Early

Journal Content on JSTOR, Free to Anyone in the World

This article is one of nearly 500,000 scholarly works digitized and made freely available to everyone in
the world by JSTOR.
Known as the Early Journal Content, this set of works include research articles, news, letters, and other
writings published in more than 200 of the oldest leading academic journals. The works date from the
mid-seventeenth to the early twentieth centuries.
We encourage people to read and share the Early Journal Content openly and to tell others that this
resource exists. People may post this content online or redistribute in any way for non-commercial
purposes.
Read more about Early Journal Content at http://about.jstor.org/participate-jstor/individuals/early-
journal-content.








JSTOR is a digital library of academic journals, books, and primary source objects. JSTOR helps people
discover, use, and build upon a wide range of content through a powerful research and teaching
platform, and preserves this content for future generations. JSTOR is part of ITHAKA, a not-for-profit
organization that also includes Ithaka S+R and Portico. For more information about JSTOR, please
contact support@jstor.org.

VOL. III.

MAY 24, 1906.

No. 11.

THE

JOURNAL OF PHILOSOPHY

PSYCHOLOGY AND SCIENTIFIC

METHODS

ON THE NATURE OF INDUCTION1

NY propositionis susceptibleto two sorts of proof. We can

adduce premises that directly imply it, or we can adduce
premisesthat indirectlyimply it because they imply the falsityof
alternatives. In inductivereasoningwe prove uniits contradictory
versal propositionsby adducing as premisesthe particularpropositions furnishedby experience. Formal logic tells us that the value
of a particularpropositionconsistsin its power to disproveits contradictoryuniversalratherthan to prove its subalternateuniversal.
We mightnaturallysuppose that the evidentialfunctionof experience as a knowledgeof particulars,was to disproveuniversalstatementsratherthan to prove them,and that if a universal conclusion
was proved true by appeal to experience,the proof would be based
upon the disproofor eliminationof alternatives. That inductionis
actually and always of this indirecttype of inference,and that as
such it is properlyexpressedby a disjunctivesyllogismin the negative mood (modus tollendoponens), is what I wish to show.
There is, of course,no noveltyin the conceptionof inductionas
because they
a process of elimination. Mill's canons are efficacious
embody implicitly the eliminative principle. In Hobhouse and
Aikins, to mentiononly two of the modernlogicians,the principle
is explicitly recognized,and the chief problems of induction are
treated,especiallyby Hobhouse,fromthatpoint of view. Yet so far
as I am aware therehas been nowherean attemptto identifyinduetionin all its phases withthe kind of indirectinferenceknownas the
reductioad absurdum,and it has seemedto me worthwhile to make
that attemptfor two reasons: First, because the several inductive
methodswhenviewed fromthis standpointappear not as a group of
disconnectedprinciples, but as an organic system and hierarchy
whichis applicable in its entiretyto everyinductiveproblemand in
which each principle has its own functionand virtue by which it
'Read at the annual meeting of the American Philosophical Association,
at Cambridge, December, 1905.

