The Science of "Muddling Through"

Author(s): Charles E. Lindblom

Source: Public Administration Review, Vol. 19, No. 2 (Spring, 1959), pp. 79-88
Published by: Wiley on behalf of the American Society for Public Administration
Stable URL: http://www.jstor.org/stable/973677 .
Accessed: 26/01/2015 08:30

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp

.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.

Wiley and American Society for Public Administration are collaborating with JSTOR to digitize, preserve and
extend access to Public Administration Review.

http://www.jstor.org

This content downloaded from 147.8.204.164 on Mon, 26 Jan 2015 08:30:52 AM

All use subject to JSTOR Terms and Conditions
The Science of Muddling Through
By CHARLES E. LINDBLOM
AssociateProfessorofEconomics
Yale University
SUPPOSE an administratoris given respon-
sibility for formulatingpolicy with re-
spect to inflation.He mightstartby try-
ing to list all related values in order of
importance,e.g., full employment,reasonable
business profit,protection of small savings,
preventionof a stockmarketcrash. Then all
possible policy outcomes could be rated as
more or less efficientin attaininga maximum
of thesevalues. This would of course require
a prodigious inquiry into values held by
membersof societyand an equally prodigious
set of calculationson how much of each value
is equal to how much of each othervalue. He
could then proceed to outline all possible
policy alternatives.In a thirdstep,he would
undertakesystematiccomparisonof his multi-
tude of alternativesto determinewhich at-
tainsthegreatestamountof values.
In comparing policies, he would take ad-
vantage of any theoryavailable that general-
ized about classes of policies. In considering
inflation,for example, he would compare all
policies in the light of the theoryof prices.
Since no alternativesare beyond his investi-
gation, he would consider strictcentral con-
trol and the abolition of all prices and mar-
kets on the one hand and elimination of all
public controls with reliance completelyon
the freemarketon the other,both in the light
of whatever theoretical generalizations he
could findon such hypotheticaleconomies.
Finally, he would try to make the choice
thatwould in factmaximizehis values.
An alternativeline of attack would be to
set as his principal objective,eitherexplicitly
or without conscious thought,the relatively
simple goal of keeping prices level. This ob-
jective mightbe compromisedor complicated
by only a few other goals, such as full em-
79
This content downloaded from 147.8.204.164 on Mon, 26 Jan 2015 08:30:52 AM
All use subject to JSTOR Terms and Conditions
> Short courses, books, and articles exhort admin-
istratorsto make decisions more methodically,but
there has been little analysis of the decision-making
process now used by public administrators.The
usual process is investigatedhere-and generallyde-
fended against proposals for more "scientific"meth-
ods.
Decisions of individual administrators,of course,
must be integrated with decisions of others to
form the mosaic of public policy. This integration
of individual decisions has become the major con-
cern of organizationtheory,and the way individuals
make decisions necessarilyaffectsthe way those de-
cisions are best meshed with others'. In addition,
decision-makingmethod relates to allocation of de-
cision-makingresponsibility-whoshould make what
decision.
More "scientific" decision-making also is dis-
cussed in this issue: "Tools for Decision-Making in
Resources Planning."
ployment.He would in fact disregardmost
other social values as beyond his presentin-
terest,and he would forthe momentnot even
attempt to rank the few values that he re-
garded as immediately relevant. Were he
pressed,he would quickly admit that he was
ignoringmany related values and many pos-
sible importantconsequencesof his policies.
As a second step, he would outline those
relativelyfewpolicyalternativesthatoccurred
to him. He would then compare them. In
comparinghis limitednumberof alternatives,
inostof themfamiliarfrompast controversies,
he would not ordinarilyfinda body of theory
precise enough to carryhim througha com-
parison of their respectiveconsequences. In-
stead he would rely heavily on the recordof
past experiencewith small policystepsto pre-
dict the consequences of similar steps ex-
tendedinto thefuture.
Moreover,he would findthat the policy al-
ternativescombined objectives or values in
differentways.For example, one policy might
offerprice level stabilityat the cost of some
8o PUBLIC ADMINISTRATION
risk of unemployment;another might offer
less price stabilitybut also less risk of unem-
ployment. Hence, the next step in his ap-
proach-the final selection-would combine
into one the choice among values and the
choice among instruments forreachingvalues.
It would not, as in the firstmethodof policy-
making,approximatea moremechanicalproc-
ess of choosing the means that best satisfied
goals that were previously clarified and
ranked. Because practitionersof the second
approach expect to achieve their goals only
partially, they would expect to repeat end-
lessly the sequence just described,as condi-
tionsand aspirationschanged and as accuracy
of predictionimproved.
ByRootor by Branch
For complex problems, the firstof these
two approaches is of course impossible. Al-
thoughsuch an approach can be described,it
cannot be practicedexcept forrelativelysim-
ple problems and even then only in a some-
what modified form. It assumes intellectual
capacities and sources of informationthat
men simplydo not possess,and it is even more
absurd as an approach to policy when the
time and money that can be allocated to a
policy problem is limited, as is always the
case. Of particular importanceto public ad-
ministratorsis the fact that public agencies
are in effectusually instructednot to practice
the firstmethod. That is to say, their pre-
scribed functionsand constraints-thepoliti-
cally or legally possible-restricttheir atten-
tion to relativelyfew values and relatively
few alternativepolicies among the countless
alternativesthat mightbe imagined. It is the
second methodthatis practiced.
