An Abductive Theory of Scientific Method - The Monkey Cage

18 Pages·2005·0.14 MB·English

PsychologicalMethods Copyright2005bytheAmericanPsychologicalAssociation
2005,Vol.10,No.4,371–388 1082-989X/05/$12.00 DOI:10.1037/1082-989X.10.4.371
An Abductive Theory of Scientific Method
Brian D. Haig
UniversityofCanterbury
Abroadtheoryofscientificmethodissketchedthathasparticularrelevanceforthebehavioral
sciences.Thistheoryofmethodassemblesacomplexofspecificstrategiesandmethodsthat
are used in the detection of empirical phenomena and the subsequent construction of
explanatorytheories.Acharacterizationofthenatureofphenomenaisgiven,andtheprocess
of their detection is briefly described in terms of a multistage model of data analysis. The
constructionofexplanatorytheoriesisshowntoinvolvetheirgenerationthroughabductive,
or explanatory, reasoning, their development through analogical modeling, and their fuller
appraisalintermsofjudgmentsofthebestofcompetingexplanations.Thenatureandlimits
of this theory of method are discussed in the light of relevant developments in scientific
methodology.
Keywords:scientificmethod,phenomenadetection,theoryconstruction,abductivereasoning,
researchmethodology
[T]heattempttounderstandandimprovemethods,andtodoso specific research goals (Nickles, 1987), not broad accounts
viatheorizingthem,isatthecenterofanintelligentlyevolving of method that pursue a range of research goals. In fash-
cognition(CliffordHooker,1987,p.291)
ioning empirical generalizations, the inductive method un-
This article is concerned with scientific method in the doubtedly addresses an important part of scientific inquiry.
behavioral sciences. Its principal goal is to outline a broad However, it is a part only. Of equal importance is the
theory of scientific method by making use of selected de- process of theory construction. Here, however, the hypo-
velopments in contemporary research methodology. The thetico-deductive method, with its focus on theory testing,
timenowseemsrighttointensifyeffortstoassembleknowl- speaks only to one, although important, part of the theory
edgeofresearchmethodsintolargerunitsofunderstanding. construction process (Simon, 1977).
Currently, behavioral scientists use a plethora of specific Thetheoryofmethodoutlinedinthisarticleisabroader
research methods and a number of different investigative account of scientific method than either the inductive or
strategies when studying their domains of interest. Among hypothetico-deductive theories of method. This more com-
this diversity, the well-known inductive and hypothetico- prehensivetheoryofmethodendeavorstodescribesystem-
deductiveaccountsofscientificmethodhavebroughtsome atically how one can first discover empirical facts and then
order to our investigative practices. The former method constructtheoriestoexplainthosefacts.Althoughscientific
speaks to the discovery of empirical generalizations, inquiry is often portrayed in hypothetico-deductive fashion
whereas the latter method is used to test hypotheses and asanundertakinginwhichtheoriesarefirstconstructedand
theories in terms of their predictive success. facts are then gathered in order to test those theories, this
However, although inductive and hypothetico-deductive should not be thought of as its natural order. In fact, scien-
methodsarecommonlyregardedasthetwomaintheoriesof tificresearchfrequentlyproceedstheotherwayaround.The
scientificmethod(Laudan,1981;and,infact,aresometimes theory of method described here adopts this alternative,
regarded as the principal claimants for the title of the facts-before-theorysequence,claimingthatitisasearchfor
definitive scientific method), they are better thought of as the understanding of empirical phenomena that gives ex-
restrictive accounts of method that can be used to meet planatory theory construction its point. With this theory of
method, phenomena exist to be explained rather than serve
as the objects of prediction in theory testing.
I thank Fran Vertue, Tony Ward, and Claire O’Loughlin for
Two Theories of Method
helpfulcommentsonearlierversionsofthisarticle.
Correspondence concerning this article should be addressed to
Brian D. Haig, Department of Psychology, University of Canter- Before presenting the proposed theory of scientific
bury, Private Bag 4800, Christchurch, New Zealand. E-mail: method, the well-known inductive and hypothetico-deduc-
[email protected] tive accounts of scientific method are briefly considered.
371
372 HAIG
This serves to define their proper limits as methods of countofmethodhasbeenstronglycriticizedbybothphiloso-
scienceand,atthesametime,provideusefulcontraststothe phers and psychologists (e.g., Cattell, 1966; Glymour, 1980;
more comprehensive theory of method. Rorer,1991;Rozeboom,1999).
Thecentralcriticismofthehypothetico-deductivemethod
Inductive Method is that it is confirmationally lax. This laxity arises from the
fact that any positive confirming instance of a hypothesis
Inpopularaccountsofinductivemethod(e.g.,Chalmers,
obtained by the hypothetico-deductive method can confirm
1999), the scientist is typically portrayed as reasoning in-
any hypothesis that is conjoined with the test hypothesis,
ductively by enumeration from secure observation state-
however plausible, or implausible, that conjunct might be.
ments about singular events to laws or theories in accor-
This criticism has prompted some methodologists (e.g.,
dance with some governing principle of inductive
Glymour,1980;Rozeboom,1999)todeclarethatthehypo-
reasoning. Sound inductive reasoning is held to create and
thetico-deductive method is hopeless and should therefore
justify theories simultaneously, so that there is no need for
be abandoned. Although this is a fair assessment of the
subsequentempiricaltesting.Somehavecriticizedthisview
confirmational worth of the orthodox account of the hypo-
of method for placing excessive trust in the powers of
thetico-deductive method, it should be noted that the
observation and inductive generalization, and for believing
method can be recast in a more sophisticated form and put
that enumerative induction is all there is to scientific infer-
tousefuleffectinhypothesistestingresearch(Giere,1983).
ence.Inmodernbehavioralscience,theradicalbehaviorism
Althoughthehypothetico-deductivemethoddoesnotfigure
of B. F. Skinner is a prominent example of a research
asamethodoftheoryappraisalinthecomprehensivetheory
tradition that uses an inductive conception of scientific
ofscientificmethodpresentedhere,itcanplayalegitimate
method (Sidman, 1960; Skinner, 1984). Within this behav-
roleinhypothesisandtheorytesting.Itshouldthusbeseen
ioristtradition,thepurposeofresearchistodetectempirical
as complementary to the broader theory of method, not a
phenomena of learning that are subsequently systematized
rival to it. I comment briefly on this matter toward the end
by nonexplanatory theories.
of the article.
