An Overview of Political Science (Policy and International Relations Primer for EA, Part 3)

This is the third in a se­ries of posts on In­ter­na­tional Re­la­tions and Policy. First was the in­tro­duc­tion to this se­quence, which dis­cusses why this mat­ters, and who should care. Se­cond was defin­ing Policy and In­ter­na­tional Re­la­tions, and talk­ing about the sub-fields of poli­ti­cal sci­ence. That post deferred dis­cus­sion of differ­ent ap­proaches, and in­stead looked at sub­ject ar­eas. That leads to the cur­rent post, which is a more de­tailed overview of poli­ti­cal sci­ence speci­fially, rather than in­ter­na­tional re­la­tions and policy, which will be dis­cussed in later posts.

This post is pri­mar­ily go­ing to dis­cuss ap­proaches to poli­ti­cal sci­ence, such as Be­hav­ioral­ism, Ra­tional Choice The­ory, and In­sti­tu­tion­al­ism, as well as Con­struc­tivism, Marx­ism, and Fem­i­nist Poli­ti­cal Science. This back­ground is par­tic­u­larly use­ful be­cause these differ­ent ap­proaches will be im­por­tant in dis­cussing how to use poli­ti­cal sci­ence and similar tools in spe­cific types of EA de­ci­sion­mak­ing.

First how­ever, I (again) want to deal with the is­sue of poli­ti­cal sci­ence as a soft sci­ence. In ad­di­tion to be­ing use­ful for later un­der­stand­ing how poli­ti­cal sci­ence can and can­not be used in Effec­tive Altru­ism, I think it gives use­ful back­ground on what the differ­ent ap­proaches are do­ing, and why the differ­ent types of poli­ti­cal sci­ence differ in more than just method­ol­ogy.

The Soft Science of Politics

Poli­ti­cal Science at­tempts to deal with “sys­tems of gov­er­nance, and the anal­y­sis of [ac­tivi­ties, thoughts, and be­hav­ior] that ap­ply to mem­bers of a group.” Ac­cord­ing to most ap­proaches, it is a sci­ence be­cause it at­tempts to provide a sys­tem­atic frame­work and the­o­ries for pre­dict­ing and ex­plain­ing what hap­pens and what will hap­pen.

It is pos­si­ble that this sci­en­tific ap­proach builds on psy­chol­ogy the same way chem­istry builds on physics. If so, once we un­der­stand the in­di­vi­d­ual parts, we can dis­cuss what groups of those parts do when or­ga­nized in differ­ent ways. This would make it an ex­er­cise in de­duc­tive think­ing.

Is poli­ti­cal sci­ence a sci­ence in the same way chem­istry is? Un­for­tu­nately, as I dis­cussed in the in­tro­duc­tion, psy­chol­ogy is a soft sci­ences, where we can­not make sharp pre­dic­tions. As a re­sult, de­spite Asi­mov’s hopes, inas­much as poli­ti­cal sci­ence deals with the same is­sues, poli­ti­cal sci­ence is a soft sci­ence as well. We can build psy­cholog­i­cal mod­els of be­hav­ior, and look at how these ag­gre­gate into group de­ci­sions. Alter­na­tively, we can look at group de­ci­sions di­rectly and try to see how they work, build­ing the­o­ries more di­rectly. Both ap­proaches are use­ful—both in chem­istry and in poli­ti­cal sci­ence, but they are very differ­ent. This has im­pli­ca­tions for how differ­ent ap­proaches see poli­ti­cal sci­ence, so it’s worth ex­plain­ing.

The model where fun­da­men­tal un­der­stand­ing like physics or psy­chol­ogy in­forms chem­istry or poli­ti­cal sci­ence is use­ful be­cause it has a sort of guaran­tee that if the un­der­ly­ing model is cor­rect, the re­sults are cer­tain. If we un­der­stand how in­di­vi­d­u­als will de­cide, per­haps we can un­der­stand how groups will de­cide. In ad­di­tion to the fact that psy­cholog­i­cal pre­dic­tions are im­pre­cise, the prob­lem with this is much like the one in physics. Fun­da­men­tal mod­els are com­plex and scal­ing them up re­quires de­tailed knowl­edge of all of the parts, and how they in­ter­act.

