23 Pages

Les enseignements théoriques et pratiques des microsimulations en économie de la santé (version anglaise)


Gain access to the library to view online
Learn more


En matière de santé peut-être encore plus que dans d'autres domaines des sciences sociales, le terme de microsimulation est employé pour désigner des modélisations très différentes : application d'un nouveau barème de remboursement à des données de dépenses de santé, agrégation de comportements individuels théoriques en information imparfaite, modélisation des interactions entre environnement socioéconomique, santé et soins, etc. Le type de modèle construit dépend évidemment de l'objectif poursuivi, et donc en général de l'utilisateur présumé du modèle. Pour schématiser, les organismes proches des pouvoirs publics privilégient le calcul de l'incidence d'une réforme sur les dépenses de santé socialisées, les statisticiens utilisent la microsimulation pour mettre en cohérence les données dont ils disposent ou éventuellement pour générer celles qui leur manquent, les épidémiologistes modélisent la survenue d'une maladie et parfois son traitement clinique, les économistes théoriciens s'appuient sur la microsimulation pour lier comportements individuels et agrégats macroéconomiques. Un examen de cette littérature, centrée sur les modèles analysant plutôt la demande de soins, permet de faire ressortir l'intérêt de ces diverses utilisations et la richesse des résultats qui peuvent d'ores et déjà en être dégagés.



Published by
Reads 45
Language English


Theoretical and Practical
Lessons of Microsimulations in
Health Economics*
Breuil Genier** In health matters, perhaps more than any other field of social sciences, the term
microsimulation is used to designate very different simulations, such as the
application of a new reimbursement schedule to health care expenditure, the
aggregation of individual behaviours under imperfect information, the modelling
of interactions within the socio economic environment, health and health care, etc.
Models are built according to their purpose and thus generally to suit the
presumed user of the model. Roughly speaking, government organisations focus
on calculating the impact of reforms on public health care expenditure,
statisticians use microsimulation to explain available data or else to generate
unavailable data, epidemiologists model the occurrence of disease, along with its
clinical treatment in some cases, and theoretical economists use microsimulation to
connect individual behaviours to macroeconomic aggregates.
A literature survey focussing primarily on models that analyse health care
* Originally published as demand reveals how beneficial these various uses are and the wealth of results that
“Les enseignements
théoriques et pratiques are already being produced.
des microsimulations en
économie de la santé,”
Économie et Statisti
que, no. 315, 1998 5.
** Pascale Breuil Genier
is the head of the Health
Economics Office at
the Social Security hree contrasts can be used to rank modelsbehaviour models used for more theoretical
Directorate. T by decreasing order of complexity (Mot,work are based on a stylised description ofThe work presented in
this article is part of a 1992). This is done from the outset to place the reality and on assumptions about consumer
research project funded objectives and characteristics of health related rationality. This type of model has been used to
by a grant from France’s
1microsimulation models into context with study the links between insurance and healthGeneral Planning
Commission. The author regard to other microsimulation work appliedcare with a formalisation of the choice of
would also like to to social policy. The contrasts are between: coverage level to account for potential adverse
express her heartfelt
selection effects (see Box 1) and health carethanks to everyone who
helped her in conducting -exogenous vs. endogenous behaviour : since consumption patterns at given coverage levels
this survey, with special reactions to a change in public policy are being to deal with the issue of moral hazard issue (see
thanks to L. Gatewood
assessed, endogenous behaviour models seem Box 1). These models stress the mechanismsand D. Blanchet.
The names and dates a priori to be preferable. However, their main
in parentheses refer to drawback is that they are more complex and
the bibliography at the
need to be based on assumptions that areend of the article.
1not universally accepted. Most endogenous For a general definition of microsimulation, see Box 5.