281

282

THE

JOURNAL OF PHILOSOPHY

supplementsthe defectsof the principlesthat precede it; second,on

account of the new light thrownby the indirecttheoryof induetion upon the generalepistemologicalproblemof derivinguniversals
fromparticulars.
And now,by way of introductionto the more positivetreatment
of the subject, let us considersome of the difficultiesinvolved in
what is still, I think,the usual conceptionof induction. Induction
whentreatedas a mode of directinferenceis divided into two kinds,
-perfect and imperfect. In perfectinduction,we reason that as
theseA's are B's and as theseA's are all the A's, it mustfollowthat
all A's are B's. It is clear that what is called perfectinductionis
only possiblewhen the total numberof individualsmakingthe class
is limited. Thus we can proveby this methodthat all the monthsin
days, or that all the flowersin the
the year have less than thirty-two
gardenare fragrant,but not that all bodies gravitateor that all men
are mortal. In these latter propositions,which are genuine universals, the classes contain an unlimitednumberof members,and experience can never supply us with more than an insignificantfraction of them. In imperfectinduction,whichis supposed to be a degenerateformof perfectinduction,we boldly concludethat because
portionof a class has been observedto possess a ceran infinitesimal
tain propertythe whole class will have that property. The methods
or canonsof inductionare the principlesthat informus whenwe can
and whenwe can not take the inductiveleap.
Now thereis one eircumstancein particularwhichmightlead us
to suspectthat therewas somethingradicallywrongwith the notion
that inductionis a degenerateformof perfectinduction. Neither
the actual numberof positiveinstancesobservednor the ratio of that
to the total numberhas anythingwhateverto do with the degree of
validitypossessedby the induction. Perfectinductionis essentially
quantitative,depending,as it does, upon observationof all the members of a given class. The canons that guide us in making the socalled imperfectinductionare, on the otherhand, essentiallyqualitative, and not, as we mightsuppose,imperfectlyquantitative. That
is to say, it is never a question of observingalmost all, or a bare
majority,or even an appreciable fractionof the whole number of
materialbodies,forexample,as evidenceforthe inductivegeneralization that all bodies gravitate. We contrivein the few cases under
our controlto eliminateby the methodsof differenceand especially
of agreementall the characteristicsof bodies that could possibly
cause their gravitationexcept those of extensionand inertia, and
on the strengthof this eliminationwe unhesitatinglyconclude that
a materialbody,merelyas such (and hence all materialbodies), will
gravitate.

PSYCHOLOGY

AND SCIENTIFIC

METHODS

283

And now that we have brieflyconsideredthe contradictionbetween inductivetheoryas exemplifiedin the supposedlyarchetypal
syllogismof perfectinduction,and inductivepractiseas exemplified
in Mill's canons,we may look to see how this contradictioncan be
removedby treatinginduction as belongingessentiallyand exclusivelyto the indirecttype of inference.
Every inductive problem indirectly,and the usual inductive
problem directly,concernsthe determinationof a causal relation.
A phenomenonoccursin whichwe are forsomereasoninterestedand
we at once seek among its antecedentsand consequentsfor phenomena whichare relatedto it as cause and as effect. Defininga causal
relation as the relation of universal concomitantpresence,absence
and variationof two phenomena,we mustassume as the basal postulates of all induction (1) that every eventhas an antecedentand a
consequentwith which it is causally or universallyrelated,and (2)
that we can enumeratethese possible causal relations,by the aid of
perceptionand previous knowledge. Now let M be a phenomenon
whose causal relationswe are seekingto discover,and let A be an
antecedentor consequentphenomenonwhichwe suspect,or provisionally assume,to be, and whichin realityis, causally relatedto M; we
can thenclassifythe possiblecausal relationsof M withrespectto A
under fiveheads. This divisionmay be brieflystated in the formof
a disjunctivepropositionwhich will constitutethe major premiseof
a typical inductivesyllogism. Thus we can say that
The cause or effectof M is either (1) a phenomenonsymbolized
by X that is related to A only casually or by chance; or (2) a
phenomenonsymbolizedby B, C or D, which is collocatedwith A
but not indissolubly;or (3) a complexphenomenonsymbolizedby
A B, A C or A D, of whichA is an indispensablepart; or (4) a phenoinenonsymbolizedby a whichis an aspect,phase or degree of A;
or (5) A itself. The fourinductivemethodsof simple enumeration,
difference,
agreementand concomitantvariationexpressthe typesof
particularnegativepropositionsfurnishedby experience,and as such
they constitutethe complex minor premises of the syllogismand
serve to contradictor eliminateall but one of the alternativesset
forthin the major premise. The conclusionis, of course,the cateof theonlyalternativenot eliminated. I shall now
goricalaffirmation
tryto showhow each of the inductivemethodsis especiallysuited to
eliminateone of these alternativesand that the eliminativefunction
is the only functionthat they could or do perform: First, then,
to remove the possibilitythat A and M are connectedmerelyby
chance,we use the methodof simple enumeration. We observethe
frequencywithwhichM occursin conjunctionwithA, and compare
this with the frequencywith which M mightbe expected to occur

