Topic: What's Wrong with Rational Choice?
By: Dr. Richard M. Pious, Barnard College and Columbia University
What is "Rational Choice" in Political Science?
Almost every day policy advocates receive advice from experts that is premised on rational choice theory: the idea that people always pursue a pocketbook self-interest, and that political behavior of voters, government officials, and elected politicians can be understood through simplified models of choice. Political scientists and economists who have developed these rational choice models are fast becoming the dominant forces in their academic disciplines. It is fair to ask what, if anything, they can contribute to the toolkit of activists interested in changing public policy.
The rational choice approach was originally developed during and after World War II as part of the defense mobilization. It provides a set of simple rules, which eliminate complex descriptions about political life that supposedly do not help to predict change. Simplification and model building makes rational choice consilient with developments in other social sciences, especially with cost-benefit public choice in economics, game theory in international affairs, and operations research in engineering. It purports to offer the most effective approach to decisionmaking under conditions of risk (when odds are known) and uncertainty (when odds are assumed but not known). Its models are both predictive and prescriptive — in the sense that it presumes that one can and will act rationally in pursuit of one's interests, and that the results can be known in advance. All this sounds, well, rational, but practitioners are actually heir to historical tendencies that are anything but value neutral. The substitution of economic rational choice criteria for traditional political analysis and action goes back in American history to the Progressive era, and it is important to recognize that current modelers are following the old "goo goo" approach favored by the business community, as updated by post-World War II "defense intellectuals" and their domestic "policy analysis" counterparts.
In fact, rational choice models are not consilient with other levels of empirical observation and generalization, and although they purport to provide mathematical rigor, the methodology comes at the expense of what anthropologists call "thick description" of the way societies function. Rational choice models omit — by design — a great deal that traditional political science found essential. They eliminate political culture: the way people think about politics and their customary way of interacting in political institutions. They leave has no room for constitutional and legal norms and constraints, or prescriptions of international law. They offer an a priori, deductive approach, which strikes at the core meaning of a political act: the substitution of influence for other decisional rules, whether they be religious and natural laws, cultural norms, scientific laws, military or paramilitary force, or bureaucratic command. And of course none of the models have any use for the type of "framing" that is studied by experts in rhetoric and communications and is used by FrameWorks practitioners.
Do Rational Choice Models Work?
In the post-World War II period the Pentagon and Rand Corporation used modeling to rationalize nuclear strategy and weapons procurement and deployment. Nuclear strategists used these models in an attempt to avoid Armageddon while still allowing for compellence of smaller nations (i.e., using the threat of force to get them to accommodate to our interests). The approach was used in the early stages of the Cuban Missile Crisis, but when the risks escalated for both sides on October 27th of 1962, it was replaced by secret backchannels and a deal based on old-fashioned political horse-trading (removing missiles in Cuba for missiles in Turkey) that ended the crisis, though the fact of the trade was kept secret for more than three decades. Similarly, Robert McNamara and his Pentagon whiz kids developed an approach to the civil war in Vietnam that they thought would force North Vietnam to end its escalation and negotiate an end to insurgency in South Vietnam. "Every quantifiable indicator tells us we are winning this war," McNamara said at the time. Later, in his memoirs, he laid blame on the North Vietnamese, by arguing that the reason we couldn't get them to the negotiating table was because they weren't "rational." And so we lost. Perhaps then we should look more closely at three rational choice models, which will enable us to see what they are missing — and why they so often miss the boat.