Curiously,however,the literaturesof deci-
sion-making,policy formulation, planning,
and public administrationformalizethe first
approach ratherthan the second,leaving pub-
lic administratorswho handle complex deci-
sions in the position of practicingwhat few
preach. For emphasisI run some riskof over-
statement.True enough, the literatureis well
aware of limitson man's capacitiesand of the
inevitabilitythat policies will be approached
in some such styleas the second. But attempts
to formalizerational policy formulation-to
lay out explicitlythe necessarysteps in the
This content downloaded from 147.8.204.164 on Mon, 26 Jan 2015 08:30:52 AM
All use subject to JSTOR Terms and Conditions
REVIEW
process-usually describe the first approach
and not the second.'
The common tendency to describe policy
formulation even for complex problems as
though it followed the firstapproach has been
strengthened by the attention given to, and
successes enjoyed by, operations research, sta-
tistical decision theory, and systems analysis.
The hallmarks of these procedures, typical of
the firstapproach, are clarity of objective, ex-
plicitness of evaluation, a high degree of com-
prehensiveness of overview, and, wherever
possible, quantification of values for mathe-
matical analysis. But these advanced proce-
dures remain largely the appropriate tech-
niques of relatively small-scale problem-solving
where the total number of variables to be
considered is small and value problems re-
stricted. Charles Hitch, head of the Economics
Division of RAND Corporation, one of the
leading centers for application of these tech-
niques, has written:
I would make the empirical generalizationfrom
my experience at RAND and elsewherethat oper-
i.e., of
ations researchis the art of sub-optimizing,
solving some lower-levelproblems,and that diffi-
culties increase and our special competence di-
minishesby an orderof magnitudewitheverylevel
of decision makingwe attemptto ascend. The sort
of simple explicit model which operations re-
searchersare so proficientin using can certainly
reflectmost of the significantfactorsinfluencing
trafficcontrol on the George Washington Bridge,
but the proportionof the relevant reality which
we can representby any such model or models in
studying,say, a major foreign-policy decision, ap-
pears to be almosttrivial.
Accordingly, I propose in this paper to
clarify and formalize the second method,
1 James G. March and Herbert A. Simon similarly
characterize the literature. They also take some im-
portant steps, as have Simon's recent articles, to de-
scribe a less heroic model of policy-making.See Or-
ganizations (John Wiley and Sons, 1958), p. 137.
Research and National Planning-A
2Operations
Dissent," 5 Operations Research 718 (October, 1957).
Hitch's dissent is from particular points made in the
article to which his paper is a reply; his claim that
operations research is for low-level problems is widely
accepted.
For examples of the kind of problems to which op-
erations research is applied, see C. W. Churchman,
R. L. Ackoffand E. L. Arnoff,Introduction to Opera-
tions Research (John Wiley and Sons, 1957); and J. F.
McCloskey and J. M. Coppinger (eds.), Operations Re-
search for Management, Vol. II, (The Johns Hopkins
Press, 1956).
 THE SCIENCE OF "MUDDLING
much neglectedin the literature.This might
be describedas the method of successivelim-
itedcomparisons.I will contrastit withthefirst
approach,which mightbe called the rational-
comprehensivemethod.3 More impressionis-
tically and briefly-and thereforegenerally
used in this article-theycould be character-
ized as the branch method and root method,
the formercontinuallybuilding out fromthe
current situation, step-by-stepand by small
degrees;the latterstartingfromfundamentals
anew each time,building on the past only as
experience is embodied in a theory,and al-
ways prepared to start completelyfrom the
groundup.
Let us put the characteristicsof the two
methodsside by side in simplestterms.
Rational-Comprehensive(Root)
la. Clarification of values or objectives distinct from
and usually prerequisite to empirical analysis of
alternativepolicies.
2a. Policy-formulationis thereforeapproached through
means-end analysis: First the ends are isolated,
then the means to achieve them are sought.
3a. The test of a "good" policy is that it can be shown
to be the most appropriate means to desired ends.
4a. Analysis is comprehensive; every important rele-
vant factor is taken into account.
5a. Theory is often heavily relied upon.
Assumingthat the root method is familiar
and understandable,we proceed directlyto
clarificationof its alternativeby contrast.In
explaining the second, we shall be describing
how most administrators do in fact approach
complex questions, for the root method,the
"best" way as a blueprintor model, is in fact
not workable for complex policy questions,
and administratorsare forced to use the
methodof successivelimitedcomparisons.
Intertwining
Evaluationand Empirical
Analysis(lb)
The quickestway to understandhow values
are handled in the method of successivelim-
'II am assuming that administratorsoften make pol-
icy and advise in the making of policy and am treating
decision-makingand policy-makingas synonymousfor
purposes of this paper.