284

THE

JOURNAL OF PHILOSOPHY

with A if they were quite independent. If the formerfrequency

greatlyexceedsthe latterwe considerthe conjunctionto be something
of observationsrequiredto eliminate
morethan casual. The niumber
the hypothesisof chance is thusstrictlydeterminedby the joint independentprobabilityof the eventsin question.
Having eliminatedchance by the methodof simple enumeration,
we next eliminateby the methodof difference
the possibilitythat M
is causally related,not to A, but to the antecedentsand consequents
withwhichA is collocated. We find,let us say, cases in whichB, C
and D are simultaneouslyor successivelypresentwhen the event M
is absent. The resultsof such observationsmay be statedin the form
of a particularnegativeproposition,as follows: Some cases of B, C
and D are not cases of M, whichmeansthat the universalaffirmative
proposition: All cases of B, C or D are cases of M-which expresses
a possiblecausal relation-is eliminated.
Supposing,now,that the phenomenonA has not been foundpresent in the absence of M, and consequentlyhas not been eliminated,
it becomesnecessaryto apply the methodof agreementin order to
decide whetherM is not causally related to a complexphenomenon
such as AB, AC or AD, of which A is only a part, for the weak
point in the methodof differencelies in the fact that it can only
prove thatA is at least a part of the cause or effectof M, not that it
is the whole. We observeby the methodof agreementthatB, C and
D can be simultaneouslyor successivelyabsent when M and A are
both present. This again eliminatesB, C and D, but it also eliminates the possibilitythat A needs to cooperatewith B, C and D in
orderto be causally relatedto M.
We have now proved that M is causally related eitherto A or to
some phase of A which we called a. And to secure the elimination of this fourthalternativewe use the last and mostpowerfulof
the inductive canons, viz., that of concomitantvariation. If we
findthat M and A vary in perfectconcomitancewe knowthat every
phase or degreeof A, ratherthan some particular phase such as a,
is causally related to everyphase or degree of M. If, on the other
hand, we had discoveredthat A did not vary withM in any manner,
we should have proved that the true cause or effectof M was a and
not A, as such. Or, again, if we had foundthat M did not vary directlywithA, but withsome functionof A, we should concludethat
the cause or effectof M was A in conjunctionwith a.
The universalaffirmative
conclusionof an inductivesyllogismis
thus in any case the result of successiveeliminations-in the form
of the particularnegativepropositionsfurnishedby experience-of
all but one of the alternativeuniversalsset forthin the disjunctive
major premiseas hypotheses. And each of the inductivemethodsis,

PSYCHOLOGY

AND SCIENTIFIC

METHODS

285

as we have seen, adapted to the eliminationof a certaintype of alternative.