Spatial Positioning Models. These are models designed to specify the most rational positioning an elected official can take along a set of conservative to liberal values, in order to maximize the votes he or she will win in the next election. They are used not only in campaigns, but also to advise the congressional parties how to develop legislative proposals that will play well in elections. The problem with these models is that often multiple games are being played, and some involve values (susceptive to framing) rather than just spatial positions on a liberal-conservative dimension. In the Bush 41 budget summit held with Democratic legislative leaders in the summer of 1990, for example, the Democrats had an advantage because President Bush not only had to play the budget game, but also had to prepare the nation for potential war with Iraq. Though Bush forged a compromise that brought the deficit under control, going back on the pledge "no new taxes" became a valence issue (an issue of deep-seated values) that ruined his credibility with the right wing of his party — an outcome that spatial positioning models could not take into account. In contrast, when President Clinton in December 1995 played a budget game with a very different outcome (he rejected Republican offers of compromise, abandoned the budget process entirely, and forced Republicans to fund the government with continuing resolutions) he gained public support and revived his flagging presidency. In this instance, by abandoning the game entirely, and framing the issue as "the Gingrich who stole Christmas" from furloughed government employees, Clinton found a winning strategy that had little to do with a rational choice approach. Spatial positioning models fail to take into account the new political culture, which is in fact an anti-politics culture in which movement to the center in order to forge a compromise has a negative valence with the electorate, since it is viewed as a "political game" worthy of censure and condemnation — the "Just Politics" frame, in FrameWorks' parlance. Movement toward compromise itself must be effectively framed to avoid such negative judgments by voters. And so must any decision to eschew a compromise and stand on principle: taking a "principled action" does not explain itself to voters, but must be framed for them.
Game Theory Modeling. These are models that attempt to provide maximum payoffs with minimum risk (minimax outcomes) under conditions of risk or uncertainty. Originally developed by mathematician John Von Neumann and economist Oscar Morgenstern, they derived from Von Neumann's desire to win hands in poker games and other games of chance during World War II, and there is little evidence that he ever took them seriously. They later were adapted for deterrence and compellence purposes through work by the Rand Corporation for the Department of Defense. The problem with these models is that games can get out of hand because they are not self-contained. The fallacy involves setting up the game with only two sides (usually there are more), that is symmetrical (usually each side has different resources and advantages), that assumes a great deal of information (usually much of the information a player "knows" is wrong or skewed), that assumes a single game (usually multiple linked and nested games occur simultaneously and sequentially), and that assumes that there are "rules" to the play (in most political games the side that creatively breaks or bends the rules or can establish a new overarching framework that all must adhere to, is more likely to win). The result is that many "angles" as Franklin Roosevelt would have put it (or dimensions, in modeling parlance) are not considered. The mathematics underlying these games are primitive, so in that sense rational choice mathematics is to real math as military music is to music. There are issues of complexity and chaos that produce mathematically absurd results. Even in precisely crafted games, there are impossibilities, particularly in calculating odds, and converting uncertainties into risk. The Prisoner's Dilemma, perhaps the best known of the games, cannot really be used as a model for international negotiations or legislative bargaining. When researchers (I include myself) use variants of the game in classroom simulations, using either trivial or non-trivial rewards (i.e., grades) as incentives, the results are completely out of sync with anything that occurs in the real world.
