American Political Science Review Vol.94,No.2 However such issues of concept formation are ulti- be possible.We cannot predict ahead of time which mately resolved,there are two compelling reasons for of these possible end-states will be reached. focusing special attention on processes that exhibit 2.Inflexibiliry.The farther into the process we are,the increasing returns.First,such processes characterize harder it becomes to shift from one path to another. many important parts of the social world.Second, In applications to technology,a given subsidy to a social scientists are developing theory that makes the particular technique will be more likely to shift the investigation of the causes and consequences of in- ultimate outcome if it occurs early rather than late. creasing returns a particularly promising area of in- Sufficient movement down a particular path may quiry.These two reasons are developed in later sec- eventually lock in one solution.4 tions of this essay. 3.Nonergodicity.Accidental events early in a sequence The basic logic of increasing returns processes can do not cancel out.They cannot be treated (which is be captured in a simple mathematical illustration.? to say,ignored)as "noise,"because they feed back Imagine a very large urn containing two balls,one into future choices.Small events are remembered. black,one red.Remove one ball,and then return it to 4.Potential path inefficiency.In the long-run,the out- the urn,accompanied by an additional ball of the same come that becomes locked in may generate lower color.Repeat this process until the urn fills up.What pay-offs than a forgone alternative would have. can we say about the eventual distribution of colored balls in the urn?Or about a series of trials in which we To this one can add a general point of particular fill the urn and then start over again one hundred interest to social scientists:These are processes in times? which sequencing is critical.Earlier events matter much more than later ones,and hence different se- In each individual trial we have no idea what the quences may produce different outcomes.In these eventual ratio of red to black balls will be;it could be processes,history matters. 99.9%red,or 0.01%red,or anything in between.If If these characteristics are common in politics,then we were to run 100 trials,we would probably get 100 they carry major implications,both for the kinds of different outcomes. questions we should ask and the kinds of answers we In any particular trial,the ratio will eventually reach should expect to find.Most important,they suggest the an equilibrium.Later draws in a series contribute need to focus on the temporal dimensions of social only minutely to the distribution of balls in the urn. processes.In searching for explanation,we need to Thus,the distribution settles down onto a stable think about causes and effects that are often separated outcome. in time,rather than focus exclusively on synchronic Sequence is thus crucial.Early draws in each trial, explanations (Harsanyi 1960;Stinchcombe 1968).Ex- which have a considerable random element,have a ploring these implications requires a review of recent powerful effect on which of the possible equilibria work on increasing returns in economics. will actually emerge. Mathematicians call this a Polya urn process.Its char- INCREASING RETURNS ARGUMENTS IN acteristic qualities stem from the fact that an element ECONOMICS of chance (or accident)is combined with a decision rule that links current probabilities to the outcomes of Economics has traditionally focused on the search for preceding (partly random)sequences.3 Polya urn pro- unique equilibria.5 The goal is attractive because it cesses exhibit increasing returns or positive feedback. suggests a world of potential predictability and effi- Each step along a particular path produces conse- ciency.Given knowledge of existing factor endowments quences which make that path more attractive for the and preferences,equilibrium analysis will point to a next round.As such effects begin to accumulate,they single optimal outcome.Moreover,because econo- generate a powerful virtuous (or vicious)cycle of mists assume a context of decreasing marginal returns, self-reinforcing activity. this goal is potentially achievable.With decreasing Increasing returns processes have quite intriguing returns,economic actions will engender negative feed- characteristics,which Arthur (1994,112-3)has sum- marized as follows 4 This emerging stability represents a critical distinction between increasing returns processes and chaotic processes,which may gen 1.Unpredictability.Because early events have a large erate no equilibrium.For an interesting discussion of this quite effect and are partly random,many outcomes may different framework,with applications to politics,see Fearon 1996. s Some social scientists,such as Stephen I.Gould,also have been drawn to arguments about path dependence,critical junctures,and 2 The following discussion relics heavily on Arthur 1994,a collection punctuated equilibria in evolutionary biology (see especially krasner of his ground-breaking essays on increasing returns and path depen- 1989;Spruyt 1994).Without denying the relevance of this literature, dence. I find it a less useful point of departure than the economists'focus on 3 This case depicts a specific type of increasing returns process,in increasing returns.Most aspects of politics lack anything like the which the probability of a particular draw precisely equals the ratio mechanism of natural selection that drives Darwinian theory (inter- between the two alternatives in the existing population.Arthur national relations and certain characteristics of electoral systems (1994)shows that many of the features of this case have a greater constitute important exceptions).Furthermore,socially created con- range of application,but not all of them.It is easy to model path structs of norms and formal institutions have no real analog in dependent processes with only two equilibria (e.g.,Hill 1997),which evolutionary theory.These constructs,however,are crucial features probably comes closer to capturing the essence of many path of politics and,as we shall see,are a critical element in social dependent processes in the social world. processes subject to increasing returns. 253