INSEE Studies no. 36, June 1999 1and not the macroeconomic results (or -closed ended vs. open ended financing:
forecasts). Microsimulation exercises that are microsimulation models rarely incorporate a
limited to applying spreadsheet methods to closed ended financial equilibrium condition.
exogenous expenditure data are a useful Models applied to health economics are
adjunct for assessing redistributive effects ex open ended since they do not introduce any
ante. element of macroeconomic financial
Box 1
Imperfect information: adverse selection beneficiaries to pay less attention to preventive
and moral hazard behaviour, but this is actually contrary to observed
behaviours (Caussat and Glaude, 1993; Genier
Some markets are characterised by asymmetrical and Jacobzone, 1998; Menahem, 1997). Or else it
information between sellers and buyers. Not would lead them to engage in more r isky
sharing information can lead to opportunistic behaviours, such as practising dangerous sports.
behaviour on the part of one or the other party. However, given the non financial consequences of
Insurance analysis has focused on two of these health problems, this type of behavioural
behaviours in particular: adverse selection and response still seems unlikely.
moral hazard.
The ex post moral hazard in health insurance leads
Adverse selection to greater consumption of health care in case of
illness. In fact, all the empirical research concurs
Adverse selection comes into play when a voluntary that beneficiaries of complementary health
insurance policy is purchased. Individuals who think coverage use more ambulatory care services than
they have a high probability of making a claim non beneficiaries, even though the structure of their
("poor risks") have a greater incentive to take out consumption is quite different.
insurance (Rotschild and Stiglitz, 1976). An
insurance company that offers a contract that has Utility under uncertainty: risk aversion
been designed to fit the average r isk of the
population is therefore likely to incur losses, since Under uncertainty, economic agents’ expected utility
the average risk of its clients is greater than that ofdepends on their ut ility level (Von
the general population. However, empirical findings Neumann Morgenstern utility function) in each of
show that adverse selection effects are relatively the possible states and the probabilities of entering
limited where complementary health coverage is each of these states. For example, if two states are
concerned in France. Even though the links possible depending on whether or not a risk with a
between insurance and state of health usually seem probability p causing loss d occurs, agents’
to be significant, differences between the health of expected utility, which is assumed to depend solely
beneficiaries and non beneficiaries are still small. on income R, can be written as:
The amplitude or these differences, and even their
sign, sometimes vary depending on the health E(U)=p*U(R d)+(1 p)*U(R).
indicators or the populations being considered
(Genier, 1998). Agents are said to be risk neutral if their utility U is
an affine function of its argument r (e.g. U(r)=r). In
From the insurer’s point of view, the existence of this case, expected utility is equal to expected
sub populations with different risk levels may be an income, or R pd. Agents are just as w illing to pay
incentive to select potential clients so that only for insurance as they are to go without coverage, if
"good risks" are covered. Selection techniques may the premium is actuarially fair, which means equal
include medical questionnaires, targeted advertising to the expected loss ( pd), since they have the same
or exclusion clauses in policies. In this case we level of expected utility ( R pd) in both cases.
speak of cherry picking or risk selection.
On the other hand, if U is a concave function,
Moral hazard agents are said to be "risk adverse" and this makes
buying insurance advantageous for them. Insurance
Once a contract has been signed, beneficiaries enables them to reach an expected uility level oft
may, unobserved by the insurer, engage in certain U(R pd). By virtue of the properties of concave
behaviours that are likely to influence the probability functions, this utility is strictly greater than:
that a claim will be made ( ex ante moral hazard) or
to increase the value of claims made ( ex post moral P*U(R d)+(1 p)*U(R).
hazard) (Ehrlich and Becker, 1972). These
behaviours are assumed to reflect the fact that Risk aversion is therefore linked to the concavity of
beneficiaries are less inclined to prevent losses if function U. The an absolute index of risk aversion is
they know they are covered. In the case of health defined (for a point r) as the value of the ratio of the
insurance, ex ante moral hazard would lead derivatives U"/U’ at the point r.
2 INSEE Studies no. 36, June 1999equilibrium, although it sometimes appears inmodels. Some of them can be used for
the form of a financial equilibrium constraint simultaneous simulation of health and health
imposed on insurers. Individuals’ behaviour care.
affects their health spending and thus
influences the premiums paid by all. No claim is made as to the exhaustiveness of
Individuals then take premiums into account our survey of existing microsimulation models.
when making their own decisions relating to We cite a selection here merely to illustrate the
insurance. variety of applications and the contribution that
these methods can make to planning health care
-dynamic vs . static models: in static models, systems. However, we did chose to cite only
aggregated results are obtained by reweighting models dealing with health care demand, use
observations according to benchmark and financing, to the exclusion of models
2information from outside the model. Such dealing with health care supply.
