You are on page 1of 45

CHAPTER 1

The Science in Social Science

1.1 INTRoDucTION
TRIS BOOK is 1bout research in the social sciences. Our goal is practical:
designing research that will produce valid inferences about social and
politicallife. We focus on political science, but OUT argurnent applies to
other disciplines such as sociology, anthropology; history, econornics,
and psychology and to nondisciplinary areas of study such as legal
evidence, education research, and clinical reasoning.
This is neither a work in the philosophy of the social sciences nor a
guide to specific research tasks such as the design of sUTveys, conduct
of fie1d work, or analysis of statistical data. Rather, this is a book about
research design: how to pose questions and fashion scholarly research
to make valid descriptive and causal inferences. As such, it occupies a
middle ground between abstract philosophical debates and the hands-
on techniques of the researcher and focuses on the essential logic un-
derlying all social scientific research.

1.1.1 Two Styles of Research, One Logic of Inference


Our main goal is to connect the traditions of what are conventionally
denoted "quantitative" and "qualitative" research by applying a uni-
fied logic of inference to both. The two traditions appear quite differ-
ent; indeed they sometimes seem to be at war. Our view is that these
differences are mainly ones of style and specific technique. The same
underlying logic provides the framework for each research approach.
This logic tends to be explicated and formalized clearly in discussions
of quantitative research methods. But the same logic of inference un-
derlies the best qualitative research, and all qualitative and quantita-
tive researchers would benefit by more explicit attention to this logic
in the course of designing research.
The styles of quantitative and qualitative research are very different.
Quantitative research uses numbers and statistical methods. It tends to
be based on numerical measurements of specific aspects of phenom-
ena; it abstracts from particular instances to seek general description
or to test causal hypotheses; it seeks measurements and analyses that
are easily replicable by other researchers.
4 . The Science in Social Science
h' ontrast covers a wide range of approaches, Introduction . 5
Qualitative researc ,m c , hes relies on numerical mea- research can be systematic and scientific. Historical research can be an-
but by definition, nonehof ~e~e~pFor~~~us on one or a small number alytical, seeking to evaluate alternative explanations through a process
surements. Such work as en. depth analysis of historical ma- of valid causal inference. History, or historical sociology, is not incom-
of cases, to use intensive inter~l~ws o~ to be concerned with a rounded patible with social science (Skocpol 1984: 374-86).
terials, to be discursive in met o ,an t or unit Even though they
h' ount of sorne even . Breaking down these barriers requires that we begin by questioning
or compre enSIve acc f qualitative researchers generally un- the very concept of "qualitative" research. We have used the term in
have a small number o ca~nfes, t' from their studies. Sometimes
ounts of I orma lOn d Our title to signal our subject matter, not to imply that "qualitative"
earth enormous am . . linked with area or case stu _
this kind of work in .the social :~~:-~::~ent, decision, institution, loca-
research isfundamentally different from "quantitative" research, ex-
cept in style~
ies where the focus lS on ~ pa. A' also the case with quantitative Most resea~ch does not fit clearly into one category or the other. The
tion, issue, or piece of l~gls~ahor:' so~ant in its own right: a major
research, the l~stance 15 ~ ~na
change in a nahon, an electlO , J
:!
or decision, or a world crisis. Why
dd ly I'n 19897 More gener-
best often combines features of each. In the same research project,
some data may be collected that is amenable to statistical analysis,
. ollapse so su en . while other equally significant informabon is not. Patterns and trends
did the East German regIme c . t gimes of Eastern Europe col-
in social, political, or econornic behavior are more readily subjected to
ally, why did almost ~11 the COtn:::~:r r~ot always, the event m~~ be
lapse in 1989? SometImes, bu . 1 r of event such as a politIcal
chosen as an exemplar of a partlcu ~r {P~ommuni~ to reject a waste
quantitative analysis than is the flow of ideas among people or the
difference made by exceptional individualleadership. If we are to un-
derstand the rapidly changing sodal world, we will need to indude
revolution or the decision o~ a I:artlcuf ar k's linked to area studies
. c: et' es thlS kmd o wor 1 f h information that cannot be easi1y quantified as well as that which can.
disposal sIte. _o:n 1m. d culture of a particular part o t e Furthermore, alI social sdence requires comparison, which entails
where the focus 15 on the hlstory an t' analyzed dosely and in full
world. The particular place or even 15 judgments of which phenomena are "more" or "less" alike in degree
(Le., quantitative differences) or in kind (i.e., qualitative differences).
detail. .. 1 cientists have debated the merits of Two excellent recent studies exemplify this point. In Coercive Cooper-
For several decades, .p~hhca s. es area studies versus comparative atian (1992), Lisa L. Martín sought to explain the degree of interna-
case studies ver~us ~~a;;stIc~. St~l ~litics using quantitative methods tional cooperation on econornic sanctions by quantitatively analyzing
studies, and "sclenhÍlc s~u l~S PI' rich textual and contex-
ninety-nine cases of attempted economic sanctions from the post-
versus "historical" inVeStIgahon~'::t?v~~e~:archers believe that sys-
World War TI era. Although this quantitative analysis yielded mueh
tual understanding. Son: .qu~n l I d to truth in the social sci-
e
valuable information, certain causal inferences suggested by the data
tematic statistical analysls. lS .t e on y TOha hemently disagree. This
were ambiguous; henee, Martín carried out six detailed case studies of
ences. Advoca ~s . o
t f quahtatlVe researc ve . 1
livel debate; but unfortunately, lt a so sanctions episodes in an attempt to gather more evidence relevant to
difference of opm~on le~ds to . to Ya quantitative-systematic-general- her causal inference. For Making Demacracy Work (1993), Robert D. Put-
bifurcates the sOCIal sCI~nc~s m 'stic-discursive branch. As the nam and his colleagues interviewed 112 ltalian regional councillors in
. . h d qualItatlve-h umam . f f
lzmg branc an a o histicated in the analysls o sta 15- 1970,194 in 1976, and 234 in 1981-1982, and 115 community leaders in
former becomes m?re a~~ :~~~etle5s comprehensible to those who 1976 and 118 in 1981-1982. They also sent a mail guestionnaire to over
tical data (and. the1r wo . es) the latter becomes more and more 500 Community leaders throughout the country in 1983. Four nation-
have not studled the techmqu ' h lyses to the seemingly non-
d f th . relevance of suc ana wide mass surveys were undertaken especially!pr this study. Never-
replicable and nongenerahzable even s m which its practitioners are
convince o e Ir ' . t . the1ess, between 1976 and 1989 Putnam and hiscolleagues condueted
detailed case studies of the politics of six regions~eking to satisfy the
interested. . b k' t show that the differences between "interocular trauma tic test," the investigators "gained an intima te
A major purpose of th1S . o~ lS tr~ditions are only stylistic and are knowledge of the internal political maneuvering and personalitíes that
the quantitative and quahtahv~ l unimportant. AH good research
have animated regional politics Over the last two decades" (Putnam
methodologicaHy an~ sub~a~:I~:!t understood-to derive fro:n ~he 1993:190).
can be understood---:mde: '
"""n'lP llooerlving- 10glC of mference.
Bo th quantitative and quahtatIve The lessons of these efforts should be c1ear: neither guantitative nor
qualitative research is superior to the other, regardless of the research
i
address-ed. Since mdnv nI sntefC::1t h~~ 50GJl SC1-
f lntroductíon

t;:ntists canno! be meaningfuHy fonnul<1ted in \\'i)V~ th"t permil statístl·


cal of witn quantitdtive dnta,. ,Vi.> do flor wish to
encouragt' the exdusive use of quantitative ri'Clmiques, VVE' are not tr."-
tú aU social sdentists out oI thE' library and into the o)mputt~r
center, 01" to replace idiosyncratie eonversations v·;ith struc!ured ínter-
vrevo/5. \v!:' ,1rgue that nonstatistieal researen h'il! prQduee more
reliable result.'i if researehers pay attentíon to the rules of st'ÍenU{k in-
fercnce-·-rules thuí are MJmenmes more dearly stated in the style or
Vr."r1<o;,"IV defined stntisticili melhods that under-

,;¡bstract fomlai modeIs applkable


to aH kínds nf rt:seardl, even that ror whkh vari"bles carU10t bt> meil-
sured quanhtativdy, The very i'lnd even unrcil1istic, nah,lfe oí
statistical modeLs 15 vI/har lTUIkes the rules ni inierenc,,' shine through so
de"rly,
The nlles oí inference that \<\,'12' di::;cuss are not reievant to all íssues
that are of lo social sdentisi5L Many nI' ¡he most irnn<,rL"1
pohtical life-abour 5uch as agency, ob-
dtizenship. and the proper reiation-
betvveen national socicties anó intemational po!itics--are phi1ü-
mthcr than But the rules are rdevant to alI n.:seardt
lhe goal is to learn fads about the real worM. jnda~, thc dis-
tinctive chara<:tenstic that sets soda! setenee apart from casual ooser-
vation is tIMt social saenee seeks to arnve ay valíd i21ferenc('S bv toe
of weU-estaolisheu pmcedures oí inquíry. Our foens
researd\ means that we sidestep lssues in the
sdence as ''.'en as controversies the role of a.n
the nature and existence oi tmth, ilHU fl'.'-
!ated We asstmw tnat ir is possible to have sorne knc)\vledge
oi toe external world but thar 5uen knowk>d~e Í5 uneertaín.
nothing in our set of ruk:s tnat we mU3t run
eXfX:~rímel1t Uf sueh a or coHect <1.11 relevant
w!? can maK€ valió sdentifk inferences. An impor-
tant topic is worth studying even H very !ittl.e infoffi1atíon ís avaHable.
The rt~ult oí uny research design in ¡his siruation wiH be
ul1c€rtain condusíons. out so long as we honestly report our
chis kind of study can oe very usefuL Umited intt'1rmation
tüm;¡ke
15 often a na~t.:ss.'lry feature of sodal inquiry. Because lhe social worl.d
tllat us understand those r~ inforrn,1t1011 about the
Dhen,,,,..,,,,,, afe ,,{ten
to understanó them
neous¡y~ even ,,{!ten "bout cm cnndusions ís high.
gene}' of a problem mar be so tilat data gathered by the
userul sdentific metnods mig.ht oe ohsolete befúre it can be accum~
lated. If a distraught person ís nmning al HS swínging an ax, admi
S TI.e Scúmce in Social Scíence
able t~) scientific n:~~"2n.-'(h, but thc ~}c\:unluidti$,.H) oi ú~cts úhH1t-' is no·t sufü~
If In!roduef1on 9

('ÍtnL Facts. C.JD be f.'c1Ht?cted \lf quantit;¡tí\'l? r-t>~ean::her:s~


reasonabit.~ ~;'Stírnat( oE trnct~rtainty, (~ ,d the ret1I \\'odd or 011
more ot" ¡..'SS systematícaliy. ano lhe fünm.c:r is heiter than th,,·
¡aHeL huí OUT particular defínitíon (>f sóelKe rtC(lUlres the addítíonal ster ¡nferenel! abou! <1 c;nb;d dÚ't'l in l'he f1'\11 world le; unintcrpretabl€, A re-
~¿\~;ljcher \vho L"ül~ tn Lhvt! tht~ issuc (d un(ertainiY i...; FÜner a~~t~rf­
llf iluempting tn infer the immedíatc d,11a lo :;omething brodder
\hat 1:; not directiy ubservoo, Thar somethjng may involve j¡¡
íng tlmt he' O[ she knows everything perfeetl\' 'tl[ [ha! hé "l!' ~¡~e ~:,~ ';1:1
f,:rcnce-.using ob!ierv¡¡tions frmn the world to learn abour other unob- idea how certain or UH,'ért1ín the results "re. Etther wav, inferences wlíh-
servoo facts. Or that sümething .may ínvolve (IlUsa! il!faCl1(c-leamíng ouí uncertai.nty eshHlales ¿¡re not science ,);. we define 'jI.
abnut causal effects from tue data observed. 'Hle domain of inferenn~ can 4. Toe content ís the method. Fina!!;', scientífic research "oh,,'l'('s to " $el ('jf
be restricted in spacte and behilvíor in American election,; mIes (lf inference on v",hích íts vaJiditv Explicating the mOSl im-
sinee 1%0, so<.ial me)'vemt:nts ín F;¿¡stem Europe since 198t:)-··or ít can be portan! nJles is a task of dlis oool<.': The n'l1!eni' of ",\"1C!1<:\"< ¡s
extensíve--h1.llnan beha\'ior sim:e the ¡nventíon of In either the meihods and rules. lwt the ~ubject m,lttef, "ínce \A;P can use
case,lhe mark of sdentifk research Js (he oi rnak- these methuc]:. ro This W,1S

ing infereno2s that go the ol;!sen',ülüus coHectoo. over a C,'nftlrv agu when Kar! Pcarson UtlQ1: lb} e).rl;'l!m~,j lh,ll "¡lw fl,'ld
:t The pmcedu:res a.re public. Scientific resean:h uses co(hfieó,< and 01 scienc" b unl.ímíted; its mah"ria¡ í" endless; en:!\' gn>uP of natura!
methods lo and data whooe reliablHty can there- !10mena, l'verv of 5r'cÍi,! lite . (·very s!age nf i;ll.~t nr '
fore be assessed, Much social research in ¡he qualítdtín:, follows me'nt ig material f(\1' science. Th{· of ,¡¡l sdence consists "kme in It;:
fe\Ver ruk'S oE research proCedufl' or üf ¡nf.:rence. As Robert K merhod, flot in íts materíaL"
Merton ül94911%8:71-(2) pui il, "'1'he of qualilatíve (our íeatures 01' science have il further imnl¡·c·'tí('l'· ' ..""'" u~t
~ Thes"
L,., r ~A. -' l~ "\'J' "u.\,..J:,. ,J A'

,lata üfren resides in a unfathom¡lble lis v€'st ís il S(lcil1i . rL:searcher 01' teitm oi reseilrchers
and inélfabIe ' . is pub- lilbors undel' hmitiltions uf aod a.nd mistakes are
líe, no! prívate." /'vterton' s statemeni is not [roi' oí al! umnroidablc, yet such errors .vin be poínted out others. LTn -
still tme oí sorne but the social ch,:¡r.Kter 01' scíence can be sinee it
¡-mmy had no method-snmetimes as íf the use uf ex- means that out" "york need no! to be an im-
methüds wouki díminÁsh theÍr Nevertheless they cannot portnnt contributinn-·whether tu [he d~"",'''¡''''
but U$e some methnd. St.,mehow ask ques- lo s"
tions, ínfer informatiun alx>u! 'rhe w()rld fmm thes€ observatíons, "ud to redirect) the concems uf
m,,!ke ínfererKi::s "DOU! cause ¡¡nd eH!;'cL H the method and scholars and uses methods tI.> arrive al ínfc'f'
sea rcher' s observa tíons "nd ¡níerenee" are ¡dt ences lhat are ccmsis!ent w1th rules oí sdenee and the tnformation ,1t
the oí what tvas dúne. \Ve cannot it is ,'"
'~k"i:: a ü.ll1lnDutlOl1. 1 contrfbutíun
OUf t)
1. ¡,,<1 i~nu
A 1 t,c
ev,tluate lhe (lf seit:ctivn Iha! wen.' used to n:cord übservatíons, uí even a minO!' ;¡rtide is than ¡!taL P( ¿he wnrk" ¡hat
the ways Íll whíCÍl observathm" \vere and tbe \vhích fOft'ver in 2 desk Jr¡nver or v\'íthin tht' cnnfínes üf a compu!er.
(ondusions Viere drawll, '.Ve (annat ¡eara from th",ir mcthods l'r n:u"","<:
their TL"$ults. SucÍl research is l10t ,1 "el. V\'hether or l10t it makes
":",,un,,"-,' ít ¡s no! a contril::ll.!tÍOH to soda! sóen..:""
1,1.3 Sdl'l1ce ami
explidr or n.oi·-hiwt' Hrnitatíons. Tht: ¿¡dvan-
Sodal sdence constitutt:.'s an aUcmpt to make sense ~!f s{}Clal sítth!u.ons
tage oE b ¡har those limitatlons can b(' understood M\d, if 1'0&- that we as more
. ,'Ir ¡"ss
~,.. \<'~,
vv,. ¡''''L>.
".C UJ h)
" h Oh'"
síDle, addressed. in addition, the methLx1s can be and shared. This ever, tÍlai wha! \Ve perct:Í\"'" as complexlty is not entirelv inllerent in
protess aHows n:::;:;ean::h fl.>sults to br lí'search-
phenomen¡¡: tIJe vvorld is l10t naturall -'v divid'"~'U" I'n"" ' aHúJ cnrn-
ers ;md research :;tudíes in b,' F""I·;,-"¡,,.,'¡ and scholars lo learn,
,1, The (cmdusions are uncertain.
an .1~t rrldr>..-lrÜ~? uf thi' in)~XH'L'¡}'d ~¡:f ,,+{~ienhh(' ln.ft.::ren('~\

prneess, tts
d¡Ü3 tu learn abtYllt lnd{vtL rrUJst ds~t·;-t' Ú1t~t ~·t l\,lrtrpit~t(~ e\h~}tJsti\'e in·'
ís ímpossible, even in prindplt'.
the world tha!

lntroduction . 11
1{} , The s,.-imct' in Social Sdence
pIe, ~('ts or events. On the cüntr.lr\. the p,-,rceíved complexíty of a.si.tu- course of \?yents ,'.'hich Ís altered through modifications in one or more
aílün depends in part on lit)\"; weil we úHi simplir).. ,m..:l ,1U~ 'conditiolls'" ü\·'ebt'f [190Sj 19'19:173), The applicatitm oí t11is idea in
capudty tu simplify depends on whether we can spearv outeomes ano a sy5tematic, scientific wa)' is illustrated in a particularl)' ¿~xtreme ex-
explanatory vi1ri<~bk>5 in a coherent \-vay. Ha\'1.r:g . mort'~ obs:rvatlOns ample of arare event fmm geology and evolutionary biolngy, both
mav iissist us in lhis but is usually lt1Sl.üÚoenL 1 hus ccmpicx- historically oriented natural sciences. Stephen J. Goulcl has suggestt>d
ity'; is ¡:mrtiy nmefitümal an ¡he state (lf our theory., . . that oue way ro distinguísh systematic features oi evolution from sto-
Scíentific methuds can be as valtl<lbie fOT mtnnslCally eomplex chastic, chance events mal' be to imagine what the wodd \VouId be
events as for simpler onCS. Complexity is Hkely lo make OUT inferences like íf aIl conditions up to a spedfic p¡.)int were fixe-i and then the rest
less certain but should nof make them any 1e5..'> scientifk. Um:'ertainty of history vvere remu. He coutends that if it were possible to "replay
and limited data shou1d not cauSE' us ro abandon scientific reSE'arch. the tape off lite," to leí evolution ocrur again fmm the beginníng, the
On ¡he contrarv: the biggest payoff for using the rules of scientifk ín- world's mganísms toda)' would be a completely different íGould
ference CKcun; 'predsely \vhen data are limited, observation tools ~re 1989a),
aTe undear, and rclatíonships are uncertatn. A unique event on vv'lüch students of evolution have recentlv [0-
vVith dear and data" method may be lest? cused is [he sudden extinct1.on of tile dinosaurs 65 million vcars' agt1,
ímportant, since c'l/en partial1y mies of inference may producE' GouIel 0989a:318} "\ve mus! assume thar consciousn~~ss wo~ld
not have evolved on our planet jf a cosmic catastrophe had not
ansvlers that are rol.lghly corred,
sorne complex, and in some scnsc tmiquc evenís vvíth
f
daimed the dino&'1urs as vktims," If this statement is truc, the extinc-
enormous ramíficatÍons, The eollapsc of the Roman Empire, the bon oi the dinosaurs was as imp¡.,rtant as any historÍ(al event rOl'
Freneh Revolut1.on., the American Civil War, World Wal' 1, the 11010- human beings; however, dinosaur extinction dtx>s nor fal! neatlv into
caust and the of Germanl' In 199(\ (.re al! examples of a dass oi events fhat eould be studied in a systematíc, eompa~ative
such events.. TIlese events seem to be the resul.t of complex interachons fasnion througn the applk"ltíon of laws in a straightforward
oí many forces whose conjuneture crudal lo the event having way.
taken plan:, That is, índependently caused sequences of events ~nd Nevcrtheless, dinosaur cxt1netion can be studied after-
forces at a glven plaee and time, their interaction appeanng native hypotheses can b(~ developed and tesied with respect ro their
to brinO' about the events fFfirschman 1970). Ftlrther- observable ímplicdtions, Orw hypothesis to account for dinosaur ex-
tht~e evenís ''''ere inevHable
b
more. it IS oiten diJ{kult to tinctiol1, developed by Luis Alvarez and eol1aborators ai Berkelev in
produds of l;¡r,ye··<;e;ale the late 1970s (VV, Alvarez and Asaro, a cosmk coHi¿ion:
on f,·",,,,,vn a meteorite crashed rnto the earth ¡lt aboul 72JX10 kilorneters an nouL
of our chance oHen creating a blast greater than that from a fuH-scale nuclear \Var. H this
5eems to have played a outside the 01' the theorv hypothesis is corred,. ít would have the observable that
providt>d crocial links in thc of events. iriditlm element ('omrnon inmeteorites but .are on should
Oue \vav to understand sueh events i5 by """,'k:'j",,,, be found in the oí the eartn's crust tha! corresponds to
com:cptuaíiúng eaeh case as a member of a das:; events ahout which st.>diment lnid clown míllion years indeed, ihe discoverv
meaningful ge~eralization5 can be made, This method often ",,~orks 01' iridium at predkted in the earth been t~ken as parti¡~J
\Vdl ter ordinarv v\'ars or revolutions, but some wars and revolutlOns, confirming evidence fer (he Although this is un unambígu-
being much mo~e extreme than others, are "outHers" in the stanstkal ously unique event., there are rmmy other observable implicatíons. For
di.stribution. Furthermore, notable early wars or l'eVO!utlOns may exert one example" it shouJd be possible lo find the metorite's crater some-
írnpact on subsetluent events of the same dass-we where 01'1 Earth (and severai candidates haye been foundl.
think of tite Freneh Re\'olubon-that cauhon is necessarl' in The issue of the cause{s) oí dinosaur extinchon rernains unresoh'..'C1,
comparing them \vith their successors, which rnay be to sume ext~nt although the has genel'ated mueh valuable research. For
¡he product of ímit.'ltion the dass oí E'venís can be usetu1, .~ l-k)t\-'e\d&f.. an aHerrL~tiVt:· that t;~);tinctÚ.H\ \\'d~ (\1tL~~'J t:"v \:',)kanic crup..
but ir is not appropriate. t..,,'ith the preS{~nú,~ ot tridiun1, dnd g<-"\~ms rl1or~ l,)A)!1Sb!ent th~rt
tklrLS f !;>; 31S<J Cf)t'tsÜHent.
Another wa\' scientifkaHv \vith rare' large-scale events is !he meteuri!e hypoth(>$ls wíth the findil1g that aH the specic'5 extim1imb did !lO! (KCUr
to engage in ~ounterfactual anall's!s:' "the mental construction oí a simultarwously.
12 . The Scicl1C'e in Sodal St.ience :-'1a¡or Components of Research Design 13

our pnrposes, the pnint nf thi~ example t..;; that sClentific gc)neraliza- dance \\,ith lt. an tr1\'(~stigJtor hL15 c(,He~:ted dat.;) ~1S. rfOYided ~~y
L)r1Ce
tkms dre usehJl in . CYeH highly t1!1u:;udl evento; th;l! do no( tal] ti researcb he or shc \\~dl o.ften find an in1perh:ct tit aUl,ong U1C
¡ntp i1 large ditss nf evenls. The AlvilR'J: hypothesis cannot be test!:\:! main research quesbons, the tbcory dnd tIle data at hand. j\t thlS o,tage.
wíth reference to a ser uf common evento" buí ii does ha\'e ohsen:abte researchers oiten bet~ome discouraged. They mistakenty hdie"e that
imp!ícatiol1s for orher phenomena th"t c;m be cv"luilred. We should other scientists find dose, immedi"ie fi!s between data ,md re-
nüle .. hmvever, tha! a hypothesis is no! considered a reasonahly ct~rtain Thís perception is fiue to the faet that ínvestígators often take
explanation until it has been evaluated empiricaHy and passed a num- do"\,vn t:he scaffolding after puttíng up their íntellet~tual buildings, lea\'-
ber or demanding tests. At a minimum, iis implications musí be con- ing iittlc trace oí the agony ana uneerta.inty of con5tructiO!:, Thus tn.e
sistent \vim OUT knowledge of (he externa! wodd; at l:x:st, ir shouid pro<:ess of inquir\' seems more met~hamcaJ ;md cut-and-dned than Ir
preclíct what 1mre Lakatns 097m refers lo as "new facts," th"t those acroaUy 15.
fornlerlv unobserved. Some of OUT advice is direded toward researchers who ané' tr:ring to
TI1e poíut is ¡!tal even apparently t~venl" such as dinosaur make cmlnections l:x:tween th0)í\/ ;.md data, At times, they can design
extioction can he sfudiL-a sdentiiícalIy 'Ve p<ly attention ro impnw- more data-collectinn proc01ures in order tn evaluare a
theory, dala, and om use ni the data. lmprovíng OUT theory theory al other times, the)' can tlSe the data tht'v h;:rve and rt,:'Cast
through conceptual darification and specification of var¡"lbles can a tlteÓretical questiün (or cven pose an t.'ntírely diff('renr question that
generare more observable implications and e'len test causal thcurtes was not originany foresl'Cn) to produce a more importan! reseaTeh
uf uníque events sueh as dinosaur extindion. lmpnwing our data al- project. The ¡f it adheres to rules of 'wiH stiH be
JO\\'s us to observe more of these observable implications, ami imprü"- sdentific and produce relíable inferences aaour the vvorld.
ing our use oí daí.) permits more oi these implíeatinns ro be extrac'ied VVherever possibJe, researchers should also improve their research
from existing data. That a set of events to Di' studíed is highly complex m.;i<'llS bdore conductin<> anv field research. I'lO\vever, data has a
d '-.. 17 o ~ h .
dnes not render careful resean:::h irrc!evant \\fhether ,ve study disciplining thoughL lt is exrremdy wmmon lo find t at me
milny phenomena or fe\v-or even one-the study \vil! be improved ¡f rescarch dl'sign fans apart ,vh('n the very first obscrvat1on5 are
we eoned data on ilS manv observable oE our thenry as co!lel::ted--lt is not tnat the throry i5 \vrong bui that the data are no!
suited to answeling the qUi..'5tiOns originally posed. Understanding
from the outset vvnat can and \vhat cannot be done ar this later stnge
L2
can he:!p the rescan:::her al least 50me of the problems \vhen
first designing the research.
""¡P""'" research at lis best is a creative process of and For ¡malytical purpose5, "ve divide al! rese<1rch ínto four
within a weH-estab1isned struttUTe oí scient¡fk the research thl:' IheoTY, the da!ll, the use [he
düfll. These components are nor usual1y separately ¿¡ud
ftlr a metharucal process of U"""''''",,,, and evalua- scholars do not aUend to them in an)' order. In fact., for
the seholar musí nave tue ""AH"'''"' o! mind ro qualitative researehers who begin their neld work bdore choosing a
oi looking al the \vorkt ro ask new questions, ro precise research question, data comes follO\VL-a by the ot~ers.
revÍse research desüms appropriately, fu,d then ro coHect more data of Hmvever, this particular breakdow'n, \\'hkh we explain in sect:ons
a diiferent l:ype than inlel1ded. However, if !he researcher's 1.2.1--1,2.4, ís particularly usel:ul for understanding the nature (~l. rt.'"
findings are to be valíd and aceepted by scholars in this field, aH these sean:::h designs. In urder to darify precisely wnat cvuld he done tí re-
revisions ana recol1siderations rrlUs:t take accordi.ng to expíicit sources were redireded our ad,dce in the remainder of this sedion
t

consistent vvith the rules of A dyl1amic process as..sumes that researchers have unlimited time and resources. OE
ínquiry necu"TS \'vithin a stable structun" oi rules. course, in any actual research sltuahon, OT\t? must always make eom-
&xial sdentisls often research \vith a considered dt,sign, co1- promises. W~ believe that understandíng the advice in the four cat~,,­
lect some data, and drav."' condusions. But mis process 1S rarel}' a gories that fol1ow will help n."Sean:hers mi1ke these compmmlses :n
smooth une and ís not !x"'St done in this ordcr: conclusions 5uch a way as ro improvE' their research most, even \vhen 0"1
rareiy foHm'\' easiiy from a resean:::h design afld data coHected in ateor- fad thdr researeh 15 subject to external constraints.
The SOalce in SoCÍal Sdence Majen Components of Rescarch Design 15