This content downloaded from 147.8.204.164 on Mon, 26 Jan 2015 08:30:52 AM
All use subject to JSTOR Terms and Conditions
THROUGH" 81
ited comparisons is to see how the root
method oftenbreaks down in its handling of
values or objectives. The idea that values
should be clarified,and in advance of the ex-
amination of alternativepolicies, is appeal-
ing. But what happens when we attemptit
for complex social problems?The firstdiffi-
cultyis that on many criticalvalues or objec-
tives,citizensdisagree,congressmendisagree,
and public administratorsdisagree. Even
where a fairlyspecificobjective is prescribed
for the administrator,there remains consid-
erable room for disagreementon sub-objec-
tives.Consider,forexample, the conflictwith
respect to locating public housing,described
in Meyersonand Banfield'sstudyof the Chi-
SuccessiveLimited Comparisons(Branch)
ib. Selection of value goals and empirical analysis of
the needed action are not distinct from one an-
other but are closely intertwined.
2b. Since means and ends are not distinct,means-end
analysis is often inappropriate or limited.
3b. The test of a "good" policy is typicallythat vari-
ous analystsfind themselvesdirectlyagreeing on a
policy (without their agreeing that it is the most
appropriate means to an agreed objective).
4b. Analysis is drasticallylimited:
i) Important possible outcomes are neglected.
ii) Important alternative potential policies are
neglected.
iii) Important affectedvalues are neglected.
5b. A succession of comparisons greatly reduces or
eliminates reliance on theory.
cago Housing Authority4-disagreement which
occurreddespite the clear objectiveof provid-
ing a certainnumber of public housing units
in the city.Similarlyconflicting are objectives
in highwaylocation,traffic control,minimum
wage administration,developmentof tourist
facilitiesin national parks,or insectcontrol.
Administratorscannot escape these con-
flictsby ascertainingthe majority'spreference,
for preferenceshave not been registeredon
most issues; indeed, thereoftenare no prefer-
ences in the absence of public discussionsuffi-
cient to bringan issue to the attentionof the
electorate. Furthermore,there is a question
4 Martin Meyerson and Edward C. Banfield, Politics,
Planning and the Public Interest (The Free Press,
1955).
82 PUBLIC ADMINISTRATION
of whetherintensityof feelingshould be con-
sidered as well as the numberof personspre-
ferringeach alternative.By the impossibility
of doing otherwise,administratorsoften are
reduced to deciding policy withoutclarifying
objectivesfirst.
Even when an administratorresolvesto fol-
low his own values as a criterionfordecisions,
he often will not know how to rank them
when theyconflictwith one another,as they
usually do. Suppose, for example, that an ad-
ministratormust relocate tenants living in
tenementsscheduled for destruction.One ob-
jective is to empty the buildings fairly
promptly,another is to find suitable accom-
modation for personsdisplaced, another is to
avoid frictionwith residentsin other areas in
which a large influxwould be unwelcome,an-
other is to deal with all concerned through
persuasionif possible,and so on.
How does one state even to himself the
relative importance of these partially con-
flictingvalues? A simple ranking of them is
not enough; one needs ideally to know how
much of one value is worth sacrificingfor
some of another value. The answer is that
typicallythe administratorchooses-and must
choose-directlyamong policies in which these
values are combined in differentways. He
cannot firstclarifyhis values and then choose
among policies.
A more subtle third point underlies both
the firsttwo. Social objectives do not always
have the same relative values. One objective
may be highly prized in one circumstance,
another in another circumstance.If, for ex-
ample, an administratorvalues highly both
the dispatchwith which his agencycan carry
throughits projectsand good public relations,
it matterslittlewhichof the two possiblycon-
flictingvalues he favorsin some abstractor
general sense. Policy questions arise in forms
which put to administratorssuch a question
as: Given the degree to which we are or are
not already achieving the values of dispatch
and the values of good public relations,is it
worthsacrificinga little speed for a happier
clientele,or is it betterto risk offendingthe
clienteleso thatwe can get on with our work?
The answer to such a question varies with
circumstances.
The value problem is, as the example
shows,always a problem of adjustmentsat a
margin. But there is no practicable way to
This content downloaded from 147.8.204.164 on Mon, 26 Jan 2015 08:30:52 AM
All use subject to JSTOR Terms and Conditions
REVIEW
state marginal objectivesor values except in
termsof particular policies. That one value
is preferredto anotherin one decision situa-
tion does not mean that it will be preferred
in anotherdecision situation in which it can
be had only at greatsacrificeof anothervalue.
Attemptsto rank or order values in general
and abstracttermsso that they do not shift
fromdecision to decision end up by ignoring
the relevant marginal preferences.The sig-
nificanceof this third point thus goes very
far.Even if all administrators had at hand an
agreed set of values, objectives, and con-
straints,and an agreed rankingof these val-
ues, objectives, and constraints,their mar-
ginal values in actual choice situationswould
be impossibleto formulate.
Unable consequentlyto formulatethe rele-
vant values firstand then choose among poli-
cies to achieve them, administratorsmust
choose directlyamong alternativepolicies that
offerdifferentmarginal combinationsof val-
ues. Somewhatparadoxically,the only practi-
cable way to disclose one's relevantmarginal
values even to oneselfis to describethe policy
one chooses to achieve them. Except roughly
and vaguely,I know of no way to describe-
or even to understand-whatmy relativeeval-
uations are for, say, freedom and security,
speed and accuracyin governmentaldecisions,
or low taxes and better schools than to de-
scribe my preferencesamong specificpolicy
choices that might be made between the al-
ternativesin each of thepairs.