Althoughtheinductivemethodhasreceivedconsiderable
The theory of method introduced in the next section is a
criticism, especially from those who seek to promote a
broader theory than both the inductive and hypothetico-
hypothetico-deductive conception of scientific inquiry, it
deductive theories. However, it should be acknowledged at
nevertheless stresses, in a broad-brush way, the scientific
theoutsetthatithasitsownomissions.Mostobviously,the
importanceoffashioningempiricalgeneralizations.Shortly,
method begins by focusing on data analysis and thereby
itisshownthatthealternativetheoryofscientificmethodto
ignores the matters of research design, measurement, and
be presented uses the inductive method in the form of
datacollection.Thisisalimittoitscomprehensivenessthat
enumerative induction, or induction by generalization, in
it shares with the two theories of method just canvassed.
order to detect empirical phenomena.
Hypothetico-Deductive Method Overview of the Broad Theory
For more than 150 years, hypothetico-deductivism has
According to the broad theory of method, scientific in-
been the method of choice in the natural sciences (Laudan,
quiry proceeds as follows. Guided by evolving research
1981), and it assumed hegemonic status in 20th century
problems that comprise packages of empirical, conceptual,
psychology (Cattell, 1966). Psychology’s textbook presen-
andmethodologicalconstraints,setsofdataareanalyzedin
tations of scientific method are often cast in hypothetico-
ordertodetectrobustempiricalregularities,orphenomena.
deductive form, and the heavy emphasis psychological re-
Once detected, these phenomena are explained by abduc-
searchers have placed on testing hypotheses by using
tively inferring the existence of underlying causal mecha-
traditional statistical significance test procedures basically nisms.1 Here, abductive inference involves reasoning from
conforms to a hypothetic-deductive structure.
The hypothetico-deductive method is standardly portrayed
inminimalterms:Theresearcherisrequiredtotakeahypoth- 1Thetermcausalmechanismisambiguous.Inthebroadtheory
esisoratheoryandtestitindirectlybyderivingfromitoneor of method being proposed, the generation of theories involves
explanatory inference to claims about the existence of causal
moreobservationalpredictions.Thesepredictionsareamena-
entities.Itisnotuntilthedevelopmentofthesetheoriesisunder-
bletodirectempiricaltest.Ifthepredictionsareborneoutby
takenthatthemechanismsresponsiblefortheproductionoftheir
thedata,thenthatresultistakenasaconfirminginstanceofthe
effects are identified and spelled out. Also, in this article it is
theory in question. If the predictions fail to square with the
assumed that the productivity of causal mechanisms is distinct
data, then that fact counts as a disconfirming instance of the
from the regularities that they explain (Bogen, 2005; but cf.
theory.Althoughtacitlyheldbymanyscientists,andendorsed Woodward,2003).Ofcourse,thisdoesnotprecludethemethod-
in different ways by prominent philosophers of science (e.g., ologicaluseofgeneralizationsthatdescribenaturalregularitiesin
Hempel, 1966; Popper, 1959), the hypothetico-deductive ac- ordertohelpidentifythecausalmechanismsthatproducethem.
THEORYOFSCIENTIFICMETHOD 373
Table1
SubmethodsandStrategiesoftheAbductiveTheoryofMethod
Theoryconstruction
Phenomenadetection Theorygeneration Theorydevelopment Theoryappraisal
Strategies Abductivemethods Strategies Inferencetothebest
Controlforconfounds Exploratoryfactoranalysis Analogicalmodeling explanation
Calibrationofinstruments Groundedtheorymethod Theoryofexplanatory
Dataanalyticstrategies Heuristics(e.g.,principle coherence
Constructivereplication ofthecommoncause)
Methods
Initialdataanalysis
Exploratorydataanalysis(e.g.,stem-and-leaf,
boxplot)
Computer-intensiveresamplingmethods(e.g.,
bootstrap,jackknife,cross-validation)
Meta-analysis
Note. Forthemostpart,particularmethodsandstrategiessubsumedbytheabductivetheoryareappropriateeitherforphenomenadetectionorfortheory
construction,butnotforboth.Exceptionsincludeexploratoryfactoranalysisandgroundedtheorymethod,bothofwhichhavedataanalyticcomponents
thatcancontributetophenomenadetection.
phenomena, understood as presumed effects, to their theo- hereobviouslyendeavorstothrowsomelightonthenature
retical explanation in terms of underlying causal mecha- ofscientificinquiry. It also has someclearimplicationsfor
nisms.Uponpositivejudgmentsoftheinitialplausibilityof thewayresearchiscarriedoutwithinitspurview.However,
these explanatory theories, attempts are made to elaborate partly because of its incomplete nature, the theory is not
onthenatureofthecausalmechanismsinquestion.Thisis accompanied by a set of instructions for its ready imple-
donebyconstructingplausiblemodelsofthosemechanisms mentation. Such an accompaniment awaits a fuller account
by analogy with relevant ideas in domains that are well ofthemethodandwouldhavetobemodifiedasafunction
understood.Whenthetheoriesarewelldeveloped,theyare ofthenatureofthesubmethodschosentooperatewithinit.
assessedagainsttheirrivalswithrespecttotheirexplanatory Becauseoftheprominenceofabductivereasoninginthis
goodness. This assessment involves making judgments of broad theory of method, I henceforth refer to it as the
the best of competing explanations. abductivetheoryofmethod(ATOM).Theexpositionofthe
An important feature of the broad theory of scientific methodbeginswithanaccountofphenomenadetectionand
method is its ability to serve as a framework within which then considers the process of constructing explanatory the-
avarietyofmorespecificresearchmethodscanbelocated, ories. Toward the end of the article, two pairs of important
conjoined,andused.Operatinginthisway,theseotherwise methodologicalideasthatfeatureprominentlyinATOMare
separate specific research methods can be viewed as sub- examined. The article concludes with a discussion of the
methods of the parent method. In turn, the submethods nature and limits of the method.
