Usefulness of the article: This article focuses on the importance of the normative approach (which seeks to determine what is desirable) and how it complements the positive approach (which seeks to understand what is).It highlights some of the issues and challenges associated with this approach, using the example of an extension of the Medicaid program (public health insurance for low-income individuals) in Oregon in the United States.
Summary:
- Positive analysis, which seeks to understand economic relationships and mechanisms, has made major progress in economics. However, « understanding what is » is often insufficient;
- Assessing whether a change, reform, or economic policy is desirable or not requires a normative approach, which formulates value judgments and evaluates the desirability of changes based on these judgments.
- This type of approach opens up many debates but is necessary to separate questions of fact from those of value, and to make the terms of the debate and the choices to be made more transparent.
- Ultimately, positive and normative approaches are complementary.
Over the past two decades, economic research has seen a significant increase in empirical analysis relative to theoretical work. This has been made possible mainly by improvements in computer tools, which have made it easier to perform sophisticated econometric analyses. At the same time, growing interest in empirical studies has led to improvements in econometric methods and the identification of causal relationships. This development represents a major step forward, as a better understanding of the causal relationships between different variables allows for a more accurate assessment of the potential impacts of different policies.
A prime example of this empirical « revolution » is the advent of « randomized » experiments, in which a change (treatment) is randomly assigned to members of a subset (random or representative) of a population of interest. For example, as discussed below, insurance can be offered randomly to certain individuals in order to assess the effects of having insurance.More specifically, this method allows us to obtain an estimate of the average effect of a treatment without having to rely on prior assumptions or knowledge that may be questionable. This last point is, as noted for example by Deaton and Cartwright (2018), an advantage from the point of view of estimation and internal validity (i.e., it facilitates the measurement of the effect of the treatment on the variables of interest in the specific context of the study) but does not guarantee external validity (i.e., the results cannot necessarily be generalized to other contexts). Another limitation[2] (the one that interests us here) that applies more generally to all studies that seek only to determine a causal relationship or to understand economic mechanisms (so-called « positive » analyses) is that determining the effect of a change on different variables (often) tells us nothing about the desirability of that change. To do this, we need a normative framework that allows us to assess whether, according to certain criteria, this change is desirable or not.
An example of the « ambiguity » of empirical results
To illustrate this point, we will take a recent example from a large-scale « randomized » experiment that took place in Oregon, USA. In 2008, the state of Oregon wanted to expand the Medicaid insurance program (for low-income individuals). However, as demand for this expansion exceeded supply (determined by the available budget), the state of Oregon decided to randomly allocate the remaining places in the program among those who had applied (for more details, see Finkelstein et al., 2012). This expansion and its random allocation made it possible to assess the impact of improved insurance coverage on a range of variables. In principle, having insurance has positive effects (less financial risk, better health, etc.) but can also generate moral hazard (i.e., overuse of certain medical services). The empirical results are consistent with the expected effects (see Finkelstein, Hendren, Luttmer, 2018). The program did indeed have positive effects on the health of beneficiaries, as the risk of depression and health-related financial risks were reduced. There was also an improvement in self-reported health, although neither mortality nor objective measures of health were affected (at least in the short term). At the same time, the program also led to an increase in overall health spending, which may have been due to the overuse of certain services (moral hazard).
As Amy Finkelstein shows (see this video – 32min50s), media coverage of these results has been substantial but has mainly offered diametrically opposed interpretations. For example, the headlines of articles relating to these studies range from « Four reasons why the Oregon Medicaid results are even worse than they look » to « Oregon’s lesson to the nation: Medicaid works . » In light of the same results (from the same positive studies), we therefore have conflicting conclusions about the desirability of the program under study (in other words, opposing normative judgments). This shows the limitation of confining oneself solely to assessing a program’s impact on a set of variables when measuring both desirable and less desirable effects: the assessment of the program’s desirability is then subject to partisan and biased interpretations.
The normative approach
Except in the (rare) case where a change improves the well-being of some without deteriorating that of others (this is referred to as a Pareto increase in well-being), any normative approach requires making assumptions and choices. These assumptions and choices are, of course, debatable. The advantage of conducting research on normative evaluation is that it makes these assumptions and choices clear and transparent, thereby allowing them to be debated. It also allows for a clear distinction to be made between questions of fact and questions of value, whereas a partisan and biased approach will often tend to mix the two, for example by reinterpreting the facts to suit one’s opinions. It also allows for a clear and transparent highlighting of the trade-offs that may have to be made between different dimensions of well-being.
The most common approach in economics to assess the impact of a reform in terms of « well-being, » such as the extension of a public insurance program, is based on individuals’ willingness to pay. In this case, we can, for example, compare individuals’ average willingness to pay for the reform (derived from the insurance offered by the program) with the average cost per individual of the reform (the taxes needed to finance it). In this case, the weight placed on each individual is the same. However, for reasons of fairness or the correction of inequalities, for example, it may be considered that the well-being of some individuals should be given more weight than others (see, for example, Saez and Stantcheva, 2016). The question of « how much weight to give to whom? » is of course not straightforward, but the advantage of this approach is that it makes explicit the choice of weighting individuals and the normative principles on which it is based. It also allows, for a given weighting, reforms with a similar objective to be compared.
Of course, this approach is not without its critics. For example, estimates of willingness to pay are often based on the assumption that individuals are rational. Under this assumption, choices tell us what individuals prefer, and this is what allows us to identify them. The field of behavioral economics has shown that in many situations this rational framework is not a good approximation of reality. This raises the question of how to conduct welfare analyses without falling into paternalism (see Bernheim and Taubinsky, 2018). This inevitably leads to a set of conceptual and methodological difficulties that are nevertheless important to address so that the assessment of the well-being of a particular reform is based on a clear and precise, if not peaceful, debate, where questions of fact and value are clearly separated.