models are often preferred to dynamic models
where changes in the population are simulated
Spreadsheet models endogenously period after period. However,
time is not a key variable in health matters, iand public acceptancen
contrast to pension plan simulations for
example, and static models offer the advantageThe sheer volume of finely detailed data is one
of simplicity. of the strong points of spreadsheet models,
which enables them to simulate economic
In this article, microsimulation is taken to mean policy alternatives with precision. In particular,
any simulation based on microeconomic they are easy to use for simulating fairly
descriptions of individuals in all their diversitycomplex eligibility rules and entitlements to
and the use of this information for elaborating benefits.
and interpreting results. From this point of
view, theoretical models that describe complexOne of the key contributions of this type of
interactions between the demand for insurancemicrosimulation model is to provide an
and that for health care using a simplified assessment of the efficiency of a measure and
representation of the diversity of individuals its acceptance by the public. In fact, all of these
can be considered to be the earliest models lead to the same conclusion: efficiency
microsimulation models (see Box 2). These (in terms of savings for the system) is broadly
older models combine a highly stylised incompatible with public acceptance. Pfaf etf
repres al. (1992) concluded, "if we consider the fact
with a very complex definition of their that half of health expenditure is financed by
behaviour patterns. These models are approximately 10% of individuals, it seems
characteristic of a time when fewer data were very unlikely that anyone can come up with a
available and computers were less powerful. contribution system that is both efficient and
The present survey focuses more on models socially acceptable." Similarly, the model built
using databases to give a more comprehensiveby Fiume Lachaud, Leteno Largeron and
description of the diversity of individuals. Rochaix (1996) shows that deductibles in
Amongst such models, spreadsheet models are health insurance coverage, even if we were to
the simplest. They usually assess individuals’keep the current exemptions under which
expenditure according to a highly detailed deductibles are waived, would be too high to be
classification with an accurate simulation of acceptable.
the rules for reimbursing health care costs or
calculating contributions. These models make
An evaluation of financing reforms init easy to simulate the mechanical effect of a
change in legislation. Other more elaborate Germany’s health insurance system
models simulate behaviour patterns or
introduce macroeconomic financial The German microsimulation model
equilibrium assumptions. There are also Gesetzliche Krankenversicherung (GKVSIM)
models that go further than merely simulating
health care consumption and financing. The
introduction of health modules in
"general purpose" models makes it possible to
See Klein et al.. (1993) for some references to microsimulation
simulate social policy as a whole.
models dealing with the location of medical facilities, choices as
Epidemiologists use other microsimulation to the size of medical teams, frequency of receiving patients, etc.
INSEE Studies no. 36, June 1999 3Box 2
Some the models developed in the 1970s and more accurate estimate of the impact of tax
1980s combined a highly stylised representation of subsidies. Huang, Cartwright and Hu (1989) supply
the diversity of individuals with vast complexity in detailed results according to binary variables
1the description of behaviours. These can be seen relating to income, age, sex or state of health.
as the forerunners of simulation models. Four These choices obviously influence the models used
models dealing with the links between insurance for microsimulation of the health insurance market.
and the use of health care services are presented The types and forms of insurance vary from one
here as illustrations. They all have more or less the model to the next (see Table). The analysis focuses
same structure (see Diagram). They model either on basic insurance or on supplementary
individual reactions to the supply of health insurance. In the latter case, government insurance
insurance, with a particular focus on moral hazard is assumed to exist and to be mandatory. The
and adverse selection behaviour (see Box 1). Then coverage offered by the insurance under
they model insurers’ budgetary constraints so that consideration comes with deductibles or
the relationships between premiums and health co insurance rates, which may also be associated
expenditure can be calculated. Then the model is with a cap on out of pocket expenditure by families.
