Efficiency of the Welfare State

Pierre Pestieau and Claudine Gouyette

Institute for European Studies

January 1998

1. Introduction

In general, when the concepts of efficiency and of welfare state are coupled, one first thinks of the effects of the welfare state, notably including the taxes it implies and the benefits it generates, on the efficiency of the economy. This topic has been widely discussed in recent works. ₁ One of the main charges addressed to modern welfare states is, indeed, that they would hurt economic performance and international competitiveness. Another charge just as widespread is that they would be inefficient in the provision of social services, and be responsible for the proliferation of transfer programs that are costly and miss their target populations. This charge is thus different from the first one, though not totally unrelated. It concerns the economic efficiency of the welfare state per se, and this is the topic of this paper.

Each type of social expenditure has what could be called a target population which it reaches via a more or less lengthy process. To take an example, the fact that 100 million ECU are devoted to social housing does not mean that the natural beneficiaries of this type of program will receive a service equivalent to that amount. At each stage of the process providing such a service, dysfunctionings can appear with the consequence that the amount initially intended for this service is greatly reduced. At the production stage corporatism, self-interest and mere incompetence can lead to the first failures in functioning. Then at the allocation stage, the benefits, if any, may be attributed to population groups which do not really need them, neglecting those for whom they would be vital. This is another type of failure. We have just mentioned two forms of failures in functioning, ''productive inefficiency'' and ''distributive inefficiency.'' The first has been widely studied by production economists and applies to components of the welfare state involving production of services. This is the case of hospitals, schools, day-care centers where services (outputs) are produced out of some human and material resources (inputs.) The second has been studied particularly by economists and sociologists and concerns income transfer programs, particularly means-tested ones. Note however that agencies in charge of these transfer programs, e.g., the social security administration, can also be inefficient when they require excessive administrative costs.

In this paper, we will discuss those types of inefficiency with a particular focus on productive inefficiency. We will present a methodology that is not often applied to the welfare state: the efficiency frontier approach. Also, we will report some unpublished results. Our main purpose in this paper is to show that one can measure inefficiency in the welfare state, and not just talk about it in abstract terms, as is often the case. In so doing, we want to address a number of policy questions. When inefficiency is detected, can it be corrected? Can it be explained? Is it inherent to the public sector? In other words, would privatizing some of the activities of the welfare state, where it is possible, increase their efficiency? Our motivation in the present study arises from a conviction that removing these efficiency slacks could alleviate part of the financial strains that burden most welfare states. Too often we hear unqualified statements as to the inefficiency of government action. These, in turn, can be used to justify hasty dismantlement or privatizing of a specific program or service. We believe that the welfare state, like any accused, is entitled to a due process and particularly to a fair trial. If then, its inefficiency is proved without extenuating evidence, then measures of rolling back or of privatizing can be taken.

The rest of the paper is organized as follows: the next section is devoted to distributive inefficiency that arises in the allocation of social transfers or services. Section 3 deals with the administrative cost of transfer and social insurance programs. Section 4 introduces the efficiency frontier approach. This approach can be applied to a number of specific social services (section 5) or to the whole welfare state (section 6). The final section discusses the causes of inefficiency in social spending and presents a few recommendations.

2. The Matthew Effect

One can speak of distributive inefficiency of social policy when the programs implemented do not help to remove the insecurity of existence from beneficiary households or individuals, whereas there are enough resources to achieve this objective. How can this phenomenon be explained? Simply by the fact that needy households do not exercise their right to benefit from these social protection programs, while other households not suffering from precarious conditions do benefit from them. Furthermore, this inefficiency is not limited to social protection in the strict sense of the term, but also to a number of social facilities and services: study grants, social housing, social assistance, etc. This phenomenon of unequal and inappropriate distribution of the state's social expenditure has been studied under the name of the ''Matthew effect''. ₂ For cultural and institutional reasons, well-off social groups often benefit from social provisions theoretically intended for disadvantaged groups. These are stopped by the administrative complexity and by the fear of stigmatization. Admittedly, the Matthew effect does not explain everything; some new types of poverty and exclusion are not being provided for in the existing social assistance programs.