Cost-Benefit Analysis and the "Core Solution". These are models, derived originally from the utilitarian theorists of the 19th century by way of the Progressives of the 20th, that attempt to derive a "core solution" in public policy, whereby the benefits in taking a decision exceed the costs, so that those incurring costs can be compensated by those receiving benefits, and so that no further change in public policy can provide a better distribution for the winners or compensation for the losers. The problem with these models is that the subjectivities involved in cost-benefit analyses make calculation of a core solution impossible. To begin with, there is "mental accounting," which involves people keeping their resources separate in their minds (i.e., not spending an inheritance, or being willing to play with "the House's" money at a casino, but not risking other funds). And so legislators will propose a Social Security "lockbox" as well as many other separated "accounts," none of which make sense in terms of finding a core solution. There is "prospect theory," which social psychologists and economists have developed to explain why people often value the prospect of losing what one has at a higher value than the prospect of making gains (and sometimes vice versa). Framing is important here, because the placement of a benchmark defines subsequent gains or losses. (For example, you go to a casino and win big on the first day: if the benchmark is the small amount you started with, you will continue to play the entire stake; if you shift your benchmark to the amount you had after the first day, you may be averse to "losing" what you now have and you will stop playing. No rational choice model can make this determination, which is a matter of individual psychology.) And so everyone has their own "core solution," and unless framing is somehow developed persuasively for a large number of voters, there can be no societal consensus about costs and benefits in public policy. Mental accounting and prospect phenomena affected the development of Clinton's health insurance plan of 1993-4, as well as the current Bush administration's Medicare "reforms" of 2003-4. The Clinton plan attempted to use "mental accounting": Clinton proposed to give every individual a "health security card" and talked about "health insurance that can never be taken away," as an echo of Franklin Roosevelt's shrewd understanding that if Social Security recipients were given a card and an individualized account, no Republican majority in the future could get support from the public to dismantle the program. Ultimately, Clinton was unsuccessful because the debate shifted away from health security and focused on a Republican frame: the complexity of the program and the erroneous assumption that patients would not have the right to keep their own doctors. President George W. Bush's success or failure with Medicare reform will be determined by what benchmark the vast majority of Medicare recipients choose: if they focus on the "doughnut" gap in coverage (after a $250 deductible is met, Medicare would pay 75 percent of drug costs up to $2,250, then there would be no coverage for drug costs between $2,251 and $3,600, then most drug expenses would be paid for people enrolled in the plan), prospect theory suggests they will not be happy with the Bush plan. If their prospective calculations are that they spend much more on drugs, they are likely to favor the plan, because when out-of-pocket spending exceeds $3,600, insurance covers 95 percent of drug costs or requires a modest co-payment. The Bush administration frames the plan in terms of catastrophic coverage; opponents frame it in terms of the doughnut -- and prospect theory suggests that each individual will make a subjective calculation about his or her "starting point" under the plan. (It also suggests that people are much more averse to potential loss than they are to possible gains, and so the doughnut could be a big political liability for Bush, even though the additional spending will be significant. Finally, mental accounting can be used to frame these additional expenditures: the administration's accounting is that billions more is being spent on the elderly; the opposition's accounting frames these expenditures as a subsidy for Big Pharmaceuticals. In sum, the subjective factors matter far more than a simple "liberal-conservative" calculation about how much additional money is being put into the system.
Models vs. Thick Description
The models utilized in rational choice studies betray the most limited understanding of politics and policymaking, and they usually don't even understand the processes that they purport to explain. Consider the "veto game" models, which have been developed in the past decade to explore the phenomenon of "veto bargaining." It is well known to empirical researchers that the presidential veto, rather than ending the legislative process, is a prelude to bargaining between president and congressional leaders over the final content of a bill that the president might sign. Modelers create a game with two players: president and Congress, and one kind of play, which involves the threat or use of the veto. The problem with this model is that it assumes that the president has only two choices: veto or allow a bill to pass. But this is an incomplete understanding of presidential choice: the president can sign legislation and issue a "signing statement" which indicates his intention to interpret a controversial provision of law in his own way; he and his subordinates can engage in nonfeasance (dispensing with the execution of the law); he can issue executive orders limiting the reach of a law; his appointees in line agencies can issue memoranda or promulgate regulations to limit the reach of a law; he can delay implementation due to "technical difficulties." He can reinterpret existing statutes and fuse them together creatively in a "mass of legislation" providing him with additional powers, and can then interpret a restrictive provision of a law flexibly. If a president does veto a law, he can still implement the provisions of it he favored by relying on constitutional powers, or appropriations authority in lieu of statutory authority, or the paralegal authority of his own executive orders, so that he can take actions even without statutory authorization. And so a president may sign a bill when he has no intention of honoring some of its provisions, or he may veto a bill when he has every intention of carrying out some of its provisions. This is just one example where the tendency to reduce the complexity of a range of action to a simplified choice (veto or veto bargain) betrays the validity of the model.