121 ro beemnc J sZlCial scientist ;:lnd, ¡ah'!', ro cho()se a particular rt~Sblrch


questiOlL As such, thc\' 111.1)' ((\nstltuk ¡he ",hll" ~f\b,mS [zlr
Ihroughout thís book, \Ve con.síder what to do once v;e ídt:'ntíÍ\' ¡bt' in: ¡} particular rescarch pmject-and approprít'ltdy so, But, ll(' matter
object of researen, Givcl1 a rescaren questi(lT), what art: tlw \va~'s ro ho\\' persona! or idiosyncratic the re"son;; tor choosing ¡; topic, the
conciue! tbat rescaTen 5(1 that We can obtilin "dUd explanatíons nE ~Ociá¡ methods of "óenc(' and rules of infere!1cc díscu~sed in thís book wíll
and politic,11 plwnümcn" 7 ()m Jiscussion begins with d research qU("S- help schülars ..h;'vise more poweriul rescareh dtesígns, Fmm ti,,"
tlOn and then proceeds in the stages oÍ desÍgníng ilod eonductin'"
~ L. t"'f
the spective oí a potential mntribution to social personal rei1-
rest,~arch, But ''lhere do rescaTen yuestiol1s originate? How dOes a sons are neilher necessary nor sufficient justíficatítll1s for the choice
scholar choose tn/:' tapie for There is no simple ans\ver to thís oi a tüpíc. In most cases, they should not appt.>ar in om senolady
question. Like others, K"d Popper (1968:32) has argued that "tltere ís writings. Toput ít most direcHy but quitt;> ¡ndelicatelj~ no 011':: nm~
no suen thing as a logieal methud oi hi1ving !lew ideas ... , what we think-the schnlady communitv caH:S H'hat \Ve can
eontalns 'an irrahona! element' m " 'creat1ve intuition.' u fhe nlies ni aemonstrilte.
at (he earliest of tne I'eseafch process are les" rules fUi choosíng ,) c!G not exisÍ, th<c'tc are
th"n are the rules ror otñeI' rL'Se;lrch adivines. filere are texts on Jc- individual detern1ining thc
signing laboratory expe-r¡!t1I::'nts un social statistical criteria 1m of a research ¿'nterprise io ¡he scholarly community. IdeillI:v,
drawing a {or " sun'cy oi attitudes 011 pubhc policy, ana man- aH rt.'Semch in the sodal sdences should [\'.'0 critt'ria,
ua!s on eonducting participanl observaoül1 of a buream::ratk offke, But FirsC a rcsearch 5!n)uld pr)5C a tha! i~ in the rC;J!
there is no mJe foI' 'whích researeh pwject tu conduct. nor if The topk should be conset.¡uential inr política!, sociaL or eco-
we should decide to w()rk, are theI'e rules governing l1ümÍC ior understandíng something that affects
\vhere ,ve shüuld conduct it rnany ¡x'oplc's Uve's, or for understanding and predíctíng eVl:'nts that
"Ve can pro pose ways to select a sample oi comrmmities in order ro might be harnlful or beneficíaJ Shively 1990:15). Second, a j-""''',,,.,,,,,
the impact Di alternative educational polkies, or ways to coneep" project should makc a mnfrióufiul1 tu un
ethnic conflict in a t11anner conduclve to the forrnulation and flm: out c0llecfin'
oE as to its inddence. Hur there are no rules tnat tell ftilfíans af the woríd. Tnis l¡¡Her crlterion does out imply
educatíonal poliey or ethnic conflict. ¡n terrns off !hat al! n>seareh that mntribl.ltes lo our stock of soda! sdenee
sodal sdenee tnere are bet!er ana \VOTSe '¡.vavs ro the natioos in tact ¡¡ims at makíng Gms;¡l inferences, Snmetimes
of ¡he Easl German govemment in 1989iust as -thert' are hetter the statc oí knO\vledge in ;¡ field is sl1ch lhM tnl1ch ilnd
and worse ways ro study the be!\'\'een i\ C;1n ..",.-." description ls nt"f'\led bdore \Ve can take on the ehúIlenge of
tion on laxes ~md the lik.::lihCK)(¡ ot electoral suecess, Sur there 15 no nation. Often ¡he úmtributíon of a pl\)¡t:'ct wil! be descríptivc
\vay to dettlrrnine whether ir ls better lo the of the East ínierence. Sumetimcs the may no! c"cn be descriptl\?e ¡nft~rence
reginw 01" the role ni taxes in U.5. electoral politícs, bui rathel' will t'X' the dos!:' oDser\'ation of partículilf evcnts <JI' ¡he 5um-
Ihe specific topk that a social "cienns! studíes may nave ¡} personal mar)' of his!orkal detaíL meet OUf second críterion
and idiosyncratic lt is no acddent that reseárch on particular because are prerequisites to
~'oups 1S likely to be pion(~ET(,,(] by people oi that group: 'vmnen have Our firsi criterion directs our attention to tl10 rteal worle! of poHtics
oHlm led the ,vay in the history of women, blacks in the historv oi ,1nd social pbmomeoa and to rile current and historical re('ord (lf tht:,
blaeks, immigrants in the oí immigration. Topícs ma)' aJ~) be ('vents and problems Ihat Jives. VVhether el rescaTen
mfiuene€l.i by and Vah.1é'S. The student ()f ¡hird- question meets this eritt:'rion is a sodetal judgment. The "ec·
world polines ¡S likeJy to hdve a dL'Sire for trave! and a ond criterion direets our attention Í<) the literature uf
tolerance for difficult living condítions tnan the sl:udent oí' mngres- sderKe, to ¡he inteHeetual puzzlcs m)t püsed, lo th;,t re-
"ioIlal polky m¿lking; the annlyst of intemational eoot't'ration '~ay maín to be ,1l1d to rhe scientific theorics ¿md melhods 2\'ajlablc
nave a particular distaste for violtmt conflícL ' lo süIn' them.
These personal ¡ma ,'alm,'5 "iten provide lile motivation
16 . The Sci¡'7¡ce in Social Sdeu<."'e ivlajor Cornpüncnts of Research Desígn 17

mt~ets OUT first criterÍon. Ten major wars during tlw last fOUT hundred ~ C:-h<)t)st.'> dn hyp~)tht~sis in the htC'Taturt~ that \-\.~(~
llave killed almost thir:y mi1lion peop1l,' tLevy 19S5:372j; snme \{;fone \.n~ h.::'¡ievt' has nüt l:wen
"limltt.>d \'\'iUS," such as those between the Uníted States and North whether it ís inJeed fais,e or vfhetlwf s"me otht:'f theor\' l.'i n'frl'CL
Vietnam and between lran nnd fraq, nave eaeh dainK>t1 over a míllion 3. Atternpt ro resolve or pren.·id .. further evídenct' ni one síde ni a contnr
and nuclear war, were it io oeeur,. cOUld km billions of hum,m verSy in the rUeratun.::.it-perhaps dt~n1iH1str~1~} that the (pntro\'crsy i/'/3S
being5. Polit.ical mismanngement, both domestk and international, has unfounded fmm the sta.rt.
led tú economic prívation on a global basís-as in the 19305-as well 4. Design research to iHurninate nr evall1ate unquestioned assumptions in
as to regional and local depressl0n, as evidenced by the t:ragic expe- ¡he lítemrure.
riences of much of Africa and Latin Amerka during the 19805. In 5. A.rb"U€ that an importallt topie has heen overlooked in the literature and
generaL emss-nation¡11 variation in political institutions lB associated then pro~"t~d to cpnITibut:e a systematíe studv 1'0 the area.
vdth great variation in the conditlons of ordinarv human Iife, whieh 6. Show tha! Iheorit'S 01' evidence for sorne purpose íll un\é' lih~ra-
are reHectcd in differences in lite and infant mortal1ty be- ture could he applít>d in another literJture tu solve "n but appar-
tween countrít.>s with similar le veIs of economic deveJopment (Rus5ett cntfy unrebtL4:1 problern.
1978:913-28), Within the lJnÍtL't.1 programs designed to aHeviate
Focusing too mueh on making a r:ontríbution to a litera-
po vertYor social disorganization seem to have varied greatly in their
ture without some attention to topics that have real-world importance
efficacy; It camwt be doubted that research \vhich contributes even
nms the fisk of desn.mding ln politicaUy Con-
marginaHy to an understanding üf these ls5ue5 is ímportant.
versely, attention to the curren! política! to i;:,;-
\Vhile SOt-:iaJ sclentists have an abundanee of sígnific.cmt questions
sues of the amenabi¡ity of a subject to syslem'ltlc study within tht.?
that can be im/estigated, the r001S fnr understanding them are &Caree
framework a body of social sCÍeI1ce knowledge leads to careless
aOO ra!:her crude. Much has been written about vvar or mísery
..vork that adds litile to our deeper understanding.
that adds 1mIe lO the underst.md.ing uf these issues because it raíls
Our two erireda for choosing research questions are not
either to describe tht'Se phenomena systematically or to makl~ valid
in opposition to one another. In nm, understanding
causal Oí descriptive inferences. BriHiilnt insights can contribute to un-
'''TOrId phenomena is enhanced by tile generatiol1 and evaÍuation
derstanding by yielding new bUI brilliance is
explanatory hypotheses through the use of the method. But
not a method of .1\11 hypotheses need h) be eva1u-
in the shorf term, t11ere mav be a contradiction betvveen use-
at€'d empirically befor(' Gm a contribution ro knowledge.
furness and kmg-term sci~ntifíc VeliLle. For instanee, \1ankiw (990)
This book offers no advice on """'"'.V.<HH brillíanL What it can do, how-
poínt's out that maaoeconomic theory 2nd applit'd macroeconomics
ever, is to the importanee cnnducting researcn so lhaí tt
divergeti sharply during the 1970s ,md 19805: modds that had been
eonstitutes a contribunon to
ShtH\'Il to be lheoretical1v incoherenr "it~re still llSt'd ro forecast the
OuT secand l"riterion lor a research question, "'making ti
direction oí the ES. • ,vhile the rte,,, tht:oreticaI de-
contribution/' means exp¡icitly a research des¡gn \vithin the
to correcí tuese flaws remained speculative and vI/ere not suffi-
framework oí the existing sooalSclentific literature, This ensures that
dentIy refined to make acrurate pretHetíons.
the investigator understand the "sta te of the art" and minimizes !:he
The criteda of practica! applicabiIity to the real world and
dü'lnce or duplicating what has aIready b(~en done. lt also
tion ro to one another when a
that the work done \,;'ilJ be to thu5
,dI begin with " rea!-
success oi' {he of scholars h,ken as a v\'1101e.
¡ht:> ¡hrear uf
contribution ro the líterature can be done in mi1ny diiferent \"'<'a1's.
l:}t~twf"{'>n men and 'VOlTH"I1.. ¡he lransition lO de-
We líst a few oí the here:
Others may stalt \vílh ,.n
1. Choose a thé soda! science mcrature: .1 coniradiction between
studif's of under llr
an ínconsistency between thcories \oting and recen!
contribution. election outcÓmt>s. The distinct10n hehveen rhe criteria oí (ourse,
19
18 The Setene!' in Social ':Xiencl:'
not hard ilDd bsL Snme rese,y¡ch questim15 Siltisty both criter;a tn)!1) Eter;):url' ",houkl ~in1jL1r1v l~e (h-dnht~d, f'L1\'-
the but in n.:.~st.~Jr('h, ft..:searchers oJtcn ing cho:sen a \\'« ente!' il "n/ith tht:: litcrál.:U[\.:',
nearer (me than the other. 4 What questínns oí interest to U5 have aln,;\ldy been aH5wert. j? Hüw M

Whercver it begins, the ¡m)ccss of dt?sígníng rescareh ro ¡mswer a can we imd refine out' question su ¡hal lt seeu1S capable o)
spedfic questlon 5hou1d mc>ve h)\\'ard the satlsfaction of OUT two c.ri- .. ~,."-m·,,d witn ¡he iun!s ;lYdildb1t<> \-Vl: ma\' "tart v;ith a lS511t',
teri,L Ana obviously our dírt'ction of movernent wíll depend on wlwre but we wíH have to curne te h"th ,,¡th ¡he literature uf
,ve start. Lf we are motivated by a sodal sdentific puzzle, we mus! ask science and ¡he problems ni inference.
ho\v io make thaL research topie more relevant lo reaJ-\Norld topics of
signíficance-for instanee, how might ¡aboratory experiments beUer
illuminate real-world strategic choices by poIitica¡ decísion-makers nr,
what behavÍoraJ cousequences might tbe theory have. lf ,ve vvith
a rea!-world \ve should ask huw that problem can be studied
\vith modem scientifíc methods so that ir contribuíes to the stock oí
sodal sdencE' Ir be i.hilt VI'e w¡¡¡ decide tha!
too far from (me criteion or ¡he jo; not the mn5r truitfu! approach.
Laboratory eX¡'?erimenters may argue tnat the &earen fnr externa! refer· the
en!s 1S premature and rhat mure progres" \vHi be made by refining the- such an oxymown sh,mld no\ be caBed ,1
urj' ana nlethoJ in the more controlled environment oi the laboratory. 1991:4; see a150 \Vmx..15 ,md Wallnn 1982).
i\nd i.n terms of a long-h:,m1 rescareh tney may be righL 'file development ot a as tlw fir:'it 01
the schoLar mot1va!E'd a real-,vorJd pmblem rnay argue It 50m.etimes comes in but it ne.d nor. in fac:t.,
that aecurate description 15 needL,(l before moving to explanatiOl'L And we cannot a theof)' ,vitlwut of prior work 011 tlle
sueh a researrher mav also be right Ac"Curate description ís an impor- and the 'lince even tIle re"earel!
tant step in researrh pmgrams. don \vould \VhiÜeVer d!110unt oí'
in either case, a research ana íf possibJe a specifk researcn data has bL"CU there are sorne
pmjecr, should aim lo our two aiter¡a: it snould deal witn d ate ilnd the usefulncss oE a theorv. VVé
rt'al-world to contribute. or of these here but SilYe ¡¡ more detailed discussio)l ior
to i1 ¡iterarure. Si.no: OUf main concem in choose theoríes thdt could be wmnf!;. indL't'tL more 1"
this b(,ok is research more scientiflc. we wHl learned from theories thar are wrong rhan frO!'n theories th"t ,¡re
who sta.rts wilh the "real .. v,'orld" pt'Y- $(1 lhar could not bto' \vrnng I;;~ven in : \'Ve need ro
of a din.:'Ct answer ro ti",.' VVhdt ('vidence
rathel" ¡han wilh COl1\'lm::e us iha! Wt' are H lhere 1S no anf-'wer ro thi;~
literature. it 15 ,,;ssentlal to devise ti workable plan fOi then v;e du no! have a theon,r
fhal (anno! bt into t1 re5carch ;::hnüse one that is capil-
¡Ji Ciwsal 3hould be as Thjs
¡ha! will make nO contri- choice váll allenA' more tests nI' the \víth more d"ra ¡¡nd a
of will the al risk üf falsifi;:d nwre tunes,
"nd wiIi make it L"o'~"'~''':: to cullect data 50 a~ ro tmíld
r'nr the
20 . TIw Sdcna: ín Socia! Sdence Maior Components (lf Researdt Design . 21,

Tllin.1, in the~)rj~"::--:, 1~~~ ~~s cnnCft.:te ¡)S possible~ \~Jguely /\11\ schobr CHl úHl\C up \\'íth ,1 "plau:.iblc" 111,,>
st<lled theor!e,; ,md hypüthesl':-' SelY,.' ll\l purpnst' bUi hh)biuscah?:. The- for <"In\' ~et (li data alter the ¡"eL tti di) so (k'm()nstfdtc~ 1k)(h-
Orlt'S th,1t are static't.i precise!y ilnd make specific prtxiictions can be about lhe verad!)' uf the theorv, Th0 thcory wil! fi¡ [he data
shown niore ro be \VTong ana dre therefore beUer. still Inav ¡x:' wildh,' wrong-índeed, demonstrahh' h'mng ,,·¡th
Sorne researchers recommend follmving the principIe of "parsi- other d:lta. llurrul1 al'(' verv aí pattdl):'
rnonv." LJnfortunatelv the word has bcen used in so manv ways in
I
t very at recognizing nonpattems. (Müs! oi' us c':en se\:
casllál conversatinn ~nd schülariy \vritings that the principie ha~ be- .,',._~...',....," in 'random ink b¡ots~J Ad hoc adjustmen!s in a theory th<"lt
come obscured (see Sober [lQ881 for i\ complete discllssion). The dear- unt fit existing data IYtUst be usted rarely imd \,,:ith considerable
est definition ol parsimony V'<1S gh'en by Jeffre~'s (]961:-t7): "Simple discipline.8 . .
theories nave highef prior probabililies."7 Parsimony is thcrefore a rilere is stiH the problem of Wh,lt lo do when ,ve hilVe fnushed Otlr
iudgment, or even assurnption, about ¡he nature of the ,vorld: í! ís as- coHe,~tiol1 and and wish tn work on improving ¿¡ theory,
sumed 1(1 be simple, The of theori",s thar a In this sittwtion. ",,'e fpl10vdng !wn rules: if our pre-
simple \yorld is a rule thar applies in situations ~'\'her(' there is ",',,-.>,,,,,, 1S condítional on variables and we ilre willing to drop
" high degree oi certaínty ¡hat the workl is indet'Xi simple, Scholnrs in oue oi the condití¡)ns, ,ve may dü so, For example, if vve hypothesizt"C1
physics seern to find parsímony appropriate, but ¡hose in biolog.\r often oricrinallv tha! demoCTatic countríes with social vvettare
think oi ir as In tlw soóal some forcefully ddend par- te;;'s do' not fíght ench other, ir \úmld be 1:0 extend that
simony in theír subfields Zdincr 1984J, but we believe ít ís hvporhesis lo aU mndcm dcrnonades and thus !:'valuatc our theor)'
occasionaJ!r appropriate. Gi\'en the precise ddinition ni parsimony as against more cases and íncrease its ehances ot being .. '
an assumption about lhe .,'ürld, ,>,ce should never insist on parsirnony püint is th¡lt <lfter the we mav modi!y our 111 a
as n uf !ht'orit~s, but it i5 usdul in ¡hose 'way that make$ it apply to a of phenomena. SinoS' such "n
situations wherc \\'e ha,,€' sorne oí tIte símpHcity of the aJteration in our thesís it mon, fuUv to modifíca-
'levorld 1-ve an' oon in l:his direction should not lt~ad to ad hili: that
Our poínt is that we do nor advise researchers to se\:~k parsilTlony as merely to "sdve" dr1
an essential good, sínce there seems HUle reason lo adopi iI: unless ,ve ro phenomena that llave
alrl'adv know ajo! about a sUbiecL We do no! even need tn The opposít{' practke, ínappropríate, AHer ob-
dvoid since ít is directl.v implied by the data, we should add a restridi\'t: conditíon and
[he maxim as aH our ~~w,.·""A as j¡ our th,,'ory, with ihat qualifkatiol1,. has been shown
evidence relative to the to be corred, lf our oríginal" \vas that modern do
eS¡:lgi:Hell can ¡ead ro what \ve call not fight \vars wílh one al10ther due to their conslitutional ít
reS{~arch (see SL'Ctlon 4,1), out [hese are prob· \vould be less found lo ,mr
lems nf research and not assumptions abour the world. . restrict the pm¡:X1sition to dernocracies with advanced
/\.11 our advice ¡hus far applies ir we han: no! coHected our data once íf has bccu ¡he data tlmi such ti
and begun imy Hmvever, ifwe have gathered the !o make PUl' corred, 01' suppose that
\ve can certain!y use these mies to modify Ollr and gather \vas that ren>lutions only occur under
He\V data, and thus new observabie of the new oi severe econornk bUÍ we find that this ís not true in one