Some applications
Following her work on the expansion of Medicaid in Oregon and the highly contradictory media coverage of its results, Amy Finkelstein, in collaboration with two co-authors (Finkelstein, Hendren, Luttmer, forthcoming), sought to assess the impact in terms of well-being of being covered by the program or not. They were particularly interested in assessing whether the value of the insurance provided by the program (i.e., the value that individuals are willing to pay for it) was greater than its cost. They thus leave aside the question of the program’s redistributive value but discuss how this issue can be addressed within their analytical framework. A major interest of this article is that it considers different approaches that vary according to the assumptions made, which are clearly stated. In general, an approach requiring fewer assumptions requires more information, i.e., it requires knowledge of the program’s impact on a larger number of variables. This type of study highlights the strong link between empirical analysis and normative evaluation. The latter needs the results of the former (estimates of causal relationships) in order to be applied. The credibility of causality studies is therefore a prerequisite for normative analyses, which require a greater number of assumptions but are essential.
Another area of application for normative approaches is cost-effectiveness analysis in health economics (Drummond et al., 2015). A question that often arises for public health system decision-makers is whether or not a treatment or intervention should be reimbursed. This is a complicated decision, particularly because the health budget is allocated to treatments for very different conditions (cancer treatment, a type of prosthesis, etc.). In order to compare the cost-effectiveness of treatments for different conditions, it is useful to have a measure that applies to all of them. In health economics, the most commonly used measure is the number of quality-adjusted life years.[3] We then compare the effect of a treatment using this measure with the additional cost of that treatment.[4] The idea is that if the effect is sufficient in relation to its cost, then the treatment should be reimbursed.[5]
The threshold set depends on normative considerations that must be addressed at the societal level, which are by nature debatable. Similarly, the use of quality-adjusted life years as a measure is debatable and may not take sufficient account of important ethical considerations. Despite these limitations, it remains important to have a systematic approach based on clear principles. This does not mean that a « blunt and harsh » application is required. On the contrary, while it is important to use normative measures based on well-considered principles, it is also necessary to understand their limitations in order to potentially adjust decisions (which can then be based on clear principles of exception). Not basing decisions on any clear and systematic approach means basing important decisions on arbitrary and ill-defined decision-making methods. The normative approach has its limitations, but not using it means refusing to make rational choices, even if they may be difficult.
Conclusion
This article has highlighted the importance of normative analysis in addition to positive analysis in economics. This is as exciting as it is necessary, because it is not enough to understand economic mechanisms. We must also try to understand what is beneficial and what is not. This involves making choices and assumptions, but the advantage of addressing them in academic debate is that it makes them clear and transparent, and separates questions of fact from questions of value.
References
Bernheim B.D. and Taubinsky D., « Behavioral Public Economics, » NBER Working Paper, 2018
Deaton A. and Cartwright N., « Understanding and Misunderstanding Randomized Control Trials, » Social Science and Medicine, 2018
Drummond M.F., Sculpher M.J., Claxton K., Stoddart G.L., Torrance G.W., Methods for the Economic Evaluation of Health Care Programmes, Oxford University Press, 2015
Finkelstein A., Hendren N., Luttmer E.F.P., « The Value of Medicaid: Interpreting Results from the Oregon Health Insurance Experiment, » Journal of Political Economy, forthcoming
Finkelstein A., Taubman S., Wright B., Bernstein M., Gruber J., Newhouse J.P., Allen H., Baicker K., Oregon Health Study Group, « The Oregon Health Insurance Experiment: Evidence from the First Year, » Quarterly Journal of Economics, 2012
Saez E. and Stantcheva S., « Generalized Social Marginal Welfare Weights for Optimal Tax Theory, » American Economic Review, 2016
[1] Economists are particularly interested in causal relationships because, from an economic policy perspective, it is often useful to know what effect a change in economic policy will have on one variable X (e.g., education) through another variable Y (e.g., income). In the data, X and Y may, for example, be correlated simply because they both depend on another variable Z (e.g., greater facilities). If the correlation between X and Y is only the result of the effect of Z on these two variables, exogenously modifying X would have no effect on Y. To simplify things a little, econometrics mainly seeks to identify the part of the correlation between X and Y that is due to a direct effect of X on Y.
[2] The use of the term « limitation » here in no way implies that these studies are not (in many cases) of great value. To our knowledge, all studies, even the best ones, have limitations.
[3] This measure also has the advantage of being a cardinal measure and not just an ordinal one. This means that an increase in the measure from 0.1 to 0.2 is, for example, equivalent in terms of well-being to an increase from 0.8 to 0.9. It should also be noted that, in addition to these positive aspects, a treatment may have side effects. It is important to be able to weigh up these different aspects, which this measure does (with varying degrees of success).
[4] Note that we are simplifying things somewhat here and that the assessment is based on an incremental (or marginal) approach. See, for example, Drummond et al. (2015) for more details.
[5] Note that here, as in the previous example, a credible measure of the effect of the treatment is necessary, hence the importance of measuring this effect accurately.
[6] Note that we are not arguing here for economists or researchers to replace politicians. Ultimately, deciding on questions of values is a matter for democratic debate. However, research can help to make more transparent what belongs to the realm of facts and what belongs to the realm of values, thereby clarifying the terms of the debate and any trade-offs between different dimensions of values.