In some places, we can rely on statis­ti­cal guaran­tees the way chem­istry re­lies on the­mo­dy­nam­ics (i.e. statis­ti­cal me­chan­ics.) That’s how we pre­dict elec­tions. In most places, how­ever, there are lots of fac­tors that need to be known to move from in­di­vi­d­ual to group be­hav­ior. Th­ese fac­tors in­clude the in­sti­tu­tions, the way differ­ent groups or classes of peo­ple in­ter­act, the im­por­tance of in­di­vi­d­u­als in power, and the com­plex way that even perfectly ra­tio­nal agents can be­have. Each of these roughly cor­re­sponds to the fo­cus of one of the be­low ap­proaches. (In­sti­tu­tion­al­ism, Marx­ism, Psy­cholog­i­cal the­ory, and Ra­tional Ac­tor the­ory re­spec­tively.)

The sec­ond model is where physics (or psy­chol­ogy) is effec­tively ig­nored, and in­sights about the sys­tem need to come from a more di­rect em­piri­cal ap­proach. If we see that, for in­stance, so­cial co­he­sion pre­dicts de­gree of democ­racy, we can quan­tify and test that the­ory di­rectly. This of­ten re­quires us­ing higher-level ab­strac­tions, how­ever, and these are of­ten hard to mea­sure. This model al­lows ap­proaches that are not con­tra­dicted by the first one, and the two are of­ten used to­gether.

Lastly, there are the mod­els where all of the above is bul­lshit, and peo­ple can cre­ate mean­ings and have per­sonal agen­das that drive poli­tics that can’t be use­fully cap­tured by pre­dic­tive mod­els, and can only be ob­served. This is some­times the view of Fem­i­nist, Con­struc­tivist, and var­i­ous Post-Modern ap­proaches. When this is as­sumed, “the­ory” is not con­sis­tent with the Pop­pe­rian or Bayesian un­der­stand­ing of what a sci­en­tific the­ory or ex­pla­na­tion must be. In­stead, the­ory means a con­ve­nient or in­sight­ful way of group­ing ideas or ob­ser­va­tions par­si­mo­niously, an ex­pla­na­tion in the sense Oc­cam prob­a­bly origi­nally in­tended.

Un­for­tu­nately, which of the three ap­proaches is be­ing used is of­ten not clear to read­ers. In some cases, it seems they are not even con­cep­tu­ally dis­tinct to writ­ers and poli­ti­cal sci­en­tists, who use the term the­ory in ways that skip back and forth be­tween these ap­proaches. Given that, I’ll dis­cuss the pri­mary “the­o­ries” and ap­proaches in poli­ti­cal sci­ence.

Ap­proaches to Poli­ti­cal Science

For this overview, I’m go­ing to treat the ap­proaches as if they are com­pletely sep­a­rate. They aren’t, and most good poli­ti­cal sci­ence is not dog­matic, and takes in­sights from differ­ent ap­proaches where they are helpful. I’d even ar­gue that once we get past the fairly ma­jor ob­sta­cles of ego, nar­row­ness of ex­per­tise, and con­fir­ma­tion bias, there are rel­a­tively few dis­agree­ments be­tween poli­ti­cal sci­en­tists about the places where each ap­proach is limited, and where each is es­pe­cially helpful.

But it is helpful to con­sider them sep­a­rately, just to un­der­stand that the ap­proaches do bring differ­ent ideas to the table. This is true even though so­phis­ti­cated analy­ses of each type be­gin to con­verge in many ways. Ra­tional choice the­o­rists may ac­count for in­sti­tu­tional and Marx­ist ar­gu­ments in the struc­ture of their mod­els, or the mo­tives of the ac­tors may ac­count for norms and do­mes­tic con­cerns of each ac­tor. Similarly, “mod­ern” be­hav­ioral ap­proaches are free to con­sider Nash equil­ibria and other ra­tio­nal ac­tor tools. And given that caveat, the be­low table, also linked, has a good overview of some of the view­points I’ll dis­cuss.