Of CTlUTse, this process ¡s time consurning, and oí our case studies, In sítualioll ¡¡ wnuJd flor be
\'\'ilsteful nf the d¿üa collected. VVhat then about the ro ¿¡dd general such as, rcvolutlons never uelllr
situatíon where our IS in nbvinus net'tÍ oÍ but uds (lf ,... rn"npntv \vhen ¡he milítarv Ís the
\ve cannot afford in collect additíonal data? Thís sltualion-íl1 \vhích ershíp is "P1'''''''''''''' the f.'Conomv is bast>d" OH J smiill
- H \\"(1- thn'(' (hos~;n ~1 k~pk' <Ji n\11·\\'orki irnpnr:~uK't' ~~nJ '~H' ~l(W \x,'h~~:h rnake~ :'-'l~~1H"
(ontribution ro ii scholadv htCTJttlrt~t the ~;ü('~al natuH::: 4,)1 dliH..leU1hl VI- HI corn::\ct thÚ1 snu-
. This has comt' (o l:w KnUWl1 dS the ¡¡tioa: $Olnt,'tme wil! rq:¡j¡~i11e our ,,'¡lh Mw!her ;¡eí 01 ¿¡¡ti! ,md uelno!l"t'r;¡!e UM!
cnncept ís similar tú CXcanú razor. we wefe wrong,
,'an~l111e'-clirn~1te
i5 \\-:1DT'L Sucb ,1 f{~rr~1uL1ti~)n í~ Iil(lfC!Y ~)
_misleading) Vi/ay o f ' m y b correct, e:xcept in (()Un-
Since we nave discovered that our theory is lncorrect
fÓÍ"cmmtrv x, ji dQ€s nor help to fum thís ialsífication into a spuríous
generaiization. WiLhemt éforts to cnl1ect Dew we wil! han? no
admis$ble evidence to support rhe new versíon (lf tht, theory.
So OUT bask rule "vith resped to altering our theorv after observing "Data" are svstematicaHy collected dements of ínformation ~bout the
fue data is: me am ma/(¿' the theof!! les;; rcstridivc (50 tf1l1t ii nruers a broadcr world, They 'can be qualiíative or quantitative in style. Some~lmt'S d,lta
nmge pltenomeml üud ís expDst'd lo mure Dpporhmities {or fl115ificatimrJ. are colleded to evaluate a ver}' specífic theory, but nor sO ínt:-equently,
[mI me s!u:ruld flof maite rt more rc::;tricfi,w WitilOut CO!!ecfing new data fo test scholars col1ect data behm.' knowing predsely \.vhat they are mterested
the neu' Vi?fSiOft the tl/f;ory. lf >ve cannot collect additíonal data, then in finding out Moreov.:'r. cven if dJtil ,me co!1edea ro evaluate a ~pe-
we are stuek; aud .ve do not pmpose any magica!way of getting un- 'f' - h"nt~thm;;l'"
oH: 'r-' ", .. ::>, ma\'•
ultimatelv• be interf'sted in questlíms
stuck. At some point, dedding thnt \Ve are wrong lS indeed, nega- that h~d nor üccurred to them prevÍnus!y.
uve tindings can be valuable for a scholariy literature. Who In either case-,when data afe gathered for a purpose or
vvould not prefer (mE' finding o',[er any number nE flimsy when data are used for some purpose not dearly in mina \vhen the)'
positive findings based on ad. boe theories? were gathereLl-certain rules \'\tíll impmve the quallty 01 those data. ln
!',4oreover. if \Ve are wrnng, we !'leed not stop \<vriting after admitting princtple, Wt' can think aDOU! these ~ules fo~ ¡mpm\·ín.g e b t a ,
defeat, V'v'e muy acid a section to our artlde or a chapter lO our book [mm the rules in section 1.2.2 fUf llnpmvmg theory. ln . am
about tuture empinen! resean::h and curren! rncclH'ticaI SpeCUL"tion. ín data-colh."Ctíon efforl requires some (lE thcory, just as formulat-

this contexto we have consíderab!y more fn.'Cdom. We ma)' 5uggest ad- lng any sorne data (see . . 19(4): ' .
ditional conditions that be attnchea tú our theorv, ii Our first ¡¡nd most important tor unprovmg data LJtIahty
we beileve they mighr solve the probIem¡ propuse a modification uf i5: record and fi'lmrt the ;mxess whid¡ ¡he data are \Nithoul
p",id·",.,n thcüry nr pmpose a range of entirely different theo- this information we cannot determine lvhether standard pnxe-
lve cannot conduae anything ,vHh a great deal of dures in analyzing the data \viH produce biasea inferen~es. On1y by
o?rtainty perhaps iha! lhe ,ve stated nt the outset is knovdng the process by whích the data were generated W1Jl ,ve be able
out \\'e do have Ihe of !lev\" researen tú - valid nr causal inferences. In a
ü:r data-cnllection that could be used to dc'cide v\'!lcther our opinion polI. recording the data-generation pmcess t!,at ,ve
speculations iire COlTié'ct These can be very in sug- kmnv the E'x.¿¡ct method v\'hich the v;as dr:nvn and ti).:
oP;:.ti'"v areas vvhere futun: researchers can !oo!c dile questiofls that were asked, In a qualítative {',lse
as \Ve discllssed sodal sdence aoes not studji, reporting the mies ""hiel: v;e choose the smaH num-
to rules: the need tor sometimes mandates ber nf cases for IS critica!. We addlt10nilJ in
that lhe textbook be discarded f And dar" can discipline thought. d1apter 6 for case in qualitative but even more im-
llenee rescarchers wm somebmes, after data, have i.nspí- portant than choosing a good method is being C~rel111 lo record dnd
rat¡ons abour loen-v they should have consfmcted the theorv in the first report ,\-'hatt'ver method was lIsed and aH tl1(> inrormatlOn neces'k1fy
Such a moditkation, even if restrichve, may be \vorthwhHe iE we for someone else lO ¡¡PpJy it'l
can convinee oUrSt'fVes ánd nthers l:hat mcldífying the theory in the in section 1.2,2 \ve for theories th,ü are df
\vay that We' propuse is h'C could have done beiore vve co1-
¡ecred the data ¡f 1ge had it, Buí unhl tt.'sted v'(üh 11(,,1' Jata.
the status uf such a \ViH remain ver\' tmcertain. and ir should be
labelcd 35 such.
Onc COlbcqucnce off ¡hes"! mies Í" !hát
¡jiten \tT\ useiuL "S[!l:"i. m 't's«'úrch \-';hctt~ datá must be
or other l1h?¡ms. Preliminary datil-
questíons or modHy thc
24 The S.:it'lVf in Soda! 'Sci('!Kt' \1ajC'f ('ompont'TIts nf l{t>seilrch Design 25
()Uf ~> \' t: ~.,.
fr'l ...... ~:)n \\'hi·ch deterrt:n()'<
C(}(lper,h.l ~!.. ) l.,
v
" , .
data qunHty is in un!{"T l~etfL'r .f.{, e;\¡h-i{¡!t~ t; L'c:lI~'(t di.1'fa eH a,o.: lJhIU'i! sludy 5uch pmblcms 35 cnef\' mio market> dmi
of its (ll'St'TUablt, . This me3rtS cullectíng as muci~ (Fud~nbt>rg and TimJe jqK,n, Ci\'en l.he elose sim¡¡ilr1ty ¡'t>tween. tite
aata in as nMn\' diverse wnh',ts as possible. EdCh addit1c;nal ímpli- theories, empírical e\"idence supportíng game lheorv' s predKholls
cat1cm of our th00ry which wc "bsern! pw\-kh:s aHotlwr cnntcxt in about firm behavior wauId ínCTCa5t' ¡he phnlsibilitv oí reJ,ltt'cl hypoth-
which to 1..'\'all.1ilte Í(s The more ObSi:'fVilblt, implicatkms t.>SeS aboul stilte behavíor in íntemation,)j politícs, Uncertainty \A:ould
which ilré hmnd to be COI1SLstent \-virh the theory, the more powerful rema in abuut the apphcabílity of concluskms tmm Ol1e domaín to an-
the explanatíol1 ami tbe more ccrialn the results. otiler, but the issue is important ennugh lo \<V<1rrant attempts lo g,1111
When adding ddta on ne\\' obseJ'yable irnplícations oi a theorv, vI/té' inSight and evidence \vherever they can be found. ,
can la) coHect more obscrvations on the same dependent variabíe, or Obvioush~ In C(~llect data forever without doing anv arhlJysis wouki
lb} record addithm<ll dependent vari"bles, VVe can, for inst.mce, dis- precJude rather than faciJítate complctioll oí useful rescarch, In PfiK-
to shorter time perinds nr ,1reas, We can rice, iimited time and resoufees wil! ahvays cOHsl:rúín datú~wlledion
collect ínt'orrnatüm on dependen! variables of dire{~t ínterest: efforts, AHhllugh more informatiol1, addítional eas,:;,s, extra ínlcrvic\''''s,
lf the results dre as t1w pn:dicb, we wíll hin:e more confidence anothcr afia other relevant fonns of data collection \vil! al-
in thE' improve rhe oí OUT inferencps to sorne dcgree, prornis-
For cOTlsider the fational tneory: potentíal ínití- l'lOtential ;;chola!':: can Ix' ruined bv too much inforrnatÍon as
ators ot \varfare calculate ¡he cos!s and bendits of attacking other as too little. Insisting on rcadíng another bnok or slill
sta tes" and rilese cak'ulatlofis can be influenced bv cn:dible threats oí one more data sd wit110ut ~'ver \vriting ti w(lrd ís a prescriptíon fur
reíal¡.,tiofi, The most dirt'C't test oi tnis ¡heorv ~"ould be ro asscss unproductive.
threats ol ,var, dedsions to atta~k are assocíakd h'Hh OUT third guideline is: !lltlximi::c tile oí' om mc:vUn'111cn!,;, 'v'a
suen factors as (he oí militarv forces behveen the pütential ato liditv refcrs ro what we think \-ve are measuring, The unlL·m-
t.,cleer and the defendt'r nr tht' ính,rests ai "take fm ¡he def~nder (Huth pioyment rate nM)' be a indiz-ator of the state uf the economy, !:lut
1(88), even cases in v.;hích threats Me the hvo are not In 1t is cask'Sf lO mdxirnize vahd·
íssued corLstifutes a set oJ implimtions of the theorv, thev ildhering to the and nol alknving unobserved or unmeasur-
¡,¡re on1y pan oE the observatiollS thi'lt cnuld be gathered (¡ul¿i use¡i in ¡he wa,', ff an informant re~ronds lo our queshO!l
"lone mal' ¡cad to selectíoll :;inee sítuatíons in Vd1ich ¡hreóts th~n we knO\\' he Si/id tilat he was Oí
tnemsclvcs are deterred vn)Uld be exduded ITorn lhe data seL 1-lenee it that, \ve have a ';aJid 111CaSurement VA1dt he mCimt ¡s
might be tvorthwhile aIs(, to colled data Oll an additinnal de'PE~n(lel1t rm altogdher different that canllo! be measured ,,,¡th <1
variable a ser nf bdsed on a high (lf confidence, For in countríes w1th
measurement of whether mdy be a W2y oí
incentives tü do so, poiitkal staternent fm sorne rol" it is <l ,vav nf
Insofar as suffÍLlent datd ()fl dcterrenee in intern¡¡tíonal """"<"'''_~ donft kno\'\',"
is it could al50 be helpfuI to test a different nne wiih Our fourth gu¡dehne is: msurc tlwf daia-cof1eclion mcillOds ¡¡re rdiaMe.
similar motivational rOl" a differenl dept:'udent variable ReliabHitv mean!; that app1ving the same in the same WiW
under different conditions ,,,híeh ís still an implication wm' the S<1~,e me,lStm:. \'\/11en ;¡ relíable pmcedure is
of the sume For ex- applied at ¡imL'S and nothing h<1s in the meal1tlme
r""'TH1'li"''H to see undcr símul,lted "threats" are dc- to the "¡me" state oí the object \NE' ,lrc the san,e
t0rred rather ¡han iKcentudted power and firm b,lrgaíning result 'will be obscrved, Rd¡"blc measures dls<'
(Jr \n~ coulJ exam!r'le v-'hether other actors in e sit:.
sueh as flrms eompeting for milrket share or 01'- \Ve can cht."Ck Z~Uf~f+;\.~5 b\' HH!d<:;,t.H'ing th(~ ".1nh:·~ q\ktntlty

{;:¡rnilies (or USt' deterrence


wheth1.... r th{~ !n(\l;-;U.r('~ .-1-rt' tht~ <a11L Su~ctinH.>~ thi~ ~-t't:'rn:-..~ -C;k";~:< ~u·ch :,:"l~
thl> ~H11C qHt.'''..t¡n!'', .1t J¡tÚ:H:nt ;l~th"> ,juring
ane! 110\'1/ successful are undel' rndt'ed, econo' iH'h...\.-~ lHay ind~.h'·n\.\.-' th",' tu
misís working in the fieid of industrial orgilni7A1t!on have used nor¡- ~n V;'t~ nt>k~d in be (,an~fú!
Major Compünents oí Resean::h Desígn 27
26 . Hle SCÍi:nce in Social Science
','c sh:dy is thc ",-wiologic11 study ofMiddlett'wl1, Indiana,
suJts when applied by dlffert'nt resean::hcrs . and this outcome
by Roberl and Hden LV1KL Tht,ír fir;;t "I\1iddlettl\-vn" 5tudv \\';1::-
F-~end~-, ot coursc" u¡Jon theTc bcing c'plicit procedurE's that can be fol-
. in 192Y ;md was ;-{'p¡katt~d in a book publislwd Ín' 1937.
lo\\'ed. i1
fifty yearr; after the original stmlv, a long series ni books and
Our final guíddine Ís: al! data ami should, as possible
are published !hat n:plicatt' these NigínúJ studies bee
be Replicability ilpplies not cmly lo data, so that we 'can ~
t~t aL,1983a, 1953b ;md tlle dtatiüns thereÍnJ, AH qualítative
whether oul' mensures ilre rehable, but to rile cntíre reasonin<> process
net.--d not be this extensíve, but this major resean:h projed
used in producing condusíorL". On the basis of our resea:rch ~eportf
serve as an exemplar for 1Nhat is possiblc,
ne'v resean::her should be able to duplicate OUT data and trace the
research should attempt io achíe,-e as mudo replicability as pos-
?yv\'hich we reached OUT condusions, Replicability ís important even'
SChOJ,'fS should alvvays record tbe exad methods, mIes, and pro-
tí ~~: :me actu~lly repHcates our study~ On!y by reporting the study in
used tn gather ínformatíon and draw inlerences so that an"
surÍloent detall so thilt it can be repikatcd ¡s jt possible to cva!uaie the
can do the <¡arlle thing and draw (one hopesJ the same
foUO\-'i'ed i1nd methods used,
sion. Replicability also means that scholars \\'ho use unpub-
Replkability of data may be difficult or impossible in some kinds of
or private records shnuld endeavor lo ensure t11dt tuture seho!-
rL"Search: intervieweL"S may die or disappcar, and dired observations
wiH have access to the material on similar terms; taking advantagc
(jf re~I-"~o..r1d E'vents by wítnes~"S or partidpants cannot be repeated.
access withour seekíng access for otheTs predudes repli-
Rephcabtbty has also come to mean diHerent things in different re-
cation and calls into ques!ion the scicntific qualítv ¡jf the \-York. Usuallv
sea,n::h traditions: In, quantitatl've resean::h, scholars tocus on repH-
our work wiH not be replicated, but we have the responsibilíty lO ilt:t
cahng I'he analys!s arter starting \vith the &ame data, As an\,one ;,vha
H someone may .vish to do so. Even if tIte work is not replkated,
has ever tTied tu the quantitativc rt."Sults ni even .
providing the rnateríals for such repiication will endble readers io un-
publisht>d \vorks knows 'l/veR 1t 15 usually a lot harder than it should
derstand and evaluate \vhat '.'ve have done,
be and always more valuable than it seems at the outset {see De\vald
et al. 1986 on repHcation in quantiti1tive research)'
The <1na10g)'" in traditkmal qualitative research ís provided bv foot- Data
notes and biblingraphic essays. Using these tooIs, succeeding < •
data pwblems by !lew ano better data ís almos! "l-
should be ,lble lo locate the SOUrces used in published vvork and tnake
an improvement on trying to use t1awed data in heHer
their m"<.'n evaÍllations of the inf0'rences daimed from this information.
the former appmach is nut always possible. Soda 1
For fesearch based on díred replication is more diffkult.
scientists nften find \víth el,lta and líttle ellanee
On0' $Cholar could borro,,,, another's field notL"S or tape recorded inter-
to ilr1ythíng better; thus, han: to make the bt~t of what
VleiVS to Sf'e whel~er lhev support the condusions made by the origi~
the)'
nal llWestl bBtor. Smce so much oí rhe data in field reseal'ch involve
lmprovíng tite use oí previousJy coIlected data 1S ¡he lTlai.n topíc
con:'erSr1ti~ms, ímpressions, and othel' unrecoroed partkipatory inful'- ..
nmght in 011 statistkal methods and ís, indee:::L tite chid contri-
mahon, thlS of results lLsing {he S..lme data is not often done,
bution oE Ínferentii;11 statístics lo th(' sodal sdences. The precepts on
However, some important advances might be adtieved ir more schol-
this topic that are so dear ¡n ¡he stud¡! of inferentíal statistics al::>o
ars tried this type of i1nd it '\'GuId pmbably also encourage '
apply io quaIHative research. Thl~ remainder (lf this book deals wíth
others lO keep more neld notes, Occasionallv, an entire re- :
these precepts more hlllv, f-'lere .ve merely <1 brief outline oí
induding has been replicated. Since we
tne guiddines lor ¡mpm~'ing ¡he use oí' collected data,
cannot back in the replicati~m cal1not be but can be
First, whenever possible, \\'e shouid use data to gerwrate ínterences
quite dH.ldl,ne nonethe!ess, Perhaps the most extensíve replkation of
that a.re "ul1biased:' th"l corred cm average. To understand this
/\'n is t hs,~ US(I ni tTh'nv Lh~~ n {'~IH~ c¡Y.1t,"r to t>'<Jruft very idea fmm staiístical imagine,. the S<1m¿
!ron, tr;¡n'"-"("'ripts ,lt intervlev,/& fr {lV(? methodology (in quan!ita!ive or qualitatín" research) for analyzing
_ how "fkn tlw same judgme!1L H they do no! prodw,:e reliaol" mt'a- and drawing condusions {mm data across Inimv data sets. Because of
thel1 "le ca!! m¡¡ke the cnding mIes more preciS<., ,md try agaín, E,·entually; ¡¡ se!:
5llnL'S, smaU errors in the data or in the application oí the procedure, a single
al mIes CAn often be gt'neratoo so tila! ¡he app!kation üf !ht~ same procedure by diJ'fere!.lt appliú'1tion of this methodology would probably never be exactly c~r-
roders will yield the same result. '
'~
Theme:; uf Thb \tdunw

\-\"111 be ('()rrt}ct \\d"h,~n tdkvll ~1:~ an aver-


a¡-'Pllcdhr'''''' ........ ,'"·,',, ji nt) single appH,,-atifHl ü; correcL _h1b,,·t1,H facts. In

not systematica!lv tilt Ih" oukom<..' in 0111: dírectíou !:ion, as Vh~n ilS hmv data dísdphne~ . " 11en:, we
want tu strcss thnt tlwnrv "nd em¡:nncal rcse.1n::h must be ttght!y COI1-
.-h'",'¡-,.,,,· unmas0d í¡¡ferentes depends. (){ Cllurse., br,tll un ¡-he nected. ¡har ddes real ViOr}; fOT us lus , . rOl' em-
nal cüUectioT1 uf the dat;¡ ami ib later U:-'10; and, ilS \\<e pointt::d out be- pírica! inyestígation;. no . . im'l-:tig,iltíon OH,' t)t'Sllcct:s~tuI,,~,¡th"
fore, ir is always Oest t,l anticípate problems befoH: dala COHectiOl1 be- out ro gtllde lis chOICe oi qUt'stmns. TI1eoI} ,md data ,.O!l:,ttOD
HOVVt,ver, \.\'e mcntion thest' issues brietlv here becauS(' when are büth i1spects oí the bv which \Ve s0\7k to dt"'-"1de
using the data, we need to be particularly caref~¡l to anal}'ze whether whether (l tneory should be provisíonalJy viewi;'d inle or ¡dlse,~5ublect
sources oC bias were ovel"iooked during dala collectinn. One such as it 15 in both cases to the unc\~rtainÍy that charit(terízes all inlerencl:
source., wnich C<ln 1e..ti lo biased infenmres, is ¡hat o( seledion bias: We should ilsk nI' anv t!wo¡"v: \Vhat are íts observable lmphcatlOl1s:'
choosing Ub:il~rvaLiüns in d manneT that syslemabcally distorts the We z¡sk abuu! ¡my en~piri(\11 Are ¡he (,bsen'a-
fmm \vhich were drawn. Although M1 obvious exam- nons rele\'i1J1t ro ihe imrlications of our and, Íí sn, v,h,lt do the)'
pie is delíberately cases whkh support our rheory, selec- liS ro ínier aboul the (orrectness of the thc~(}rv? In ilnv

tkm bias calO occur in much more suotlt' wavs. Another diffícultv can ""-,,,,~n, .. studv, the implícations of the ilnd Hit' obsen'iltion uf
resdt from omítted variable bías, w!üch rcfe~s lo the exdusion of :some fads need tU l~,esh vdtn one another: sociúl sderKe condusíons cannot
control variable that míght infiuence a seeming causal comwctioll be- be rdiable ii are nol based on ano data in
tween Ul,lr explanatory variables and ¡b,lt \vhkh \-ve ",'aní to explaín. one another ,me! ilnd
VVe (ÍiSCU5S ¡hese ilnd numerou5 other pítfalls in prool1dng ímplications of a theorv.
unbiased jnfert.~nct..'S in chapters 2-6.
The second 15 based 0n the statisticaJ 01 "'effi-
. ¡¡n efficient use of data im'olVi..'S maximizim: the information
llsed for or Ci'lusal infcrence. ?vtaximízi~g re- The :;cholar who seuTches fol"
nor onl)' using aJI our d¡:¡ta, but aIso usíng aH th~' relevant infor- is
mllt10n in toe data lo ínferences, For eXilmple, if tlw data artO' ence: as t!Wcfl as ,,,,,;;,;;,IlI,'
into smaE \Ve should use íl tila!: Sde!KC seeks (o inGeilSe ti",
Ine smal ler will llaye ro the inrormaüon used in the
of assodated with thenL bul if toe',' are, what ai first to be a
in ob5erv"ble implications of the thenn,", fhe¡.- "vilÍ con- callsfdvariabJe nr il few variables, the
¡..,in sorne inforrnation whiell can be brought lo oear 'on th~ inference is ven:
une Of '" variables ,ve al50 han:'
¡he social sciences in dna even more so in !)i:lrtirular
areas. Thís may be becanse do not know how to íncrease
1.3 TJ1E\l1"S Of Tms VOl.l'ME flot ro be in d cmwenicnt fash~,
it or becanse n"tllre
vVe conclude t!ti" o\'ervít'\v ,,,,,~~f,"~ ¡he {OUT importan! ion ur for both (Jf these reasons. Ancas
thernes in h,¡ve discussed here are oiten titose in \Nhích ol' il.ny-
,1 h\,.lst nf Y3riab!es: \Vi;' use a lot in
ú hule. in such ('"se<;, uur "buuid bc to

73J !1m! {fútil


with more
There ¡¡fe ,',Hl0tlS W<1,'" in VI/hic!, we edIl innc;¡~'t: ,¡ur (lver
In thís to be worth .. ,1 n'St>1tch Tl~e IV"V is t" íncn'éhC Hit' lHitnbvr oí
wh¡le, lrH:st 11,,\1:' :Ir r",':,;::: (,t
fínd Ji the 1hose }\s w" have dcscribed "00ve, ¡his 13Sk CDn invo!vc
The Scicm::e in Social Science Themes of This Vo!ume . 31
n) írnprovíng the theorv so that it has me>fC observable ímp1ic;1tions, , t'Jtt€'fl g,ün len::r"gt' alxml uur then.)' > bv ;lt t[¡Ie' dat.:l
(2) impn)ving the data so more of these implications are indced oh- ihese othe. levds,
SerVE.'t.1 and used to eva}uate the theory, and (3) impnwlng the use of example, if h'C develop a tneory k) explain n:nJlutions, ,ve
the datD 50 that more oí these implicDtions are extracted frcon) exístíng suui.'llulook for observabk" ímplícatjons of that theory n01 only in over-
dat,. None of these, nor the general concept nf maximizing lcverag; aU outcm:nlC'S but also such phenom12'na as the r~:sponses to in-depth
are lhe same as the concept oí parsimony, whích, as we explained in intenriews 01' revolutionaries, the reac¡ions al' peoplc in smaH commu-
section 1.2,2, is an assumption about the nature of the \vorkl rather niHes in minor parts nf the country, and offidal stalemenl<; by partv
than a mle for designing research. leaders. We should be wiHing lo take whatever inJormatíon \ve can
Maximizing leverage is so important and so genera] that we sfmngly acquire 50 long as it helps liS leam about Lile veracHy of OUT theory~ lf
recummend {halresearc}¡crs routínely li::;f aH po:;sibfe obst'rmbíe ímpJicatitms '\.\"€ can k>st our theory by examining outcomes of revolutions, fine. But
thtir tila! be obscrced in t/reir data Ot in o¡ha data. It ~l most cases ver)' littl~ ínformatíon exÍsts al: tha! ¡evel, perhaps just
may be possíble to test some of tbese new implications in the original ene or a few observatíons, and theír values are rarely unamhíguous or
data set-as long as the implícation does no! "come oul of" the data measured without error, Manv different theoríes are consístent \v1th
but is a hypothesis índependentJy suggested by ibe theory or a differ- the existence nf a revolution. Únly by deeper in the prcsent
ent data S{~t. But it is beHer stilJ to fUffi to other data. Thus \ve should case, or bringing in reievant informatíon exishng in olher ca5{;."5, i5 it
also consider implkations that might appe,u in (lther data-suco as possibie ro distinguish among prcviousiy indistinguishable IhíJories.
dilta about other units, data about other of the units under Tile only issue in USillg inJürmatíon ai other level" and from other
study, dala from different of aggreganon, and data fttJm orher 50UTces to study a theory designed at nn aggregate level is whether
time such as predictions ab::mt the near futul'\c>'-and evaluate these new observations contain SO!Ht' ínJormation that is relevant to
the hypo!hesis in those seUings. The more evidence 'Ne can find in evaluating implkations of OUT theory. If these new observations help
varied crmlexts, the more powerful our explanation become5, and the to test 01.11" thenrv, lhev should be llsea even if thev are not the implíca-
more ccmfidence we and others should in aur Hons oí greatesÚnh=r~sL Por example, we may m;t care at aH about tlle
At first thought some researchers ma)' object to the idea oi collect- views of revolutionaries, but if their ansvvers to our questions are coo-
. ob::rvable implications aH)' 5OUr-ü~ 01" at any level 01' agsrega - sistent with our theon' of re'volutlons, then the theorv itself wil! be
tWl1 difierent fmm that for whkh the theorv was For exam- more líkely to be ('(lrrl~i, and ¡he colIection of additim~al information
Ut'berson (985) to qualitative ~t'search the statistkal idea \,,'ill have !:leen usefuL 1n ill1 observation at the must
"n" .. ,.,,,",",,,h, dala to make in- (X'Ctunmce ni a predicted revolution. foI'
wam cross-lE'\'el one observed implkation of the theory, and be-
lha! \ve can use data lo make íncorred cause oí the small amoun! 01' infom,ation in ít, ¡t should not L1€ pn'\·i-
about Índiv¡duaIs: if we are interestetl in then leged OVer other observable implicatíons. \Ve need io colled infoffilil-
individual'> is generalhi a better if we can obtain non 011 as many observable ímplications of our theory as possible.
datn, t'''Íovvever, ir the infer~nce "ve seek tn make is more rilan a
C<l.st hypothesis, our theory may implicaríons at
man~v ieveis (lf analysis, and we wiH oHen be able to us..: data fmm aH 133 Rc¡;orting
these leveIs to provide some our theory Thus, even AH knowiedge and aH quantitative and in Cjualitative
ir Vv'te an: priman!y interested in an leve! ni \ve cm research-is unceríain. measurement is as
~~. Tl1t~ ". ~ 15 bttG~U$t.~ tbt.~ process> of rCds{)ning freHn ís quant¡tativ€, but the sources of error may ditEer. TIle qualitative in-
ilg,greb~lle- ro mdJ\,'¡dUal-h:ve! proC('l'St5 15 Heitlwr nllr .1 terviewer conductíng a long, in,depth intervie,y with a respondent
b $.1n unhWH.Hl(lte choüx~ of \vürd in dt:~-cribt' tl-u:- h:!\'f~l nf ,... hose background he has studied is less l1kely to mismeasure the sub-
R~ )bin;fYt1 ;': 0~ n (:oflclUÓt'"d in hb (~rigintü artidt! ággregate
ject's real polítical than IS d researcher conduding a
ft'l n:'\'t"--nn: :1bout individuab bit qUiHititatl'd: sil-cía; scicntb-ts and statisti ..
d,m~ nn\\', re~'úgnize Iha! süme íniofmatÍon ,¡bout índiFídu"ls dOt,$ "xi::;! ,'1 ,'ggTL~
strudured interview with a randomly selecled respondent abüut
gate levds <JI ,lna])'S!::;. <Uld numy metlwds of unbíased "ecoJogkal" inferenre have been 'whom he knows nothing. (Although the opposíte is aIso possible if,
develúlx"<l, for instan ce, he reties too heavily on an infurrnant who is not trust-
Themcs of Thís \'ülume ;;3
ín-
~'1jrUwJHow<:ver, the ;:¡Uiq'V researdwr í:; h':,;s likelv te· generalize
ti\' cau:-~' dnd <.'ftecL' rt is h¿¡rd tu imagine he,," not
inapFoprjah;~fjt (rom the CáSeO' int¿n·iewed to th\! broader
a hear! aUt1ck might ,-aUSe c,ne lO ea! less red mea! bui ít ís
popttÍ<!lticm than 15 the in-depth ff'Se.1fCneL Neither ís hnmune from
~ uncertainties of measurement or tht, undt'dyíng pmbabilistíc na- . Pernar::; penpk lose th",ir appctile for h<lmburgers and
túr€' oí the ,,,mId. in lf this were tt-\\: case, tbose W!lO did not 11,1\,<: " heart atiack
AH good social sdentists··-whetner in [he quanlitatin: or qu¡"litative whatever reason) would live longer "nd ei1t less mí'at This fad
traditíons-report estimates oí rhe uncertaintv of their inferences. Per- roduce the "ame rdatiol1ship that lcd the H..'St7drchers lo con-
haps the single mos! seríous problem v\'Íth q~¡alí!atiVt' researrn in po.. Hwi mear \Vas the cuiprít in heart
Jitic.. l ::.cience ís the pervdsive failure lo prm-ide reasonab!e estimates is not our purpose lO caH su eh m~>(iíci11 studics ínin questíon-
(lE rhe ni the investigator's inferences (see King 1990), We we wish mere!)' to illuslrate huw scientists appnhlch t¡w
in almos! anv situation, no matter how lim- oi causal inferen~e: with ilnd il cunL'ern (or alternati\:c
ited the follml/ing the n¡i~s in this but \VE' shouid that n1dV h,wt' been overlooked. Cau~al inferenct' thu~
. bero mes a ' condusion becomes the occ2tsion ror
avoíd forging s,veeping condusions from weak data. The roinr ís not
!hilt relíab!e inferCl1ces are in but further research tu refine ilnd h:st iL Through succeSSIV\"
racher that "ve snould report a reasonab!e 0stimate oI ¡·he !o come dO~'r i:lnd doser !o accurate ínfercnce.
we have in ead¡ of our lnferences. Neustadt and rv1ay
ue"u,,,, vdth <11'(',15 in which quantitative estimares
are propose a uscfuJ method nf encouraging pt,lkymakers
C\<vho are üÜen faeed witl1 the necessitv oi condusions about
\-v'ha! to follow out of inadequate data) tn the uncertaintv
ol their mndusions, They ask "How mucn of \Tour o,,\'n monev ,vHuid
you wager on ie' Thís ~lakes sense ¿'s ~s vve also as!.::. ":At what
odds?"