The Matthew effect has been widely studied in a number of European countries, including Belgium (Deleeck and Berghman (1985)) and the UK (Le Grand (1982)) in the 1970's. It has received less attention over the last decade. This might reflect the fact that today very few people believe that expenditure in social services benefits primarily the less well off. When Peter Townsend concluded his survey on poverty in the U.K.: ''contrary to common belief, fewer individuals in households with low rather than with high incomes receive social services in kind of substantial value.'' (1979, p. 222), it came as a surprise. Today, most social scientists admit that social spending in areas such as education, health, transport, housing, is distributed in favor of the higher social groups.

As Legrand (1982) puts it, there are different kinds of equality pertaining to social services: equality of public expenditure, equality of final income, equality of use, equality of cost and equality of outcome. He then studies whether public spending on health care, education, housing and transport promotes equality in any of its interpretations. Focusing on equality of public expenditure in Britain, he shows that the top socio-economic group receives up to 40 per cent more national health service expenditure per ill person than the bottom group, accounting for difference in age and sex. The top fifth of the income distribution receives nearly three times as much public expenditure in education per household as the poorest fifth. Public expenditure on housing favors the better off, with the highest group receiving nearly twice as much as the lowest. Finally, the richest fifth of the income distribution receives about ten times as much subsidy per household on rail travel, and seventeen times as much on private transport as the poorest fifth.

As another example of the Matthew effect, one can mention the use of subsidized child care in Belgium. In 1992, 28.2% of households with children under 3 used that facility, with a wide dispersion across classes of disposable income. The rate of participation was of 15.5% for the bottom decile, 21.8 for the fifth decile and 47.8 for the top decile. This rather regressive effect still holds even after controlling for working mothers. ₃

This evidence can readily be applied to other countries with strong social policy. It calls however for two caveats. First, as has often be noted, the political survival of the welfare state might require that not only the poor benefit from social policy but also the middle class. In other words, without the Matthew effect, there could be no social program at all. Second, one has to keep in mind that, fortunately, the Matthew effect does not apply to the entirety of the welfare state. There are many redistributive programs that work, and that explain the relatively good performance of the welfare state at fighting poverty and social equalities.

3. Administrative Cost

In welfare and social insurance programs, the productive activity is limited to one of financial intermediation. The question then raised is that of the relative administrative costs associated with public and private intermediation. This has been particularly studied in the field of retirement and health care insurance for which there exist a number of empirical studies. ₄ The current evidence pertains to international cost comparison of social insurance and cost comparison between private and social insurance.

The administrative costs of social insurance vary greatly across countries and institutional settings. Table 1 reveals that countries in the European Union spend about 3.5% of their annual budgets on social security in administrative costs. This ranges from a low 1.22% in Austria to 6.72% in Greece. Part of the explanation for these marked cost differences is that social security systems vary across countries in terms of the particular mix of social assistance and insurance, and in terms of the types of payments offered. It also lies in scale economies. It appears from a multivariate comparison of social security costs in a sample of countries that a one percent increase in participation raises costs by only 0.6-0.9%. The same study shows that a universal demogrant system is significantly less costly to administer than a means-tested or an earnings-related program. ₅

Table 1: Administrative costs as a percentage of social security benefits expenditures in the European Union and the USA

limited to one of

Austria	1.22	Italy	2.2
Belgium	4.55	Luxembourg	2.74
Denmark	2.98	Netherlands	3.10
Finland	3.96	Portugal	4.86
France	4.18	Spain	2.81
Germany	2.86	Sweden	4.24
Greece	6.72	UK	3.10
Ireland	4.88	USA	3.28

Source: Mitchell et al. (1994)

A comparison of publicly managed insurance systems with privately managed alternatives indicates that the latter are in general considerably more costly. In the U.S., it is estimated that roughly 12% of the revenue of the health insurance industry goes for administrative expenses. This percentage is well above the cost of state-managed health insurance in European countries and even in the US. As noted by Diamond (1992), this high percentage is primarily a reflection of returns to scale in transactions including advertising and commissions. It is also, in part, due to adverse selection because of the need to underwrite in details and to the high turnover that characterizes private health insurance. The same cost difference is observed in retirement insurance. It is now clear that the new Chilean private system that many consider as a model is very expensive, a lot more than well-run, unified, government managed social security systems. ₆ Life insurance industry has costs that run 12-14 percent of annual benefits, while social security administration reports costs at less than 1 percent.