Table from “The­ory and Meth­ods in Poli­ti­cal Science” by David Marsh and Gerry Stoker

Be­hav­ioral­ist Poli­ti­cal Science is the at­tempt to look at ac­tual poli­ti­cal be­hav­ior (hence the name,) and find falsifi­able laws that ex­plain and pre­dict it. Be­fore the “be­hav­ioral rev­olu­tion” that hap­pened around the 1940s and 1950s, much poli­ti­cal sci­ence was le­gal the­ory and spec­u­la­tive and untestable the­o­riz­ing. Be­hav­ioral­ism (and not “be­hav­iorism,” which is a quasi-re­lated ap­proach in psy­chol­ogy,) in­stead fo­cused on what in­di­vi­d­u­als did, and what the re­sults were. The new ap­proach was value-neu­tral, and was in part a re­sult of, and re­sponse to, the claim that poli­ti­cal sci­ence wasn’t a sci­ence at all. This be­came the dom­i­nant ap­proach in poli­ti­cal sci­ence in the 1950s, ’60s, and ’70s.

Be­hav­ioral­ism pro­moted sys­tem­at­i­cally col­lect­ing data and test­ing ideas. Case stud­ies and ex­per­i­men­ta­tion were able to be used at small scales, and these be­came key tools. Un­for­tu­nately, they were not able to an­swer many key ques­tions in poli­ti­cal sci­ence, which usu­ally dealt with larger groups. Sur­veys quickly be­came a key tool for poli­ti­cal sci­en­tists. Be­hav­ioral­ists were also a key group in de­vel­op­ing statis­ti­cal ap­proaches to anal­y­sis ap­pro­pri­ate to so­cial sci­ence, where con­founders and lagged effects were key challenges. But it turns out that there is re­ally very lit­tle that can be done with only data. In­stead, gen­er­at­ing hy­pothe­ses to test re­quires the­ory, in the sec­ond sense above, to at least drive the choice of data to gather.

In the 60s and 70s, there was a push­back against Be­hav­ioral­ism. It was at­tacked as mind­less em­piri­cism, and the in­sis­tence on value-neu­tral­ity meant that many in­ter­est­ing ques­tions, such as what ap­proaches would be bet­ter for the pop­u­lace, were ig­nored. It was also effec­tively use­less for look­ing at in­ter­na­tional re­la­tions, which was heav­ily re­li­ant on the­ory, con­sid­er­ing state mo­tives, and look­ing at in­sti­tu­tional de­ci­sion­mak­ing. As a con­se­quence of these crit­i­cisms, mod­ern Be­hav­ioral­ism ac­cepts the im­por­tance of the­ory, and is far less in­sis­tent on only con­sid­er­ing value-neu­tral ques­tions.

Ra­tional Choice The­ory was an ap­proach adopted by poli­ti­cal sci­en­tists from eco­nomics, and poli­ti­cal eco­nomics, start­ing in the late 1950′s. Start­ing from the as­sump­tion that groups are self-in­ter­ested and act (ap­prox­i­mately) ra­tio­nally, it can make clear pre­dic­tions about what groups will do. This is close to the deuc­tive ap­proach men­tioned ear­lier. Ra­tional Choice was a fairly suc­cess­ful al­ter­na­tive to Be­hav­ioral­ism, and it was still jus­tifi­ably sci­en­tific, while ad­dress­ing ques­tions that Be­hav­ioral­ism typ­i­cally did not. This was speci­fi­cally crit­i­cal in In­ter­na­tional Re­la­tions, where the equiv­a­lent ap­proach is called Ra­tional Ac­tor the­ory. (This will be dis­cussed in the next post.)

Ra­tional Choice The­ory uses math­e­mat­i­cal tools, and re­lies heav­ily on game-the­ory, since al­most all of poli­ti­cal sci­ence is about the in­ter­ac­tion of peo­ple, groups, and states. This means that per­haps the most im­por­tant ques­tion for ra­tio­nal choice in poli­ti­cal sci­ence is the level of anal­y­sis—who the ra­tio­nal ac­tors are. This can also in­clude qual­i­ta­tive eval­u­a­tion , for ex­am­ple, to di­rectly elicit from ex­perts what the con­se­quences of a given de­ci­sion will be, rather than to model the sce­nario di­rectly. And even within quan­ti­ta­tive mod­el­ling poli­ti­cal sci­en­tists use ex­pert elic­i­ta­tion and sur­veys, as well as var­i­ous other tools from the so­cial sci­ences, to find in­puts to their ra­tio­nal ac­tor mod­els.