solved to derive each consumer’s insurance choice The choice of insurance supply model is not
and expenditure. neutral, since consumers react differently
depending on the type and form of coverage they
Nevertheless, implementation of this theoretical enjoy. The expenditure insurers expect to cover and
model varies depending on the authors. Keeler, thus the premiums they set vary greatly between
Newhouse and Phelps (1977) and Keeler, Morrow basic insurance and supplementary insurance. In
and Newhouse (1977) keep to a highly theoretical return, the consumers’ decisions may be different
plane. They calibrate their microsimulations solely depending on whether they are dealing with basic
on the basis of a few parameters, such as risk insurance or supplementary insurance. Regardless
aversion or individual distribution parameters for of premium levels, the incentives for beneficiaries to
health expenditure. The consumer behaviour consume health care services are closely linked to
descriptions by Feldstein and Friedman (1977) and the provisions of their insurance policy. In the case
later by Huang, Cartwright and Hu (1989) are also of basic insurance covering expenditure in excess
theoretical. But they approximate the distribution of of a given annual deductible, Keeler, Newhouse
individual states of health with the health and Phelps (1977) show that the closer their
expenditure curve for the population. total expenditure since the beginning of the year gets to
Microsimulation of the insurance supply
The results are established for several categories of
consumers instead of for an average individual only.
For this purpose, Feldstein and Friedman (1977) 1 Other models dealing with similar concerns are based on a large
divide the population up by income brackets for a number of empirical individual data (Marquis, 1992).
The interactions represented in a standard simulation model of the insurance market
4 INSEE Studies no. 36, June 19993 Box 2 (continued) health care expenditure. The model by Keeler,
Newhouse and Phelps (1977) is the only exception
the annual deductible amount and the more time to this rule. In their model, which is centred on the
they have left till the end of the year the more accumulation of health care expenditure over the
consumers’ incentive to consume increases. This is year, disease is represented by a binomial
because the marginal cost of further consumption is distribution. This means that the intensity and
less than the price actually paid. The difference is probability of onset are assumed to be constant
interpreted as the marginal value of the reduction in from one individual to the next and from one
the sum to be paid before reaching the deductible episode of disease to the next.
and receiving further health care at zero cost.
For a given supply of insurance and a given state of
The price of insurance depends on premiums and health, insurance demand and health care demand
tax subsidies. In most models, premiums are have to be simulated simultaneously. Consumers’
calculated as the actuarial cost of covering risk insurance choices stem from maximising behaviour,
2 4adjusted by two correction factors. The first is making allowances for risk aversion, premium levels,
called the loading rate, which simulates the expected reimbursements and a valuation of the extra
insurers’ administrative costs. It ranges from 1.09 to medical consumption induced by insurance coverage.
1.35. The second factor accounts for the effect of This method was introduced in the two articles by
tax subsidies on the price consumers pay for Keeler and Newhouse (1977), adapted by Feldstein
insurance. This rate ranges between 0.4 and 1. The and Friedman (1977) and later used by Huang,
effect of these two factors is far from negligible in Cartwright and Hu (1989). It is based on many stylised
the United State. In fact, this is the main conclusionassumptions, but it reduces the information
of the article by Keeler, Morrow and Newhouse requirement to a few parameters, such as risk aversion
(1977). The authors conclude that in the case of and the price elasticity of health care demand.
government insurance that reimburses expenditure
above and beyond a deductible, and in view of the The demand for medical sevices depends on ther
loading rates applied by American insurance beneficiary’s state of health. However, moral hazard
companies, supplementary insurance covering part means that it also depends on the level of coverage
of the deductible would be too expensive to attract provided by the beneficiary’s insurance. Insurance
a perceptible demand from consumers unless it was is incorporated via the out of pocket cost of health
subsidised by tax breaks. The conclusions in
France are likely to be different since tax subsidies
are the exception rather than the rule and the
loading rates observed are somewhat lower.
2 The article by Keeler, Newhouse and Phelps (1977) is the
exception. It reconciles supply and demand by taking into account
The weak link: modelling states of health an administrative cost that is poportional to the fraction of ther
population whose expenditure exceeds the deductible under
Modelling states of health is often implicit in health government health insurance.
3economics. In the absence of relevant information, The average of the distribution from which the drawing is made
may vary from one individual to the next (articles co authored byan individual’s health is approximated by its
Keeler, 1977). The distributions used for the estimate areconsequences. Yet, these consequences are
log normal (articles co authored byKeeler , 1977) or gammaprecisely what we are trying to ana lyse: in other
distributions (Feldstein and Friedman, 1977; Huang, Cartwright
words, the individual’s expenditure on health care.
and Hu, 1989).