In summary,two aspects of the process by
which values are actuallyhandled can be dis-
tinguished.The firstis clear: evaluation and
empirical analysis are intertwined;that is,
one choosesamong values and among policies
at one and the same time. Put a little more
elaborately,one simultaneouslychooses a pol-
icy to attain certain objectives and chooses
the objectives themselves.The second aspect
is related but distinct: the administratorfo-
cuses his attentionon marginal or incremen-
tal values. Whetherhe is aware of it or not,
he does not find general formulationsof
objectivesveryhelpfuland in factmakes spe-
cific marginal or incremental comparisons.
Two policies, X and Y, confronthim. Both
promisethe same degreeof attainmentof ob-
jectivesa, b, c, d, and e. But X promiseshim
somewhat more of f than does Y, while Y
promiseshim somewhatmore of g than does
 THE SCIENCE OF "MUDDLING
X. In choosing between them, he is in fact
offeredthe alternativeof a marginalor incre-
mental amount of f at the expense of a mar-
ginal or incrementalamount of g. The only
values that are relevant to his choice are
these incrementsby which the two policies
differ;and, when he finallychooses between
the two marginalvalues, he does so by mak-
ing a choice betweenpolicies.5
As to whether the attempt to clarifyob-
jectivesin advance of policy selectionis more
or less rational than the close intertwining of
marginal evaluation and empirical analysis,
the principal difference establishedis that for
complex problemsthe firstis impossible and
irrelevant,and the second is both possibleand
relevant.The second is possible because the
administratorneed not try to analyze any
values except the values by which alternative
policies differand need not be concernedwith
them except as they differmarginally.His
need for informationon values or objectives
is drasticallyreduced as compared with the
root method; and his capacity for grasping,
comprehending, and relatingvalues to one an-
other is not strained beyond the breaking
point.
RelationsBetweenMeans and Ends (2b)
Decision-makingis ordinarilyformalizedas
a means-endsrelationship: means are con-
ceived to be evaluated and chosen in the
light of ends finallyselectedindependentlyof
and prior to the choice of means. This is the
means-endsrelationshipof the root method.
But it followsfromall that has just been said
thatsuch a means-endsrelationshipis possible
only to the extent that values are agreed
upon, are reconcilable,and are stable at the
margin. Typically, therefore,such a means-
ends relationshipis absent from the branch
method, where means and ends are simul-
taneouslychosen.
Yet any departurefromthe means-endsre-
lationshipof the root methodwill strikesome
readersas inconceivable.For it will appear to
themthatonly in such a relationshipis it pos-
sible to determinewhetherone policy choice
is betteror worse than another. How can an
administratorknow whetherhe has made a
'The line of argument is, of course, an extension of
the theory of market choice, especially the theory of
consumerchoice, to public policy choices.
This content downloaded from 147.8.204.164 on Mon, 26 Jan 2015 08:30:52 AM
All use subject to JSTOR Terms and Conditions
THROUGH" 83
wise or foolishdecision if he is withoutprior
values or objectivesby which to judge his de-
cisions?The answer to this question calls up
the third distinctivedifferencebetween root
and branch methods: how to decide the best
policy.
The Test of "Good" Policy(3b)
In the root method,a decision is "correct,"
"good," or "rational" if it can be shownto at-
tain some specifiedobjective,wherethe objec-
tivecan be specifiedwithoutsimplydescribing
the decision itself.Where objectives are de-
finedonly throughthe marginalor incremen-
tal approach to values described above, it is
still sometimespossible to testwhethera pol-
icy does in factattain the desired objectives;
but a precisestatementof the objectivestakes
the formof a descriptionof the policy chosen
or some alternativeto it. To show that a pol-
icy is mistakenone cannot offeran abstract
argumentthat importantobjectives are not
achieved; one mustinsteadargue thatanother
policyis moreto be preferred.
So far, the departurefromcustomaryways
of looking at problem-solvingis not trouble-
some, for many administratorswill be quick
to agree that the most effective discussionof
thecorrectnessof policydoes take the formof
comparison with other policies that might
have been chosen. But what of the situation
in which administratorscannot agree on val-
ues or objectives,eitherabstractlyor in mar-
ginal terms?What then is the testof "good"
policy?For the root method,thereis no test.
Agreementon objectives failing,there is no
standardof "correctness."For the method of
successive limited comparisons, the test is
agreementon policyitself,which remainspos-
sible even when agreementon values is not.
It has been suggestedthatcontinuingagree-
ment in Congress on the desirabilityof ex-
tending old age insurancestemsfromliberal
desiresto strengthenthe welfareprogramsof
the federalgovernmentand fromconservative
desires to reduce union demands for private
pension plans. If so, this is an excellentdem-
onstrationof the ease with whichindividuals
of different ideologiesoftencan agree on con-
cretepolicy.Labor mediatorsreporta similar
phenomenon:the contestantscannot agree on
criteria for settling their disputes but can
agree on specific proposals. Similarly,when
84 PUBLIC ADMINISTRATION
one administrator'sobjective turnsout to be
another's means, they often can agree on
policy.
Agreementon policy thus becomesthe only
practicable test of the policy's correctness.
And forone administratorto seek to win the
other over to agreement on ends as well
would accomplish nothing and create quite
unnecessarycontroversy.