providetheparenttheorywiththeoperationalbitethathelps
itmakescientificinquirypossible.Comprehensivemethods Phenomena Detection
are often constituted by a number of submethods and strat-
egiesthatareorderedaccordingtoanoverarchingstructure Scientistsandphilosophersoftenspeakasthoughscience
(Ross, 1981). In characterizing the broad theory, I indicate is principally concerned with establishing direct relation-
how a number of specific research methods are deployed ships between observation and theory. There is empirical
within its compass. Table 1 contains a variety of research evidence that psychologists speak, and sometimes think, in
methods and strategies that can be placed within the struc- this way (Clark & Paivio, 1989), whereas philosophers of
ture of the comprehensive theory of scientific method. A science of different persuasions often say that scientific
number of these are discussed in the exposition of the theories are evaluated with respect to statements about
method that follows, but most of them are not required for
its characterization.2 The majority of submethods selected
2Note, however, that the strategy of analogical modeling is
for consideration in the article have been chosen primarily
essentialfortheorydevelopmentintheabductivetheoryofmethod
to facilitate the exposition of the processes of phenomena
and that the theory of explanatory coherence does heavy-duty
detectionandtheoryconstructionwithoutattemptingtogive work in the abductive theory of method because it is the best
an essential characterization of these processes. developed method of inference to the best explanation currently
As a theory of scientific method, the account presented available.
374 HAIG
relevant data (Bogen & Woodward, 1988). Despite what scientific explanation because they are stable and general.
they say, scientists frequently behave in accord with the Amongotherthings,systematicexplanationsrequireoneto
view that theories relate directly to claims about phenom- show that the events to be explained result from the causal
ena, such as empirical generalizations, not data, while in factors appealed to in the explanation. They also serve to
turn,claimsaboutphenomenarelatedirectlytoclaimsabout unifytheeventstobeexplained.Becauseoftheirephemeral
data. That is, talk of a direct relationship between data and nature, data will not admit of systematic explanations.
theoryisatvariancewithempiricalresearchpractice,which In order to understand the process of phenomena detec-
often works with a threefold distinction between data, phe- tion, phenomena must be distinguished from data. Unlike
nomena, and theory. phenomena,dataareidiosyncratictoparticularinvestigative
As just noted, ATOM assigns major importance to the contexts.Becausedataresultfromtheinteractionofalarge
task of detecting empirical phenomena, and it views the numberofcausalfactors,theyarenotasstableandgeneral
completion of this task as a requirement for subsequent as phenomena, which are produced by a relatively small
theoryconstruction.Thissectionofthearticlediscussesthe number of causal factors. Data are ephemeral and pliable,
process of phenomena detection in psychological research. whereas phenomena are robust and stubborn. Phenomena
First,thedistinctionbetweendataandphenomenaisdrawn. have a stability and repeatability that is demonstrated
Then, a multistage model of data analysis is outlined. This through the use of different procedures that often engage
model serves to indicate one way in which a variety of different kinds of data. Data are recordings or reports that
statistical methods available to psychologists can be com- areperceptuallyaccessible;theyareobservableandopento
bined in phenomena detection. publicinspection.Despitethepopularviewtothecontrary,
phenomenaarenot,ingeneral,observable;theyareabstrac-
tions wrought from the relevant data, frequently as a result
The Nature of Phenomena
of a reductive process of data analysis. As Cartwright
Bogen and Woodward (1988; Woodward, 1989, 2000) (1983) remarked in her discussion of phenomenal and the-
havearguedindetailthatitisclaimsaboutphenomena,not oreticallawsinphysics,“thedistinctionbetweentheoretical
data, that theories typically seek to predict and explain and and phenomenological has nothing to do with what is ob-
that, in turn, it is the proper role of data to provide the servable and what is unobservable. Instead the terms sepa-
observational evidence for phenomena, not for theories. ratelawswhicharefundamentalandexplanatoryfromthose
Phenomenaarerelativelystable,recurrent,generalfeatures thatmerelydescribe”(p.2).Examplesofdata,whichserve
of the world that, as researchers, we seek to explain. The as evidence for the aforementioned psychological effects,
more striking of them are often called effects, and they are are rates of operant responding (evidence for the matching
sometimes named after their principal discoverer. The so- law), consistent intergenerational IQ score gains (evidence
called phenomenal laws of physics are paradigmatic cases for the Flynn effect), and error rates in psychological ex-
ofclaimsaboutphenomena.Bycontrast,theso-calledfun- periments (evidence for recency effects in short-term
damental laws of physics explain the phenomenal laws memory).
about the relevant phenomena. For example, the electron The methodological importance of data lies in the fact
theoryofLorentzisafundamentallawthatexplainsAiry’s thattheyserveasevidenceforthephenomenaunderinves-
phenomenological law of Faraday’s electro-optical effect tigation.Indetectingphenomena,oneextractsasignal(the
(Cartwright, 1983). Examples of the innumerable phenom- phenomenon)fromaseaofnoise(thedata).Somephenom-
enaclaimsinpsychologyincludethematchinglaw(thelaw enaarerare,andmanyaredifficulttodetect;asWoodward
ofeffect),theFlynneffectofintergenerationalgainsinIQ, (1989) noted, detecting phenomena can be like looking for
and recency effects in human memory. a needle in a haystack. It is for this reason that, when
Although phenomena commonly take the form of empir- extracting phenomena from the data, one often engages in
icalregularities,theycompriseavariedontologicalbagthat data exploration and reduction by using graphical and sta-
includesobjects,states,processes,events,andotherfeatures tistical methods.
that are hard to classify. Because of this variety, it is
generally more appropriate to characterize phenomena in
A Model of Data Analysis
terms of their role in relation to explanation and prediction
(Bogen & Woodward, 1988). For example, the relevant Inordertoestablishthatdataarereliableevidenceforthe
empiricalgeneralizationsincognitivepsychologymightbe existenceofphenomena,scientistsuseavarietyofmethod-
the objects of explanations in evolutionary psychology that ological strategies. These strategies include controlling for
appealtomechanismsofadaptation,andthosemechanisms confounding factors (both experimentally and statistically),
might in turn serve as phenomena to be explained by ap- empirically investigating equipment (including the calibra-
pealingtothemechanismsofnaturalselectioninevolution- tion of instruments), engaging in data analytic strategies of
ary biology. both statistical and nonstatistical kinds, and constructively
Phenomena are frequently taken as the proper objects of replicating study results. As can be seen in Table 1, these
THEORYOFSCIENTIFICMETHOD 375
proceduresareusedinthedetectionofphenomena,butthey cernedwiththeeffectiveorganizationofdata,theconstruc-
are not used in the construction of explanatory theory (cf. tion of graphical displays, and the examination of
Franklin, 1990; Woodward, 1989). The later discussion of distributional assumptions and functional dependencies.
the importance of reliability in the process of phenomena Thestem-and-leafdisplayandthebox-and-whiskerplotare
detection helps indicate why this is so. two well-known exploratory methods.