4 Or from an administrative cost that reflects the excess
In concrete terms, a proxy for the individual’s state consumption of beneficiaries, for Keeler, Newhouse and Phelps
of health is obtained randomly from a distribution of (1977).
Type of insurances considered in the four models
Authors Basic insurance Supplementary insurance
Keeler, Morrow, Newhouse (1977) Annual deductible Partial coverage of the deductible
under government insurance
Keeler, Newhouse, Phelps (1977) Co assurance above -
and beyond a deductible
Feldstein, Friedman (1977) Co assurance with no deductible -
Simple model Two co insurance rates -
Realistic model Two deductibles (ambulatory
care/hospital care)
1Huang, Cartwright, Hu (1989) Medicare Co insurance for expenditure
not covered by Medicare
1.American insurance plan for the elderly
INSEE Studies no. 36, June 1999 5Box 2 (end) health care for the consumer, which means the
marginal cost of added medical consumption with
care after reimbursement, which is the cost regard to exceeding the deductible, and the price
individuals take into account in their maximisation with no insurance.
programmes. Feldstein and Friedman (1977),
followed by Huang, Cartwright and Hu (1989) The models presented above can be seen as the
assume health care demand under constant forerunners of today’s models. Even though they
price elasticity, which they derive from a distribution take the diversity of individuals into account, they
5 . of states of health. Keeler, Morrow and are more concerned with establishing theoretical
Newhouse (1977) assume that health care demand results than with evaluating inequalities within the
is independent of supplementary insurance, but that population.
it varies as a function of the deductible under the
basic government health insurance. In the specific
5 To be more precise, the health care expenditure Q is obtained
case of government health insurance with an nfrom the state of health X° by Q=X°*((C*P+)/(1+ )) , where C is
annual deductible, Keeler, Newhouse and Phelps the co-insurance rate, P is the price with no insurancl e,is the
(1977) assume that health care demand is a non cash cost of health care and n is the price elasticity of health
function of the ratio between the actual cost of care demand.
uses 1981 microdata obtained from a health GKVSIM has also simulated the impact of a
33 insurance plan. measure currently being tested under which The data are
statistically aged to beneficiaries are refunded for the difference if
reflect the current GKVSIM was used to evaluate the impact of the their annual health care expenditure comes to
situation, i.e. 1991, in
reform proposed by the Federal Minister for less than the value of one month’s healththe analysis of the
impact of the 1993 and Health, Seehofer, in 1992 and the reform that was insurance contributions. About one fifth of
1994 reforms in the eventually implemented in 1993 health insurance beneficiaries would receive
study on financing
(Gesundsheits Struturegesetz) (Pfaffet al. , 1992). refunds under the measure. The averagereforms.
These reforms, like most of those that went amount refunded would be quite low at 216
before, favoured action to curb demand. They Deutschemarks per beneficiary, but it would
called for a greater financial contribution fromvary between 249 Deutschemarks for men and
beneficiaries, which was supposed to make themonly 161 Deutschemarks for women. The total
more responsible health care consumers, and forcost would be equivalent to just over 1% of the
a possible refund of contributions to total benefits paid out by the health insurance
beneficiaries whose health care expenditure is system. It could be financed by a 0.15%
low. Compared with the initial proposal put increase in the contribution rate; not counting
forward by Seehofer, the compromise solutioned administrative costs. An unexpected
that was finally adopted limits the responsibilityresult of the simulation is the relatively large
of the insured. Their out-of-pocket costs, whichproportion of pensioners who would be entitled
stood at 6.6 billion Deutschemarks in1991, to such refunds. In addition, the simulation
would have risen by 1.9 billion Deutschemarks underlines the anti redistributive nature of the
under Seehofer’s proposal, instead of the 1.3 measure. Since contributions increase with
billion Deutschemark increase produced by the income, the better off would be more likely to
reform that was finally implemented. Compared receive bigger refunds, even though the positive
with the 1991 simulation, the increase in the correlation between income and healthcare
beneficiaries’ financial contributions for their spending would tend to reduce this effect.