Failures of Prediction
Rational choice and game theoreticians have an almost perfect predictive record — of getting it wrong:
- Downsian models (named after the economist Anthony Downs) in the early 1960s predicted that candidates and parties would "rationally" move to the center in formulating public policy and in making rhetorical appeals to the electorate. In retrospect, this seems to be simply an ex post facto explanation for the positioning of the parties in the 1950s and early 1960s. But since the mid-1960s, parties have diverged ideologically far beyond any "rational choice" electoral calculations. There isn't any such model capable of understanding or predicting why the Republicans in the House pursued the Clinton impeachment in 1998, and why Republicans continued to do so in the aftermath of their electoral losses in the midterm election. Nor could a spatial positioning model explain either Bush 43's positioning on most domestic issues — or the Democratic contenders' responses in the early primary season.
- Budget and expenditure studies in the late 1970s predicted that both expenditures and taxation would "rationally" be increased by lawmakers interested in distributing additional benefits to their constituencies. These theories were developed just when President Reagan and Britain's Margaret Thatcher developed a strategy of tax cuts and targeted domestic cuts (to exploit public attitudes toward "undeserving and lazy" recipients). It cannot explain the Bush 41 budget framework for reigning in projected spending increases in 1990, nor the Clinton framework for lessened increases in 1993. Nor does it explain developments involving the hollowing out of the welfare state's social insurance schemes in much of Europe.
- Rational choice models of legislative behavior in the 1980s predicted increases in pork barrel distributive expenditures, because individual members of Congress would find it in their electoral interest to vote for such measures. In the past few years, studies of pork barrel votes have demonstrated that in the 1990s it was more effective in elections to vote AGAINST some pork barrel (and claim credit for cutting government expenditures) than to vote FOR such distributions. In sum, a behavioral and empirical analysis of what legislators are actually doing trumps an a priori model based on "free rider" logic attempting to develop a rule about political behavior that would hold up for all distributions all of the time.
- Rational choice models of voters and public opinion assume that voters, even without perfect information, are likely to make decisions based on rational calculations involved in maximizing their economic interests. This flies in the face of decades of behavioral political science research indicating that voters usually know little about the issues, are not always consistent from year to year in their stance on issues, can hold views on different issues that are incompatible (e.g., in the 1980s large majorities wanted arms control treaties with the Soviets, and large majorities believed the Soviets could not be trusted to adhere to such treaties), and they tend to rely on "clues" for voting to simplify their acquisition of information. It is not always the case that voters will make decisions that are clearly in their own economic interest, or that demonstrate a close congruence between their positions on issues and the candidates whom they are supporting. The 1940s and early-1950s studies conducted at Columbia University by sociologists, which argued that social class was the best predictor of voting behavior, were subsequently supplanted by survey research by political scientists and social psychologists at the University of Michigan, which demonstrated that party identification, group identity, and a sense of "goodness and badness of the times" were much better predictors of the vote. Even when voters cast ballots based on the economy, studies demonstrated that they tended to focus on how the overall economy was perceived, rather than on their individual circumstances. Or they supported a president even when unemployment increased and times got worse — if they believed that the president was trying hard to remedy the situation. Thus, while voting one's self-interest is tautological, when subjected to close analysis, it turns out that voting choice is not closely correlated to individual economic circumstances, but rather to psychological factors highly susceptible to election-year framing.
Toward a Consilient Approach to Social Sciences
Rational choice is based on outdated and inflexible 1950s technology. Let me propose the analogy here: the 1980s attempts to develop a clunky form of artificial intelligence is to 1990s fuzzy logic and distributed neural networking based on complexity theory, as 1960s to 1990s rational choice is to 21st-century approaches to decisionmaking. Rational choice will suffer the same fate as deductive economics, which is being supplanted by behavioral economics, a field that finds psychological or social psychological mechanisms for economic behavior, and then provides mathematical descriptive language for greater precision and prediction. What will the decisionmaking theories of the future contain? Below are some possibilities:
- Identifying the brain chemistry that substitutes power criteria for other factors. Within a few decades we should better understand what triggers this substitution, and we should be able to analyze it much as we have analyzed "fight or flight" mechanisms.