It is important to recognize those cost differences. Too many people think of privatisation as a route to greater efficiency and lower costs. Yet, as early as 1942, Beveridge (1942) referred to a ''markedly lower cost of administration in most forms of state insurance.'' In defence of private insurance, a number of observers note that it is likely to offer better and more diverse services in exchange for these higher costs. There would be a trade-off between diversity and transaction costs that requires further study.

In this section the focus was on the efficiency of social insurance programs and their administrative costs. In this particular activity, the index of performance is unidimensional; thus, measuring the efficiency of insurance programs is rather easy. The difficulty arises when there are several kinds of services produced or provided and of resources used. We now turn to this problem.

4. Doing Better With Less

For an economist, an activity is called productively inefficient if the same production of goods and services could be carried out with fewer resources or if, with the resources used, more could be produced. To illustrate, this concept a figure can be helpful. Being restricted to two dimensions, we consider an activity a that produces a service y_a out of an input z_a . Figure 1 represents such an outcome, as well as a curve that represents the productive efficiency frontier. This curve is the yardstick with respect to which the efficiency of a can be assessed. The relative vertical distance between a and this frontier measures what we call the productive inefficiency of a. Note that such a measure can apply to a setting with multiple outputs and inputs, possibly including quality indicators.

Measuring productive efficiency would be a simple exercise were the efficiency frontier known. Unfortunately, this is not the case; the true frontier cannot be merely found in the blueprints of a social engineer. It thus must be inferred from the reality, that is, constructed from a sample of possibly inefficient observations. For example, the dots on Figure 1 represent such a sample.

Figure 1. Productive Efficiency

There are different types of methods which vary from the estimation of a prespecified production function to the construction of a stair-shaped envelope of the input-output points. Whatever the chosen method might be, parametric or not, stochastic or deterministic, one should realize that the productive inefficiency so measured is, of course, a relative one. Hence, one often uses the term ''best practice'' frontier. This frontier is made up of those observations that appear to be the best ones in the sense of some posited assumption of dominance.

For that reason, the choice of the sample of observations is crucial. It is important that they originate from similar conditions as to the technology. To take the case of a cross-section sample of retirement homes, the question of spatial homogeneity is quite relevant. It is not impossible that geographical or institutional differences go a long way towards explaining variations in performance. Part of the efficiency assessment exercise consists in accounting for these differences.

One can see right away the limitations of the efficiency frontier approach for the purpose at hand. First, it only applies to components of the welfare state in which there is production of services: hospitals, social security administration, social work. Second, it has to be concerned with activities involving a large number of productive units since the method is comparative. Note however that in certain instances, intertemporal or international comparisons can enlarge the sample of activities the performance of which is to be assessed. As it appears clearly from Figure 1, productive inefficiency measures waste of resources. Why produce only a when b is attainable?

It is rather clear that productive efficiency should not be the sole objective of social policy makers. There are other objectives for sure: first of all, achieving some redistribution as we have seen, but also fostering employment and growth and keeping within the financial constraints. Some of these objectives are not always compatible. Productive efficiency is, indeed, the only objective whose achievement does not impede that of the others. Producing too few services or employing too many resources, as compared to what is technically feasible, cannot be justified in terms of any of the other objectives that are traditionally assigned to the welfare state.

5. Survey of Productive Efficiency Studies

For a long time, there was no evidence but anecdotal on the efficiency of the welfare state. Fortunately, over the last decade, there has been a number of studies using the productive efficiency approach to measure the efficiency of various sectors of the welfare state. Going back to Figure 1, the procedure is quite standard. We collect data on a sample of activities that can be represented by the dots. Then, we construct a frontier that essentially envelops that data. For each observations, e.g. a, one calculates its degree of productive efficiency: the ratio az_a/bz_a . This ratio would be equal to one if the observation were lying on the frontier. The last stage consists in attempting to explain, when possible, some of the observed efficiency slacks.

Table 2 provides a number of results obtained quite recently. They concern four areas of social spending: health, education, day care and retirement homes.