If agreementdirectlyon policy as a test
for "best" policy seems a poor substitutefor
testing the policy against its objectives, it
ought to be rememberedthat objectivesthem-
selves have no ultimate validity other than
they are agreed upon. Hence agreement is
the testof "best" policy in both methods.But
where the root methodrequiresagreementon
what elementsin the decision constituteob-
jectives and on which of these objectives
should be sought, the branch method falls
back on agreementwhereverit can be found.
In an importantsense, therefore,it is not
irrational for an administratorto defend a
policy as good without being able to specify
whatit is good for.
Analysis(4b)
Non-Comprehensive
Ideally, rational-comprehensive analysis
leaves out nothingimportant.But it is impos-
sible to take everythingimportantinto con-
sideration unless "important"is so narrowly
definedthat analysis is in fact quite limited.
Limits on human intellectual capacities and
on available informationset definitelimitsto
man's capacityto be comprehensive.In actual
fact, therefore,no one can practice the ra-
tional-comprehensive method for really com-
plex problems,and everyadministratorfaced
complex problemmustfind
with a sufficiently
waysdrasticallyto simplify.
An administratorassistingin the formula-
tion of agricultural economic policy cannot
in the firstplace be competenton all possible
policies. He cannot even comprehendone pol-
icy entirely.In planning a soil bank program,
he cannot successfullyanticipate the impact
of higher or lower farm income on, say, ur-
banization-the possible consequentloosening
of family ties, possible consequent eventual
need for revisionsin social securityand fur-
therimplicationsfortax problemsarisingout
of new federal responsibilitiesfor social se-
curityand municipal responsibilitiesfor ur-
This content downloaded from 147.8.204.164 on Mon, 26 Jan 2015 08:30:52 AM
All use subject to JSTOR Terms and Conditions
REVIEW
ban services.Nor, to follow another line of
repercussions,can he work throughthe soil
bank program'seffectson prices for agricul-
tural productsin foreignmarketsand conse-
quent implicationsfor foreignrelations,in-
cluding those arisingout of economic rivalry
betweentheUnited Statesand the U.S.S.R.
In the method of successive limited
comparisons, simplificationis systematically
achieved in two principal ways. First, it is
achieved through limitation of policy com-
parisons to those policies that differin rela-
tivelysmall degree frompolicies presentlyin
effect.Such a limitationimmediatelyreduces
the number of alternativesto be investigated
and also drasticallysimplifiesthe characterof
theinvestigationof each. For it is not necessary
to undertakefundamentalinquiryinto an al-
ternativeand its consequences;it is necessary
only to studythoserespectsin which the pro-
posed alternativeand its consequences differ
fromthe status quo. The empirical compari-
son of marginaldifferences among alternative
policies that differ only marginally is, of
course, a counterpartto the incrementalor
marginal comparison of values discussed
above.6
Relevance as Well as Realism
It is a matterof common observationthat
in Westerndemocraciespublic administrators
and policyanalystsin generaldo largelylimit
their analyses to incremental or marginal
differencesin policies that are chosen to differ
only incrementally.They do not do so, how-
ever,solelybecause theydesperatelyneed some
way to simplifytheirproblems;theyalso do so
in order to be relevant.Democracies change
their policies almost entirely through in-
crementaladjustments.Policy does not move
in leaps and bounds.
The incremental character of political
change in the United Stateshas oftenbeen re-
marked.The two major political partiesagree
on fundamentals;they offeralternativepoli-
cies to the voters only on relatively small
points of difference.Both parties favor full
employment,but they define it somewhat
differently;both favor the development of
6 A more precise definition of incremental policies
and a discussion of whether a change that appears
"small" to one observer might be seen differentlyby
another is to be found in my "Policy Analysis," 48
American Economic Review 2g8 (June, 1958).
 THE SCIENCE OF "MUDDLING
waterpowerresources,but in slightlydifferent
ways; and both favorunemploymentcompen-
sation, but not the same level of benefits.
Similarly,shiftsof policy within a partytake
place largely through a series of relatively
small changes, as can be seen in their only
gradual acceptance of the idea of govern-
mental responsibilityfor support of the un-
employed,a change in partypositions begin-
ning in the early 30's and culminatingin a
sensein theEmploymentAct of 1946.
Party behavior is in turn rooted in public
attitudes,and political theoristscannot con-
ceive of democracy'ssurvivingin the United
States in the absence of fundamentalagree-
ment on potentially disruptive issues, with
consequent limitation of policy debates to
relativelysmall differencesin policy.
Since the policies ignored by the adminis-
tratorare politicallyimpossibleand so irrele-
vant, the simplificationof analysis achieved
by concentratingon policies that differonly
incrementallyis not a capricious kind of
simplification.In addition, it can be argued
that, given the limits on knowledge within
which policy-makers are confined,simplifying
by limitingthe focus to small variationsfrom
present policy makes the most of available
knowledge.Because policies being considered
are like present and past policies, the ad-
ministratorcan obtain informationand claim
some insight. Non-incrementalpolicy pro-
posals are thereforetypicallynot only politi-
cally irrelevantbut also unpredictablein their
consequences.