- Matching psychological studies of behavioral economics with behavioral political science to create political counterparts to prospect theory, mental accounting, and other factors that influence calculations of political gains and losses (for example, access to decisionmakers, and other "intangible" rewards of participating in politics.) These will build on the Nobel Prize-winning work of economist Kahneman, himself a refugee from a "rational choice" department of economics, whose influence is now helping to reshape economics from its emphasis on formal deductive models into new approaches based on solving empirical questions of economic behavior.
- Continuing of the "small group" approach to the dysfunctions of small group decisionmaking developed by Alexander George and others, and applying the new knowledge of brain chemistry and social psychology in these arenas.
- Merging the study of public law with behavioral political science, so that we have a better understanding of how case and statute law, particularly framework laws involving war, intelligence, domestic security matters and budgets, are utilized to determine political outcomes.
- Using the mathematics of defining problems to determine those that are solvable and those that are unsolvable in a formal modeling approach. This should reduce the impact of many of the models because they will not be applied to problems that are so complex, or so non-linear, that solutions do not make sense. Indeed most political problems will NOT be susceptible to one or two-dimensional or even three-dimensional mathematical solutions. In that sense they are part of the set of what mathematicians refer to as "non-satisfiable" problems.
What does this critique of rational choice models mean for the work you do and the advice you receive?
First, whenever pollsters or other experts tell you that voters will act in their own self-interest and pursue their pocketbooks, question the data. What suggests this is so? (It is interesting to note that formal models that predict election results based on economic conditions have not had much success, and the political science journals often publish articles in which modelers explain after the election why they got it wrong, or why the model "worked" but some other messy real world factor intervened to throw off the prediction.) In reality, voters are looking for cues in the way the issues are framed and will respond to these framings within a context that has been elaborately developed by the campaigns. So the important point is to look for research that focuses on how the "explanations" a voter develops for the behavior of those running for office will enable them to make a decision about whom to support. Then it is possible to influence the voters' construction of these explanations.
Second, remember that party identification and group identities remain powerful predictors of the vote. Winning candidates and parties often create "reference groups" to enable voters to identify themselves in ways that stir powerful emotions: "taxpayers" "forgotten Americans," "middle Americans," "Patriotic and God-fearing Americans" have been used in past elections to reorient union members and cross-pressure them into switching party identification (particularly in the South) or third-party voting, or simply tuning out and staying home on Election Day. The players who can develop strong new reference groups and relate issues and emotional valences to these groups are usually those who win on Election Day. This worked for the Republican Party in the South, which reoriented regional politics not on the basis of economic self-interest, but on the basis of race, social issues, and religious values, and it has often worked elsewhere for Democrats who can attract upper-income voters on issues other than pure economic self-interest.
Third, remember that, in order to make the models work mathematically, the models used in rational choice deliberately omit a great deal of the description of politics - ideology and political culture, for example — as well as oversimplifying descriptions of the processes of government. Some of the practitioners can compensate for the shortcomings of their models with a lot of common sense and street knowledge; others, as one administrative assistant in a graduate department of political science noted, "couldn't find their way in Washington with a street map." Know which modeler has real world experience and empirically based knowledge to serve as a corrective to the formal model — and hire consultants accordingly.
I hope this note has been helpful in enabling you to think about the limits of so-called rational choice and game theory models, and why their inherent mathematical and logical limitations makes them unsuitable for any explanatory or predictive use in understanding politics.
About the Author
Richard M. Pious is the Adolf S. and Effie E. Ochs Professor of American Studies at Barnard College and a professor of political science at the School of International and Public Affairs at Columbia University, where he has taught courses on presidential and parliamentary political systems, American politics, and comparative government. His books include The American Presidency and The President, Congress and the Constitution. He edited the centennial volume of the Academy of Political Science, The Power to Govern, and a 10-volume series of classic editions in public, comparative, and international law.
January 2004