Table 2: Survey of a number of productive efficiency studies

Areas	Production unit	Studies	Country	Productive efficiency (in %)
Health	Hospitals	Bosmans/Fecher (1995)	Belgium	54 to 78
		Dervaux et alii (1994)	France	75
		Burges/Wilson (1993)	USA	85
		Holvad/Hougard (1993)	Denmark	68 to 78
Social	Day care	Donni (1994)	Belgium	91
		Bjurek et alii (1992)	Sweden	89
	Retirement homes	Boveroux et alii (1995)	Belgium	93 to 96
		Sexton et alii (1993)	USA	69 to 79
		Kooreman (1994)	Netherlands	80 to 94
Education	High schools	Distexhe et alii (1994)	Belgium	77 to 86
		Wickoff/Lavigne (1991)	USA	81 to 86

Those studies reported on Table 2, as well as an increasing number of others ₇ lead to three conclusions: (i) there are serious productive inefficiencies; in other words, more and better could be done with less; (ii) these inefficiencies are not specific to a particular type of organization (non-profit, state-owned, for profit); (iii) competition, autonomy and flexibility would help to reduce a substantial proportion of these inefficiencies.

For a long time, there was a strong belief among economists that public sector managers, unlike their private sector counterparts, do not have to worry about making losses or becoming the victims of takeovers. Hence, they have little incentive to carefully monitor the activities of their enterprises or services. It is now widely accepted that the productive efficiency of a public or private activity depends on the market environment and the institutional setting in which it operates. A privately owned monopoly or regulated firm may produce very inefficient results, while a publicly owned operation that is subject to a lot of competition and to few legal restrictions may produce quite efficiently. This is a theoretical statement that is backed by a number of empirical studies. It implies that to enhance efficiency in the welfare state it is important to open its operation to competition and if needed, to allow private companies to bid along with public ones for contracts to supply particular services.

This conclusion is somehow at odd with that reached about financial intermediation. The reason for this contradiction lies in the underlying technology. Activities such as day care centers or retirement homes quickly reach their optimal scale; in other words, multiplying them is scale efficient and furthermore allows for some form of competition. On the other side, in activities such as social security or unemployment insurance, fixed administrative costs can be spread over a large group of people and their scale economies dominate the potential gains arising from competition.

6. An Aggregate View at Efficiency

The modern welfare state encompasses a variety of activities ranging from pure financial intermediation, such as social security, to the production of services such as day care. For each of these, one can use a specific measure of performance. Yet combining all these performance indicators is not easy, as they do not measure the same thing. As an aggregate measure, we suggest seeing how the resources allocated to the welfare state achieve two of its main objectives, namely, alleviation of poverty and reduction of economic uncertainties. We thus first proceed by looking at the cross-country relation between social spending and either poverty or income disparity. Poverty is measured by the percentage of households with adjusted ₈ income below half the median adjusted income; income inequality is measured by the Gini index. Figures 2 and 3 provide the regression line for the two relations at hand ₉ ; it is pretty clear that social spending exerts a clear-cut effect on both poverty and inequality. The estimates are given on Table 3. A test on the time stability of the estimated coefficients suggests that the impact of social spending on both dependent variables has not changed over time.

Table 3: Impact of social spending on poverty and inequality

Dependent variable	Intercept	Social spending	R2
Poverty rate	7.927	-1.863	68.89%
	(6.493)	(-4.936)
Gini index	4.846	-0.493	52.72%
	(10.638)	(-3.502)

Note: Logarithmic regression; t-statistics in parentheses

We should be cautious in interpreting these relations for a number of reasons. It is clear that reducing social inequalities and fighting poverty are not the only goals of social spending. At the same time, traditional measures of poverty and inequality that are based on the income of the households do not correctly reflect some of the outputs of social protection: health quality, feeling of security. Even more important, it is possible that countries with low poverty rates and low inequality have a strong preference for social protection, or that even before social protection intervenes, they have a rather equal distribution of wages. ₁₀ Despite these qualifications, it is tempting to use the statistical relationship observed between social spending on the one hand, and poverty or inequality on the other to construct an efficiency frontier and to measure the efficiency of various welfare states.