The second method of simplificationof
analysisis the practiceof ignoringimportant
possible consequences of possible policies, as
well as the values attached to the neglected
consequences. If this appears to disclose a
shocking shortcomingof successive limited
comparisons,it can be replied that, even if
the exclusionsare random,policies maynever-
thelessbe more intelligentlyformulatedthan
throughfutile attemptsto achieve a compre-
hensivenessbeyondhuman capacity.Actually,
however,the exclusions,seemingarbitraryor
random from one point of view, need be
neither.
Achievinga Degree of Comprehensiveness
Suppose that each value neglected by one
policy-makingagency were a major concern
of at least one other agency.In that case, a
This content downloaded from 147.8.204.164 on Mon, 26 Jan 2015 08:30:52 AM
All use subject to JSTOR Terms and Conditions
THROUGH" 85
helpful division of labor would be achieved,
and no agency need find its task beyond its
capacities.The shortcomings of such a system
would be that one agency might destroya
value either before another agency could be
activatedto safeguardit or in spite of another
agency's efforts.But the possibilitythat im-
portantvalues may be lost is presentin any
form of organization, even where agencies
attempt to comprehend in planning more
thanis humanlypossible.
The virtue of such a hypotheticaldivision
of labor is that every importantinterestor
value has its watchdog.And these watchdogs
can protectthe interestsin theirjurisdiction
in twoquite different byredressing
ways: first,
damages done by other agencies; and, second,
by anticipatingand heading offinjurybefore
it occurs.
In a societylike thatof the United Statesin
which individuals are free to combine to
pursue almost any possible common interest
they might have and in which government
agencies are sensitiveto the pressuresof these
groups,the systemdescribedis approximated,
Almosteveryinteresthas its watchdog.With-
out claiming that every interesthas a suffi-
cientlypowerfulwatchdog,it can be argued
that our systemoftencan assure a more com-
prehensiveregardforthe values of the whole
societythan any attemptat intellectualcom-
prehensiveness.
In the United States,for example, no part
of government attempts a comprehensive
overviewof policy on income distribution.A
policy neverthelessevolves,and one respond-
ing to a wide varietyof interests.A processof
mutual adjustmentamong farmgroups,labor
unions, municipalitiesand school boards, tax
authorities,and governmentagencieswith re-
sponsibilitiesin the fieldsof housing,health,
highways,national parks,fire,and police ac-
complishesa distributionof income in which
particularincome problemsneglectedat one
point in the decision processesbecome central
at anotherpoint.
Mutual adjustmentis more pervasivethan
the explicit formsit takes in negotiationbe-
tween groups; it persiststhroughthe mutual
impacts of groups upon each other even
where they are not in communication.For
all the imperfectionsand latent dangers in
thisubiquitous processof mutual adjustment,
it will oftenaccomplishan adaptation of pol-
86 PUBLIC ADMINISTRATION
icies to a wider range of intereststhan could
be done by one group centrally.
Note, too, how the incrementalpatternof
policy-makingfitswith the multiple pressure
pattern. For when decisions are only incre-
mental-closelyrelated to knownpolicies,it is
easier forone group to anticipate the kind of
moves anothermightmake and easier too for
it to make correctionfor injury already ac-
complished.7
Even partisanshipand narrowness,to use
pejorative terms,will sometimesbe assets to
rational decision-making, for theycan doubly
insure that what one agencyneglects,another
will not; theyspecialize personnel to distinct
points of view. The claim is valid that effec-
tive rational coordination of the federal ad-
ministration,if possible to achieve at all,
would require an agreed set of values8-if
rationala" is defined as the practice of the
root method of decision-making.But a high
degree of administrativecoordination occurs
as each agencyadjusts its policies to the con-
cerns of the other agencies in the process of
fragmenteddecision-makingI have just de-
scribed.
For all the apparent shortcomingsof the
incrementalapproach to policy alternatives
with its arbitraryexclusioncoupled with frag-
mentation, when compared to the root
method, the branch method often looks far
superior. In the root method, the inevitable
exclusion of factorsis accidental, unsystem-
atic, and not defensibleby any argumentso
far developed, while in the branch method
the exclusions are deliberate,systematic,and
defensible.Ideally, of course,the root method
does not exclude; in practiceit must.
Nor does the branch method necessarily
neglect long-run considerations and objec-
tives.It is clear thatimportantvalues mustbe
omittedin consideringpolicy,and sometimes
the only way long-runobjectivescan be given
adequate attentionis throughthe neglect of
short-runconsiderations.But the values omit-
ted can be eitherlong-runor short-run.
"The link between the practice of the method of
successive limited comparisons and mutual adjustment
of interests in a highly fragmented decision-making
processadds a new facet to pluralist theoriesof govern-
ment and administration.
"Herbert Simon, Donald W. Smithburg,and Victor
A. Thompson, Public Administration(AlfredA. Knopf,
1950), p. 434.
This content downloaded from 147.8.204.164 on Mon, 26 Jan 2015 08:30:52 AM
All use subject to JSTOR Terms and Conditions
REVIEW
Successionof Comparisons(5b)
elementin thebranch
The finaldistinctive
methodis thatthecomparisons, with
together
thepolicychoice,proceedin a chronological
series.Policyis not madeonce and forall; it
is madeand re-madeendlessly. Policy-making
is a processof successiveapproximationto
somedesiredobjectivesin whichwhatis de-
sireditselfcontinuesto changeunderrecon-
sideration.
