and they do tend to yield similar answers in practice.

- joe

[Michael Radlick]

After reviewing Nathans, L; et al (2012) Interpreting Multiple Linear Regression: A Guidebook of Variable Importance I have been working through the options for post regression analysis in Stata. I believe that the Stata ado program regcoef addresses structure coefficients and the domin ado program addresses relative weights and dominance analysis. However after searching I don't see specific programs for either product measure (Pratt, 1987) or commonality coefficients. Any idea where these might be found?
Michael Radlick, Ph.D. Researcher

[jose nogueira]

Hi, I have read Zingales 1994 and see that they use the Shapley value of votes held by small shareholders (those who own less than 5 percent of the votes). It says "The Shapley value of votes held by small shareholders is the main proxy for the control value of these votes. This can be thought of as the probability that those votes are pivotal in a random coalition formation." I believe this is NOT the shapley decomposition, so I am afraid I can not use neither shapley, nor shapley2 nor shapleyx. I believe I need to compute Shapley value for each observation of my database (for the shareholder structure of each of the companies composing my database), not for each variable. Do you know how can I compute this in Stata? (its out of my reach to manually construct the tree for each of the observations in the sample). thanks in advance and Best regards, Jose
Zingales, L., 1994. The value of the voting right: a study of the Milan Stock Exchange. Review of Financial Studies 7, 125–148.