1.3.·1 El]:!' a St}(.'·iaf Srienti~t.'


¡md RÍ{YIJ Hypothest's
The of caus.c¡l inferencl's means that good scíentists
them. VVhen ¡(lId l\ ca uses B, someonc v'lllo
social sden tist" asks that cormection is a true
causalone. H ¡s eas}' to ask stlch qUe5ticms abüut the research of others,
but it 1S more important tu ask them about our own research. There are
man}' reasons lvhy \Ve might be skeptical nf a causal ilcconnL plausible
though it sound at f¡¡-si We rcad in the nevvspaper that rhe
Japanese e,]1 !ess red meat and have fewer heart attack.s than Ameri·
can$. This observation alome is il1ten~stil1g. in additiol1, rl.E' explana-
han-too much leads to the high Tate of heart d¡sease in the
United 5tatt's-is plausible. The skepticaJ social scientist asks about the
accuraey of the data (how do we know ;lbour e¡:¡ting hilbits? ,vha! sam-
pIe vvas used? are heart attacKs dassífí",'<.l similarlv in ]apan and the
Unik'll States so that we Jre comparing sÍmihlr ph~nomena?i- As-sum-
ing that the data are accurah.', lvnat else mlght explain the effeds: Are
tl'\eI'e orher variables (other dietary difference5, genetic features f Há'-
_-Genera! Knowledge ,md Particular Facts . 35
nl scnolar::;hip: discovering
knowledge and particular fads. \Ve are then
to explai.n in morl' detail tite concel~t of inference ín s"ction 2.2..
d

Descriptive Inference Our approach in thz' i'vmainder of the bOOK is ro present ideas both
vcrbaHv and thnmgh \'ery simple algebnlk mo{iel::; 01' rescarch. In
. <l>I'",UM" . 2.~ we nmsíder ¡he nature oí these models. \Ve then discuss
irR"'ldels for data coHection, for summarizing histoncal dc,taiL aud lor
SOCIAL SCIENCE RESEARCH, whether quantitanve 01' qualitative, in- descripti\'e inferencf' in secnons 2.4, and 2.6, respedivdy. Fínall~',
volves thc dual go¡ds of descríbing ilnd explaining. Sorne scholars set we provide StlmE' spt'Cific critería for iudging descriptive infen:nccs in
out to describe the \''lorld; others to t'xplain. Eaeh LS essentíal. \Ve C3n- "'l~
!,/"",,"'",\,!H .:..../ •

no! construd mcaningful causal explanations without


dcscription, in ¡um, losz's mos! of it:;, inlerest unless linked to
2.1 GENERAL K!\lOWUmGE AND I'A1U1U.iLAR FACTS
some causal rdationsnips. Description often comes fírst; it is hard tu
dcvelop explanations before \ve know something abour the \",orld <lnd TI'le world tha! sodal scientists study ¡s madi'! up of particulars: indi-
what needs ro be explained cm the of \""hat But the vidual v<Jters, partlcular government specifk tri bes,
relanonship behveen descripticm and explanatíon is interactive. Süme- groups, shltes, and natíons. Good sodaJ science attempts tú
times nur explanatíons lead us lo ¡ook for déscriptions of different SO beyond titese particulars to more general knO\vledge. Genem!¡za-
parts 01' the \vorld; conve1'sdy, our descriptions milV lcad to nev" tion, hü\vever, does no! elíminate the impnrtance oi the particular. in
causal explanations. . mct, the very purrose oí müvíng fmm the partícuhlr tü the is
and expianation bt)th depend upon rules of sdentific ro imprüve our understandÍJlg of both. The specific entlties of ¡he
~nference. In this we toens 00 description .md descripti\'c in- soda ( \.·vorld-"()f, more preciseÍy., spedfic facts about these entíties-
1e1'ence. Description ls far frorn mt."C'nanical 01' unpmblemat:k since ir provide the basis nI! 'which genernlizations must rest. In addition., ,ve
involves selection from the ¡nfiníte number of fads tnat could be rc.~ almost leam more aoout a spt:cific case: by studying more gen-
corded. There dre several fundamental of scientifk description. eral condusions. If we wísh to kmn\' \.... hv the foreign minister of Brazil
One ís that it invol\'!;:'S part of the descriptíve task is to infer it ,vin help to leam wh~' othe~ ministers in Brazit
informatlo!1 about unobserved facts [rom tnt' facts ,';'e have fnreign ministers in oiher cmmtrit~S have resígned, or \vhy peopll'
Another involves dislingl.lishing thar whkh is in general resign frm11. or I;'ven nongovernmental joos.
arle abtmt ¡he observed and th"t whicÍ1 is n(Jnsvstcm¡ltic. E:1.ch oí these v,·m lIS understand different of facts
As ShOllId be we ,vith who denl G Ti1te "mere'" ano principies of human behavior, but they are very important even ir
description. Even ¡f explal1ntion-connectíng causes an~1 our one and only goal is to understand the most n:.'\.--:cnt BrazBiilD
the ultimate goa.l, descrjption a central role in an explanation, and f<treign minister resigned. For eXitmple, by' st"lJdying other ministers,
it J.s fundamentaHy important in and nI" ítseH. It ¡s not description ver- we rnight learn that al! the ministers in Brazi.l resigned to protest the
sus explanatian tha! distinguishes scientífk research from other re- ac:nons of the presidenL something ,ve m.ight nüt have 1'('aiized ex-
search; ii lS whether systematk is conducted accanlinn to amining onl}' tbe actions of the foreign minister.
vaIid pmcedures. Inference, 'ivnether descriptíve or causal, qmmtita- Sorne scKÍal science resenren tries lo say sOJ1.1.ething about a dass of
tlve or qualitative . ls the nltimnte uf al! social sdence. events or tmlts vdthout snying anything in particular about a specific
tematicaHy coHecting fads ls a very importm1t endca\,,'or witho1.l1 event 01' unit. Studies nf v'oti~g b~havi~r using mass L'Xplain
t-vhkh sdence ,v(mld noí: be possible but whkh deleS not by HseU con- the oecisiol1s of pi..'{)ple in general, no! the vote of an]' particular
stitute Good ardtlval work or wdl-donc summaries of híston- indh-'iduaL Stlldies of finance explain the of
cal mil)! mak€' gCh.Xl hbh)r)~ but neither are suffkient monev on electoral outeomes across di! districts. 1'",lost
to constitute sodal science. such ;tudies ,vould nui mcnHon the Seventh Congressional Dístrict in
In this cnapte1', we distinguish description--the coUection oi faets- Pennsylvanía (}[ any other district except perhaps, in passing 01' ¿lS
fmm descripthre inference:. In section 2.1 we discuss the relationship exceptions to a general rule. These studies foilow the injunction of
Ct'flt'ral Knowlt;:'{ige anJ Particular Facb
36 Descríptjn~ InferenCt.'
thcpnes lo c;lU~:ll o¡' eh'nts
PrZt)V~t()r~ki and Ir'une zlYH]i: ("'Jnrun{::L,~
F)T'()!>cr ndn1~,:~v
<'in ,KCOU¡'\t oí the n2:¡"i,lES ftlf nr ot sudal ih.:t10n,
thOllgh tites" "ludie" !:1ilV nor sL'ck hl UIhJ,:rst;¡nJ any particular
to know no! onlv ,""llar ca1.lsed the agent to pedorro some act
trIce they should not 19nürt.'--;¡s sometimcs ís unfor!unatelv done the rN~~ms fnr taking the adion," Geertz OQ73:17)
rhis tradition-:-,the requirement th¡!t rhe facts abouí ¡he \',l~inus dis~
",,"rites that "it is not in nur ínterest tn blead, human behaYÍor
tncts that go mío tht~ general analysís musí bt' ilccuratt',
¡:m,1perties tha! inkr~~st us beiore :~e begin ~o e>.amíne it." ,
Other research tries t:) tell tl"
SilIllethíng abou! a partícular \\'ho "mterpret"tlOfI seek to IlIummate the m-
stance. H focuses on the Frenen Revolution or sorne other
aspects human Dehavior by empl()yin~ Vcrstdwl1 ("em-
event and attemprs to pmvídc an expbn"tion of hu\\' or vvhv
understanding the meaning (lf actíons and interactions fmm
event came aDout. Resedrch in this tradition \v(mld be .
o'\'\'n points oí vi!.',>v" ! 1975:81 n, interpretívísts
cerrain!y unínteresting ro most of fhe usu<1l re"ders of such ..,,"A~,~_
without names. A politícal scientist wrile dfcctívelv
o explaín lhe for intentional adínn in .< ro the
se! of in vl/hien if ¡g embedded. llwy Jbo
uf ilcmss the set of - <-<Thf' m()st obvious are ({}'"'
\,,:·ithout.!(lo~íng M specific districts or candíd,ltes bu! ¡m"gine
scope: al1 interprelative "ccoua! províde ma'\ll11i1!
Rnbert Caro s d¡SCtbSll1!1 09B3l ¡¡i ¡he 1948 Senare fiKC in Texas w1th-
nI intelligibíli!'y tn a set of sodal pradíces, and an interp:c-
out , Johnsnn and Cok" 5tt'venson.! Particular events such as
accouni of a set of should b;¿, VV¡! h
th,,> fn:nch Revolufion or the Democratic Senate primnrv in Texas in
1114S m,,}! inaced be of intrinsk interest: ¡he}' our ~uriosih~ and practíct's or tradHions üf ¡he {~loonI97S: i , .
PerItaps the single most important operatlCmal recommenLÍatlOn 01'
lt t,hey' .vere prenmditions for event'i (such as the N~pole~ ís that should leam a dea! abnut a
OJ1!C Wa:!"S or presidenc~'), \ve ma}' n0<.."(1 to know about them
prior to fonnu!¡üing Für only \Nilh a deep
t:} unders!imd those {"rer l'n~nts. knnw!edge about revo¡tl~
immersion and understanding of a subject can a
hon, ,rebeIlion, or ~iviJ 1NJr in \'\'111 pmvide invaiuable informa~ the hypotheses. Far
~1(ln !Or any more rocused study of ¡he cau5l.S oí' the French Revolution of workíng-dass black
111particular. hím~
We 'will consider these issues !hat
daimed alternath'(:' tn m!en:l1CI! bectll m 2. l. ¡ 1, Ihe that
of uniquencss ¡¡lid complcxÍly oí the oi
and thc area of case grudíes (section 2J .3).

In the .human some historienl and anthmpological rcseareh-


ers darm to seek knovv!edge through what thev caH "in-
terpretatinn." $l~ck ilccur;~te $umn~arjes nI' historícal de-
taiL They also seek !o place the 0\;ents they descríb,e in an infelli"ible
c,ont~xt within vvhkh the ()f acti¿ms bt:comes b .As

FerE'}ohn COldstein and Keohmwl has ;.vnflen, "We want

l Nnr «lB t\'€, díSD1b5 Caro a~ ~t}n)(~üne in anothuf busincss: d


differ~; frnrn that {t[ tnf: SiK'íld scit"ntJ.st l'h~ %york addn:sse~ S0111e (ji" the Sárne
~rentf5t \Ydtltt.1: \Vh;H h\ads h ~ ~\lH.~t't.h:;'S ,~r faHuft' in ~Ui. dection
. iVbat f~; tile r01f~ ~H' rno0ey <.UhJ {¡n,;;nce in t:ll'-([nfdl '-'th'(.e<..,~: \Vh.Jt
mntwat0; ",:mpaign cO!1rríbu!o!'3' TI1<' disnJssíon íOCU5eS (In ,1 parliculi\f (anélida(y in J
pa~('1J¡ar alstncL bu! the ~ubj<~t maHer .:tl1d the ovedar with st,\l~dilrd
p{)htu::al scie,lR~,
38 . D€scriptive Interence
General Knm'\'ledge ,md Partímlar Facts . 39
lt is ..-:w(Ía! to UIh::lerst¿¡n~j ,1 culture' deeplv bdurc fürmulahni'; hvpoth-
, "f*~:een a t\'\.Y1tch ::nd ..1: \\'1'nk is \-ast~ .;1$ anvont~ unfc~rtunate en()u~h ü) h{:tvc
ese~ or de~ígn¡ng a sysh~nMtic research project ro fínd ;m fl¡~S\\:e~, \Ve , t,he fírst tak.t~n Íllf ¡he s~;(ond knüws. '1"11,; \dnkt'f i;,. ,me:!
only \'lÍsh tu Bdd that evaJuatillg the v¿r<Kity of daíms based on meth- <:.:~,,-1.~ communicating in a and "recial way: (l '; delibt'r;ltely, C.l to
ods such a~ partk1p~mt observation can bt' accomplished through :;;omevnc in Filrticu!;.¡L (3) lO ímp;¡rt d parlicul,lf mt.'ssage. (4) ilCC()!\:líng lO ,1
lile ot SClt'lltlÍlC vvhich ,ve describe, Finding the right establisht.'d mde. and (5) witllout 01 Ilw rest "f thc Gnu'
.Jns\vers lo th<: \vrong questions b a futile ~Kti\'ity. lnterprewtion ba~(.':d As Ryle points out, dw winker has done 1\\'0 things, contracted ni!'
on . is often a rkh souree oi ínsightful hypotheses. For in~ eyelids and \vinked, .",hilc the t;vitcher h"s done unl)' one, cuntracted. his
stance, RIChard FeMo':,; dose obserl/atlons of Congress (Fenno 1978),
'~clids. )10m un purpüse v,hen there exisls a public
madt: Ihr~:lUgh ~vhat he caHs "soaking and poking/' llave made major code in ''''hich doíng s(' counts as d conspimtnri;¡J signa] ¡C' wlnking.
n)ntnbuhons ti} the s~dy ()f that ínstitution, parüeularly by helping
to fram~ bener que~tíor:s ior r:~gE'arch. "Soaking and pokíng," an importmt conceptual poínt. Without th0 con'
Putnam in a study oi lti'lhan reglOns (1993:12), ¡he """""'H~"h " given meaning by a uf communication, thc
to marina te in the mümti,}¡} of an institution--to experience irs study of "t'yelid contracting by human be-
customs and its successes and its as those who Uve \yould be rneaníngles5 for studenís of social relalions. In Ibis ex-
~t every day do. This ímmersion sharpens OUT intuitions and pnwidti.'S limpIe, che vvhích fmm months of "soaking and
mnumerabJe about no\\' thc institution tits and how it fUg" imd cultural 15 to the propeT qucstion oi
adapts to íts envimnment." [\ny dcfinition oi science thal dnes no! in~ ;whether eyelid conudetion even could be "hvitches" or ".vin"s." The
dude ro(lm for ideas tne gem:ratio¡¡ of hypotheses 15 as fooI- m.agnifkent importance of interpretdtion by this is
ish as an account that does not eare aboul it provides new of looking al tite world-ne\v concepts lo
truth. considered and hypothe:St.'S to be evall..lated. Without deep immer-
Yet on<:<.: . have been tormulated, demonstrating their cor- \",12 might not even think of the right theories to

1"ectness ('Yllh ¡m estima te of uncertaintv) valid scientific in- n""""",nt example, if we did not think oí the dífference
ferenc(.'S. The procedures for inference foÍlowed bV ' sodal n"'l»t"i'I"í"n twich(,"S and winks,. everything \vemld be losL l.f interpreta-
"","."L"'''''''', furthermon~, must incorporare the sa~e standards as those' tlCI'fl--Clr an:vthing t:!.se--helps us arrive ai ne,,\! coneepts or hypothe-
followed other i:md researrhers. TIlat then it is ando interpretabon, and similar
.~hile that insightfui ¡nterpretn~ hil\'e been proven
tlOn or other we aIso ¡nsís!
that sdence b essential foI' accura!e inteI'pretation. H we could under- made" rdevane theoretkal such as that behveen
stand human behnvior on1y through we wouid never be the researcher thcn need15 to cuallmfe ¡he hypothe-
abje lo our hypotheses or c,'i.dence for them It 1S in su,:h evalua!ion !.hat the logk oí
our experience. Ou1" condusíons \vould never go bevond the . sc::lentific inference is That lhe best >';ay uf determin-
status of untest:ed and our interpretations '~vouId remain t.~e meaning of eyelíd contrdctions is through tite meth-
personal rather than sCÍentific. descrihed in this book If distinguishing a tw1tch from wink were
. ?ne ,of th~ best and most famous in t.he interpretative tra~ pivota.!, we could desih"1\ a research ro do so. lf, for
dlhon 115 Chfford Geertz'15 analysis of GHbert discussion oE the mstance, \Ve belJeve that particular contractions are winks im-
difference a h.vitch and a wink Geertz (1973:6) '\Tires bued W1th meaning, then other similar ínstances musí also
be since a soph::lsticated device such as ¡his (a
Consider ... ¡wo boys rapidly contractíng the of their C'ves. In
once developed, 15 used Given this
ooe, thís ís ;m invohmtary twitch; ín the other, a conspiratorial s¡gn~I to a
record instance ín \\'hieh this actor":; eyelid
frwnd. 11)(> hyo movements are, as movements, ¡dennes.!; from an r.am-a-
contmcts. observE' vvhether tht: (,ther actor is !ooking al the right
«¡mera, "plwnomenalístic" nbsef"ation oí thCIl1 ilIon.:, one could flor tdl
time, and ,,\'hether he rt:.'Sponds. V...re cC'uld E'ven a series o.t ex-
which Wib twiren ami whkh \VilS ,,,,in k, Of indeed \vhether both or either
periments to SL,{~ if individuals in th1S culture are accustomed ro com-
was twitch or 'wínk. Yet the difference, hmvever tmphotogmphable, jx....
municating in this fashion, Understanding the culture, carefully de-
40 O('Scriptive Inference Generdl Knowkdge and P,uticular Facts . .+1

~..:rib¡ng the t'\'énL i1nd ¡',winf.' il g tu {hl~nl h) the puH~;? The dt:<.Ai~i~)n of J ~jti?t~n nt..)t tt~ \'\~tt' .
tions ,viIl aH help us ask the rjght and C1.'e11 gíve us .,\".ính or.l di}11nmatíc messóge. can ¡¡kan man\' ihings, Th,· "o-
tional confidence in out condusions. Bu! onlv \vith tlw methods of n:S6,feher shuuld alv:.lvS work hard to as!.: thc ¡ight que,;·
entifíc .ínference H'íll we be able té' eY,1¡ua!~ the h\:püthesis and d then careful!v design scientiric research In fínd llUi what lhe
hthether 1! is corred. . s ad did in fad mean.
Geertz's wink irÜt~rpret<1tion is best expressed as d causal \\'ould also like lo brídly <,(kiress the extreme daims (lÍ ,1 few
hvhich we define pre·císely in seLi:ton 3.1); the hypothetical ~ nf ín!erpretation who argue ¡ha! lhe ~oill oí some n'setm-:-h
fed oí the wink on the other polHical ador is tite other aetor's to be feclings <ind meanings with no observi1blc conscquences,
¡he eyeHd contradion minus his response ii tlwrc were no ~ liaren)' a fair characteriza!ion of a11 but a small rnínority oí re-
ment {a lid no other lE the eyelid contraction were a \vink, in ,this traditinn, out the d.1ims Me mad\:' suffkít'l1tJy Ú'rú'-
Cimsi11 effect would be if ir \Vere a twitch, the causal hat they seem worth addressing cAplk:itly. Likt' the over-enthu-
fect h'ould bt' Z('TO. if v/e decided tí' estímate this c¿)Usa! daims oi CMI} Píisitivísrs, \vho tuok the untenahlc ¡har
thus nnd out whelher it \>vas i'l vvinK nf a tvl/irchJ, al! ¡he problems ,-"'rh-,pnt,,, had no in scientific
inicrenc\:" disellssed ilt in the rest uf botik would nE'ed in rescilfch. For exam-
underslood it \ve \vefe to arrive at tite best ínft'rL'nce with rf><,n,'f't
[he lnterpretatiofl ni lhe oDserved Qeha,'ior.
on that which ís (w"rt ,md maníft~"t('d ín
lf \\'ha¡ w\:" interpret 'lS winks wen, ¡llvoluntarv
observable' aets is l\ilh'l" in ~y Ih,,' le3sL The tn
(iUI" to derÍve causar abt,ut evelid contractíün
Ihen, is lu tmdl'r-
the bnsis or i'l thL"Ory nf voluntary sodal intem~tíon ívould Dt'
¡11M the actor'" ¿¡et has for him.
tinely ,ve would noi be ab!e to ,." ... ,,.,,,,.,,¡,
know íLl u<!,,,tl',,,,, mal' be corree! thatsdentists who f(){,'us on
50cld! nvert
DL'Signing reseiln-::h lo winKs and a lOI, buí hOVl are we to kncnv if we
tn be ~ , part of most polítkilI ;;dene!:' ,.p,,,,,,,,,,.,-,, see? for exmnple, if two theorles uf self-conception have identi-'
methodologícaI lssue arises in much oi the nMnifestations, then 110 observer will have sufiicient in'
ka! sdentists work We are often
polie)' d!:'dsion makers send messages ro eacn otiler,
il 3 3 poínt <l statement
dt appeallng to a dt.lIncstic audience? of cultura! norm$, obser\'atiot1,
nmycntions in and of ¡he hislory' quantHica-
particular adors, as wen as dose obsen'ation 01' • or ,md methods are
the comrrnmicatioTl, \\fill aH help us make such 3n Or i¡)adequate to the task dístinguishing h'\'o theoríes wí¡hoUl diHering
consider the puzzle in res€arch: Voters in omsl'..'tluences, On the otIter if the hvo theoríes l1a\'c
Unitet1 seem to be by fll't tunling out O( some manifestations that diJfer, then rhe rneth()(ls "ve de-
polI:;, But lvhat the lOw turnout mean? Does it reHect alienation scribe in this booK pmvíde ro dístinguish behveen them,
\vlth the poiitical system: A In practicE', ethnogmphers (¡md aH other sodal scientists) do
ing v\·ith ¡he costs being Jook for observable behavior in order ti) among their 1heo-
dates or recent campaigns? Could nes. mal' lmmeTse themselves in the buí they al! rely on
rhe minimum age of 01' il various forms of o[,sert'afÍl1ll. Any further "understanding" of the cul-
tural context comes din'Ctly fmm thc'Se or other comparable observa-
For dw;¡;:¡k<:' tha! \Ve e[l!lId dl1
tion!'>. re]evant observations ¡s not On the con-
Cj¡tit'fL'nt tl1t;l'!'V in which M\ c\'dtd <ou!l'action Vi,l!' nnt ,1 wink but stm had i\ Glusal
eifa:! Oil nther a<:ton>. rol' e'clmp¡e, the twítch <,ould bave ht~~n mi:;mtl:rpreled. lf ,ve the appropriate observatíons 15 pt~rlMpS most diJfi-
WHe a!S\) mlefL>sted in wrt.'ther ¡he wi¡h ¡he eydid úmtractÍfm inleuJcd tu wink, cult part of a resemeh pmjed especiaHy (and necessarily) for those
t