own consumption and that of their dependants
penalises men and the elderly. Simulation can The same model has also been used to simulate
also be used to compare Seehofer’s proposal with the impact of changes in insurance financing
the reform that was finally implemented. (Pfaff et al., 1996). Making government health
Seehofer’s proposal would have led to a big coverage mandatory (it is now optional for
increase in beneficiaries’ contributions for individuals with the highest incomes) would
hospital treatment, with no stay limits and nomake it possible to cut health insurance
exemption for the poorest. This would have hit a contribution rates by between 0.20 points and
4small number of beneficiaries very hard, 3.85 points. In fact, high income earners who
especially pensioners. In its final version, the
reform mainly concerns prescriptions and leads to
a more equitable sharing of costs. Simulation also
shows that exemptions granted to low income
4 The cut in contribution rates varies according to the income
individuals are equivalent to one sixth of taken into consideration. The maximum cut is obtained when
households’ out of pocket health expenditure. contributions are levied on pensions as well.
6 INSEE Studies no. 36, June 1999
llopt for private health coverage account for fact, the simulations are often aimed at
most of the population that is not covered by providing orders of magnitude for the likely
mandatory system and these individuals are impact of certain measures, which are
usually "good risks," i.e. younger men. considered in isolation for analytical purposes,
but would only make sense as part of a broader
Variants raised questions about the assistancepolicy programme.
provided to families through health coverage. A
50% increase in contributions for beneficiaries
A breakdown of health expenditure bywhose non working spouses are covered by
their plan would make it possible to cut the financing sources in the United States
social security contribution rate by 1.2
percentage points. A slightly smaller cut wouldThe American Congressional Budget Office
be possible if non working spouses with (CBO) built a static microsimulation model of
children under a certain age continued receivehealth care expenditure (Atrostic, 1994 and
to free coverage. Financing children’s health 1995). The population base is the Current
care out of general taxation would make it Population Survey (CPS, see Box 3). Data on
possible to cut the contribution rate by 0.88health, insurance premiums, use of health care
percentage points. For the authors, the variantsservices and medical expenditure are
are not policy examples to be implemented. In statistically imputed onto the population base
Box 3
A specific survey of health microsimulation models representing the American
care expenditure population.
The 1987 National Medical Expenditure Survey Rand
(NMES) is a representative survey of 14,000
ordinary households interviewed five times between Rand (Research and Development) was set up as
January 1987 and June 1988. The information part of an American military project and started
relates to the use of the health care system, doing civilian research in about 1950. The Health
expenditure according to financing sources, Science Program was launched in 1968. In 1994, it
insurance coverage, state of health, diseases and employed 70 researchers and an annual budget of
the usual socio economic data. The data reported around 8 million dollars. Financing is provided by
by a large number of households were compared to the federal government jointly with other institutions
data collected from health care providers. When depending on the programmes (Gerbaud, 1995).
they were building the population base for
microsimulation models, some of the organisations The Health Insurance Experiment
were able to compare the data reported by a large
number of households to data obtained from The Health Insurance Experiment (HIE) is the best
employers, thus collecting supplementary known of Rand’s health policy projects. It was an
information about individuals’ health coverage. This experiment conducted between 1975 and 1982 on a
provided a more detailed description of coverage representative sample of ordinary households that
and premiums, along with further information about were not covered by government health insurance
the employer, such as the proportion of employees residing in six locations. One of fourteen different
with health coverage. In other models, these data health insurance policies was attributed randomly to
were imputed statistically on the basis of a separate each of the households in the sample, so as to
survey of employers. control for adverse selection effects (see Box 1).
The policies, which offered reimbursement of
The Current Population Survey fee for service payments, had different
co insurance rates and different annual caps on
The Current Population Survey is an annual survey each family’s out of pocket expenditure. The
in America collecting all of the usual families’ medical consumption was tracked
socio demographic data from a sample of about throughout the experiment. One of the objectives
150,000 households. This survey also contains was to define optimal health coverage. The
information about insurance coverage in case of structure of the experiment made it possible to
illness and about employment. These data, overcome adverse selection problems and the lack
combined with demographic information, are often of diversity in existing health insurance policies,
used to build the population base for which makes comparative analysis less accurate.