To better grasp our approach, consider two countries that are identical in all respects but for the way their welfare state operates. In the first, social spending is allocated towards the truly poor households or towards those facing unexpected income losses, transfer programs as well as the production of social services are run efficiently in terms of costs and resource utilization. In the second country, the picture is totally different. The Matthew effect prevails, administrative costs are rather high and productive efficiency in social services is low. It is pretty clear that in aggregate terms, efficiency will be greater in the first than in the second country. But this results in part from the assumption of them being identical in all other respects.

Given that the relation between poverty or inequality and relative social spending is negative, the frontier has a negative slope. It is constructed by displacing the regression line ₁₁ downward and parallel so as to envelope all data. The relative distance between this frontier and each country's performance can be viewed as a measure of its performance. Taking the example of France on Figure 2, with the about same amount of social spending, it could have attained a poverty rate of 3% instead of its actual 9%.

Figure 2. The Efficiency of Social Spending at Fighting Poverty

Figure 3. The Efficiency of Social Spending at Fighting Inequalities

The same reasoning could apply to the redistribution of income. On Figure 3, Germany and France spend about the same fraction of their GDP on social protection, and yet in Germany the Gini coefficient is lower than in France (0.235 vs. 0.265).

7. Conclusion

Among the critiques addressed to the welfare state, there is that of inefficiency in distributing benefits and in producing services. This is often followed by a call for rolling back some programs and entrusting others to the private sector, the implicit assumption being that one can do as well with a leaner welfare state, and that the private sector is naturally more efficient. In this paper, we have tried to explicitly tackle the question of definition and measurement of the efficiency of the welfare state.

The welfare state consists of two types of programs: transfer programs such as welfare, social security, unemployment insurance, and production of services such as hospitals, day care and schools. The issue of efficiency in transfer programs can occur in two ways: whether transfers are made to the right people and whether the financial intermediation is cost efficient. The issue of efficiency in the provision of services can be expressed in terms of resource utilization: can we produce the same quantity and quality of services with less resources, or conversely, can we produce more and better services with the same amount of resources?

We have seen that there exist clear inefficiencies in distributing services. Because of administrative complexity or fear of stigmatization, the most needy can fall outside the protection of the welfare state. As far as the administrative cost of social insurance is concerned, particularly health and retirement insurance, the public sector tends to be cheaper than the private sector. Finally, in the production of services, there are clear efficiency slacks but they are independent of ownership, public or private. What seems to matter is competition and autonomy.

The conclusion one can draw from this survey is clear. First, one has to fight the Matthew effect, and if this is not possible, make transfers transparent and universalistic. Second, one has to keep administrative costs at their current level, while at the same time maintain the quality of services provided. Third, one has to introduce competition and autonomy in the management of social services. With an energetic efficiency enhancing approach, the welfare state can recoup its credibility and recover resources that are desperately needed.

In the expression welfare state, one tends to focus on 'welfare' and forget 'state'. An efficient functioning of the welfare state requires that the state represented by the central government plays an important role. Decentralization of decisions can be desirable for productive efficiency reasons but the main objectives of reduction of poverty and inequality are to be entrusted in a central authority to the extent that the relevant concept of poverty and of inequality concerns the national level.

Endnotes

Note 1: For a survey, see Atkinson (1995). See also the work of Linbeck (1995, 1996). Back.

Note 2: The concept of ''Matthew effect'' is widely used by sociologists in different areas. It is called after the ''Parable of Talents'' as told in the Gospel according to Matthew: ''For into every one that hath shall be given, and he shall have abundance; but from him that hath not shall be taken away even that which he hath,'' Matthew 25-29. Back.

Note 3: See Storms (1995). Back.

Note 4: Mitchell et al. (1994), Mitchell (1996). Back.

Note 5: Mitchell (1996). Back.

Note 6: Valdes-Prieto (1993). Back.

Note 7: Pestieau and Tulkens (1994). Back.

Note 8: Adjusted for family size with appropriate equivalence scales. Back.

Note 9: The data used are given in the appendix. Back.

Note 10: See on this Rueda and Pontusson (1997). Back.

Note 11: In logarithms, the regression line is a straight one and its displacement is a parallel one. Back.

References

1) Atkinson A.B., (1995), The welfare state and economic performance, National Tax Journal, vol. 48, 171-198.

2) Beveridge, W. H. (1942), Social Insurance and Allied Services, CMD 6404, London, HMSO.