We' would ru:.~'ti to look for othH conseqUí..'Ilces of thís 5iHne theurj', t.,
ureas of inquiry traditionally dmninated. by qua li ti ve research.
Dt.'$Criptive lnference General Knüwledge amI I\¡rticu!ar Fa"is ' 43

2,11
Some qu,,!ítatívelv nrit::,nted resea.n:'hers 'Nould the positíon simplífy.
general knowledge 15 either necessdry or useful (perhap5 eyen lm:leed , rhe
bId as toe IXl.sb fUf ,1 en:nt TheÍr positi ] L~ ~till
nd tf¡ilf in tttc ,
tha! the events or uníts stuJyare "unique," in Ont' 5<:.'ns<:\ !hey bcttiX'cn thi~ tínckc<;:! d,'.;,-npti<m~ ,md tite tlW~t ab,;: mct .
right. There was onIr orw French Revolution and there ¡s onI)' mi'! )Jü dt'SCT'iption, no matter how tl'uck. dnd no ex-
jnlo tt n,'l1:W:;
Thailand, And no on0 who has reaJ the biographkal aecounts oI' nO matter 1..uO\\' many ('xp¡ana t ory f'>oC. ....ors
~" '"
li\/eJ thmugh the 19605 can Joubt tbe fact tbat there was on!v capturíng the full "bJoomíng an~l buzzm~ re¡, lit y oi' lh :
Lyndon 8. Johnsnn. But ¡hey go furiher Explanatinn, accordi is no choíe'e bu! to 5ystematlt. i
their position, is limited to ¡har unique event or unit: not \vhy sr.,"}) to useful As "n eronomíc histonan has
tions happen, but vd,}' the French Revolution happened; not ,vhy t ' ¡'} t11!' "xtreme oí
MISll:' on " .... ".
ITlocratization sonw!imes seems ro bu! iArhy it lélgs in Thailand; p the H:'TV --¡enn' 1S
~t .. .. "
candidales win, but LBJ wun in 1948 or 1964. ""'c",,,., tu ihc ¡¡lmiesssness (lf balladcl'rs"
p(lssible, símplify their » _

attain <111 of toe ridmess or.. and euI-


sdentísts mav use cmly a few oí the hlstt)r~' of stJmc
"Uniqueness," however, ls a mis!eadíng term, The Frenen even,s " 1'1'• . l'nf"rt'l""'"''
,- .",;0, N· , r'~\'''rthf'less,
'" '- . . . . , rich, unstrw.::tured
tion and Thailand and LB! are, indeed, unique, AH phenomena, thc historie,.! and cultural context nf the phenomena
events, are [11 sonw sense The Frencn Revolution v;ant tu deal in a simplified and sdentifíe' is usu-
was; but 50 was ¡he eiectíon in the Seventh Dist,kt avoiding tha! afe simply v,:nmg, Fe1-v
in 1988 and so "vas Lile voUng decision of every one trust the of a about revolu-
the mi1lions of voters \\'no voied in the presidential dectlon tha! elections if that knew little and
V¡evved holisticaI1y, every aspect oisocia! Reyolution o, the
and connech.~ in some to belie\'e th"t.< where
c\'ents. inherent ís part (u",,,p'r;;¡¡ and
ir drl.('s not dístinguísh sltuations amenable to sdentHic ;:l\'!,eSE;E'S of evenrs as \veH as "b(lUt
about whkh dre no! ,",U:;;'¡iVI't:, ant to be nnd
shmved in theories of dinosaur extinctíon in chapter Leven. her mav ,/arv trorn
studied paying attention to bu! both are likeiy in' be
n,."c,,\",."~""'''' implkations (if theoríes to ao::ount io! them. ln-
bdng opposed to
The real that tbe ¡ssue 01' uniqueness mises is tbe problem filc
tJ¡p l}L'~t ti
oí complexity. The point 1S not \vnether events are inl1erently unique.
but \vhether tne features (JI' social reality tliar ,ve want to unde:r:-
stand can be abstrac.ied fmm a mass 01' f,lds, One of the first and mosl:
difficult tasks oi in the social sciences is this ad of simplificll-
tÍfm. H is il tasI<. that makes as vulnerable tí) tIle cntidsm of oversimpli-
and 01 omítting sígnificant oE tite situation. Neverthe- Much 01' what politkal scientists do is describe
such simplkatlon i5 inevitable al! researclleTs. Simplification events svstematícallv. care ,1bout lhe
has been ao oi every knovvn \york--qwmtita- Union, thl? n.'actions "uf the in wuntries to the ,
tive and qualitatlve, zmthmpological and economic in the sodal sd- ized \var to drive lraq from Kuwóit, the resu!ts (lE the lm~st CO~-
en(::i~ ¡}nd in the natural and physical scíences-and \v111 pmbably aI- gresslonal

eiectlons
. ' l.
In tr1e '~" del
l',r'lhe tüc A1'd ·h<>v
,J a ",,y, ,rt, <")
relv
"
on
.
nohhca]
1:"
so-
cntis!s fUf th,)Í tened d more L,1Il1F'reiwnsin~ d\Vart'ne$$ interconn\!cted,
the reiatiol1ship bd\\'een these ana other reh?,'dIÜ ('venb--en
nuy i.md historicaI·~···tl1dn is found in journalistic accounts. Our f th~ otten (lverlooked "d.\'antages (lf the in-depth
scriptions oi events should be as pn>cise and systemdtíc as is that the dl'\'elopmen!: 01' gnnd is (:"W'
lo z:ood dt:S(T¡ptitm r,1tht'f than \<;ith íl. Fr;,m-
This means thiÜ ,vhen ,-\Ce are "ble to find valid quantitalivc 111
(Ji wndt \Ve ,vant to ,ve should use them: Whilt proportio e s!';dv Jmund 'an "",plan"ípfV t¡ucstkm mdV kód lO more
Soviet níticize gnvernment polky? 'v\lhat do pubik r~leYant desGiptiun, en,!! if the studv 15 ultinhlhA:'
km in Jordan and Egypt n:veilI about Jordanian and in its attempt lo províde e\'Cn a single yalid infereI1Ct'.
ve case sh¡dies \ve argm~, yíeld \'a!íd cilusal ínfer-
attHudes tO\.vara the Gulf \var? WhiÜ pern>r1tilge of con¡;ressíonaI
in thc Test oí this bnok are used,
cumben!s wen: reeleded: "
the\' oftcn do not meet the stan-
H precisíol1 ir cines noi neCf'SS,ll
f

accuracy, SÍlKt' qllimtitative indixes that do nol


dosely lO the ('oncepts or e\.·enb tha[ '\-ve pur¡:x)rt [O measun: can
lo seríous nk>i1SUrerrH'l1! error and probJems Enr ctlusnl inference
section 5, l.!. Similarly, tlwre dre rnore and less precise ways ro des
evenls tha! cannot be quantifietl Dí::;;:iplíned qualitativc rL"St'il
try lo ana!yze constítutiOl1S and L.1\vS ra!her ¡han n";'n""
porl \vnat observers abmJt them. In doing case studíes ot
ment researdlers dsk ¡heir ínformants trenchanL well-spec
questiuns to \vhkh ans\vers 'Nm be relntívelr unambiguous imd
ioHmv 1.1 p on oH-hano n?marks made b~' i.1.t1 • '
relevan! Case siudll.:,'S are ~ssentia¡ for
scription, "nd are, fundamental to stxi¡t! sdence. H 1S
less !o Stc'!:k lo ,·vhat we have not described ",,,ith a
of precision,
oí comp!ex events ¡s no
(Ir ínt-ernational

zmd cmmter others' 1s


¿¡ne! expectatitms may playas importi1nt a parí !\!mpÍlasue here so muer. as it 1S a w,y\, of
in fnr sta te benirvíor, 1\ pllrpnrted non in descriptivc case in su eh a vvny that ít
oí world th8f ássumes ¡he absenn; interaction
fp,Ktions \viH be rmJd. ¡¡;ss usdul lhxm a carefu! ~~The HterJhJh.~ un (On¡p~lr\ltivc l,:.-'¡:;t' ~~tthjh.~ is \'(.1sL Sorn{'> <.Jf tht.' b~;~t ~lckhtú.~nnj \vorks

OH evcnts that we have reason to believe E.:kstt'in (1975), L.i¡ph"r! (19711. ,md Colli<>r (lqqn.
46 Descriptive lnference 1nferen ce 47
bt: used [,Ir desGíptin: ('jí "::cn¡s,ll in{en"l1('l', i\luch \'úluablc ,,' en1pha:-;.is un ta(-t~ dS obSCr\\lhlc \.)1

"bout doing con,}';lr¡ÜlVe t'd"'e S makes the cummon gmtmd betweeIl the quantiLltive and
often igmxl!d, styles ni research much clearer In f"cl, once we get past
G;ses or units or records in the usual ,'en' narrO\v or even
2.2 lNFElU,:'\;CE: TllE SCIENTfFlC Pt:RPOSf 01'
sense, W€ fealize that most studi2S poientii\lly pro-
DATA COLLECTION
very lurge number ot observable irnplícations fOí the theüries
- ated, yet many ni these observations ma)' be overlooked by
lnference is the process oE using the facts \Ne knmv to leam aoou! Organizing tht: data into a list of [he obsen-
we do not kmnv. The facts \Ve do no! knmv are the subj~:cts oE ilLctlJlV"" oE a thenry thus helps rcyeal tnt: e&.sentía! sdentifk

research questions, theories, anel hypotheses, The fact::; .ve do of rí\uch qualHatin' rcsearch, In a sense, we are asking the
form our «(Iuant:ítative or data or ol-;sef\'iltions, \'Irnn 1:;; studving a evenr-il
In aSK: "H nw
partkular vve must sornehmv avoid being oven'\' me out the wa}' it die!, ",hat else
the mas5ivc Di and actual obs0rvanons world?" These addítional obs,,'fvahle
rhe world, solution to that problem lk>s .... ..,,,'-',""'1'\1' dedsions., bu! thev migh! also be found in other
the search knmvledge. That the bt.>st sdennfic way dedsion being shKlíed: fnr when ii \Yas hmv ít
organize fact" is as implkations of some (Ji' E:, ht.)w it \-vas jus!ified, Tne cnKj,,! maxim ro guíde b'Jth Hit'-
sis, SLientific simpliHcation involve5 tIte productive choice of a and data is: search fOf mOl"{' observable impJka-
(or hypothesis! io ('\-'aJuare: the theory tnen guides U5 lo the
of those [¡¡ets that are implieation:; oI tneo!): Organizing facts in ir L'i productive to
oí observabie implications ol a tl1ro!)' pn1duef."S be observed,
tant and beneJidal results in designing and conducting ítems ipr vvhích data
\-vith tnis criterion for the sek-ction of fae!s, we can qukkly
that more oDservations of the implkations of a theon' \viH onlv eme additíonal datum v,-m
evaltmting tne in question, Since more info~1ation o(tlús to evaluate a then to
cannot hure such data are neVi"r and the pnlCt'SS oí time, and díort constraintsJ it i5 wor!h
or otller observ"tíon might be
Viii;' need not have a 'abie ot this {or sorne pther rdc\'aniJ
nor musr our theof\r remain throughout be obvlous thar it v'l'ill not
ad. As \'\'ith and tlle egg, some is parí of the simplification
befare dat" colh:dicm and some data are Í){ftore any data 1nlo observable implications of a
l;"'vtruv,l/c O}1 research teH us tbat we use OUT data lo 'test the data. VVe can think ilhout the r,,1;\-' materia! of real-
E:mm the dala may be as important " goal into "dasses" thar afe made up (jf "uníts" or
evaluating prior theories and h:'Pútneses, Such iearning involves v·,rhkh are" in tum, made uF' of "attríbules" or al'
organizing our data into observabIe implications of the ne\ N The da.ss might be "\'pters"; the units might be él sample
This is (omrnon early in many researdl vorers" in s('vera ¡ distrícts: and ¡he attríbutes or
somepreliminary data have t1€en collectt't1.; after the
data coUa'tion Hum eontllmes in ordef to evaluate
new throry; VVe should ah','itvs trv to continue to colled
after the reorganizatinn in ord:~!' lo 'test thc ne ..\' ill1d thus a
using ¡he same dala to evaluare the that ¡ve usted lo develop
.¡ For Cnon,bs O'it4; ÓemO!1stratt;,'C! that vírhUlily ever)' tlSt'ful d;}t'a-(n¡¡'l>ct~
Desoiptivc InferelKC :\ Form.1¡ \Jodd uf Data Cóne,,'tion < SI

runnel tt?sts, ít wuuId be lrrde\<anL l¡ü\v~'Ycr, sine'e L'Ydl


Ul DAT,·\ C(}LLfC,'TlU'.
dust can (au~e aplane to V"lcígh more and tbus use more e
¡'uel, modds of this sort áH;: importanl to ¡he aidime industry ilnd pn!st:ntation pi descriptive ami CllLsal ínfér-
bet'n huilt (and swed millions oí doH.lrsL (l primary ni social "cience fl~se"rch-we 'NílI d(c-
AH models range beh\<een restrictiy{: ami unrt'stríctive '\ ' ior ¡he dMa tü be cdiectt'd dnd for these
stridive models are dearer, more par::iimonious! ilnd more model is quite simple, but it is a pt)werrul [.)ol for analv/ing
but they are also h:ss re¿distíc lunless the '.vorlJ is oi inferencc, Our aigebr<lic model will not be <l" formal as
ous), Modeb whích Me unrestrictive are detailed, contextua!, hui makes our ideas dearer and casier ro
rnore realistk, bu! lhey dre also leso, cIcar and harder tl1 estímate collcdiali, \ve reft.'f to a v:idl' range oi indud-
precision (sce King19B9: section ') VVhere on ¡his continu partidpant obs(:f'\'atí{)ft lnlensín' intervicws,
dwose to G.mstruc! a mt.xJd ,,1',"""'1,n,'!C survevs, historv recorded 1'rom SG:.llTeS, ran-
be anJ on the "v,,>",,,rnnn' - Z'onlent and anv other
vu¡¡' al!
h07C the data IUt'rt~ i~n:i.lted {Uhl lv.r ¡:¿: ¡ce carne ro
of information tllar we should
observable impHcatíons of our ¡llenIT. !t muv help us
C01.mt will be an abstl'adion or n¿w research question, but it \vi!l be nf no \.1st: in
actual judidal Sillee undersl:auding CjUt?StiOl1 if H is not an ohservable ímplicatiol1 nI' the ques"
tiOl1, the book ís as nmch ivhat is lef! eek 10 answer,
induded, data \vith ¡¡¡lit::; .. and llb:'.;{TI)aticrn~, 011e ex-
Vv'hHe researchers often use verbal annual income of eaeh oi tour people, The data might be
in our discussíon simpJy four numbers: , 521. ana
verbal models, JUSI as \vith models of ihe more case, wc could label the ¡ncome or four
studies of the French Revoiutton, our 1, 4) as ltL, .'12, }!:\, iind JM- One variable
research not be confllsed \vith tor t\VO unstructured intervÍl'\\'S takt' on the Yalt1E.'S "par-
are dear statements of ," "cooperan"e," or "íntransigent," ,:ma mighí he labeled .lit
In addition, we often In these cxamples, the '1\7riaNc 1S ti; the U/lits are the índividual
us to dÍscover íde,ls that we vvould no! have ¡he are the values nf tbe ,'ariablcs fuI' eiich unít
otherwise. for d ollars Uf oí The y Ís a
vVe assume rhar readers have had no
braic

in tl1ese "nd
Just because quantitaJive are probably more familiar
our dnes not mean that are any h¿tter al
the of scientific inference, these models do twt
more to than ro qua1ihüive r,""f.'p,}'{'!> suppose we ;lrc ínterested Ín
the modeh~ are useful abstracticms (li ine fl>search to ;vhich th(>y since 1945, Before \ve colled om data, we nc¿d ro decide ",hat
applif.:'(l, 1'0 ease their intmauction, '.ve introduce aH i'\Igebraic VVe could seek to understand the size
wíth ,'crba! followed abo'>: issu€ area Of
notation. nil.útionl in 1990; changes in the size o!' intemationa!
skipped without activity sinee 1945; or cnanges in the size distribution of
Summ..rizing HistorÍCal Detdíl 53
Descriptíve lnference
intern~1hon~1l organi?atiol1élf dcti\Ylty ~ínt~e 1t~tt3" Vdriablc5 to refe!' ü) J. {un case stlldy, \\~hh~'h ~bB has i1 faidy ilnr~re­
Tht'ret'el't', wherever poc;sible vve use lhe \",¡ni "~cas\'O
"divity (uuld índudc the number C>f únmtries
ing to íntem"tionaJ orgimizatinn:, ai a g¡n~n time, tne number of t'l'riters du ~md n:..'Servc ¡he WÜfd "obseryation" tu rdt;'r lo
by infi'rnatíonaJ organ:z¡¡¡¡,ms, Uf the sizv:,; 01' budgets of one 01' more \';¡ríabk's on exactlv one unit
staffs, In these the uní!::; oi ,1Ila.lvsis woukl indllde ' 1pt in the rt'sí uf this d,i1pter tn shp\\, Ix),,,, . ¡11,:('
tional orgimizatiüns, issue Meas, cmmtry rnemberships, and time nd unHs can inemase the daríty pi nur thmkmg abou! re-
ods su eh as years, five-Yt'ar periods, or decades. At the data dl:~iO'neven when i! may be jDilppropriate ro on quan-
slage. no formal mIes <1pply as to \vliar "ariab!es tu colket how me,~Ul'es to summarize the inform,ltíon at our disposaLTlw
unlls ¡here snould be, whether the units must outnllmb"'f tne \ye is; tI O\;' CilH ,ve milke de.;eriptíve infercnces alxmt
aS it really \<\'i'lS" without getting ¡ost in a sea (,í ílTelevant
or how "veH varía bies shonld be measured. The oHlv rule j
judgT1wnt as 10 ,",vllar \yill pmve ro be ímpurtimL \Vhc!~ \l/t' . how can \\'0 S(lrt oul ¡he essential from the
clearer ide,) nI' now the dat.a will be used, the rule becomes fínd
I'mmv implkatitms {lf a theorv as As \Ve
sized in chapter i, rese¡1rch ca;, be used both tu
priori hypotheses or to not previotlsly
but if ¡he latter l<eh' data must bt' coHected data are the fírst in aoy :mm-
el/aJuate these of the data. Summarícs JesL'1ibe \Vh¡lt mil}' be il CUHount
!t shnuld be ver}' dear {mm our disn.lssiol1 thal most 'works but are not dínxt!v rdated to infcrenee. Sinee \\'e are ulti-
"case studics" h,1Vt' l1umcrotls \'ariables measured over manv in generalíz:1.títln and exphm;11ioll, ;1 summarv of the
!ypes of uníts. Although re~arch rarelv uses m~m,~ to be explained is usually J good pl;,ce to start but is not a suffi-
handful oí cases, the total nmnber 01 observations 15 general. oí sodal sdence seholarship.
mense. H ís therefore t."Ssent1¡,¡ to between !.he number is necessary. We ean l1ever tell
cases and the number of observations. fnrmer mav be of soma ()f events; ít would be

but onl\' the l,ttter ís oí . whkh events "vere o.m"tnKf

his
be irreleva¡¡t h,
of the number of obsen'alions comes fmm h"lr or whe!1wr he friz'd tO
\vhere n IS ¡he r1umber of persons to be Cood hisloricaJ writlng aHhough it mil\' Hot be
\ve apply it much more genera!!)'. lndeed, \)ur defin:ition oí an verbal s~mrmi1ry nf " weHer uf historícal
vation" coincides vvith Hanv 0975:85} defi
\'\'hat he c"Hs a "caste." As Ecksteil; argues, NA studv nf six SUmJm¿ln;él!l,g historieal detail is a stab-
dection.s in Britaín mev but need n~)t be, an 11 "" { studv. It oí data in abbrevi¡ited formo liS
dlso bt:~ an n '" 6 . H can diso be al1 tI := 120,000,000 st~d''V. It ¡he appropriilte of the data .in a
,,11 \vhether the subject (lf 1S svstems, eJecÍions For example, Orle statlstic ¡5 t:he mean, or a\'crage:
voten;." The "ambiguity atxmt what constitutes a~ 'individual'
on be dispelJed not lookíl1g af concrete entities but
i!:= ¡¡ + y: + .. , +
the measurt's made of fhem. On this a. 'case' can be defined
as a [or <md interpret only a
gle measure on any pertinent variable." The onl)' difference in , Form,,11v, for " 5<?t oí' n units un whích il \\lri,ll:>1<' lf is me;\sUfL'Cl (V;" ,,1¡,el. ,.1 ~tdt¡st¡C
usage is that since Eckstein's artide, schola.rs have continued to use Ir 15 a rea!-\'~lut'{l íunctinn déined as ¡'ol!ow!>: h '" [¡(!I'.) '"
54 . Descriptive Inferem:e Dcsuipün: Inierence 35
;"vht):!'t" 2:~! '.,.th is a convenient 'lvay (JI u~ + t/~ + !f~ ,~. paper nn t\'-~Jr~ and aBian(t.>:; Íj:clud~ l,jatd in-
oth~:r statístk is {he Jabt'led U",y,: obs(~r\'ati()n~, In Sl1ch ,1 p,-,¡:,cr summanes ot the d;Ü,1
numbers might be justified; ('\/en for an expert
te indicators míght be incomprehensible \~'íthout sume iur-
rv. For a ¡",eture ün the to an under¡,;raduatt' dass,
ihe &,mple mean of the tour íncomes 1mm the example in might be 5uperíur.
$22,000, 521 and 554,292) is 526,573, The sampie
mum ls $...114,292. \:Ve Gm summanze the originai data C'onta
numbers "vith these two numbers representing ¡he sample
maximum. \Ve can "iso other :xlmple characteristics, 'e inr'erenc(; 1S th(' prncess ol1.mderst<mdinf.~ ,111
th(~ minimum. mode" or \/ariara:e. 011 the hasis oí a sd uf oDsery;1tí\ms, For
Each summary in this modd reduces aH the data in ,'ilrÍ;üions in the distrie! "ole for
this simple example, or our knmvledge of sorne asped of Eu Lilbnm. and Social Demoaatic in 13rítaÍl1 in
history in the other! ro a single number, Communicat!ng v;ith . hav\;' some to ho\ven,e
ríes is often easícr and more meaníngful lo a reader than we observe is 650 district eJections tn the House oí Com-
data. Of course, íf \Ve had ~;nJv four numl..Jt:rs in a data in thilt
then ít would make Hule sense h) use five different summaries; \ve might think Ihat we \\'01'0 direct!y tht' dec-
¡he four original numbers vvould be simpler, Interpreting a of the COl1servatil(es bj recording theír share of the vote
tistic is generaHy e¿¡sicr ¡han understandíng tht.: entire d"ta sd" out t and their oyeraH "Dare ot seats, But a c0rt .. in degrce oí ran-
lO5t: information dt>scribing a Iarge sel of numbers or unpredíctability is inhercnt in politics, as in al! ni sociallífe
onlya few. i of scientific inql.liry.7 5uppose that in a sudden fit of absent-
WhaJ rules govern the summary nr historkal detail? The first m:ss (or in deference to sodal 1he Brítish Parlian,ent
thilt summarlt'S should OH the ou/comes ¡hat TUl' !I.'ish [o to electlol1s en~rv 1979 and suppose
H we were intt.:rested in the 01 the ,werage thal ¡hese elediOt~s were indep(~ndent nf one another. Even
tional v;e would not be l,vise ro focus 011 the Uni """!p,,¡,,',nn support foi' rhe Conservatives rernaíned constan!,
but if vve were nmcemed about the size distributkm of replkation \Vnuld no! produce the same flumber of votes
l1at:ional from big ro :;ma11, ¡he United in each dístrícL The \\'eather
be one of the unirs nn "vaich
Unitcd is not a "01--'n~"".·,,¡'"
tant o:n0. In s!i:ltistícal terms . tu cl(enrs might happen In the ínternational nr SGm-
ganiza!ion, \ve wouid exanline mean might reaeh the mdSS media; even íf these had no long-term
bershíps, etc.), but to undersi;md the "".hUU"',,"' the'í' could aífect the weeklv results, Thus, nlJmer<)Us¡
to exarnine the varial1CC. A st·rorh.l, obvious precept is l'l"iUIFatr)r\i events cuuld effed slightly diffe~ent sets ni election rdl.lms.
mus! the al aur In quan .," Our observatioI1 uf any one election v/01.!ld not be a perfect mensure oí"
terms, ihis rule means thar we should ahvavs use [ewer summary 5trength after aH.
tistics than units in tht: original data, othen<;~ise, ,ve rouJd as easily As another example, suppüse we ilIt:: ính:rested in the of con-
sent aH the origina.! data \vithout summarv at all:' OUT - Bíct between lsradís (polke and rcsidents) and Palestiníans in commu-
should a150 he sufficicnHy simple tha! it can be" understoüd by . nitres on the IsmeH-rn:cupíed 'vVest Bank of the Jordan River. Offidal
No phenomcnon can he summanzed perfecHy, so stal;d repons by both sídes seem suspect nr are censort-d, so ,ve decide to
adeqtl.1cv mus! depend on our purposes and on tht: imdiE'nce conduct our own study ..ve (',m ascertain the level
conflíct in diffcrcnt communities íntensh'e in!er:ieh's or
e Thí~ püint is dosdy rd"teo to tlw cnnn:'pt of índetermÍnan! research
we díscus& in section 4. L
Dest.TÍptive lnfurence Des.:riptive lnú:,rence 57