INSEE Studies no. 36, June 1999 7from the National Medical Expenditure Surveyby social security beneficiaries. Their work
(NMES, see Box 3), and completed by a final used the Credes Health and Welfare Survey
comparison with health insurance data (enquête Santé et Protection Sociale (ESPS)) in
provided by employers. Taxes and a continuation of a European project on
contributions are calculated using the CBO taxinternational comparisons of equity in health
model. Macroeconomic aggregates are used tosystem reforms (COMAC HSR project). It
calibrate the model. provides original and interesting results and
helps inform the ongoing political debate on the
A unique feature of this model is a breakdown merits of deductibles.
of health expenditure according to final
financing sources, in which employers’ The deductible amount is adjusted in each
contributions are deemed to be financed byscenario to produce a savings of 6 billion
households. The model builders took 1988 francs, which is equivalent to a
particular care with validation, following the 0.2 percentage point cut in the contribution
recommendations made in the Committee onrate. This is the savings that is usually expected
National Statistics report (Citro and in the various consolidation plans. Simulations
Hanushek, 1991). They tested the validity of shown that it is unrealistic to consider offering
their methods for statistical imputation of free health coverage over the deductible
expenditure and insurance premiums by amount. This would mean imposing annual
applying them to the National Medical deductibles of about 5,000 francs on each
Expenditure Survey and comparing the family of beneficiaries and it would rule out
distribution of the imputed values to the any reimbursement of expenditure for more
observed values. They also tested how the than half the families, since their non zero
choice of mathematical linkage between health expenditure does not exceed the
simulated and observed variables influenceddeductible amount. Maintaining current
results. This influence of this choice appearsco insurance rates for expenditure over the
to be significant, since it determines whetherdeductible amount makes it possible to bring
or not insurance premiums and health this amount down to approximately 950 francs
expenditure are determined simultaneously per year, to which we must add approximately
and how each is imputed to the other. 175 francs if we wish to maintain certain
exemptions below the deductible amount. Such
The Health Care Financing Administration a system of deductibles would rule out
built a similar model, which is also based on reimbursements for 20% of the families
the Current Population Survey. It imputes covered who made some payment for health
medical expenditure data from the Nationalcare. The deductibles per individual, instead of
Medical Expenditure Survey and aligns them per family, would work out to approximately
on health care accounts. This model shows500 francs per year. However, introducing the
each individual’s health care spending same deductible amount for everyone would
broken down into 13 items along with their impose a proportionally greater burden on the
decomposition by financing source. This less well off. This regressive effect could be
model was used to evaluate the proposal forattenuated by making the deductibles
Medical Savings Accounts combined with proportional to income. Thus, to produce a
insurance for major expenses. If we take 6 billion franc saving, the deductible would
account of risk adverse individuals’ costs have to be equivalent to 0.7% of income. The
when a risk is not insured collectively (seeinherent drawbacks of deductibles, such as
Box 1), we see that the introduction of higher administrative costs, penalisation
Medical Savings Accounts does not seem of preventive care, the likelihood that
very beneficial (Actuarial Research expenditure will be increased to exceed the
Corporation, 1995a and b). deductible limit and the unacceptable exclusion
of households from reimbursement, diminish
the attraction of this dice for making ev people
Deductibles do not seem to be more responsible. The authors plan to simulate
the right solution in France. other reforms, such as income related
out-of-pocket fees for users. They also plan to
Fiume Lachaud, Leteno Largeron and Rochaix incorporate behavioural responses into their
(1996) simulated the effects that introducing model, even though this theoretically satisfying
deductibles in France would have on the refinement will only have a limited impact on
distribution of out of pocket health expenditure findings.