And now a wordmustbe said in regardto the two methodswhich
we have not mentioned-themethodof residuesand the joint method
of agreementand difference. The methodof residuesis confessedly
a method of elimination; it is, however,hardly worthyof being
ranked with the othermethods,for it is applicable only when both
the antecedentsand consequentsare quantitiesof matteror energy,
for in no othercase can we apply the conceptionof a cause exhausting its causality in the productionof a given effect. As for the
joint method,its continuedexistencein logical text-booksaffordsa
good illustration(1) of the fact that logicians have failed to recognize the exclusivelyeliminativenature of inductionand (2) of the
results of that failure. The joint methodbids us supplementthe
instances
methodof agreement,by the collectionof as manydifferent
of
absence
with
the
as possibleof the absenceof a phenomenonalong
of
cases
that
these
be
shown
its supposed cause. Now it can easily
concurrentabsence are as such quite worthlessas evidence of causal
connection. If we are consideringwhethera protectivetariffcauses
national prosperitywe do not adduce as evidencethe generationof
Roentgenrays or the constructingof a sonnet,and yet theseare differentcases of the concurrentabsence of protectionand prosperity
and as such perfectlyconformto the requirementsof the joint
methodas wordedby Mill and as symbolizedby Jevons. What we
actually seek to find in such an investigationare always cases in
whichnot merelythe supposed cause is absent,but in whichthealternatives to the supposed cause are presentalong with the absence of
as
the effect,and hence are eliminated. In the methodof difference
usually schematizedthis is done in a singlepair of instancesin which
ABC followedby M is comparedwith BC followedby the absence
of M; but it can equally well be done piecemealor by a successionof
instances,one showingsimplythe presenceof B, a second the presence of C, a thirdthe presenceof D, etc.; along with the absence of
M. Now as the so-calledjoint methodis, when rightlyunderstood,
nothingwhateverbut a combinationof the methodof agreementand
whereeach is applied successivelyin several
the methodof difference
instances,ratherthan simultaneouslyin a single pair, it does not deserveto be classed as a separate canon.2
The claim was made at the beginningof the paper thatthe identi2
By way of illustration I subjoin (1) the incorrect and meaningless
symbolization of the joint method given by Jevons, 'Lessons in Logic,' p. 247;
(2) the correct symbolization substantially as given by Aikins, 'The Principles
of Logic,' p. 275; (3) the symbolization in a single pair of instances of the

286

THE

JOURNAL OF PHILOSOPHY

ficationof inductionwith the indirecttype of argument,or reductio

ad absutrdum,
possessedtwo advantages,(1) the unificationof the inductivemethods,(2) the exhibitionin a new lightof the generalepistemologicalproblemof derivinguniversalsfromparticulars. I have
said what I could in regard to the firstof these advantages,but I
should like in conclusionto speak furtheras to the second.
The attacks upon the possibilityof reasoningfromparticulars,
that have been made by the scepticson the one hand and by the extreme aprioristson the other,are based in the main upon a quite
properrealizationof the gulf betweena subalternpropositionand its
subalternates. The number of cases exemplifyinga genuine universal or law of natureis, as we have said, always infinite,and hence
the direct inferencefromsome to all is not only uncertain (which
would be admittedby inductivelogicians), but would seem to be not
even probable-to be, in fact,infinitely
improbable. For we can, of
course,neverobserveeven an appreciable fraction,to say nothingof
a majorityof the membersof an infiniteseries. Now when we give
up this attempt at direct inferenceand exorcise from inductive
theorythe specterof a so-calledperfectinductionas an ideal to be
approximated,the whole problemappears in a less paradoxical and
more hopeful light. For the experientialevidence in the formof
particular propositions,which was worthlessas a means of direct
proofof the subalternate,is perfectlycapable of disprovingthe contradictoryand thus indirectlyestablishinga hypothesisas a survival
of the fittest. Of coursethis does not mean that we have merely,by
the substitutionof the intensiveforthe extensiveview of the subject,
removeduncertaintyfromgeneralizationsfromexperience,but only
that fromour point of view we may more clearly see why it is that
the degreeof probabilityof any inductiveconclusionis measuredby
the numberof antecedentlypossible alternativeconclusionsand by
the ease withwhichtheycan be isolated,enumeratedand eliminated
ratherthan by the merenumberof instancesobserved.
W. P. MONTAGUE.
COLUMBIA

UNIVERSITY.

method of agreement and of the method of difference. The substantial identity

of (2) and (3) is obvious.
(1)
ABC - abc
ADE -ade
AFG
afg
AHKI ahk
PQ
RS
TV
XY

-pq
-rs
-tv
-wy

(2)
ABCD - abcd
ABGH - abgh
ACGK- acgk
BDE -bde
CDG -cdg
BEF -bef

ABC
ADE

(3)
-

abc
ade

ABC -abc
BC-bc