tion in farnHy (ir group ~?\-ent~, If \Vl~ do this tor ~) \\'eek in eZlcn be(\lUSC thl' nt)n:~\'sh:mdti<'> chan('c eÍt:D1t'ct in thf'
mtmíh', uur condusiol1s "bout the h'\'d uf l:on¡'!ict ll1 ('a eh onc wíll tend ro ()','erwhdm Uf disí;1r; tlw ~YSh~mdti(' ek~l1wnL jf our
a {unction in part of \Vhdtevef d1dnct' events occur the \veek we w0ek bad nCluned immt'jí;:¡tdy atter ihe Israelí ínvúsÍ(m
pen to vísíL tven if \ve conduct tlll' study o\'er a veal', \\'e stilI wiH Leoan01L \\"0 wixlld similarlv ~o¡ t";ped result~ that are
perf,,'ctiy knm." tht.' true lt'VE,j nf contlict. \.'\'('n tlwugh nur unccrtainty ve o! whi1t USUdlly happt:ns on thc West Bank.
abl'ut ít \vill dn1p, polítical .vorld Ís theoreli('¡,l!y c"pable nf producing multíple
In these examples, tfte variance in lhe Conservittíve vote 3CHlSS ets tor everv pmblem but does no! foHow Hw n(~eds oí
tricts or the variance in conflírt between \Vest Bank communities sdenlists, \lVe are usuallv on11' fmtuna!e enough In übsern' (1ne
be conn:ptualized as arising from tiNO separare fadors: sysfematic d,:¡t,l, ror purposcs of a modeL we wiii lel thís une set ni data bl'
twnsystemafic differences, diHerences in our voter eme \'ariablc lf {::oa\', the vote fur Labor) mcasured o\'t~r
ínelude fundamental and chMilcteristks uf the d t1 ::: 650 tmHs ¡districts): ;¡¡, ,11;, ' , , , yry (for example, v: rníght bl,'
such as differences in in il1come, in campaígn peoph: \'oting fpr Llbor in Jístríct ¡), Tlw set DÍ ()h~i'rZIt1ii,});~
or in traditional for cach of the In hypothetkal weekly we ¡abe! Ir i5 a nvfi:cd ¡mriablc. lt~ v(¡lues \far}' UH'f the 11 umb,
{1f the sanw would per- a:ddítion, wc: dcfínl' '¡ as a rrmdom i'firi;)/ilc oenmse ít hlríes ran-
but the nnnsvtematic diffen'nCí~ such as rurmmt variations due aeros:; hypothetkaJ rcp!ícatíons of the same electinn, Thus, ys is
the ,ye3!her, ,,-'o¡11d vary In our \'\iest Bank dif-' the number of people v()tíng for Labor in dístrict S, and Ye; is tlle Tan-
\v¡mld indude the deep cultural differcnces bet\veen Isradis dom \.'aTí¡¡ble representing ¡he vote acros~ many h:T1o¡hetical cledions
and Palestinians, mutual kmrwledge of e<1ch nther, and tilar could havE' heen heId in dislrid S l.mder essentially the Silme (on-
pattems ni resídential housing 11' '¡.ve could star! our •• ¡¡finans, The observed votes for ¡he L<1bor par!y in th", une \ve
vatlol1 week a dozt'n differen! times, thesc ""'CH'''''' observe, 1f¡, 1t2, ' ' , , l/I!' díffef ilcmss constítuendes because of sys('em-
tween communities would contim.l0 to ;tffL'Ct the observed level of 1201\- ane ilnd ;¿m¡iom faetofS, Tha! to distinguish the t\vo forms (lf "vilrí-
mct differences, stlch as terrorist incidents or " ",ve otten use the ¡emi r,',lli::cd ,'{"iahle !o refer to ,¡I and nmdorn
instances of Israelí brut;,Iíty, WQuId noi be predktable ilrHl """FIt1,"¡,> to rcfer ro y
\vemld nnly aHect the \vt"ek in whieh they happened to oceur, \"'/1tl1 same lo our qu"llrative We
inferentia! techniqth's . vve can leam about tne na:- wüuld han' no nr dcstre of quantifvíng the h?\'el of tension be-
differences even with the arnbiguíty that occurs in tViietm Israelís and in part becatlse ís a
one ser ni' real data due to nunsvstem.1t¡c, or differences, catee! iSSllC that 1nv01v0s the feelings of numerous indidduals, o1"g,\-
{he nizational C'onflicts, ;md othcr fcatuft's,
in ¡nis sit.uation, li'; ís a realíz(x'l variable which stilnds ¡he total cor¡-
is no! more Hict observed ,:h;rim; PUl' \vcck in ¡he liflh cnmmunitv, sav El·Bireh,"
and our attention shnuld not bl' The random Yari<1bl~: y; represents both w!1al ,ve obs~;rve'ín Ei-Bin:h
''''''P\,',,,,. distinguishing be!v\'een ¡he and whaL Vil!: couid have observed; the randomness comes fronl tIte
1:1"'015 an essentíal task oE sodal sdenc€, One to think about ínfer- variation in chanee evcn!s OVé'r the vveck::o ,VE' Cdldd have
enee is to regard the data set \ve compile as one oi many possible eh osen lo observe,'!
data as the actual 1979 British elt~Liion retums constitute One uf the "111-
only one of many sel:; of results for dífterent hypothetkal dmn variilblcs bu! st,md,ud, ¡ermi-
on whkh electicms coulel Í1ilve been or as our one ,veek although in distinguísh {mm
of observation in eme small Ís one of manv possible weeks, nonsystematíc components in OUT data, in il case u'e vdsh lo
In inferem:e, we seek lO understand rhe dl~gn~e tü whieh
the san1e to f1Il th(~ ~}ther n~mn'tunitic's \-\~(> nligbt :-;tudv
our observations reflett either or outliers, tíild the
a Noh:' thnt tht~ rand~'~nn(~s i~ nnt exactiv n\'cr Jitfer-ent ~lctuai \\'et~k"L ~,>jrH.:t,~ bt.,th
1979 British eiections üü:uITcd during a tlu epidemic ¡hilr ch¡:¡m:c eVPl15 ,¡na ~y,;tem<ltil: Jifferenef'S migh! accmm! ror ntserved Jiff",rellce", We
thmugh 1.vorking-dass houses but tended to span' the rkh, our obser- therefor", <:re¡¡!e ¡he more ideal ;:.im,,!ioí! in which v.'e imagíne nnming ¡he wnrld ag"in
vations might be rather pOOl" measures of underlying Conservatlve wlth sys!em<ltíc feature':l hcld cünstant "na cllanee factors "¡¡,,wed tu \'¡wy.
58 Descriptíve lnferenn" Lkscriptive lnfenmce

t;:¡kv ~1 r.)~ldCH~1 \'~~rjdi"~-e Jrtd e\tT~h.v:


it" ~\'~h:n1~)ti( %t\1turt:\sJ Fur ~;?xa
we l11ight ",ish ti' the
klK'\\' ralue of the Labor \'ote
clistrict :; ({he average Labor q1te '(5 across " 1"r<2;e number oí h the Sdme random variablt~l produce divergen¡ results, This
thdic11 eie~~tions in this ,:!istridl. SÍlwe thí~ b a ~v:4'~matic fe,1ture o'f the size of th0 nonsysh?n1<1tic comJ)t.1nenL
dectoral the ex:pected vaIne is of considerable 1:n. con11-:1unÍt\V t he \~ari:H'('0 (jnsreatj
terest to sod"l scientists, In contr;;1st, the Lloor vote in one nb;;erved
election, ys, is off considerably less long-term inten:st since it 1;; a
tion of fea tu res mI;! random error. it1 ¡23l
The py'nt>r!p,1 \'¡j!ue tone fedture oí the
titth VVt'st Bank El-Bireh, is ¡"tter ,knotes lhe n:sult ni
t(J the random m a \Ves!

where n,} is the \.','!U(' be ,,,'orse. ln an\' l'\'cnt, both


acTOSS an jnfinite number oí hvpothetkal researchcrs,
observe in ::;, El·Bireh, The these isstles we distin¡::ui¡.;h hvo
mu ,vith a 5) the answer to tlw me""s of nmdom \'aríatíon n Thest' two ¿¡re extremes on
Cillcul;:¡lion (nlevd oí' conmct between Palestinians ano Israelí s) far Although l1umbers of sdmlars can be !cmnd
community 5. Ihis is of om model {or 3 fea- are comfor!able wirh each extn!lTk', mo¡.;r scíentists have
tun, of the Tandom variable y" One might use file observai lever somewhere bet\vt"t'J1 the two.
,lf"" as al1 l'Stimate uf fi5' but beca use }j-, canfains man)' chance
elements with information aboul ihis l:'>t~tt"'r
estímators exist bee sedion 2,.7},
.i\nother featurt~ off these random \',¡riables
h, knO\'\,' 15 the h~\'d of cnnHíct in ¡he iTiynlY;c West Bank

ín variul1:S pdrts of th~·

One estím,1tor of Ji might bt.~ tht: average oi the observed leve15 uf con- ::: A I)ctermi¡¡is!1c V'iúrld, Random vilrÍi¡tíol1 i" ¡ha[ p{\rtíon ni
flk! aeross aH the communities studit't1, y, but other estimators tm tMs Hit: workl ¡úr which we 11,H'" En The division hctwz~t'n
feature exist, too' {Note that the &anle smnmary uf data in abe ;md S[Odla¡;tíc \','1iatíml ís and dppend5 on n:l1at
our discussÍ(ll1 of ,'aríables are ilv¡¡¡¡ahli;;' (l!1J indm1<,d in the Cíven tlw
for the 1'ur1'os0 oi inference.) Other ",,,,,,,,,,,,..,::tt'Ir variable;;, ti"" W<.lrlJ ;s
oí the random varulbles indude [he variance ami i1 vanetv of
causa! introduced in sec!.ion 3,1. . In the in ..
StiH anoHwr feature 01' these randnm vnnables that might
be oi interes! is th0 variation in the level of nmtlict v:ithín a commu-
í: ~'t,~ (lY'91b} ror iH1 \:~L~boftltú~n cd' thl~ dhti!.Í\~tHHL

, ~ ()f CülLl~;f!" ti" tna \' bt> nf tn<!lendous H1ten~~t h:~ the peonh.~ t!1 dtstnct ::; h"1¡ tito! \'~~af< ¡~Ecnnomists tend t~} t~" ~.Jü~~r tí.' r1:ers;1{.·('bvt~ i, t'\' ht'lV.1S ':.>tzüisÜ!..-unt< dr=t t'lu~{;"r lO Pt.'r~
3

ill\cJ thus ooln Ule mndom .Hhi sys!emalí..: compün<,nt~ (ji


Ihí" <,veni be \~ur!h 5p¿Cti\'(~ 2. 1 is aJs{'~ cornmo.n in the filAd ~)t engu1(h¿ring G1B~"d
'''"''''''.'>''1<', Nt'verthde5S" \Ve shouid try lo Ihe mndnm ¡rom Ü'w WS- "<JuditY control/" nave even dehaloo ¡hb di"tinction in tl:w lídd oí 'l"antum
m€'Chanícs, Early propon{~ntsof P,:;'rspective:2 $ubscríbed to lhe "hidden variable t]wmy"
6ü Descriptive lnfcrcnce Descriptivc lnfercncc
klHd uf Si)c1~)1 :::.cicncc re~e~1rch ~h()tdd :sab~ty the l-h1~il:
i1: \\«~ tlSSUnlt>: under In th~~ bnoky FinJing l~\~il..icncc
uf
\'dri¡¡bIe~
femain UnkJH''''n, vvith some kinds oí <,~\'¡denct', bu! ir is 11\)
!cl1\X' o<.'\:uro- ',\'hen ttwse unknO"'H explilneltorv "¡¡fiables in
tiye 2 becnme tlle for the r¡md(1m vJriatíon in As a11
tive 1, Bec;w:;e nI tlw lack oí dltV ObSCiyable ímplic1tions with which ._<tn~", suppnse th¡ü we aH! ínterestéd in ¡he ouiconH:S (lE CS-SovieT
to distinguish bet'ween t!wm, d choice bdh'een !lw two pCispecti > betwet:?n ¡ 955 "nd 1q9{) OUT ultim,lte purpuse ís to
d('pend~ 011 bith <Ir beJk( r,llnier ¡han on empírica! vCfüication, question: un ..:h~r whi1! conditions .md to \yhat t"xt¿'nt
As iJnothcr ('xumple, wilh buth distinguíshíng whether summits lcad to in(reibed ('ooper¿ltitm? Answering thal
el partÍcular 01' sodal l"Vént IS the resuli of a a number oí díificuH iSSU0S of causal
nrnong a ~et c~f
From tJw ·\-Vt.~ rest fi ('t
t~f¡'t'd ,,5
ser d da!a (nr t','en anuther case) to check fur the
etfect eH' pilttt'n1, il is nory dífficult lO makt' ¡he judgment
From [he e\Jremc version of 2, ,ve can do no mort' th"n: <,venle; a cnmbinatínn of these measurement

:m l'vvnt ,1;, ",tochastíc (lr sys~ whicn summits \Ven' íoHoWf'J inncilsed su-
or írrelevant. A more realistic versíon 01 thís per- about tht: nm-
! 's corred nr i!1correct attribu!:ion of a in
but it alkl\vs us some latitudt' in Gmdítü1l1s in
and [he l'xtent !o lvhich <>v,..,,,,rt,.,hn1l,,, on both

what ,vil! remain have bt'i?l1 fulJíIled,


aH observal:ions being rl~'''''·Hn"a leve! of in ea eh y'ear, and [O assodate it
tht:n tn cvidenc0 ¡he pres.ence or absence uf " surnmit
rcsul! torces, as ,vdl as \,'ilh our (líher ,
random occurrence or V'/hat ,ve Obsc'íve (even our índices of ,.""""."",...,h",
f"r future research,
::>uC'p,'u tbe
anei VCc1rs sl,llTlmit med-
ti ve rest'archers. research is (ltten but , \>vhether thc ~tlmmits and
mos! use as sodal sdence \l:Ílen ir Is ,lIso ft'lated lO unIÓ' a¡¡{lther,
thc randOlTl váriables from which observations are ít could be rhaz the a::is(KidtÍun
ate'el nnd to lo estímate their randomne~s ciue tn funda-
the lüstoriG~¡ nr bad luck under
data coIledíons. Indef'ti, (me lTlark. of a lmidentified explanatory variables
sl1ch tmídentified variables incJude 'Neilther fluchwtions
the Soviet shifts in the b¡ll-
aH of ,.vhich could d(Cnun:t ior in
If identified, these variables are altemati'le
viuiabies that could be collect{"(l Of
,,·¡1\.'~l~t the n...,>sohtuon uf tht> rntmt":rnus :rcrn.¡ú!ün.g ("(Hlt.r¡H..'hrth.1n~ nf thb i!llfxH"LJrü 1he ~S5Ul' ~s :!ktt of "",1,,,,,,,,,,,
bt'hn\'~

,md !lo; jo. ¡he' nature ni thc wMkl. H,w:ever, lhb dispuh~ in i':(lt.lld k~lJ tn the CfH1V("ning
üf surnn'lit ¡)1t:'etlngs" ~n ·~·\.'hk:h
ks., ~,lthough uS('11 ro mw:h pi Ih", oi ~Kii\J ~:icnn~, is unllk.:'ly !o (',1St', instead oí 5ummil mt'\?ting:' explainíng cooperatíon, antídp<ltcd cCJofX'fa¡¡on \vnuki
thc off iruen:~m't~ or prilctice of resEiltch in the sodal scienres. t'xplain adual C'oojJ<'r;¡tion--han:lly ;¡ startling nnd¡ng if adors are ratínnaE
62 Desaiptíve Inferenct' ¡udging Descriplíve !nferences

to nS5CSS tla:.ir Ínt'luence on rhe SUm!Ttlt outC{1nle~ lf un.idl~nhhecL t our \.1It1, but ,ve mu~~ ,'tfs.c rni1k~" dtC\"Scrip-
variables may be trt,¿1tt"d dS e\'ents that (ould disting111shing bet\\'een ranLl(}I11 ~Jrtd pht,"~
for the observed high Ot'gree of sllperpower cooperation. Tb . Knowing wlut happened on a occasi¡lD is no!' suHkicnt
evidence the possihílÍl:' th¡ü random 0vents ümidentified Jf ~T\"': ~nakf HO iD fxtratt filt"
variables! aeCollnt for lh.: observed coopc:ratiol1, we
look al many orher vea ro,. Sine/! nmdom events ana procesS{'S
definirían no! ¡hev v,·m bl' extremelv unlike1v t
differential cooperation in y'(;:.ns with ana w¡tl{~mt supe~po"ver
mil". Once ilgain, ,,\"\:: are lea to the conc!usion that only repe¡lted
CRliERJ.\ ¡:Ol{ !l:DG¡~G DESCRlI'TlYE l'iFERENCE5
in different CO!'Ü('.xts in ¡lüs (dSe) enable us to decide whe!:h
a Liue lo lhe transienl: s..:dion, \Ve introduce threc explicit criteri,j th,ü are com-
or randorn proct'sses, in statistics for judging methnds uf infcrences-
fmm nonsysternatk pmcessL'S i5 ami fileh rclíes on ¡he r;mdom-
uf social a flu epidemic thdl introduccd in st:díon 2.6 hu! has d.in'cj and
voters more ht,>;wihr than middle-dass ones is an for and
eVf!nt that in one , replícatio
!-he 1979 dection would dccn:;lse the L~1bo;- vote. Bu! a persi.5tent
¡em (Ji" dass dífferences ín the incídence (lE a disab!íng illness woui
effect the an'rage level ()f Labor

\"Ve save
ba5is of the víctor's or an accidental research that is
a debate a randorn factor that couid have "kme for toe remainder oi this
fected the likeHhood of bt?tween the USSR and the
States rile Cold War. But ¡f thc' most effedive
to vote!'s had be€ll Üw
c0i1s1srent vktories of
il factor off infer!:'IKt? imd l,ve wi!! esb-
are sometÍlTh.'S t(lO and smnetinles too smalL A.eross 3-
\v'hen do ,.ve the ansvver 1m !1Z'emgc 7 If
" ís said tu be unbiased. Thís prop-
constants, appea15 ma]' about ho\\' far removed fmm ¡he
behavior, but that fad does ()f the methnd might hut correct
mean thi'it campaign do 110t It 15 the
off appeals 011 an electíon outcome thar Ís consta11t-~{)r, if oí
is variable, it is VVllcn
'. times one vvav~ sometimes the other. Bias occurs when there is a sys-
tema tic error in the measure Ihar shifts the estimalc mure in une direc-
Hon than <1l1other ove1' a sd of LE in our of conHict
in West leaders h3d cn'dted
Dad wCJ.ther ah\',H's Jc,ld~ lO f,,'wer votes fuI' the studv"s results lo fur\ht:f theír
favoring conciliatory ¡:miidcs).. llien the leve! 'oi
conflict 'Ne observe in every cornmunity \-\,(luld be
In short, surnrnanzing historienl ís an important ,,',t.;>rl'Y'"u1 biascd tOlA'ard greater confIíct, on average_ If tne n'plications of our
hdped O!1{~ side "nel nul the othL'r
th,,"d ~1ttenda.n('e \\'a~
fuI' instdnCt~.
C,msen'd!1,'l"S wen' more nductant lo n,[c <'11 v,,~ert::). then 1i~tl~d J~ hH' ~p\.'cn
l"e¡¡"OHs!. (JI' oul" cstim,1tes mígh! bt' bilS€d attcndal1i::e WdS f,H lll1C ,Ll\< CJoser "cruíÍnv
fmm '/ote cmmters \\'no favor one pany Gn~r the
figuI\: to l)t~ tTHKh hn\'('f.
Ir, hnwe\'er, tl1(' rep!icated dections \".'ere Í1eld on ",,¡rious days
in á mannel" Lmreldted to rhe variable we are ínterested in., any
measl.lremen! v\'ould flor for IV('
01' al1o!her might {a'/or (me ti' use tlw i1'ú>[age il"
counts dne h, r"ndom da!;}, Ci ¡", lhe
of estima tes wl1uld bE' unbía"ed. {ur th~: ,1\'e[-
lr ¡he Britlsh electínns ;Y'::,re
n,ethnd
intü the eIectinn c'"d,·.w,"

círcumstances [ha! jnduded tnt:'sc


on the that 13 and cines mi! jusi exfst
¡he data alone. H makL"S littie sens,e ro &lV thar a data set
evel1 though it b-e fiHed y,'¡¡h many errors.
In this we our ddinition of "
tistical bias" in ;111 estimafor from bias" in an ch'fmnJ
of ¡he larter are
to '\1ote--:} no! uncomm\.m ,"ubstanbn' bias uf
electoral As '.ve mav \vi3h to estimare ¡he
vote oí the actual electora! ""'er,;""
but \Ve

are
bias in the \Vhichever mean "ve are
!lnve a sta!Ísticalh' unbiased estlmator,

ihat \ove use ror (1,;'<;crU,f1VP 111tl",'PI1ICF><;

estimilte~ tnat are ''''''+'.''''''''''''"-''


=/1
snnre up
i> ¡" ;m estimatm pi u, ís a
to find <'5t1m211,,"3 tnat are eXBmplt' 'titan ,1PP<':;US in formnl s\;\tlsti..::':' lc,¡s, bu!
VVciner'5 uf education and are lht~ s¡¡me.}
labor in India 0991}. In tryíng to expIain the ¡evel of commitmel1t
to compulsoJ'y education in india a)mpan:~i ta toat in atoer countrk"S,"
Dt?scrlF'tivc 1nrerence

¡H)\\"l~\'t:L bz..' Jf;·l;.,,~.,t\:'d b\' rtHl."-..:(H11 \.'~u¡ai·';k\..... Ihv J.ndn~t'


'25, th~lt !S~ the \.-~~1r-í{3nCt> Uf ~h~" ,~·<:1rn-
V\'t: U~tL1Hy d{J not ha ve' an dpp(lrtunity to ap~:i:~Y out' l~tirnator ID' ll~tin~atür j~ hn1trs
nmnGer uí essentialIy identíG1J applications, (Le" less diicient) thitn tlw estím,l!c when 1; = 2:" HelKe,
ve the (}bvipu~ re~uH that Hl0Ti: obSer\\lhlH15 a:r¿~ bettt.~L
ness is oí ínterest. but we would ¡íke mure cunfídt!.lKE' thi1t the are tht' conJitions undcr whkh a more d{~tailed
estin-ubé' \ve 15 ck'St: ro the right 011E', Efficíency provídL"S a vva communitv would ilS
dis!inguishing amung unbíased estimators, ¡ndeed, Hw dfícíency That is, although W(' should
ríon can "Iso hélp distinguish among "lterrlatjve e:>t¡mator~ mon, observations tlw resources necessary te> cülJed
sm,,!! amount uf t'ias, tAn estimator \vitn a bias ShCl\lld con-
be roh:d out ('ven \Ylthour irs
is d rt'Íative c'nnr'PT>r
\"ilriance of lhe t:stimator across being ('qua!. our ,}na¡y~is
sor)';v" tl\iÜ the rmm: (lbs,~r-
biased estím¡¡Í\Jrs, tlh> tht~ becdllse tand thus In
ter) tite esnmatoL A 5m,,11 variance is better becilllse our one esti b sllen th,,! as the iltHl,b"f oí obsen'.,-
,vil! ¡)(~ clüser to the troc vahw. We are nui dt'Creases to zero, and :he estimare
prnr,plnr, for aH estim,:¡íor wíth il to estima te d
,mee in !h15 situatíon wiH make it that the estimatl::' thar
near the tfue vallle most of the estímates ,vouJd bt'
dllstered armmd the v,'mn~ \alm:L As vve describe bdow, we a
in the case oi a SIl'hin (;im(}Unt of and \Vé

