X B β W βB K X k 1 X kB βk W βk B 7 However in nonlinear settings this is not

X b β w βb k x k 1 x kb βk w βk b 7 however in

This preview shows page 16 - 18 out of 30 pages.

X B ( β W - β B ) = K X k =1 ¯ X kB ( β k W - β k B ) (7) However, in nonlinear settings, this is not as simple. First, contributions of a particular covariate will be obtained by computing the marginal impact obtained when the counterfactual is computed by replacing the distribution of this covariate by the one of the other group. Then, there is two possibilities, either the researcher pursues with this strategy for the other covariates without changing back the distribution of the first covariate, or he can change back the first covariate to its observed distribution and changes the distribution of the second covariate instead. The first strategy might be qualified as ”without replacement”, while the second is considered ”with replacement”. But without replacement, the decomposition will be path dependent, while without it will not add up to the aggregate effect. With distribution regressions methods, it is possible to compute a detailed decomposition without replacement. The marginal contribution C k of a particular covariate k will thus be the difference between the aggregate decomposition with and without this covariate for respectively the composition (c) and the structure (s) effects: 15
Image of page 16
C k c = Δ c ( K ) - Δ c ( K \ { k } ) (8) and C k s = Δ s ( K ) - Δ s ( K \ { k } ) (9) with Δ s ( K ) the aggregate structure effect for the set of covariates K, while K \ { k } is the same set without the covariate k. 3.5 Shapley Value Shorrocks[49] proposes a solution to the problem of path dependence in detailed decomposition. By averaging all the marginal contributions along all the possible elimination paths, i.e. the possible order in which a particular covariate can be added to other covariates, you obtain a contribution of this covariate which is unique, adds up to the aggregate effect, and is path independent. Contributions are called Shapley Value because they are computed exactly the same way. Initially, the Shapley Value was conceived as the solution to divide fairly the reward of a cooperative game among players according to their efforts. For example, imagine we have three covariates, X = { X 1 , X 2 , X 3 } and we want to know the contribution of X 3 . Then, we have exactly 3! ways to order our covariates. Of course, the order will not matter when the model is evaluated with all the covariates. But it will change the marginal contributions. Let us say that we include first X 1 , then X 2 , and finally X 3 . We will get the particular elimination path {{ X 1 } ; { X 1 , X 2 } ; { X 1 , X 2 , X 3 }} . Then, if we denote the model we will evaluate for a set of covariates by the function V ( . ), we will get three different evaluations of this model, namely { V ( { X 1 } ); V ( { X 1 , X 2 } ); V ( { X 1 , X 2 , X 3 } ) } . But this particular elimination path only interests us for the marginal contribution of X 3 to the evaluation of the model which is in this case V ( { X 1 , X 2 , X 3 } ) - V ( { X 1 , X 2 } ). Finally, we have to repeat this operation for all elimination paths and computing the arithmetic mean of all these marginal contributions.
Image of page 17
Image of page 18

You've reached the end of your free preview.

Want to read all 30 pages?

What students are saying

  • Left Quote Icon

    As a current student on this bumpy collegiate pathway, I stumbled upon Course Hero, where I can find study resources for nearly all my courses, get online help from tutors 24/7, and even share my old projects, papers, and lecture notes with other students.

    Student Picture

    Kiran Temple University Fox School of Business ‘17, Course Hero Intern

  • Left Quote Icon

    I cannot even describe how much Course Hero helped me this summer. It’s truly become something I can always rely on and help me. In the end, I was not only able to survive summer classes, but I was able to thrive thanks to Course Hero.

    Student Picture

    Dana University of Pennsylvania ‘17, Course Hero Intern

  • Left Quote Icon

    The ability to access any university’s resources through Course Hero proved invaluable in my case. I was behind on Tulane coursework and actually used UCLA’s materials to help me move forward and get everything together on time.

    Student Picture

    Jill Tulane University ‘16, Course Hero Intern

Stuck? We have tutors online 24/7 who can help you get unstuck.
A+ icon
Ask Expert Tutors You can ask You can ask You can ask (will expire )
Answers in as fast as 15 minutes
A+ icon
Ask Expert Tutors