3) Bjurek, H., U. Kjulinet and B. Gustafson, (1992), Efficiency productivity and determinants of inefficiency at public care centers in Sweden, Working Paper, University of Gothenburg.

4) Bosmans, N. et F. Fecher, (1995), Performance of Belgian hospitals: a frontier approach, Health Economics, (forthcoming).

5) Boveroux, Ph., F. Debrule and H.-J. Gathon, (1995), Efficiency and institutional arrangements in the Belgian nursing home industry, Cahiers de Recherche du CREPP, n^o95/07.

6) Burgess, J.-F. and P.W. Wilson, (1993), Decomposing hospital productivity changes 1985-1988: a non-parametric Malmquist approach, unpublished.

7) Dervaux, B., S., Jacobozone and H. Leleu, (1994), Productive and economic efficiency in French hospitals, unpublished.

8) Diamond, P., (1992), Organizing the health insurance market, Econometrica, 60, 1233-1254.

9) Donni, O., (1994), Efficiency of day care centers in Belgium, CREPP, unpublished.

10) Gathon, H.-J. and P. Pestieau, (1996), Etat providence et efficacité économique, Revue Française d'Economie, 11, 29-44.

11) Holvad, T. and J.-L. Hougaard, (1993), Measuring technical input efficiency for similar production units: 80 Danish ospitals, European University Institute, Working Paper no^o93/36.

12) Kooreman, P., (1994), Nursing home care in The Netherlands: a nonparametric efficiency analysis, Journal of Health Economics, 13, 301-316.

13) Kooreman, P., (1995), Data envelopment analysis and parametric frontier estimation: complementary tools, Journal of Health Economics, 13, 345-346.

14) Lindbeck, A., (1981), World disincentives in the welfare state, reprinted in A. Linbeck, The Welfare State, Edward Elgar, Aldershot, 27-76.

15) Mitchell, O., (1996), Administrative costs in public and private retirement systems, in Privatizing Social Security, M. Feldstein eds., Chicago University Press (forthcoming).

16) Mitchell, O., A. Sunden and Ping-Lung Hsin, (1994), Social security costs in Latin american, the Carribean and OECD nations, Journal of International Compensations and Benefits, 2.

17) Pestieau, P. and H. Tulkens, (1993), Assessing and explaining the performance of public sector activities, FinanzArchiv, 50, 293-323.

18) Rueda, D. and J. Pontusson, (1997), Wage inequality and varieties of capitalism, IES Working Paper no^o97.6, Cornell University.

19) Sexton, T.R., A.M. Leiken, S. Sleeper and A.F. Coburn, (1989), The impact of prospective reimbursement on nursing home efficiency, Medical Care, 27, 154-163.

20) Storms, B., (1995), L'effet Matthieu dans le domaine de l'accueil des enfants, Centrum voor Social Beleid, UFSIA, Antwerpen, unpublished.

21) Townsend, P., (1979), Poverty in the United-Kingdom, Penguin Harmondsworth, London.

22) Van den Bosh, K., (1996), A new social contact? Trends in financial poverty in Western European countries, EUI Working Papers 96/40, Florence.

23) Valdes-Prieto, S., (1993) Administrative costs in the Chilean pension system. Evidence from international comparisons, The World Bank, unpublished.

24) Wyckoff, J.H. and J. Lavigne, (1991), The technical inefficiency of public elementary schools in New York, Working Paper SUNY Albany, N.Y.

Appendix

Country	Year	Poverty Rate (%)	Gini coefficient (%)	Social spending (% of GDP) 1991-1993 mean
Belgium	1992	3.3	20.89	25.7
Denmark	1992	6.3	23.3	31.1
France	1989	8.7	26.5	26.2
Germany	1989	6.2	23.5	26.7
Ireland	1987	8.1	30.9	21.4
Italy	1991	9.8	27.3	23.3
Netderlands	1991	7.2	25.2	30.9
Spain	1990	9.1	28.9	19.7
United Kingdom	1991	10.8	33.5	24.3
Austria	1987	8.6	22.9	24.0
Finland	1991	4.6	23.1	30.6
Sweden	1991	2.5	21.7	37.0
United States	1991	17.2	33.3	14.8

Sources: LIS and OECD