Giventheimportanceofthedetailedexaminationofdata Two attractive features of exploratory methods are their
intheprocessofphenomenadetection,itisnaturalthatthe robustness to changes in underlying distributions and their
statistical analysis of data figures prominently in that exer- resistancetooutliersindatasets.Exploratorymethodswith
cise. A statistically oriented, multistage account of data thesetwofeaturesareparticularlysuitedtodataanalysisin
analysisisthereforeoutlinedinordertofurthercharacterize the behavioral sciences, where researchers are frequently
the phenomena detection phase of ATOM. The model pro- confronted with ad hoc data sets on manifest variables that
ceeds through the four stages of initial data analysis, ex- have been acquired in convenient ways.
ploratory data analysis, close replication, and constructive Close replication. Successfully conducted exploratory
replication. However, it should be noted that, although the analyses will suggest potentially interesting data patterns.
behavioralsciencesmakeheavyuseofstatisticalmethodsin However, it will normally be necessary to check on the
data analysis, qualitative data analytic methods can also be stability of the emergent data patterns through use of con-
used in the detection of phenomena (cf. Strauss, 1987).
firmatory data analysis procedures. Computer-intensive re-
Initial data analysis. The initial examination of data
sampling methods such as the bootstrap, the jackknife, and
(Chatfield, 1985) refers to the first informal scrutiny and
cross-validation (Efron & Tibshirani, 1993) constitute an
description of data that is undertaken before exploratory
importantsetofconfirmatoryproceduresthatarewellsuited
data analysis proper begins. It involves screening the data
tothedemandsofmoderndataanalysis.Suchmethodsfree
foritsquality.Initialdataanalysisvariouslyinvolvescheck-
us,asresearchers,fromtheassumptionsoforthodoxstatis-
ingfortheaccuracyofdataentries,identifyinganddealing
ticaltheory,andpermitustogaugethereliabilityofchosen
withmissingandoutlyingdata,andexaminingthedatafor
statistics by making thousands, even millions, of calcula-
their fit to the assumptions of the data analytic methods to
tions on many data points. Statistical resampling methods
be used. Data screening thus enables one to assess the
liketheseareusedtoestablishtheconsistency,orreliability,
suitability of the data for the type of analysis intended.
of sample results. In doing this, they provide us with the
This important, and time-consuming, preparatory phase
kind of validating strategy that is needed to achieve close
ofdataanalysishasfailedtoreceivetheamountofexplicit
replications.4
attention that it deserves in behavioral science education.
Now that psychology has finally begun to embrace ex-
Recently, however, the American Psychological Associa-
ploratory data analysis, one can hope for a corresponding
tion’sTaskForceonStatisticalInference(Wilkinson&the
increase in the companionate use of statistical resampling
Task Force on Statistical Inference, 1999) recommended
methodsinordertoascertainthevalidityofthedatapatterns
changestocurrentpracticesindataanalysisthatarebroadly
initially suggested by the use of exploratory methods.
inkeepingwiththegoalsofinitialdataanalysis.Fidelland
Constructive replication. In establishing the existence
Tabachnick (2003) provided a useful overview of the im-
of phenomena, it is necessary that science undertake both
portance of the work required to identify and correct prob-
close and constructive replications. The statistical resam-
lems in data.
pling methods just mentioned are concerned with the con-
Itshouldbeclear,evenfromthesebriefremarks,thatthe
initial examination of data is a requirement of successful
data analysis in science, for data that lack integrity can 3Behrens and Yu suggested that the inferential foundations of
easilyresultinthemisuseofdataanalyticmethodsandthe exploratory data analysis are to be found in the notion of abduc-
drawing of erroneous conclusions. tion. By contrast, ATOM regards exploratory data analysis as a
Exploratory data analysis. Exploratory data analysis descriptive pattern detection process that is a precursor to the
usesmultipleformsofdescriptionanddisplayandinvolves inductive generalizations involved in phenomena detection. Ab-
descriptive, and frequently quantitative, detective work de- ductiveinferenceisreservedfortheconstructionofcausalexplan-
atorytheoriesthatareintroducedtoexplainempiricalphenomena.
signed to reveal the structure or patterns in the data under
BehrensandYu’ssuggestionconflatesdescriptionandexplanation
scrutiny (Behrens & Yu, 2003; Tukey, 1977).3 The ex-
inthisregard.
ploratory data analyst is encouraged to undertake an unfet-
4Statistical resampling methods can be used in a hypothetico-
tered investigation of the data and perform multiple analy-
deductive manner within ATOM in order to test descriptive hy-
ses using a variety of intuitively appealing and easily used
potheses that are suggested by exploratory data analytic work.
techniques.
However,thisuseofthehypothetico-deductivemethodshouldbe
The compendium of methods for the exploration of data distinguishedfromitsusetoevaluateexplanatoryhypothesesand
isdesignedtofacilitateboththediscoveryandthecommu- theories. The latter takes place outside the methodological space
nicationofinformationaboutdata.Thesemethodsarecon- providedbyATOM.