Makingpolicyis at besta veryroughproc-
ess. Neithersocial scientists,
nor politicians,
nor public administrators yet know enough
aboutthesocialworldto avoidrepeatederror
in predictingthe consequencesof policy
moves.A wisepolicy-maker consequentlyex-
pectsthathis policieswill achieveonlypart
of whathe hopesand at the same timewill
produceunanticipated consequences
he would
have preferredto avoid. If he proceeds
througha successionof incrementalchanges,
he avoidsseriouslastingmistakesin several
ways.
In the firstplace,past sequencesof policy
stepshave givenhim knowledgeabout the
probable consequencesof furthersimilar
steps.Second,he need notattemptbig jumps
towardhis goals thatwould requirepredic-
tionsbeyondhis or anyoneelse's knowledge,
becausehe neverexpectshis policyto be a
finalresolutionof a problem.His decisionis
only one step, one that if successfulcan
quicklybe followedby another.Third,he is
in effectable to testhis previouspredictions
as he moveson to each further step.Lastly,
he often can remedya past error fairly
quickly-morequickly than if policy pro-
ceeded throughmore distinctsteps widely
spacedin time.
Comparethis comparativeanalysisof in-
cremental changeswiththeaspirationto em-
ploytheoryin therootmethod.Man cannot
thinkwithoutclassifying, withoutsubsuming
one experience undera moregeneralcategory
of experiences.The attemptto pushcategori-
zationas faras possibleand to findgeneral
propositions whichcan be applied to specific
situationsis what I referto with the word
"theory."Where root analysisoften leans
heavilyon theoryin thissense,the branch
methoddoesnot.
The assumptionof root analystsis that
theoryis themostsystematic and economical
wayto bringrelevantknowledgeto bearon a
 THE SCIENCE OF "MUDDLING
specific problem. Granting the assumption,
an unhappy fact is that we do not have ade-
quate theory to apply to problems in any
policy area, although theoryis more adequate
in some areas-monetarypolicy,forexample-
than in others. Comparative analysis, as in
the branch method,is sometimesa systematic
alternativeto theory.
Suppose an administrator must choose
among a small group of policies that differ
only incrementallyfromeach other and from
present policy. He might aspire to "under-
stand" each of the alternatives-forexample,
to know all the consequences of each aspect
of each policy.If so, he would indeed require
theory.In fact,however,he would usually de-
cide that,forpolicy-making purposes,he need
know, as explained above, only the conse-
quences of each of thoseaspectsof the policies
in which theydifferedfromone another.For
thismuch moremodestaspiration,he requires
no theory(although it might be helpful, if
available), forhe can proceed to isolate prob-
able differences by examing the differences in
consequences associated with past differences
in policies,a feasibleprogrambecause he can
take his observationsfroma long sequence of
incrementalchanges.
For example, without a more comprehen-
sive social theoryabout juvenile delinquency
than scholarshave yet produced, one cannot
possiblyunderstandthe ways in which a va-
riety of public policies-say on education,
housing, recreation,employment,race rela-
tions, and policing-might encourage or dis-
courage delinquency.And one needs such an
understandingif he undertakes the compre-
hensiveoverviewof the problemprescribedin
the models of the root method. If, however,
one merelywants to mobilize knowledgesuf-
ficientto assist in a choice among a small
group of similar policies-alternativepolicies
on juvenile court procedures,for example-
he can do so by comparativeanalysis of the
resultsof similarpast policymoves.
Theoristsand Practitioners
This differenceexplains-in some cases at
least-why the administratoroften feels that
the outside expert or academic problem-
solver is sometimesnot helpful and why they
in turn oftenurge more theoryon him. And
it explains why an administratoroften feels
more confidentwhen "flyingby the seat of his
This content downloaded from 147.8.204.164 on Mon, 26 Jan 2015 08:30:52 AM
All use subject to JSTOR Terms and Conditions
THROUGH" 87
pants" than when following the advice of
theorists.Theorists often ask the administra-
tor to go the long way round to the solution
of his problems,in effectask him to follow
the best canons of the scientificmethod,when
the administratorknows that the best avail-
able theorywill work less well than more
modest incremental comparisons. Theorists
do not realize that the administratoris often
in fact practicing a systematicmethod. It
would be foolishto push thisexplanation too
far, for sometimespractical decision-makers
are pursuing neither a theoreticalapproach
nor successivecomparisons,nor any othersys-
tematicmethod.
It may be worthemphasizingthat theoryis
sometimesof extremelylimitedhelpfulnessin
policy-making forat least two ratherdifferent
reasons. It is greedyfor facts; it can be con-
structedonly througha greatcollectionof ob-
servations.And it is typically insufficiently
preciseforapplication to a policyprocessthat
moves throughsmall changes.In contrast,the
comparativemethod both economizeson the
need for factsand directsthe analyst'satten-
tion to just thosefactsthatare relevantto the
finechoicesfacedby thedecision-maker.