1 \.A/e are interes!ed in """.U'''«


Palestlnians and lsr,lelis in fur no otJwr re3S0n, than thilt we
make corrE'ctíons for such l'dctorS
mid simíiar obst:'rvatíons number of obsen:ationsl. \\l(, are then faced \",ith d tTade'
f¡ve conununitíes, I! be ob,'íous t!u! th"t has additíonal nbservations internal tú
<1 re better (ha 11 a cases in \vhich eaeh contaíns une ob-
into enen pi the
We \vil! oi on1)' (lnc
lo our 15-observatínn re-
vvhich are easier to understand
we could firsr sdect our communit)' very
of thc

enablc:s as ro the others, VVe míght ask él fe\v residen!s or look nt neW5-
"' Ji estimator 01' u with ",->,."".-1'", io see whdher it .vas <1n average community or \<\'het!wr
aV{~fi1ge level 01' conmct found {rom
studies in difft~rent communities on [he \Ves! Ba.nk. H
bid~<i ~~~thn.üor ¡l[ /L huí it !~ nlcz~n~d5t-enr b\~\,-/~USf~ ,l::' H'i(~ numher nf urdts ¡n.:n.\l<;c . thl:;':'
both estírnalors ¡ne unbiased !f the same mndel applies, the. cshrnatox dt~'S Hui irnprovc (OT in.:k:-t\:l change id aJO. An cst1rnah)t' C'ln dtS-t) tR" \~(H~s~stent
estimator has a vmíance pf V(I' :~: al. That out hk'lSt>d, F0f examplt\, y,-, S/n is biasezt hut ü is consisten! beG1U5f' Sh¡ ¡",comes 70m
,ve vrould have chosen ""har .ve thought \..!as a "typícal" as H appmaches ínfínity,
!udging Des.::ríptín: ¡l1fercno:~ 69

dnd tht'l1 \\"t:' rnj~i1t ir thc ruh:~ are ;,~:It1ü:(LL'nduubh:~Ji\:< thi:~ bt." ~1 bt~tt(~r nlt<l~
e5timate d the d<1ta un une rule as ¡he \"stim,ltor f()r strin.-
tht, most difiícult pdrt nf the as a v,:holt~.
h'':: be \>\.~ry c¡¡refu! that bta~ Joc~ nt)t i..Tet~p in. ()ncc \ve ~~rE.' rc'"""nn,,,,~,,,*#, wen,[, ihis prücedurc us h) lht' f\)rHul rule as
confident that bí,,:; is \Ve could ¡''''cu" on íncreasing tu lhe retll th\:.' Sic'I.:tnr under
Tu do thís, we rnight man:,' \\'t.'eks in the community Íl1\'t'stigatioIl of rule ,lpplicatÍf)n, !kl\\'eVt'r.
ducting rmmenms studies, Vv'e couJd in!erview commu varí,.tll1n in the extent lO whícÍl nominal rules are actually en-
¡e¿¡deT";, ordinarv He!Kc,. measuH.'S uf forma! rules might be bí,
ínsLmcc\ in fanJr oí' in
a case, \\ t.' ,.,.'(luid {aee the bias-effídencv traJe-,1fT once ¡¡¡olÍ
múk,' 5ense io Ca!T\' "\JI thn'c uI' fnur ínteltsiv,' (,be studics oí
ímp!enwnliltíon l o ' the hctwecn formal
and actual O n c , would be tu suh-
an e:-;t1milt01' based 011 tlwsc fhree or four G1Sl~-¡es", bíased
les;; efficient-for the cstímate', bi1sed cm ltlCl G\:·+~S. il(1wi?vvr. ¡¡
he more crt'ative, if tu use (he íntensiVt,
in a work lor the three or rour cases to corred tht.' bi.as ni om ¡ OU-GlSe ind ¡-
pf ¡he \vorld. further that there is a choice 01' tWt1 cator, and then In use d COH(y('ted version (ir ihe J nO-case indicdtor áS
il case or a iictatistical our e:-;timator. In this \'>i0 \-,-'oule! be
countries of the !t \vould seem better to o11r intensive case stuJies \.vith
Bu! Jet US Sil)' that ro ít is necessal'v (ful' think should bt: follo\.veJ muen more
H:ilSl:. l l1S1 h.l use data to a UN agene)' {rorn tl1\:' sodal s,:íerKe,
emrncnts. Th('5e lwmbel's Me known tn have little re!ationship for case stud ¡es m.ldc by ¡hose v\' hu kmn\' a
actual of since tht:,' WFfe in the \\'orld h'el! b onen t h,,' one in tl-n:
Office imd based cm ccl11sideriltions oí studies mav upün num1:x'rs lhi1! are no!
that \ve tuuld. ¡he ni'\Í\'e researcheI' \In dará base (who
make the cürrectioHs to the be unáwar2 of the in which dectíon statishcs dfe in
thiÜ a partinlbT h,cale i'md dssumes" thar thev han' som<:' reó]
\\'ould ,ve choose? re!atkmship to the votes JS
hvo 01' enree. Or 'c\'C the materi¿¡Is and !11JV 1,1(' abk lo
com::ctions, ln sedions W(' wíll
how sueh choices míght be mih,1e more
data best <lBs.ver our Our fonnal oí this in ¡he bOl< below shows pn.:-
To take still another example, suppoS(' \Ve ¡¡re studying the Eur<r how to decide wha! the results oí the trade-oH are in the e)(<1I1'I-
pean Community and \Van! to estímate the degree oí pIe of British electoral constituencies. The decisíon in anv
tion of an th(;' ('n ti re that wiH I'esult w¡jJ be beHer when logic lik0 thar sho\\"n in ¡he
of i'v!inisters. We this lssue will ,llmost
number oi mies forma!!y
industrial sector in ende thcse rules in terms of their strin- \vorth more
gene)', ;mJ then estim"lc the '1V\crage uf a nll,,, H vV\.' giüher thal sometí mes exist betwt~en
bld.& imd The mean
data on HIO rules \vith sÍmilm a stríngency, the vanal1ce of out' the first two observatíons in any larger set nf unbiased obser\'ations 15
Descriptive InfercJ1ce ¡ucigíng Descriptíve lnferenct's 71

Formal Effkiency~omparisons, The L,rianct: ul t11<' ::,anlple


y' 1S (:h::noh;~d as l'CY), and H1t' ruies for cakulating v,¡riances fandol1l sdectiort we Ch(1üSe survev rcspondt'nts inm1 ai! "dulí
d"m \:"r1aI:>1.::s in the sirnple Cil9' uf fimd(im ~(,mt'líllg thi? of v:hích has ¡m ot sdl,ctio!l,l Sup-
. , t some!.mt: else a!so did a similar with 1 cÍliYd'l:<
we indude these additional observaticms wíth ours tu crCalt' a
based on 1,020 respondents? H the ne\\' observations
andomJv sdeded, just as the first tVH.'n!V,. it should be an eas)'
to ¡ne'lude the additionaJ data \vitb mus: with tbe new obser-
{he estimator is stm unbiased and no"v mueh mOfe efficienL
.;uppose tha! 990 nf ¡he 1,non nic~V\' obsen',üi(íns \VeTe
dr"wn from the US i1nd ¡he otiler ten \Vere
alíe Dlembers oí Congres5 \vho Wt;[Z; induded in
[he random had been dri'nvn. further
if \ve dssume thni r¡w ,'"ridrlü' ¡lCroSS hypothetícal we found out th"t ¡hes!: additional observations were induded í.l1
li.catíon uf each distríct eh'dinn ís lhe same as t'verv ()ther
data but did Bol know which OHes \'\'ere ,md thus could Bot
dl1d is delh,teJ OC. tlwn tlw variancf' ni ¡he samp{e mean is
tht~m, '.Ve novv know a tha¡ Mí cstimator bibed on
respondents would produce a overestimate of tbe likeli-
i::haf a Deuwcrilt vvould '.vi n the natiomvide vote, Thus, induding
additiunal observations would bias the overaH esti-
but it \vould also substantiaHy its effidency, V,,'helner
uld ¡ndude the observations on whether the
in bias i5 the incrense in slatisbe"l oHh"u",,,\{"L'
it secms dear that Hw estímator b,1sed on the
wil] produce cstimates
fre<.juently than the esUm3tor bascd OH
The bia s í n trl'd uced wouId be small
., samplc estimalor even though in \ve would proba-
,:-.: (¡-
addition, v,'e know the direction ni the bL:lS Ín this
and ('{luId even partíally corree! for ¡U
cstimator has v ff quant:itatíve data are avaÍlable and we are ab1e to ¡'"r-
!;'sttmatm has \'ariance Un!ess ¡,ve such pwbiems as these, \ve can usuaIl;," mak" a dear decisíon,
random-t'rror correctiüns to reduce the variance enm if the qualitative nature of the rest:'arch makes e\',1I'uat-
the statistkal ihis trad("-of{ difficult or ímpossible, understanding H should help
us make more reHable inferences.

Fonnal Comparisons oi Bias and Effidenry. Considí:'r t\VO estima-


10m, one a bv someone ,y¡th il vvlto i5
rut it doe'S n:ducc ¡hereÚ're that \Ve
If \ye ¿id nüt ¡d",) u::,,,, the dficiency criterion, we \vould bdieve is unbiased out less effícient ana is done by an im-
no formal criteria for choosíng one e5timat~)r ove1' the other. partial investigator, As a fnrrnai O1o(leÍ of this examplc,. SUpp05t: "ve
Stippose "re are interesred in \vhether the Democrats ¡,youM \'vish tn L'Stimafe,u and the large-n study produces estimator d:
72 Judgínf.; Dc'S("Tiptive lniercnces 73

the t}t'tc,rt:-:- ()] ~\ur ~n'L.üi-·'}i ~tt.hj\'> :~in\'t.l i~ l~


U.nbÜ15ed, ()n d\'eri1~e~ n(ro~s an lniirúte nUD1ber s,.1f hy}·hJtht.~th\Jl n. ~p~
ljciltiül1S, h.lr ¡he im'éstígi!tO!' wíth a prtYünccptiol1, d would tlw
>,/1'ong iU1S\ú;'!', albdt ,mI\- Sil, E"timah,r " wouJd ttw

( :::
r + y: '1 >.',>-1',/"''''''''-'.' critt'riO!1 leHs a difIen:nt sh lry To

!
i late the variance of each t'stilTldtor:

where di5tricts 1 dnd 2 are iH'erage


ilnd ::') : : : 11, -;l;,Vr - tU)l
JI'''''''
VVhkh estirndtor ",h(luld \.ve 1

üus or twst éstímator is nol

Flrst, vve wii! asscss bias, We c:m 5ho\\' tha¡ the first cstimatnr <1 i$
biaseo ro the usu,11 calculatinn: :;:: a'~/!l

FU,I::: ,L~
¡¡ .:::....,;
- o.or ,
\11

, This "ariance is the Silme as ¡he vilridnü' n( ¡he silmpJc mean be-
cause (Un unes not ¡has zt'rt! vilri:mcd aeross "amplle'i, Sími,·
vve cakuI::te the varidlKE' of e as

v"',,, cm ¿,bo sho\\' ¡har ¡he senmd estimdü,r < 15 unbiased d ~hni-
lar cakulatíon:

<'!fkient than ti beClUSl'


¡eSE ::::,/2 ls 325
Thís should be intuitivelv dear dE
/1 +,u lhe ínlormiltion in tlw data set.
c:::
VVhich shotikl we choose 'J Estlnutor ti IS biast'd bu! mnre effkient
1;.S unbld~cd bu! it~s:-- et"fic}cnL !n this p¿1rticuL1r ~:,Jsef
~~,",h"h estímator d. \Ve \vClu1d thus be wíllíng to
sinee the sacrifi<.:e io; [,lirlv smali ;~
)' -u,
,<

more estím,ltur. ;\t sorne Causality and Causal lnference


pOln t tol' a Ii ttJe bias
sÍnce we end up guaranteeíng ¡hd! estímates v;il! Dt' farther froln the
truth. The formal \vay to tb!.? biaso{'.ffíciency trade-off is to .
calculare the lIlean squan.' error (i'vlSE), \vhiel< ís (l co~binati{)n oi bias>
and díidency, If g is an estimator for sorne pal'ameter ;' Une Gfl'ek \,rE HAVE OISCLSSED 01' sodal sdt.'IKC rCSl.'iH'ch: SUI11m,1riY-
two
leHer t.-lSE is ddined as .lnd dcseriplhe inferelKCS
the workl ¡nío systt'mJtk dnd nn!1sysh:nvItíc cnmpOHcnts
:::::: -+ 2.6), M.lny students oí sodal and polítical
stop aL Ihis püint, esc!1ev"'¡ng causal statements and Dskíng th('ir se-
=: varianee + bias ¡e('ted and wcll-ürdered facts ro !or tnemseln:s,"
Like historians, :5üCÍiJi scientists m,>ed in summarize histqrÍCal detaíl
square error ís Hms the sum of tht' variance dnd the and ro make descriptive inferences. ror SO!1¡C socia! scientifk
(set~ l The idea ís tu chouse the estímator b witll,)Ut causal lflfert'l1d'>, Th,ll
\-vith the minimum mean square errOr sínce it shows precisely how just as causal inference 15 ímpossible \vithout ¡nfer-
an cst1matur wíth sume bias can be preferred ¡f ít has a smalle"r vnri- e.nce, descriptive inference ilIone 15 oEten anei incomplete,
anee, To say this., however, is nuí to daim tllilt a11 social scientists musí, in al!
Por our eX2lJ'nn'lp tne t.vo a 1<:' as follows: úi their 'work, seek tú (1I::,v1se causal üf the """~''''''''''"Y>,'''TU
studv. Sometimt.'s causal inference rs too diffícult; ¡n mí1nV orher
, ínference ís the ultimate (lE the l'~search
endea vúr.
Of course, we should \vhctht'r thc
-+ (10001
of a n:.~earch pmject or Many sodal
sdentists are uneomfortable w¡th causal inference, are so Vfa!"v oí
the that "corrdation IS nlJt causation" thilt thev \vill noi state
and causa! hypothL'Ses or dra,'\' causal tn [hei,
as "studylng assodation and not causJtion," Others m<1h:
causal statements with case, labeling l.l!1cvJlualvd or
ulations as "explanatíons" on the hasi" ni indeterminate research
"igILS,l We believ(; lhat each of these positions evades the pmblem 01'
Thus, fol' most values (It (jc, :\'iSHd) <: .lnd we ,vouId causal Ínlerence.
d as an estimator lo c.
~In vlew' oí' st'Hn(;~ $Ü~~Hl] '::<L'ienh~üs' prt:~feren(·~: Ú:n pver "mere dt:'~(Tlrtiün/~
In tne'l"n', \ve should unbii1sed estimates thut are as
it 15 not surprising thi){ students üf events ~~'<::t.4·;, to dress thcir \\'nrk in th~
use as muen
tr.1PFlng~ of _¡t1rg~'rn; ~}thEr\vise< iJ:--. inferltí f
rCil1 reseuren sÍ!11ations ,ve this \\"ork. r\t its cure, reoJ 1;.: aasf'd (:in (-¡:H..b31 jnferen(0s. V\'e rt~gard
trade-off beh'lleen bias and is quite sillient rnent~ in the hterature ar,out H nO f\CciU9","d 1n
itIl GlS'-""5, th("~e ~lfgun'lents
arf" ~\r ;~n;:,
~n{,\)nsish.*t)L H St~iaI s(:i(·nh~t,-¡' f.áhlres to ('xplaln ~H'{~ nt)t due ro p-uor n.~'Se~lrt'h 01' lat k
0i ímilginiltí0t\, bul rathef lo lhe nil!ure uf lh,' Jitílw!t hut ¡ha! ti".?\,
,\re examlning, ~uch oi ínferiority are C(,o<l de!:'criptíon oí ímportarH
"venls ís helter than bad explant1tiün nf anything.
//
76 . Causality and Cau&.l Infc.rence

tigatiOl1 eHh!!r renders the rC~Cdh'h írrdc\'imt ur


u~disdpljnL>d PV th.., mIes uf scll.'ntifk interencL Our uncerl¿¡inty' :;;tatus [elf <l Dt'mnuatic c:mdid"te iúr tlw LS, House ni
aboui cnusa! will never De climindred, Bul this uncertaint)' . on tht> rr\)porbnn n( \Voh~~ H1is (t1nJid~tte re()?i\<t~~,
sh()uld nol th;ü we "void a!:tcmph ,lt causal infewl1{:t~, Rúther il Democrati<: l:alldídate thv , Ld th,.'
we should dnl\'; caus¿1i ¡nierences wh(~rt' they set~m appropriate bu! \'aridble be th~' Democratk proportíon of eh\.'
,1!sO prO\ide the rcacler with the best and most honesl estimate of fOí the House, The kev caw"ll explan,ltnry \'ilri"ble Ís then dichot-
t.mcertaínty of tÍ1dt inference. rt is dppropriate tn be bold in dra\'v cHher lhe DemcKrat is ;111 incumbenl ur IwL (Fm simplídty
Gntsal inierence;; as long as we are c,ultious in ddailing the u out this 5edion, vve nni)' (onsider districts where tht' ¡~epub!i"
I;ünty of the il1ference, H is importdnt further, thnt GlUsa! hypot candidate los: the lasí cledion.J
be as as possíbk the mies of
4-6 ís to varÍ<.bl¡," is sOlTletímes ('alled the "outCO!l1C
cIrcUmStances under v.'hieh causal inÍf:rence is appropriate and varia bles" are ofkn referrpd to as "índepcn-
mak" ¡¡ for qualita!ive n::sean:hers to incrcasé [he dividl' the variablt,s into th!:'
ihal: thcir re<;t?ilrch wil! reliable evidence <lbout their variable" caned the "causé"
thl' ''('ontrol variables," causal \'driabk
in sl'ction 3.1 vve appro- (tU two or more \',¡llll'S, which are oflcn denoted "trt't1l!rWl1f
íor qualitative and tlwn in section 3,2 \Ve and "control group,"
sf'vc-ral ahl'rnatíve notlons oE catlsality in the Hterature ana No"" consíder onlv the Fourth Dístrid in \;e\\, Yurk
dL'nHmstrate that thev do not cont1kt \vHh "ur more fundamental defi- ill1d imagine aH elecÚo!l in ISlqg with a Demelt:ratk innuub':'IÜ ami one
nitíon. In s('Ction 3.3 \\'(: disL'USS rhe assumptíons abour Rcpublkan {nonincumbentl thc Democratic can-
"vorld and the to make reliabk' GHlsal inrerences. didate receh,'ed fréKtion pf the vote in !his election nlw '*
denotes the Díslríc.i in Nt'w York imd Ihe to
(Tí teria
we '~"'" "".".,,,,,,,,,'t the fae! that rhe Demn(Tat is <1n ,'d b thel1 a valuc of tlw
",ve cnndude dep",ndent variabit>. ':ro ddinc the causal eHect Ll tI:nw"tínrl ljlhmtitv},
strw:t GlUsal imagine that \Ve go back in time lo ¡he siart oí the electiün
ana rernains the same. !hat ¡he lJenKKratic in"
no! lo mn fOl" re"t.~lcct'i(ln ilnd [he lJemuCfatk
3.1 DEflNING CAUSAUT)
nominales another candidate the winner lit' the primary
\Ve define ,15 a theurcfica! eledionL \Ve denote the tradinn of th" ,,(,te that the Dernocratic (non'·
of the d¡lGluSE·d ro learn about innllTtbentl candidate would rece!vc'
"h'Y,""'"" fmm our data, iFor prob!cms of causal cratic candidate \\'ho is a Non"lncumbent),'
see chapters 4-6.) In sedion 3,1,1 h'e give om definition or This conditíon IS (he essence behind this definitíon
causa!ity in full ddail, \vith a sim,ple quantitatíve example, and causalíty, and the difierence betwi'cn the adual vote dnd the likdv
in secnon 3,1.2 \ve revisit our definltion altmg wÍth a more sophisti-
cated whi1t he G1Ils '·Rubin'" \4udd" Hnlbnd bi1SC5 hb; hit';1~ "n (he work ol' n\!men'U~ :,eh..,;,
¿H~S, Lx}n:lld Rubin's 097-L F~7b? \.v()rk nn the subj"ect \Vdt..: Hh'''':! n11tncd¡(11d~' rl~lc\'.;lnt t~ut
he ,,10.0 dks Ari',totk ¡A,,~ke, Hume. Míf), C~rangeL Fl~her( iUl,d ülh*2>rs,
3.],l rile INI:: e'xlt'nJ H"lbnJ's ddinítion oí ;l ,<;¡uSll b.\' u(.,lng sonh~Á id'f~~tf
\A,,\~ f"nUll(l thl~ t,,"x.k'n-
N

Our theorerÍl:al ddÍnition of lTl0st ,md s~cn nt?"(t'SSJfY nu c'\~~tín!;


;,;,Ü¡Ct..">
lo a As sectipn"\ d unit ¡s une o{ tlw many rt~pc-ct to a unii ¿in.,! ~hn ,,-üh"v/~n~\ nnc to partHú,n (~1tt""a¡ cíft'>('t:>.
elements to be obsefved in a study, sm:h as a person, country, year" or "!id non"YSl'em,ük cllmpilnenb,
5ee (~e¡m¡¡n and Kíng 04<Jül rOl' detaíls oí thís example. M"rc ! and IV (MI
lOUf point t}f departl1!'€ in thí" ,,('dio!'! is Holhmd's ;.rtide 0986) un causa!ity ana "tana for tn.. -"treMmeot" anO "c'ontrn!" group 0f inf ,\ny two treatm~'r1ts f.'xperiment;¡!ly
Ca usa litY and Causa! lnft?rence Ddining C1USi1nty 79

\"uh:t 1:1 this ,,:c~llntt.~r!J('tua] sllu;ltipn {JJ(~) :;-.. ~'hf' \~.auS!1~ t~i(~:.('t
Qt:.llo\v. \·Ve ITlust l)t~ y,pry ('a-fetul in
although they are obviau"ly COtmter tú the
they musí he reasonabie dmi it should be possibl,: fUf the w Causal Effect fm unit i) ,:;c ¡¡Í ...
tua] 0\'ent to nave occurred under precisdy staü?d círcumstam:
key part uf defíning the appnlpriate count"rfadual eondition is onlv in t1worv since in til1\' on" r,,'al
fying what \ve are holding constan! "",hile \Ve aTe ['1'!An,m' \Ve might observe eifhcr pr or l~eíthef<. hut n~~ver oolh.
the \'aJue of the treatment variable. In the present example, chis simple definítion of GlUsallty demonstrates tha! \Ve Ci1n
causal (or treatment) variable is íncumbencv status, and it hope to know a causal effect fm certain. t:Í\,Iland 09So} refers lo
fmm "írKmnbent" ro "non-incumnent'" During this problem as Ihe fundamental l:aw:a! ,2nd it is ín-
We hold constant up to the moment the '-"'H'''''''' n'¡"'l'1>':I17,'H problem since no I1Mtter hüw the
nomination dedslon-the relatíve of the Demt>- no matter ho\<: much data \Ve no matter !ww pt'rcepti\'c
crats and Republicans in elections in thís district, the llature oí obsern,rs" no mi1tter lIow dili¡;t'rlt the reseafch and nn
the nomínatton ¡:'1í(l('ess, tht: (lf the n,atter hmv muen control \\0 we will never knovv
triet, and the economk ilnd políticaJ infen~nce for certain. most nf the empiric,11 issues oí
l70t cnnt.nl for nI ¡he desígns !ha! \ve discué:s in thís book ínvolH' this
and 01' tu€' problem, and mosl uf OUf constitutc aHcmprs to
else that folkHvs the party nomínauo!1. it.
Cm!ScqllCt1aS of our treatment Our working definition oí differs fmm Holland's, sinee in
ot indude name ,\i."rtion 2.6 we han: ilr~ued that soda! sdence needs to parti-
Si) lorth. H ,Ve díd tite world into systematic and nonsvsternatic components, and
and hence detinition dc'{;'s not make thí...<; disunction dearJ)'~" lo see the
and as a oE this partitioning, thínK abúut !'"ha! \.vould if \Ve
total. In faer, rtCnm ¡he! CJ98 eiection campaign in the Four!h District in Nev"
could milke one ,vith a DemocnHic íncumbent and ,A
fect at 311. 4 sHgh!!y different total vote would fea-
f'vlore tlw causal effcd of in ¡he rClUrth Dístnct tures oí e]ectlon
in Ne\v proportion of the vote re.::eived the Democratk OH id en-
cand.id,üe that is aHributable to featurt"S might indude ¿l
th<~ difference behveen Uf positíon on an íssue. ¿n
that \ví!l become deilr <:.nnr'nv had \.veilther during one
canojda!e's rall\" or on election or thé results oi some investigatlve
admlnistered in fact ni in Oí COUise, the dedskm In c~lJ (HIt: value ni an .'Yl,H"",,- We can therefnre n ":ariabJe that would
¡or" variable a (rcalmen! ,md the orher <l cüntrd is ii th!s "",,,,,,,,,,,c rhe v,llues of the [k'molratic vote acmss hypothetical replications oí'
used al al!. this same eiectíon.
4 ,lon E1"t¡;r n983:34-36) has daimetl "¡he rtw'"nllrw'
counterÍactual ~t.,.ltenH!ntsF tn :-:- Vie can for district ·1 t'tv h4"" ÚTf h'1 the
((",m!s for t""th Ih" ilppanmi eXiPWn;¡rOl, 'Th(' reason for this lS thM Ho![¡md j" a sta!ísbda!1 who conws V(~ry dose lo
Elster is f'",'"'''''''''',''' 2" rand~jn:1 vari~)ti0n, \vhich is d1..~~Tit"X.~i in secücrn
íL,,"t"'<'nti'cm ut' tht~ ~·:-tt~lb~b\.~':lr Sdfutú.}n~·· tu the rm+,[,,'f'I1 (d CiltlSill inte!1.~nc(\ h(~
hui ibb d~úrüt¡{~n b
cü'!mt.>r'Ü'i<'!n,,¡ ~t~h~n1erüs '~h"Yv{~ ~n irrliporLlnt ro~(' in (,Ju~kd dh~Y.!t u<;in~ dUh·),rent
unib ti'> ;.:.,(,h:l' thc' FUnttn1k'nL)~ Pr, }h~(>rn ~H."",L'fH:i f"t'taining tne
rh.'llce Elsk','" argument 15 mOr(' ('(lijen!. we thínk. as i) se! ni valuilble against definition nf ctluSdHty in ju~t nne, in his t/~~hH) !.:~ptEÜpr ~'1Veniges
careJ",s$ use of count~,rl¡1Ctuil¡S lhan as d critique of their fundanwntal deílnitíonal ímpm" ov\~r tlnits~ tvhereas ours úlC5¡;,:ribt,~ tlvt:~rages O\'e"r oI'
mnl.'e in causal rea,solúng. tne ;;anlt~ experimt.>nt fur just a single un!t (se~' Holland 19Só;947).
Definin¡::, Cm~a!ity
, '
E"1fen:. :n(l~~ rIH..<l!1:1 ,JfhJ \'arT,l.H(,·e:-; an,,' dl:---~} ~\-~-
;\5 rott:d ,1bO\"(~ (S()¡C ~eCt1{n1 ~ ,t ':), t l'ü~ '\:;~ l~
ft:;Bturc~ '-.:f {"¡)nd0111 \:arlables {~~,~ in ~eLti~Hl 2.2}, Se'--.:ond~~', it
\'<Jliablc" :ÜnCL' it hilS il
us to p¡lrtitíon a cHl~<11 inferenct' pwblem into
'-'<inaNes not
tic c"mpOlwnts, Althnugh mi1l1V
,'arí.1bl..:' be o( interesL the most reln':mt fpr dur run,
Demo(Tiltk candidate as ex.u1Tr'le ís ¡¡w ?l/can (dH:"i¡[
votes that ""oHld be rcceived in bV this, v,'e return lo our NcV\' Ymk dl'ctlon
~ tnat thí:' \'ariable rdcrs to ¡he vote ír,Ktion r<:cclved
era tic noníncumbent as
the Democrat (inüm1bent ur nuníncumbentJ iKt\ISS " nUBlO\'r
V\'t" nm..: ddim,' the miliÍom owsaf ior dístrict 4 ,1') the
bC!\Yt'en thest? t\Yo r,mdom variables, replications (,f the Silmc clcdínn. vVe dcfílh'
wc switch notation from district 'f LO unít í: oí this random \utc imdíon

is a causa! dfect that varíes 0\'\'1'


but «Iso m,mv
ff \Ve could obser\'\' ¡,,\,oc
{,ict .f al ¡he 5ame time,
D€nlOcrntic incumber1l' cffcc! of íncmnbenc\' ín unit i is ti

rcalized c¡msa!
we ca11nol

I, nther
the re{lli;¿¡~d causal effed in
urwl'serued reaHzi:ltinn of the randüm causa! effect
,Kf('5S m<1n)'
3.1
of the same
tion in dístrkt 4 lNith d Democratk incumbent,. and anoss manv
fffect tiJf unir i)
!hetkal of same dectinn but w;1h a Democriitic
renli7ed ciiusal dfect becnmes a nmdom