376 HAIG
sistency of sample results that help researchers achieve Theory Construction
close, or internal, replications. By contrast, constructive
replications are undertaken to demonstrate the extent to Detecting empirical phenomena is a major goal of scien-
whichresultsholdacrossdifferentmethods,treatments,and tific research, and their successful detection constitutes an
occasions. In other words, constructive replication is a tri- important type of scientific discovery in its own right.
angulation strategy designed to ascertain the generalizabil- However, once detected, phenomena serve the important
ity of the results identified by successful close replication function of prompting the search for their own understand-
(Lindsay & Ehrenberg, 1993). Constructive replication, in ing. This understanding is commonly met in science by
which researchers vary the salient conditions, is a time- constructing relevant explanatory theories.
honored strategy for justifying claims about phenomena. For inductivists, inductively grounded conclusions about
In recognition of the need to use statistical methods that phenomena are of paramount importance. However, al-
are in keeping with the practice of describing predictable though inductivists often subsequently construct theories,
phenomena, researchers should seek the generalizability of their theories do not provide explanations of phenomena
relationshipsratherthantheirstatisticalsignificance(Ehren- that appeal to causal mechanisms. Instead, their theories
functionastoolsorinstrumentsconcernedwiththedescrip-
berg&Bound,1993)—hence,theneedtouseobservational
tion,economicalordering,andpredictionofempiricalrela-
and experimental studies with multiple sets of data, ob-
tionships. For hypothetico-deductivists, theories are said to
served under quite different sets of conditions. The recom-
be generated amethodologically through free use of the
mended task here is not to figure what model best fits a
imagination (Hempel, 1966; Popper, 1959). Theories ob-
single set of data but to ascertain whether the model holds
tained in this manner are often regarded as explanatory in
across different data sets. Seeking reproducible results
nature,buttheirworthisprincipallyjudgedintermsoftheir
through constructive replications, then, requires data ana-
predictive success, rather than their ability to explain em-
lytic strategies that are designed to detect significant same-
pirical phenomena.
ness rather than significant difference.
ATOM,bycontrast,maintainsthattheoryconstructionis
Thefour-stagemodelofdataanalysisjustoutlinedassists
neither inductive nor amethodological. For it, theory con-
in the detection of phenomena by attending in turn to data
struction comprises three methodological phases: theory
quality, pattern suggestion, pattern confirmation, and gen-
generation, theory development, and theory appraisal.
eralization. In effect, this process is one of enumerative
These phases do not occur in a strictly temporal order, for
inductioninwhichonelearnsempirically,onacase-by-case
although theory generation precedes theory development,
basis, the conditions of applicability of the empirical gen-
theory appraisal begins with theory generation, continues
eralizations that represent the phenomena. Thus, as noted
with theory development, and extends to the comparative
earlier,theimportanceofinductivereasoningshownbythe
appraisal of well-developed theories. Further, ATOM’s
traditional inductive method is shared by ATOM’s account
characterizationoftheoryconstructionisabductivethrough
of phenomena detection.
and through: Theory generation, theory development, and
It bears repeating that this model of data analysis is
theory appraisal are all portrayed as abductive, or explana-
clearlynottheonlywayinwhichphenomenadetectioncan
tory, undertakings, although the form of abduction is dif-
beachieved.Inadditiontotheseveralstrategiesofphenom-
ferentineachcase.Theaccountoftheoryconstructionthat
ena detection mentioned earlier, meta-analysis is a promi-
follows articulates the abductive character of each of the
nent example of a distinctive use of statistical methods by
three phases.
behavioral scientists to aid in the detection of phenomena.
As is well-known, meta-analysis is widely used to conduct
Theory Generation
quantitative literature reviews. It is an approach to data
analysis that involves the quantitative analysis of the data Abductiveinference. Thissectionbeginswithageneral
analysesofprimaryempiricalstudies.Bycalculatingeffect characterization of the type of abductive reasoning that is
sizes across primary studies in a common domain, meta- often involved in theory generation. It is followed by a
analysishelpsresearchersdetectgeneralpositiveeffects(cf. discussionofthemethodofexploratoryfactoranalysisthat
Schmidt,1992).Byusingstatisticalmethodstoascertainthe is presented as a prominent example in psychology of an
existenceofrobustempiricalregularities,meta-analysiscan abductive method of theory generation. The discussion of
be usefully viewed as the statistical analogue of direct exploratory factor analysis, therefore, serves as an optional
experimentalreplication.Itisinthisrolethatmeta-analysis andrestrictedaccountoftheorygenerationforATOM.The
currentlyperformsitsmostimportantworkinscience.Con- characterizations of abduction and factor analysis are
trarytotheclaimsmadebysomeofitscriticsinpsychology adapted from Haig (2005).
(e.g., Sohn, 1996), meta-analysis can be regarded as a The basic idea of abductive inference can be usefully
legitimateandimportantmeansofdetectingempiricalphe- traced back to the American philosopher and scientist
nomena in the behavioral sciences (Gage, 1996). CharlesSandersPeirce(1931–1958).Morerecentdevelop-
THEORYOFSCIENTIFICMETHOD 377
ments in the fields of philosophy of science and artificial derived not just from the proposed theory but from that
intelligence (e.g., Josephson & Josephson, 1994; Magnani, theory in conjunction with accepted auxiliary claims taken
2001; Thagard, 1988, 1992) have built on Peirce’s ideas to from relevant background knowledge.
significantly advance researchers’ understanding of abduc- Third, the antecedent of the conditional assertion in the
tive reasoning. secondpremiseoftheargumentschemashouldnotbetaken
Abduction is a form of reasoning involved in both the toimplythatabductiveinferencesproducetruthsasamatter
generation and evaluation of explanatory hypotheses and ofcourse.Althoughscienceaimstoprovidetrue,orapprox-
theories. For Peirce (1931–1958), “abduction consists in imately true, theories of the world, the supposition that the
studying the facts and devising a theory to explain them” proposed theory be true is not a requirement for the deri-
(Vol.5,1934,p.90).Itis“[t]hefirststartingofanhypoth- vation of the relevant facts. All that is required is that the
esis and the entertaining of it, whether as a simple interro- theory be plausible enough to be provisionally accepted. It
gation or with any degree of confidence” (Vol. 6, 1934, p. is important to distinguish between truth, understood as a
358). guidingidealforscience(agoalthatwe,asscientists,strive
Traditionally, abduction was thought to take its place at for but never fully reach), and the justification of theories,
the inception of scientific hypotheses, where it often in- which is based on epistemic criteria such as predictive
volves making an inference from puzzling facts to hypoth- success,simplicity,andexplanatorybreadth.Asproxiesfor
eses that might well explain them. However, there are a truth, justificatory criteria such as these are indicative of
number of different ways in which explanatory hypotheses truth, but they are not constitutive of truth.