With respect to precision of theory,eco-
nomic theoryservesas an example. It predicts
that an economy without money or prices
would in certain specifiedways misallocate
resources,but this findingpertains to an al-
ternativefar removedfromthe kind of poli-
cies on which administratorsneed help. On
the other hand, it is not precise enough to
predict the consequences of policies restrict-
ing business mergers,and this is the kind of
issue on which the administratorsneed help.
Only in relativelyrestrictedareas does eco-
nomic theoryachieve sufficient precisionto go
far in resolvingpolicy questions; its helpful-
ness in policy-making is alwaysso limitedthat
it requires supplementationthroughcompar-
ative analysis.
Successive Comparison as a System
Successivelimited comparisonsis, then, in-
deed a methodor system;it is not a failureof
method for which administratorsought to
apologize. None the less, its imperfections,
which have not been explored in this paper,
are many.For example,the methodis without
a built-in safeguard for all relevant values,
and it also may lead the decision-makerto
88 PUBLIC ADMINISTRATION
overlook excellent policies for no other rea-
son than that they are not suggestedby the
chain of successivepolicy steps leading up to
the present.Hence, it ought to be said that
under this method,as well as under some of
the most sophisticatedvariants of the root
method-operations research, for example-
policies will continue to be as foolishas they
are wise.
Why thenbotherto describethe methodin
all the above detail? Because it is in fact a
common method of policy formulation,and
is, for complex problems,the principal reli-
ance of administratorsas well as of other
policy analysts.9And because it will be su-
perior to any other decision-makingmethod
available for complex problemsin many cir-
cumstances,certainlysuperior to a futile at-
tempt at superhuman comprehensiveness.
The reaction of the public administratorto
the exposition of method doubtless will be
less a discoveryof a new methodthan a better
acquaintance with an old. But by becoming
more conscious of their practice of this
method,administrators mightpracticeit with
more skill and know when to extend or con-
strictits use. (That theysometimespracticeit
effectivelyand sometimesnot may explain the
extremesof opinion on "muddling through,"
whichis both praised as a highlysophisticated
formof problem-solving and denounced as no
method at all. For I suspectthat in so far as
thereis a systemin what is known as "mud-
dling through,"thismethodis it.)
One of the noteworthyincidental conse-
9Elsewhere I have explored this same method of
policy formulation as practiced by academic analysts
of policy ("Policy Analysis," 48 American Economic
Review 298 [June, 1958]). Although it has been here
presented as a method for public administrators,it is
no less necessary to analysts more removed from im-
mediate policy questions, despite their tendencies to
describe their own analytical effortsas though they
were the rational-comprehensivemethod with an espe-
cially heavy use of theory.Similarly,this same method
is inevitably resorted to in personal problem-solving,
where means and ends are sometimes impossible to
separate, where aspirations or objectives undergo con-
stant development,and where drastic simplificationof
the complexityof the real world is urgent if problems
are to be solved in the time that can be given to them.
To an economist accustomed to dealing with the mar-
ginal or incremental concept in market processes, the
central idea in the method is that both evaluation and
empirical analysis are incremental.AccordinglyI have
referredto the method elsewhere as "the incremental
method."
This content downloaded from 147.8.204.164 on Mon, 26 Jan 2015 08:30:52 AM
All use subject to JSTOR Terms and Conditions
REVIEW
quences of clarificationof the method is the
light it throwson the suspicion an adminis-
tratorsometimesentertainsthat a consultant
or adviser is not speaking relevantlyand re-
sponsiblywhen in fact by all ordinaryobjec-
tiveevidencehe is. The troublelies in the fact
that most of us approach policy problems
within a frameworkgiven by our view of a
chain of successivepolicy choices made up to
the present.One's thinkingabout appropriate
policies with respect,say,to urban traffic con-
trol is greatlyinfluencedby one's knowledge
of the incrementalsteps takenup to the pres-
ent. An administratorenjoys an intimate
knowledge of his past sequences that "out-
siders" do not share, and his thinkingand
that of the "outsider" will consequentlybe
different in ways that may puzzle both. Both
may appear to be talking intelligently,yet
each may find the other unsatisfactory. The
relevance of the policy chain of successionis
even more clear when an American tries to
discuss,say, antitrustpolicy with a Swiss,for
the chains of policy in the two countriesare
strikinglydifferentand the two individuals
consequentlyhave organized theirknowledge
in quite different ways.
If thisphenomenonis a barrierto commu-
nication, an understandingof it promisesan
enrichmentof intellectualinteractionin pol-
icy formulation.Once the source of difference
is understood,it will sometimesbe stimulat-
ing foran administratorto seek out a policy
analystwhose recentexperienceis with a pol-
icychain different fromhis own.
This raises again a question only briefly
discussedabove on the meritsof like-minded-
ness amonggovernmentadministrators. While
much of organization theoryargues the vir-
tues of common values and agreed organiza-
tional objectives, for complex problems in
which the root method is inapplicable, agen-
cies will want among theirown personneltwo
typesof diversification:administratorswhose
thinkingis organized by referenceto policy
chainsotherthan thosefamiliarto mostmem-
bers of the organizationand, even more com-
monly,administratorswhose professionalor
personalvalues or interestscreate diversityof
view (perhaps coming fromdifferentspecial-
ties,social classes,geographicalareas) so that,
even within a single agency,decision-making
can be fragmentedand parts of the agency
can serveas watchdogsforotherparts.