In the 1990s, some con­fi­dently pre­dicted that all of poli­ti­cal sci­ence would be sub­sumed by ra­tio­nal choice the­ory. And be­fore giv­ing the area more thor­ough ex­am­i­na­tion, this was cer­tainly my view. But they were wrong—and even to the ex­tent that they are cor­rect, it would be a mis­take to ig­nore other the­o­ries. I’ll ex­plain more about why this is a bad idea in the next post on in­ter­na­tional re­la­tions, but at the very least it ig­nores key is­sues that mat­ter for ac­tual de­ci­sions. That’s at least part of why ra­tio­nal choice the­ory is cur­rently a minor (albeit in­fluen­tial) ap­proach in poli­ti­cal sci­ence. (Again, it is much more im­por­tant in in­ter­na­tional re­la­tions.)

In­sti­tu­tion­al­ism fo­cuses on in­sti­tu­tions, norms, and the struc­ture of gov­ern­ments. In an ear­lier form this was the ver­sion of poli­ti­cal sci­ence that talked about how to write con­sti­tu­tions, and con­tinued the philo­soph­i­cal tra­di­tion of Aris­to­tle’s “Poli­tics,” Machi­avelli’s “The Prince,” Hobbes’ “Le­viathan.” (In­ter­est­ingly, US Pres­i­dent Woodrow Wil­son had a PhD in poli­ti­cal sci­ence, and was very much an in­sti­tu­tion­al­ist.)

The ap­proach has been rein­vi­go­rated more re­cently as a way to re­spond to some of the short­com­ings of be­hav­ioral­ism and ra­tio­nal choice the­ory. Early be­hav­ioral­ists wanted to move past le­gal fic­tions and org charts, and get to what was hap­pen­ing based on “real” be­hav­ior. When taken to an ex­treme, this ig­nored the ex­tent to which those real be­hav­iors we in­fluenced by laws, norms, and the struc­ture of gov­ern­ment. Ra­tional choice the­o­rists, on the other hand, start with the idea of prefer­ences. This meant that they would some­times ig­nore the way that these prefer­ences were turned into de­ci­sions by the struc­ture of the sys­tem.

In­sti­tu­tion­al­ists ar­gue that in­sti­tu­tions are more than ag­gre­gates of in­di­vi­d­ual prefer­ences and in­di­vi­d­ual be­hav­iors, and they should be treated as such. In short, re­duc­tion­ist ap­proaches were un­helpful be­cause they were not the cor­rect level of anal­y­sis. The prob­lem is defin­ing what the cor­rect level of anal­y­sis is, and what in­sti­tu­tion­al­ism has to say.

Modern In­sti­tu­tion­al­ism has dozens of view­points about what should be an­a­lyzed, each tak­ing a differ­ent but plau­si­ble set of ideas and ques­tions. A few ex­am­ples of these ap­proaches should give a sense of what types of ap­proaches these are, His­tor­i­cal in­sti­tu­tion­al­ism looks at how in­sti­tu­tions de­vel­oped, and how that in­fluences their struc­ture and the way de­ci­sions are made. Net­work In­sti­tu­tion­al­ism looks at how in­for­mal in­ter­ac­tions be­tween groups and in­di­vi­d­u­als can shape de­ci­sions, and how power is wielded by those per­haps with­out offi­cial po­si­tions. Nor­ma­tive in­sti­tu­tion­al­ism con­sid­ers how norms and be­liefs shape de­ci­sions and con­strain of guide in­sti­tu­tions. And ra­tio­nal choice, Fem­i­nist, and Marx­ist in­sti­tu­tion­al­ism takes each of those view­points and ap­plies them to the study of in­sti­tu­tions.