can be abductively obtained. In focusing on the generation Fourth, it should be noted that the conclusion of the
of hypotheses, Thagard (1988) helpfully distinguished be- argumentschemadoesnotassertthatthehypothesisitselfis
tween existential and analogical abduction. As he put it, true, only that there are grounds for thinking that the pro-
“Existentialabductionpostulatestheexistenceofpreviously posedhypothesismightbetrue.Thisisaweakerclaimthat
unknown objects, such as new planets, ... [whereas] ana- allows one to think of a sound abductive argument as
logicalabductionusespastcasesofhypothesisformationto delivering a judgment that the hypothesis is initially plau-
generatehypothesessimilartoexistingones”(p.54).Exis- sible and worthy of further pursuit. Assessments of initial
tentialabductionisthetypeofabductioncentrallyinvolved plausibility constitute a form of justification that involves
inthefactoranalyticgenerationofexplanatoryhypotheses. reasoning from warranted premises to an acceptance of the
Later, it is shown that the theory development phase of knowledge claims in question. This form of justification is
ATOMadoptsamodelingstrategythatinvolvesanalogical discussed later in the section on ATOM and Scientific
abduction,anditsapproachtocomparativetheoryappraisal Methodology.
uses a further form of abduction known as inference to the Fifth, the schema depicting abductive inference focuses
best explanation. on its logical form only. It is, therefore, of limited value in
Existential abduction can be characterized in the follow- understanding the theory construction process unless it is
ing general schema: combinedwithasetofregulativeconstraintsthatenableus
toviewexistentialabductionasaninference,notjusttoany
Thesurprisingempiricalphenomenon,P,isdetected. conceivableexplanation,buttoaplausibleexplanation.The
But if hypothesis H were approximately true, and the relevant description of research problems presented later indicates
auxiliaryknowledge,A,wasinvoked,thenPwouldfollowas
thattheconstraintsthatregulatetheabductivegenerationof
amatterofcourse.
scientific theories comprise a host of heuristics, rules, and
Hence,therearegroundsforjudgingHtobeinitiallyplausible
andworthyoffurtherpursuit. principles that govern what counts as good explanations.
Exploratory factor analysis. Unfortunately, there is a
This schematic characterization of existential abduction, dearthofcodifiedabductivemethodsavailableforreadyuse
asitoccurswithinthetheorygenerationphaseofATOM,is in the behavioral sciences. A notable exception is the
coarse grained and far from sufficient. It should, therefore, method of exploratory factor analysis. Exploratory factor
be understood in the light of the following supplementary analysis is designed to facilitate the postulation of latent
remarks. variablesthatarethoughttounderliepatternsofcorrelations
First, as indicated in the discussion of phenomena detec- innewdomainsofmanifestvariables.Itdoesthisbyusing
tion, the facts to be explained in science are not normally multiple regression and partial correlation theory to model
particular events, but empirical generalizations or phenom- sets of manifest or observed variables in terms of linear
ena, and, strictly speaking, they are not typically observed. functions of other sets of latent, or unobserved, variables.
Second,confirmationtheoryinthephilosophyofscience, Althoughthenatureandpurposeofexploratoryfactoranal-
and the nature of the hypothetico-deductive method in par- ysis is a matter of some debate, it can plausibly be under-
ticular, make it clear that the facts, or phenomena, are stood as an abductive method of theory generation (Haig,
378 HAIG
2005; Rozeboom, 1972; Stephenson, 1961).5 This charac- (2002)maintainedthatourability,ashumans,toengagein
terization of the inferential nature of exploratory factor explanatoryinferenceisalmostcertainlylargelyinnate,and
analysis is seldom given in expositions of the method; he speculated that it may be an adaptation selected for
however, it is an interpretation that coheres well with its becauseofitscrucialroleinthefitness-enhancingactivities
general acceptance as a latent variable method. ofourancestorssuchashuntingandtracking.Whateverits
On this interpretation, exploratory factor analysis facili- origin, an informative methodological characterization of
tates the achievement of useful existential abductions, al- theabductivenatureoffactoranalyticinferencemustappeal
though for this to happen, the method must be used in an
to the scientist’s own psychological resources as well as
exemplary manner (cf. Fabrigar, Wegener, MacCallum, &
those of logic.
Strahan,1999;Preacher&MacCallum,2003)withcircum-
Exploratoryfactoranalysis,then,canusefullyfunctionas
spectinterpretationofthefactors.Asnotedearlier,existen-
a submethod of ATOM by being located in that theory’s
tialabductionsenableus,asresearchers,tohypothesizethe
context of theory generation. Although it exemplifies well
existence of entities previously unknown to us. The innu-
the character of existential abduction, exploratory factor
merable examples of existential abduction in science in-
analysisisclearlynotanall-purposemethodforabductively
cludetheinitialpostulationofhiddenentitiessuchasatoms,
generating explanatory hypotheses and theories. With its
genes, tectonic plates, and personality traits. In cases like
focus on common factors, it can properly serve as a gener-
these, the primary thrust of the initial abductive inferences
is to claims about the existence of theoretical entities6 in ator of elementary theories only in those multivariate do-
mains where there are common causal structures.
order to explain empirical facts or phenomena. Similarly,
Understood in the context of theory generation, methods
the hypotheses given to us through the use of exploratory
factor analysis postulate the existence of latent variables of existential abduction like exploratory factor analysis
such as Spearman’s g and extraversion. It remains for fur- should not be expected to achieve highly developed and
ther research to elaborate on the first rudimentary concep- well-validated scientific theories. At best, they deliver ru-
tion of these variables. dimentarytheoriesthathaveinitialplausibility.Itisimpor-
The factor analytic use of existential abduction to infer tant to realize that these abductive methods enable us to
theexistenceof,say,thetheoreticalentitygcanbecoarsely justifytheinitialplausibilityofthetheoriestheyspawn.The
reconstructed in accordance with the aforementioned very process of the abductive generation of theories has a
schema for abductive inference along the following lines: bearingonthefirstdeterminationsoftheirworth,inthatwe
appealtothesoundnessoftheabductiveargumentsusedin
The surprising empirical phenomenon known as the positive
manifold7isidentified. the introduction of theories in order to evaluate their early
Ifgexists,anditisvalidlyandreliablymeasuredbyaWechsler epistemic promise (cf. Whitt, 1992).
intelligencescale(and/orsomeotherobjectivetest),thenthe Relatedly,thenascenttheoriesbequeathedusbymethods
positivemanifoldwouldfollowasamatterofcourse.
like exploratory factor analysis postulate the existence of
Hence,therearegroundsforjudgingthehypothesisofgtobe
hidden causal mechanisms, but they do not provide an
initiallyplausibleandworthyoffurtherpursuit.
