Units of Action

Unit of ac­tion is a term I have been us­ing in­ter­nally to be more spe­cific when think­ing about groups of peo­ple. This post is for flesh­ing out and clar­ify­ing my think­ing for my­self, and see­ing if it would be use­ful to any­one else. Also it feels like there re­ally ought to be a term for this already, and I might be able to find more in­for­ma­tion if some­one knows it.


A unit of ac­tion is a group that takes ac­tions, as a group.

I take the word unit from the mil­i­tary, and also from unit of anal­y­sis, re­flect­ing my be­lief that this is the cor­rect level of anal­y­sis for big-pic­ture prob­lems. Ac­tion is in the col­lo­quial sense of “did some­thing on pur­pose.”

To be more con­crete, the kinds of things I am point­ing at are like fam­i­lies, cor­po­ra­tions, and gov­ern­ment agen­cies. The kinds of things I am point­ing away from are like race, class, gen­der, and re­li­gion.

Causal vs correlational

One way to see the di­vid­ing line is be­tween groups that cause things to hap­pen, and groups to whom things hap­pen. Fam­i­lies take va­ca­tions; cor­po­ra­tions launch prod­ucts; gov­ern­ment agen­cies sue peo­ple for break­ing reg­u­la­tions. The ex­am­ples I pointed away from are de­mo­graphic—they don’t do any­thing as a group, but races may be seg­re­gated, gen­ders have differ­ent bath­rooms, and classes get tax breaks.

Agent heuristic

Another heuris­tic is whether the group can plau­si­bly be mod­eled as a sin­gle agent. Fol­low­ing game-the­o­retic in­tu­ition fur­ther, the unit of ac­tion can be bro­ken down into other units of ac­tion the same way agents can be bro­ken apart into mul­ti­ple agents. We might model one firm as one agent when look­ing at the be­hav­ior of firms in com­pe­ti­tion, but when look­ing at the be­hav­ior of one firm model the differ­ent de­part­ments within the firm as differ­ent agents. Some­times, as in the case of the mil­i­tary, these com­po­nents are very ex­plicit.

Per­pet­ual Coordination

Units of ac­tion are sta­ble, suc­cess­ful-at-some-rate co­or­di­na­tors. I have a vague in­tu­ition that we can think of them as co­or­di­na­tion strate­gies which prop­a­gate in the same way or­ganisms have re­pro­duc­tion strate­gies, but the anal­ogy is hardly pre­cise; re­pro­duc­tion prop­a­gates ge­netic in­for­ma­tion, and units of ac­tion mostly prop­a­gate re­la­tion­ships.

Mis­cel­la­neous thoughts

  • Hier­ar­chy is usu­ally suffi­cient, but not nec­es­sary, to in­di­cate a unit of ac­tion.

  • There is no tax­on­omy; it is defined by ac­tions, not org type.

  • If a given group stops be­ing a unit of ac­tion, I ex­pect it to fall apart soon (split up, go bankrupt, etc).