Psy­cholog­i­cal poli­ti­cal sci­ence uses a psy­cholog­i­cal lens to un­der­stand poli­ti­cal ac­tors. This can in­volve un­der­stand­ing why and how in­di­vi­d­u­als make poli­ti­cal de­ci­sions, how lead­ers re­act to stim­uli, and how sys­tem­atic de­par­tures from ra­tio­nal­ity in­fluence de­ci­sions in ways that change the types of ideas es­poused by ra­tio­nal choice the­ory. Poli­ti­cal ap­proaches are differ­ent than ap­proaches like con­struc­tivism, but can be re­lated by the fact that they view psy­chol­ogy and sub­jec­tive be­liefs as a key fac­tor.

I also want to men­tion Con­struc­tivist, Fem­i­nist, and Marx­ist Poli­ti­cal Science since I think they of­ten get dis­missed in dis­cus­sions by out­siders, de­spite be­ing use­ful. Th­ese are all based on the fact that peo­ple con­struct nar­ra­tives, and their de­ci­sions are shaped by their in­ner view­points. Th­ese in­ner view­points may be sub­jec­tive, but they still in­fluence the world—dur­ing the Rus­sian rev­olu­tion, it was far less im­por­tant whether Marx or Stalin was cor­rect in their view of the world than it was that many peo­ple chose to em­brace that view­point. Given that, spe­cific wor­ld­views can be very in­fluen­tial, and provide use­ful ways to un­der­stand the world.

There are par­tic­u­larly ex­treme view­points that come un­der these um­brel­las, how­ever, that do things like dis­pute the ex­is­tence of ob­jec­tive re­al­ity, or claim that sci­en­tific knowl­edge is im­pos­si­ble. Th­ese view­points, while be­ing parts of these move­ments, have made it easy for peo­ple to dis­miss the in­sights more gen­er­ally, of­ten in ways that make them ig­nore im­por­tant is­sues.

In fact, I would guess that many peo­ple who dis­miss Marx­ist and Fem­i­nist poli­ti­cal sci­ence still em­brace many of the the­o­ret­i­cal view­points that they es­pouse. If you think that we need to con­sider the im­pacts of un­em­ploy­ment or au­toma­tion on poli­tics, you’re us­ing a Marx­ist lens which looks at differ­ent so­cial groups as im­por­tant poli­ti­cal ac­tors. Marx­ism sees much of poli­tics as a strug­gle be­tween differ­ent groups of elites and seg­ments of the pub­lic. (Iron­i­cally, the elitism ver­sus the pub­lic is of­ten a talk­ing point of the mod­ern Amer­i­can right-wing.) Similarly, if you think that minori­ties, women, and other groups are dis­tinct from other groups on the ba­sis of their iden­tities, you’re us­ing a fem­i­nist lens.

There are other ap­proaches as well, but oc­cupy smaller niches. To name a few, com­par­a­tive poli­tics looks at differ­ences in poli­ti­cal sys­tems, and there are poli­ti­cal sci­en­tists who spe­cial­ise in spe­cific types of poli­ti­cal be­hav­ior, like na­tion­al­ism or se­cu­rity stud­ies, or spe­cific is­sues like elec­tion sys­tems (psephol­ogy) or in­ter­na­tional de­vel­op­ment. Each of these is po­ten­tially worth fur­ther dis­cus­sion, but my cur­rent plan is to fo­cus on other is­sues, so they won’t be dis­cussed.


(1) This sec­tion is much less an overview of poli­ti­cal sci­ence as un­der­stood by prac­ti­tion­ers, and much more my per­sonal way of un­der­stand­ing and jus­tify­ing the (mis)use of terms like “the­ory” in these ar­eas of so­cial sci­ence. When I was first ex­posed to psy­chol­ogy, poli­ti­cal sci­ence, and re­lated ar­eas at a grad­u­ate level (af­ter do­ing my BS in math­e­mat­ics,) I was frus­trated that, it seemed, these ar­eas had never heard of Karl Pop­per or falsifi­a­bil­ity. I was wrong, but it took me quite a while to figure out what I was miss­ing—I’m hop­ing this is helpful to oth­ers bridg­ing a similar in­fer­en­tial gap.