informative characterization of their nature. Such theories
Thisexampleservestoillustratethepointthatthemethod
havethestatusofdispositionaltheoriesinthattheyprovide
of exploratory factor analysis proper should be taken to
us with oblique characterizations of the properties we at-
includethefactoranalyst’ssubstantiveinterpretationofthe
tribute to things by way of their presumed effects under
statistical factors. It is important to realize that the factor
analyst has to resort to his or her own abductive powers
when reasoning from correlational data patterns to under- 5Some take exploratory factor analysis to be a data analytic
lying common causes. Note that the schema for abductive method,only.Myprincipalreasonforassigningatheorygenera-
inference, and its application to the generation of Spear- tion role to exploratory factor analysis is based on the belief that
man’s hypothesis of g, are concerned with the form of the factorsarebestregardedaslatentcommoncausesandthatinfer-
encetosuchcausesisabductiveinnature(Haig,2005).
arguments involved, not with the actual generation of the
explanatory hypotheses. In each case, the explanatory hy- 6The term entity is used as a catch-all ontological term that
covers a miscellany of properties that includes states, processes,
pothesisisgiveninthesecondpremiseoftheargument.An
and events. Although existential abductions in exploratory factor
account of the genesis of the explanatory hypothesis must,
analysisaretopropertiesexpressedasthevaluesofvariables,not
therefore,befurnishedbysomeothermeans.Itisplausible
allexistentialabductionsneedtakethisform.
to suggest that reasoning to explanatory hypotheses trades
7Thepositivemanifoldisatermthatissometimesusedtorefer
on human beings’ evolved cognitive ability to abductively
to the striking, and well-established, fact that almost all different
generate such hypotheses. Peirce (1931–1958) himself
testsofabilitycorrelatepositivelywithoneanothertoasignificant
maintained that the human ability to engage readily in degree.DespiteitshistoricallinktoSpearman’stheoryofgeneral
abductivereasoningwasfoundedonaguessinginstinctthat intelligence,thepositivemanifoldcanbetakenasevidenceforthe
has its origins in evolution. More suggestively, Carruthers existenceoftwoormorefactors.
THEORYOFSCIENTIFICMETHOD 379
specified conditions (cf. Mumford, 1998). A move beyond development within ATOM is of central importance. The
the rudimentary nature of their dispositional characteriza- needforanalogicalmodelingwithinATOMstemsfromtwo
tion requires subsequent elaboration. It is to a strategy for featuresofitscharacterizationoftheorygeneration.First,as
developing such theories that I now turn. withexploratoryfactoranalysis,theabductivegenerationof
theories takes the form of existential abduction, through
which the existence of theoretical entities is postulated.
Theory Development
Therefore, an appropriate research strategy is required to
Models in science. The standard inductive and hypo- learn about the nature of these hidden entities. For this, the
thetico-deductive views of scientific method give little at- strategy of analogical modeling is used to do the required
tention to the process of theory development. The use of elaborative work. Second, recall that the postulation of
traditionalinductivemethodleadstotheoriesthatareorga- theoretical entities through existential abduction confers an
nized summaries of their constituent empirical generaliza- assessment of initial plausibility on those postulations.
tions, and the orthodox hypothetico-deductive method as- However, for claims about those latent entities to have the
sumes that hypotheses and theories emerge fully formed, statusofgenuineknowledge,furtherevaluativeworkhasto
ready for immediate testing. bedone.Theconstructionofappropriateanalogicalmodels
In contrast to these two theories of scientific method, servestoassesstheplausibilityofourexpandedunderstand-
ATOMisconcernedwiththedevelopmentofexplanatory ing,aswellastoexpandourunderstandingofthoseentities.
theories. As just noted, the theories it generates through ForATOM,theorydevelopmentexpandsourknowledge
existential abduction are dispositional in nature, and ex- of the nature of our theories’ causal mechanisms. This is
plicit provision has to be made for their development achieved by using the pragmatic strategy of conceiving of
before they are systematically evaluated against rival these unknown mechanisms in terms of what is already
theories with respect to their explanatory goodness. As familiar and well understood. Well known examples of
noted earlier, ATOM recommends that this be done by models that have resulted from this strategy are the molec-
building analogical models of the causal mechanisms in ularmodelofgases,basedonananalogywithbilliardballs
question. in a container; the model of natural selection, based on an
Thereisalong-heldview(e.g.,Duhem,1914/1954),still analogy with artificial selection; and the computational
popular in some quarters, that analogical models are dis- modelofthemind,basedonananalogywiththecomputer.
pensable aids to formulating and understanding scientific To understand the nature of analogical modeling, it is
theories. This negative view of the cognitive value of ana- helpful to distinguish between a model, the source of the
logical models in science contrasts with the positive view model, and the subject of the model (Harre´, 1976; Hesse,
thattheyareanessentialpartofthedevelopmentoftheories 1966). From the known nature and behavior of the source,
(cf. Campbell, 1920; Harre´, 1976; Hesse, 1966). Contem- one builds an analogical model of the unknown subject or
porarystudiesofscientificpractice,includingphilosophyof causal mechanism. If we take the biological example just
science, frequently accord analogical models a genuine, mentioned, Darwin fashioned his model of the subject of
indispensable, cognitive role in science (e.g., Abrantes, naturalselectionbyreasoningbyanalogyfromthesourceof
1999; Giere, 1988; Harre´, 1988). the known nature and behavior of the process of artificial
Science uses different types of models for different pur- selection. In this way, analogical models play an important
poses. For example, iconic models8 are constructed to pro- creative role in theory development. However, this role
vide a good resemblance to the object or property being requires the source, from which the model is drawn, to be
modeled, mathematical models offer an abstract symbolic differentfromthesubjectthatismodeled.Forexample,the
representationofthedomainofinterest,andanaloguemod- modern computer is a well-known source for the modeling
elsexpressrelevantrelationsofanalogybetweenthemodel of human cognition, though our cognitive apparatus is not
and the reality being represented. Harre´ (1970) contains a generally thought to be a real computer. Models in which
useful taxonomy of this variety. Although it is acknowl- thesourceandthesubjectaredifferentaresometimescalled
edged that there is a need to use a variety of different paramorphs.Modelsinwhichthesourceandthesubjectare
modeling strategies in science, ATOM adopts the strategy the same are sometimes called homoeomorphs (Harre´,
of using analogical models to help develop explanatory
theories. Because analogical modeling is a strategy that
8More precisely, iconic models are constructed as representa-
increases the content of explanatory theories, its reasoning
tions of reality, real or imagined. In ATOM they stand in for the
takes the form of analogical abduction.
hypothesizedcausalmechanisms.Althoughrepresentations,iconic
Analogical modeling. The idea that analogical models
models are themselves things, structures, or processes that corre-
are important in the development of scientific theories can
spondinsomewaywiththings,structures,orprocessesthatarethe
betracedbacktoCampbell(1920).Althoughanalogiesare objects of modeling. They are, therefore, the sorts of things sen-
notalwaysusedinscientificexplanation,theirroleintheory tencescanbeabout(Harre´,1976).