. .l ce

il makes our definihon of


iltic

As W\~ in mur<' d,'íail in s¡~d:inn 22,. this úm L,,~ (ontu"ing. A "'ran- v,,'here in the first Hne uf this eqmttloH, 1) (beta) lo this
dmn vari."b!c'· cpnl;¡ins S<lmc (~om¡:XWlent and t'lms is nHt tw~ causal effect. In tÍle second 'we indiGlte ¡har tbe mcan
\rn"""" .... " this H".,","""_'
[Oí unit ¡ is just the mean '-<llue) of tite randüm
«nd in the [hiTd and fourth Unes \ve show no\\' ro cakul,üe ¡nI:' mean,
The last Hne 15 ;mother ,vay vvriting thc diíference in ¡he mean!:' (,l'
lhe t\\'o sets oE e!ectíons, <1\'cr;¡ge ,,( lile díffcn:na!
behvt:ell t\VO random variables the d¡flen:no: oí tht:c
peHn FGll10mic ur tlw um:erLlin úlnSÚlUelKC nf ,1
To summarize in words: rhe (i1!iSi,l is [he bdaten¡ the
change in ",lector.1l Systt'HL Tlw !O (lar repreSt'ntatícnl 15 tlMi e"eh of ¡hese
"random" procesl,cs have systemalíc :md pmbabllistic components, temafic component of obsenlfltüms madc whcn lhe e:rplanaf"ry nniablc
82 . Causality ilnd Causa! Infen:~¡Ke Detlning C'lUsaJitv ' 1'3
,"1ftC ¡·;;lue anJ I1::' "::;A>~?~~!;:;Jtic , F\.''I'( eX~1mrle/ une \)f tht} lnai,,~r l'lucshcH1> th.;Jt tdl.,.~\;~~ thp~e in""
u{lr}tibl{ ((!k{'~ L?! {ohJther ;\"7lru.,', \,,~ith pülitics and has to cid '\\'Hh the Ct~ns0.quence~
'fhe 1.1st línt: or' 3.3 i5 similar lO ~~qu"tíon }, 1, dnd as particubr law or passc,; a tax biU ¡hal is 1n-
rhe Fundamental oí C:ms"l Jnference sUB exists in this to have a particubr to ¡xlrtieular invest-
¡,,(ion. lndeceL the this v<av IS 0ven m,lre increase revenue by a certain arnounL and Cl'!1sumption
ble because t:ven if we could gd amund the F¡~ndament¿¡l Fmblem Does it ha\'~ this dfect? V'le can ot-sene wh,lt after
a reaHzed ca1.lsal we would still have al! the usuaÍ pmblems , is pas..';(~ to sce if the íntended o.lflscquences ap¡:H'ur; bui even
inference, induding the pmblem of sepRrating out systematic do, ít is never certaln that they rC5ull (rom the 1<1\V, The change
!1onsystematic cOHlpllnents of the fandom causal eHect From he .. ""·,·,~,,,,·t políC)l might hav\': hilppened lf we could rerun
\Ve use HoUand's phrase, the Fundamental Froblem uf Causal 'wíth and 'wíthout the rte\v reguiation" \Ve ,yould h,we
,-"nce, tú tefer h¡ lht' (ha! he idc.'lltified a:, ereN tE to rhese more lt~\'eragc in estimating the causal E'Hect of t.his la\\',
dard 01' vvhích \\'0 llave added to his we Gmnot do !his, Hut ¡he l.\'íll research tC)
In the hox on pagt' we prov;de a more genera! notat:ion 1m us dl1 approximate ans'wer to our
\,,;hicl~ V.'m prove useful thrtlughout tlw í!;'st oí thís hook. nm\' rhe !oHowlng extended
oíher of thesc 1'0ndom causal effects In the vVi1ke (lf the coHapse uf the Soviet
he nf inteF'st in v"riol!" cirnlmsti1nn's, Fur wis in rhe ex-Soviet republics and in Eastem
kno'.\' {he varíancc in the (rt'úlizcd) causal effeds uf ' uted ne\\' governmental fonns.
status on Democratic vote in unít i, as \vith the variance realize-in a political (~xper¡ment they dre ¡ntroducing
¡he vote itsejf that we described ln equation 23 in section 2,6, To constitutlons, nmstítutions that chey wil! have the intended
latE' the YilrialKE' oE the causal we oi creaung stable democratk s:<'st~rns. One uf the constítutionnI
:rh'Oh~es is betvveen nnd p1'esídential iorms ni go\'(~rn-
(vari;,U1ce of the G!Usal effect in unít i1 ~: V'/h1Ch 1S more líkely tn lead lo il srabl€' ¡:.; the
of consíderable debate among schol¡¡rs in the fíeld (Um:i
in fOí tl)(., 'tz 19L/:'; Ujphart '!'}93), The debate is not lile leas! be-
varianct> , ), new íncumbenis (lf tÍ1e numerous of svs-
,,'¡sh LO know the variation in the Glusal dice! oi incumbene)! so . fems and the of that migh!
can hnw 'wJ!l be to that of ,.,ro""',, and interact 'Nith Hüs choice ísw:h as the nalure 01 rhe elec-
Ir Ís not om ro a !horough oí
from lilese choices buí rather a n~rsion of the choice in
underst,md that Ihis vMlanu.: in order (O define a causal effec! in the context of this example,
oí' ¡he ,',>,orld and is not ,m,.'Ú~"~ ,'>'e highlíght the distinction between systematíc and \1on-
svstematic of a causal effed .
3J.~ Ji . The debate about presidentíal versus parlíamentary' 1n-
irolve:s varled fearures of the two s\'sterns. We wiH focus on two: the
\Ve our precise definitjon of in section 3,1. extent to \vhich each represents the varied inte1'ests oE the dH-
sume of the in that 5c{.-:-tion are subt1e and quite soprusticated,· zenry and encourages strong and dccisÍve The is
lNt~ iHustrated oul' \yith a running example fron': th,:¡t pa1'liamentary do ¡) better of n:prcsenting the tull
quantitative i\?search. This heJ¡''Xc·d as communícaie tile con., r¡mge of sodetal group5 ¿¡nd ínterests in the 51nce there
\ve wished to stress '\vlthout aIso having ro aUend to the c()ntex~ aH' many seats to bí:~ filled, and can be fmee! repre-
detilil 2nd that L good qualitativ.l: St,'ntath.·f.:s clte'cred fmm v,1rÍnus gnmp5 !n cnntrdst, the all-or-nothinf;
rescaTe;', In thís our ddinítion oE cilus,,11- Chari:K1er 01 presídentia! mean:; (hat som(' groups \vin feel left
ity again, but this tirne vta a quaHtative exarnple. out of tIte govemment, be disaffected, ,1nd cause greatcr in.<;tability. On
Poütical scientists \vould lcaro a lot if they could rE'run history 'l\tith th,,> othel' hand, parliamentary systems---especiaHy if they adequately
everything consh'mt save for one investigator-controlled explanatory represent th!.' fun range of social groups and interests--arE' Hkeiy to be
84
id
ch(1r~)cteristics" h)\), (\H1 h:,Jd tc~ di~{1fft7(tit)n Jrh:1 ~n~.t:ibdity,'\
in tl1jS \.~ircun1~LH1(t\ it D1av be thdt
JI1\.'purros,,' uf this ",cdion 15 to d precise ,ü"-I'!.,."t-i
a causal dfe'ct Tp d" S,L ¡hat we could íns!itute a noliticdl le"der;.; \'voukl h, chousc 3rt o¡."\tlon ¡h"t
r sHghtl v lt.~->:-- .stabllít\~ \'H) il\Vl~ragt} but h .1S a Icnver \'ar1'-
<-,' 'and thU5 minirrd:lcs tht' i.:hance ni a ddsa~truu::.
degree oí demoCTalic stabílíty {perhaps ,1I:'1ual survivilJ or
dernocracy, attempted mups, Of other indícators oí instabílity},
¡'he Silnie (ountry and dI the samc time, ínstítute a presidential
aiso íts stabílity over the sm1\C period \vith the Sil
sures, The /'calí::ed mus,,! would be tht: difference bdween
gree of observed undeT ¿~ .1.1. \l,/e defincd in terrn~ of ?l
a The -is lhe diffcrcnce bd"'l.'ert the
variable \\--!len t!w
In this sedion, we use (lur ddínitinl! oí " , ro ' ___ . '{,
dnd appi1rcnHy (()mpllCi1tmg H.1ea:-, \'V\'
pv nrher aboué '\:allsal
dl'hJ
m;:my
oí the realized
¡he av l~)king
nf this pmblem to
out ilnd Jeavcs the lT\t'dn causal dfect to ¡nelude onlv ""ch-"..,
tures. Svsternatic ft:atures include indedsivencss
",,,,,,,t,'n-, or disaffection among minorities in a
temnJic tealures mí;;1ü ¡ndude t11e sud den ni él president
¡hnw¡s the ínto chaos, TIw taller event would not
it \-','ould appear in one "fe
oí t:he a b t11c diec! \1f " lWW
on n:ducíng the Uníted curren! acccmnt defícit \víth
(me oi these lo OUl' definition uf the eHect herc is
,,,eh,,,,,,, would be ínlercsled in íts mean causal dfect on curren! aC(Cllmt defídt \vith thc taA
stabilítjt; however, ¡his om: one N;n~,...",rp(1 to the same si!uation (ilt the s,1m,,~ time and for
uf this Given this situiltion, pol:itkal me v/ith the exo.:,ption th<'lr the trealy ,vas not in effecL
may be interested in more than the causal dfecL They causal mechanism her€' would indude, in tum, the
\ds11 ro understand tvhat tite maxímum and minimum causal and rntification of lhe tax ul the event,
or at least the paritmec oE the causal might be. For mee!ings of the r(:le,'ant actors vvithü: majo: multination,11 , ,a
mav be tlm! reduces dcmocratic stabilitv on in actions ro reduce the!r total iax burJe~l
(such as its transfer pricing mIes Uf mO\'tng manubduring
'Thc~\! ciis!Índions JR i hemsdv('s dvhated. ~m1V Jfgue lha! ,1
can du ti bettt~r jo.L~< And pth(~r:~ t)rgu~ that plants between countríesJ, further actíons bv Giher _ and
workers to lnke 01' the rnovements ot .. c1nd labor, bt"
hveen countries, and su un, unhl<, \ú,'- ¡vai.~ h'¡" ",e dt'n-}l "'te'"- on th\"
, L,,, ' ' ' ,"1"1,
",
nn"<l,:w,r ~·~ppt.\1 r~ iD be .,1 n"n~,_ q,'m:mc
,mee of payments Det'\ve-en ¡he United Siates ;m~ Japan. ,
oí svsk'm, On tllC other h,1i1d, the \'uhwr;;bílíiy oí
From the standpoint of pmce~sl>S through WhlCh causahty
systell1~ tu !he' v¡¡ganes of tiw Jwalth ami person¡tlity oi ,; single individual ís a
¡¡tic efk"Ci ¡hal r¡;&;e; the likt'lihood th;ü j(1fHi' nOl1s}'stl'matk featufe 'wíll ,'pF'Ii'ar,
&n E'mphasis on causal mechanisms makes inhlÍtive sense: an)' coher-
86 CausaIity and Causal lnferenct' Altemative Definítions oE Causality

C,Jus.al nl(~chanisrn~ i~ v\·a:·~ of dc\in¿; ern causal


t1t""\\" to \"<c should not
anah/ses. H has b,,~t~n in dl1ferent forms, "pro.:ess 0:)11[115('a d",fíniÜp!1 oi causality wirh the ntm.:idinítionaL albeit (,fb.:'n
íng" ttvhich Wté' disellss in sectiuu "historie,,! anall'sls:' and . operi1tíonal pmeedure uf idl'ntifying c.:H1~a¡ nwch,misms
taBea case stuJies:' of the derails of w('ll-done case
involve iJent:ifying these causal mechanisms.
''lv1ultipk Causalily"
However, identifyíng the causal mechanisms rE'(juires C<lUS¿tl
enee, tlsing the methods díscuss{'d belcn\'. That is, to demonstrate Chark'S Ragin, in a recent v\'ork argul"S for il methodol-
causal status ni ('aeh pütential linkage in su eh a positcd mecha ~"ith many explan,ünrv variahles dnd fe\\' e>bservatÍpn5 in order
the \,>,ouid have ro define and rhen estímah' ¡he causal (me can t;;lke ínto account what he calls "multiple GlUSiÜíl'n." That
js, "The phenomemm under has altemiltive JdcTmi-
more'
OUT oi ca nants--\vhat \·Hll OH4:'\1 rdcm~d tu as thte' ot nf
ci)ch link in the ch,iÍn ni causa! evenís.
fUf causes.'" This 15 tlw pmblem referred f\¡ ilS gen-
our ddinition uf prior lo Lhe id",!! 1982:1n In slruations cau-
tion 01' causa! nlL'Ch,misms, Furthennore, there exísts in the these dutnors argue tha! Hie S<1me outcome can bt' GlUst'd
cia! sciences an infinity oE G1Usai between any hvo línks in combinations ni difieren! variables, ¡¡l
chain oí CHIS,! mechanisms. lf we that an Under nmdHions in vvhich díHerent vdri"bles cm ¡1V
causes a dependent a "causal mechanisms" approach
re{luire 115 lo ídentífy a lisí oí' eausallinks betvveen ¡he two
This defínition would {liso us tn a series of caus<11 t.hese variables ha\'(' cilusdl status, is correct tilar some statlstíL<,l
ages, to causdlit:v ror each pair 01' conseculive variables in m~)(1JelS
(or ndE'\'anl qualítative researen could fail to alert an
s{~quence, ¿¡nd to the behveen anv twü of these investígator lo ¡lw existence of "multipie .. but
<1bles and the coml<.~·t¡on5 (lf' variables. This models can handle situations líke tbes,,' (50me oí
al no time it discussesJ,
definition of causalitv rOl' au\' one Cduse and Ol1e the fundamenta! features oí "multipie are
In our of the effect of a' , versus parliam compatible ,·vith out' deHnition oí Gmsillit!,. They are ,11so no d¡fh~rent
¡he h;'Tothesized qUilntit¡¡ti\'e Ihan resean:h. Thc ¡de;) cont;lins no ne\\,'
under a ".,.'oc"",,,,, reatures or fnr ccmsíder the hy-
pothesis th<l! a level nt income
nona1 attainment ünd educated
ano, in rurn, both 15 insuffident. In this case, ;ve need io compare
couId monitor the attitudt'S or causal variable: respondents who h.nve high educdtíonal attaÍmnent
differ under tbe tvvo experimental ano highly educated parents, the t\VO wllo have one but not the
oi rile under eaeh sj'stem. other, and the ¿;TouP wjth mdther. tht~ concept oi cau-
sation" puts greater demands on our data since we no';- cat-
variable t.~itht~r nt..'C(~"
t'l~ )\\ '~~\ 'c-rf
th ~s.
btXi:1U5t:" thl~diFtüxiú.rn bt~t\.Vt~pn fll\;:''lSSary and suffkient con-
3i'./,n,,,',u< ",hen we aJlrwv for tlw
,,,,,,h;,,~ NCllnsídcr the cLúrn th;,.)t F>nf>¡t UHnrnUn1(dtJun ún1\.~ng surcr··
[he plhieh a cause has its
P(l\VcT~ Juhng c["jsis jn\7i~<ts0S fht~ hhcEhOt)d oi \.v~H·, This ~~ (J daárt jt iJt.'D·
for ,1 i.HId is a verv userul opcral'iol1aJ tiCk>;; ,1 G\lh,¡1 Vilrí,¡!:Jle (¡:><1()r (ommnnÍ(i\linn) ,¡no ilSSHts ¡h¡¡{ thi$ vilri;¡bl,' ín.::n::;:¡;,vs (n..,
procedure. ldentifying causa! mecnanisms can ;'omet:ime~ gl'le u$ . probabilit\, of il outn:mw (w,u). It cannü! be tr;ms!at('(l into a daim "prmt tl:w m~:-
¡nore over a tlieory by makíng observatlons at él J?ssaryani! conuitíons [ur "",,1',however; i! is iITcducibly
(ji {lUT C<lUSal \~drL¡l'\lc~. tYLH nut
definitlon of (>,)lv··~~'·dl1·V, For our lhtil thi.~
per~(\n h3d a ditft'~Tt~nt irrt"le\'~H1t jnb ~ur
income for the S<lm~ pt'r,,,,P1L al Put diff(~rentlVt thi~ dlternatlve (\yunh:rt'ácfual 1S tht,:~ eff€ct
cxperit:ncing c¡)ch Di the four conditions, edm~atí()n cll~lparcd to that ni mme, wilh seníoritv bdd
13ut what happens if different GlLlsal FailuTe lo hold :icniority consLmt in the h~',\ e,lUsal ümJi-
vdlues üI rhe dependen! variable" FUf e\¿¡mple, would CdUS(~ any 1\,se.1[cl1 Jesign t,l yíeld esHm"ües uf pur i]rst
\vhether o, not (me gradudted from ínsh;,)d uf ¡!'tis revised ont', H the laHer wefe ¡he
cilusal variable in a popula!ion oí ',<'b",r,,'
both groups ((luId quite earn the Silme inconw
pendent variableL Om' n,ason might be that this
attendance') ha~

nf irlúlmeíOf ¡hose eduG.!ed and


earn" particular lt.'vel oí income intluenced caus"l fadors do",s not milh~
who no educatíon might pmblematic T!'tt'
their {OUT VCM!> oí údditYOH' ver\' úHnmnn situatinns is h) define tlll'
on thE' In thís situabon \vouldn't \ve iJL~ lea lo conclua makifl Ob unr edeh eiluSill effet't ven;~ VVe demonstrilte in
educatlon" has no causal 011 ¡ncome levpls 5 that researchefs need flor "all" causal on a
>-'.'no win become 1yorki2'TS? variable to pnwide estimates uf the Orle causal efft'ct of in-
OUT' definition 0f (even ít that lVt>n,," A resean:Jwr can hiell'; nn ,miv the
che condiríof1, In the {}ne effeet nI interesÉ. estdblish firm and tlwn m,)"e on t(1
chal nI;}\' be oí ¡nieres! (see ~t'd¡(lns :;,2 imd ,:::;:1). l'

son bctwecn wh,,[ he refer,; to


bad; in time tour ¡le 15 ínter·
ínstead nf jn \vhkh díffer when an
Uf her income " tn wltt'n it is dCcrC,1Sl.'d.
levels uf income rOT this une
tne CilU'i<li infllH::nce üf x; tan \'dríabltd un '¡ jil
causal dfect in ¡he 'Ve have imagined that
dependent varí.1ble!. tor (>ne has a!so lo considc'f wht:ther shitb lo
causal effcct is zero, But this croes no! me"n that "college t'.~dtJcati{}n
no on N that the average ' a V¡¡!U0 nf Xl rrüm either dír0...:tion h.1v0 the same ctlllsequences íor
y,. , ¡f tht, causa! relariol1snip bet1.vcen Al ~,m
{1 j and (2) is 2em, In therc ís
"tIte causal effl'Ct of (:nHeQc
no! define a causal diee! without al lcast teVO
tlons nect:! nü! be rhe hvo Usted !'tere, but \ve kncn·v ft 1S r,,:llh~r than detenninisHc t~1en(c.< \.ye ~lbo di~;gn~'t~ vdth
pn'mise 0987: ¡:;) th,,¡ ni ¡he cnm-
pHrath/t~ rn('t+tnd are nnt ,,~fHl(e}\'t-'i.l in
of CdUS11 condí!ions ls lo
rhenoml~nün r~ examirH:'tl ~lnd tlcc-nunted t"nr ir h:~ l"ut··
lIitte vd!h someone \vithmll '" but with thc same
kx::t i1 \x~nsn~uf intuffl\,,"ltíon on ev,,'ry in~L~:uh.:·t· ilnd t~\·er:' rt.~nrtutdtlnn
¡ob seniority as the college graduate. In one sen;:;€', this is u ¡¡oc! cümbírh1ü,m of "al m.'!, ,,1' Üw expi;m,,(ory Filrí¡¡blt>s., lhe wor1<:l stíl! wnuki h¡¡n~ pn>-
since the non-L"ollege graduate wou!d have to do something for cluero ¡hes<' d,lta lo some probabi!í"tk pro,,;;ss (,b defhwd in sec!i"n 2,6), Thb
90 Causality and Causal Iníerem::e 01

than it shtHJld be\.~J,lu~C \~' . ,' n~/,-'tl~~~1:-:1y JiS(tlrded di~tri~~ts


tlw DemoCTdt lo"! dlé' tirst elccií,'n:
us Líebers(1{l's conú'pts \}[ "svmmetrical'" and "asvmmctri('al"
c;v are imDortant
, , to ~'onsider 'in the cm1tt'xt ot (,du~a¡ inÚ-rencc, ,
As ,111 (1f Lícbt'rsnn's point, imagine th,.t the Fourth
, ,they slu,')t,lld l1üt be confust'J with a thülretical detinítílm uí
gression,11 Distrid in l\f;'w York had no irh"umbt:'nt in 1998 ilnd Ihilí in sechon 3,1,
Democratk candidate n.::ceivt'd SS perú'nt of the vote.
would definE' thE' causal effc'Ct uf incumbenc',' as lhe incn:asc
vote if the vv'inning Democrat in 199B runs as an incumbcnt in ¡he fon ESTJ:'vJATf"<¡G

e!edion in the vcar 2000, This cffed vvould bt, dsymmctrk" if the
sence of au inn;mbent i.n the subsequent e1edion (in ycar :::002) do Wt' avoid the FundamentdJ Prublem o( Causal lnfcnmce aBd
the vote to rdum to 55 TIw eH0i:1 might be "asymmetric" ih", problem l)t {!'Pm nons\'st.t,'mdtk \:omplJ"
íur example, tht.' íncmnbtmt Dcmocr¡¡t raised moneyand íntpro\'t:'d The iuU ¡¡nswer lo ¡his question wiH consume bu!
Demtx:ratic campdign as ,1 fesuIt, if nn incu we providc .In pverv¡f;w herc (1f wh¡¡i 1" in ícrms üf ¡he rwo
v,:ere in 2002, ¡he Democrabc candídilte might 'bIt: assumptíons !hi1t enaHe tlS 10 arounJ ¡ht~ fundi1mcntal
than 55 of the voh::, km 'n1('Sc dTe IiIJít \\'E' discuss ¡n seclínn 33,1 )
UebersoI1's b dever "ud ver)! and tDJlditional {~{?ction 3.3,2), l'llt;;~se l1ke any
\'icw" nis does nüt cCHls!itute a cther attempt to círcunwent the Fundamental Problem uf C¡usallnft'f-
pHes enee" ahvays im'olve Sllmc untes!able ¡¡ssumptíons, H ís ¡he
causal viiit\' ni aH fesearchers tn make the substantive implications of this
" sppt in their r("Sé¡uch extn:mely c!eilr and visible to read-
derined 011 the bas¡s of E'vents occurring Causal inferences should no! appe<1f like , TIl\? ibsumptions
1I in the 1998 election in the fourth Disfrkt in Nt~w York Om d ami slmuld be wi!h wh,,!ever síde
ís the difference in the cmnponent of the vote in t!lis d cm be mustered, bul it mus! be
\vith an íncul1.bent in election and wíthüut "n incumbent in
same election, and distrkt.
In contrast, LieberS\)!ú,; involves no hypc)thetical
aud therefore Cdnnot be d causal detlnitiou, wíth
onlv '.vhat would actuaHv o('cur if the yalu<-"S of our ,'ariabl" eaeh time-as a tTue ::;olu-
tv,'¿ real t:,lections .!Tom n~n1ilK"1nnbent tü tinn lu !lit' Fundamental Problcrn oí Causa! lnterence woukl
to nonincumbent in hvo other élections. empiriGll we can a!tempt to makc " second-bcst
this wüuld ínvolve mrmerous problem" of infenmce, We experíment in two different ¡!!lit:; that a r e " Two ¿mi!:>
cuss manv oí these problems of causal inference in chapters 4-tL in ,1fe hamogew?eus when fhr ,'(¡{ues ¡¡f [he
present e~ample, \ve might ask \vhether the estimaled each mlit are Ihe same <rilen ¡lur ;lt/rú¡ble itlkrs OH fl
larO'er on1v beeause we failed to account for a number of oalue. (That is, and For example, if In' nbserve
él , h '
registerL,<1 cítízens in th~ fourth District Dr, did t e surge m su (an incumbent) in district 1 ,uld X-::::: O {no incumbenO in district 2, <111
for irte Democrat in the elec'1ion in \vhkh <¡he or he ,va" an iDcum assumption (Jf unIt rneans that ,ve can US{' the observed
pmpnrnons nt [he votl' in !\vo sepdriltc districts fnr about thc
~l10t,¡-it.\1 n as d causal effect ¡3.. ,vhieh we dSSUlne is the same in both dístrids, Fnr a
the('n.~ticd~ nr m¿lking infen~nt'{:~; tol("..lrn fn;~rn d/da tht'" S:lmt) dat1 set \vith unit ís the that
di 5cwnlifk rnt'erem:e th,1¡ we di~tlSS in ¡!lis hOOK, However, hi\; apprtMch can "ti!¡
al! units wíth the ;,ame v,11u<:' (jf the variables have tl1.::'
"<11uabli:: ,15 ,1 lo:m oí foro1al ¡heory bee sectíon ~5,2): it en¡¡bk'5 ¡he
a imd Hs ímpiic,'ltion; in a way tha! be muen more 5."lme expeeted v,1Jue of ¡he depenJent varí,lble, course., thb is unl)'
out iL an assumption and ít can be wrong: thc hvo districts might differ ín

You might also like