(2) Defi­ni­tion adapted from Wikipe­dia. Their defi­ni­tion of poli­ti­cal sci­ence cites the Oxford Dic­tionary, and says it “deals with sys­tems of gov­er­nance, and the anal­y­sis of poli­ti­cal ac­tivi­ties, poli­ti­cal thoughts, and poli­ti­cal be­hav­ior.” This overuses the word poli­tics, which, ac­cord­ing to the link, “in­volves mak­ing de­ci­sions that ap­ply to groups of mem­bers.”

(3) Chem­istry cer­tainly didn’t start with a cor­rect model of physics and build on that—early ad­vances like Boyle’s law (dis­cov­ered a half dozen blocks from FHI) were based on in-ret­ro­spect very in­sight­ful but com­pletely wrong mod­els. Boyle thought that the par­ti­cles of air were static, and held in place via some sort of spring that kept them in place. New­ton’s work would later jus­tify the in­cor­rect model of static par­ti­cles, but it took a few more cen­turies to dis­cover a proper jus­tifi­ca­tion based on ki­netic the­ory.

(4) Bot­tom-up and top-down mod­els are some­times dis­cussed, but the terms seem re­ally am­bigu­ous in this con­text, so I won’t use them.

(5) In part, this was in­stead of try­ing to dis­cuss poli­tics in terms of what differ­ent in­sti­tu­tions did. I’ll talk about that in the be­low sec­tion on In­sti­tu­tion­al­ism.

(6) In the early 1900s, sam­pling for sur­veys was first un­der­stood and used, and the meth­ods were first used for so­ciolog­i­cal re­search in the 1930s. Note all the amaz­ing datasets from sur­veys performed on an on­go­ing ba­sis in the US over decades were largely due to the new ideas of be­hav­ioral­ism. This in­cludes the Cen­sus Bureau, which in the 1940s start­ing ask­ing about things other than just pop­u­la­tion, such as hous­ing, mort­gages, util­ity bills, ra­dio and TV own­er­ship, and in­door plumb­ing .

(7) It is pos­si­ble to ad­dress many such ques­tions with­out ac­cept­ing a set of val­ues as cor­rect, but many of these ques­tions were not in fact ad­dressed.

(8) When pre­dict­ing in­di­vi­d­ual be­hav­ior, ra­tio­nal choice is an eco­nomic and psy­cholog­i­cal model. Ap­plied to groups and in­sti­tu­tions, how­ever, it ad­dresses ques­tions in poli­ti­cal sci­ence.

(9) Re­turn­ing to the ear­lier anal­ogy be­tween poli­ti­cal sci­ence and chem­istry, we could in the­ory build a game-the­o­retic model of ev­ery in­di­vi­d­ual, each choice they make, and how this in­ter­acts with all other peo­ple to make pre­dic­tions at the group, state, or in­ter­na­tional level. In prac­tice, this is as use­ful as build­ing a model of the hu­man body at the level of quan­tum chro­mo­dy­nam­ics. In­stead, we as­sume that some higher-level group is a ra­tio­nal ac­tor.

(10) For ex­am­ple, first-past-the-post vot­ing schemes, ger­ry­man­der­ing, and in­for­mal coal­i­tions are an in­sti­tu­tional rea­son that in­di­vi­d­ual choices are not re­flected in poli­ti­cal de­ci­sions. When build­ing naive game the­o­retic mod­els, it’s easy to miss these con­sid­er­a­tions, and ac­count­ing for them can make the game the­o­retic mod­els very com­plex, make the re­sults frag­ile to ex­act as­sump­tions and very small changes in rel­a­tive util­ities, or make the mod­els too flex­ible to be pre­dic­tive.

(11) I am be­ing un­fair by group­ing these and giv­ing them so lit­tle at­ten­tion, and will note that they are plau­si­bly worth more at­ten­tion, as I will ex­plain, but I am far less fa­mil­iar with them than the other ap­proaches.

(12) It’s worth not­ing, how­ever, that Fem­i­nism was a poli­ti­cal move­ment that later turned into a view­point for study­ing and un­der­stand­ing poli­tics—un­like the other ap­proaches, which were not poli­ti­cal move­ments, or even Marx­ism, which origi­nated as schol­arly ideas, and even once it be­came a poli­ti­cal force, it still had a spe­cific set of aca­demic view­points.