380 HAIG
1976). The paramorph can be an iconic, or pictorial, repre- ogy. Here, Goffman’s (1969) dramaturgical perspective
sentationofrealorimaginedthings.Itisiconicparamorphs providesthesourcemodelforunderstandingtheunderlying
that feature centrally in the creative process of theory de- causal mechanisms involved in the production of ceremo-
velopment through analogical modeling. nial, argumentative, and other forms of social interaction.
In evaluating the aptness of an analogical model, the Thus far, it has been suggested that, for ATOM, the
analogy between its source and subject must be assessed, epistemic worth of hypotheses and theories generated by
andforthisoneneedstoconsiderthestructureofanalogies. existential abduction are evaluated in terms of their initial
The structure of analogies in models comprises a positive plausibility and that these assessments are subsequently
analogyinwhichthesourceandsubjectarealike,anegative augmentedbyjudgmentsoftheappropriatenessoftheanal-
analogy in which the source and subject are unlike, and a ogiesthatfunctionassourcemodelsfortheirdevelopment.
neutralanalogywherewehavenoreliableknowledgeabout However, with ATOM, well-developed theories are ap-
matched attributes in the source and subject of the model. praised further with respect to a number of additional cri-
The negative analogy is irrelevant for purposes of analogi- teria that are used when judgments about the best of com-
cal modeling. Because we are essentially ignorant of the peting explanatory theories are made.
nature of the hypothetical mechanism of the subject apart
from our knowledge of the source of the model, we are
Theory Appraisal
unable to specify any negative analogy between the model
andthemechanismbeingmodeled.Thus,inconsideringthe Contemporaryscientificmethodologyboastsanumberof
plausibility of an analogical model, one considers the bal- general approaches to the evaluation of scientific theories.
ance of the positive and neutral analogies (Harre´, 1976). Prominent among these are the hypothetico-deductive
This is where the relevance of the source for the model is method, which evaluates theories in terms of predictive
spelled out. As is shown in the next section, ATOM sub- success; Bayesian accounts of confirmation, which assign
scribestoaviewofcomparativetheoryappraisalthattakes probabilities to hypotheses via Bayes’s theorem; and infer-
good analogies as a criterion of explanatory worth. encetothebestexplanation,whichacceptsatheorywhenit
Analogical reasoning is important in science and clearly is judged to provide a better explanation of the evidence
lies at the inferential heart of analogical modeling. How- than its rivals do. Of these three approaches, the hypo-
ever, as noted above, because the theories fashioned by thetico-deductive method is by far the most widely used in
ATOM are explanatory theories, the analogical models in- psychology (cf. Cattell, 1966; Rorer, 1991; Rozeboom,
volved in theory development will involve explanatory an- 1999).Despiteoccasionalurgings(e.g.,Edwards,Lindman,
alogical reasoning, that is, analogical abduction. The rea- &Savage,1963;Lee&Wagenmakers,2005;Rorer,1991),
soning involved in analogical abduction can be simply psychologistshavebeenreluctanttouseBayesianstatistical
statedintheformofageneralargumentschemaasfollows: methodstotesttheirresearchhypotheses,preferringinstead
to perpetuate the orthodoxy of classical statistical signifi-
HypothesisH*aboutpropertyQwascorrectinsituationS1.
cance testing within a hypothetico-deductive framework.
SituationS1islikethesituationS2inrelevantrespects.
Despite the fact that inference to the best explanation is
Therefore,ananalogueofH*mightbeappropriateinsituation
S2. frequentlyusedinscience,andextensivelydiscussedinthe
philosophy of science, it is virtually unheard of, let alone
Darwin’s theory or model of natural selection, and the used, to appraise theories in psychology.
other aforementioned analogical models, can plausibly be Truetoitsname,ATOMadoptsanabductiveperspective
construedtobebasedonanalogicalabduction.Thegeneral on theory evaluation by using a method of inference to the
argumentforanalogicalabductionjustgivencanberewrit- best explanation. It is shown shortly that, in contrast to the
teninsimplifiedformforDarwin’scaseasschemafollows: hypothetico-deductive method, ATOM adopts an approach
toinferencetothebestexplanationthatmeasuresempirical
Thehypothesisofevolutionbyartificialselectionwascorrectin
adequacy in terms of explanatory breadth, not predictive
casesofselectivedomesticbreeding.
Casesofselectivedomesticbreedingarelikecasesofthenatural success, and, in contrast with Bayesianism, it takes theory
evolutionofspecieswithrespecttotheselectionprocess. evaluation to be an exercise that focuses directly on expla-
Therefore,byanalogywiththehypothesisofartificialselection, nation, not a statistical undertaking in which one assigns
the hypothesis of natural selection might be appropriate in
probabilities to theories. The basic justification for using
situationswherevariantsarenotdeliberatelyselectedfor.
inference to the best explanation when evaluating explana-
Themethodologyofmodelingthroughanalogicalabduc- tory theories is that it is the only method researchers have
tionisquitewelldevelopedandprovidesageneral,though that explicitly assesses such theories in terms of the scien-
useful,sourceofguidanceforbehavioralscientists.Instruc- tific goal of explanatory worth.
tively for psychology, Harre´ (Harre´ & Secord, 1972) fol- In considering theory evaluation in ATOM, the idea of
lowedhisownaccountofanalogicalmodelingtoconstruct inference to the best explanation is introduced. Then, a
a rule model of microsocial interaction in social psychol- well-developedmethodofinferencetothebestexplanation