8 INSEE Studies no. 36, June 1999deductible limit has been exceeded. In factRand models combine
consumption above and beyond deductible
realism and sophistication.
limits is still lower than consumption by people
with 100% coverage all year round and
everal models attempt to combine the expectations of exceeding the deductible limitS endogenous effects of behavioural have an even smaller effect. The construction of
responses introduced in the earliest models the microsimulation model was guided by the
with the detail of data and the rules applied in findings of the Health Insurance Experiment
spreadsheet models. The best known attempts (Keeler, Buchanan, et al., 1988; Buchanan et al.,
were based on the results of the Rand experiment 1991). In particular, the model makes use of
(see Box 3). This experiment gave rise to atthe fact that the main influence of health
least two interlinking simulation models coverage seems to be on the decision to start
(Marquis and Buchanan, 1992). a health service episode and not on the cost of
the episode (see Box 4). Individuals will
undertake all potential health service episodes if
Two models. . . costs are fully reimbursed, whereas they will
choose only some of the episodes at random if
The Health Insurance Demand Model (HIDM) costs are only partially covered. This model was
simulates households’ health insurance choices also used to analyse mental health service
between group plans, plans available through episodes (Keeler, et al. , 1988).
employers, individual plans or no coverage at
all. The choice of plans offers different
deductibles, co insurance rates over the . . .that can be interlinked
deductible amount and maximum annual
amounts beneficiaries must cover out of theirInterlinking these two models makes it
own pocket. The choice of insurance results possible to calculate the insurance premiums
from the households’ maximisation of required according to the type of insurance,
expected utility. Utility depends solely on the individual beneficiary and the number of
disposable income (excluding insurance persons included in the coverage. The
premiums and health care expenditure) and premiums are found from the average health
risk aversion, which has been benchmarkedexpenditure on the beneficiaries, as
using data from the Health Insurance calculated using the Health Episode
Experiment; see Box 3). Expected utility isSimulation Model, then multiplied by a
calculated with regard to the distribution ofloading factor that depends on the type of
the household’s anticipated health care insurance and on the size of the company in
expenditure. In concrete terms, this the case of group policies. The premiums thus
distribution is estimated by fifty simulationsfound can then be incorporated into the
of health care expenditure using the HealthHealth Insurance Demand Model.
Episode Simulation Model (see below).
Then, the family’s expected utility for each Marquis (1992) built an early version of the
insurance choice is modified by introducing a Health Insurance Demand Model. It shows
random term to account for the diversity ofthat the source of financing is a decisive
households’ behavioural responses. factor for analysing adverse selection
phenomena. The adverse selection effects are
The other model is the Health Episode substantial in the case of basic optional
Simulation Model (HESM), which simulates coverage with a deductible, whereas they are
households’ health care expenditure. The weak in the case of supplementary insurance
primary purpose of the model is to analyse the to cover some or all of the deductible that is
impact of annual deductibles (meaning the not covered by a mandatory government
maximum expenditure that households have to health plan. In fact, supplementary insurance
cover out of their own pockets) on behaviouris under. -priced, since it does not take into
Does consumption increase once the deductibleaccount the impact of the coverage provided
limit has been breached? Does present on the expected expenditure above and
behaviour change when beneficiaries expect to beyond the deductible. Moral hazard is also
exceed the year’s deductible amount? taken into account somewhat arbitrarily. The
Empirical findings have not confirmed these demand from beneficiaries with incomplete
theoretically intuitive suppositions. The coverage is estimated to be half, or 0.55 to be
increase in consumption is very small once the precise, that of fully covered beneficiaries.
INSEE Studies no. 36, June 1999 9contributions towards financing insurance forThe cost of extending health insurance
their employees are not counted in thecoverage in the United States
employees’ taxable income. This results in a de
Reforms have been proposed to extend health facto tax break on the purchase of insurance.
insurance coverage in the United States, where This tax break is proportional to the employers’
some 15% of the population still have no coverage. contributions and the price of the insurance.
One way to achieve this aim is to subsidise theTherefore, the greatest tax savings go to the
purchase of insurance by allowing individuals toemployees with the highest income, which
deduct their share of insurance premiums from their provides a further argument in favour of doing
taxable income. Marquis and Buchanan (1992) useaway with the tax break. However, the main
the Health Insurance Demand and Health Episodebenefit expected from such a reform is to curb
Simulation models to show that such a tax breahealth care sk pending. As consumers will have
would only have a limited impact. In fact, it would to bear the full cost of their insurance, they will
only increase health insurance demand from be incited to take out less costly policies that
individuals with no coverage by 2 percentage offer less coverage. This will promote cost
points. Other work by Rand investigates the cutting by encouraging competition between
possibilities for extending health coverage through insurers and help to curb consumption.
employer health plans to small businesses, either by
creating bargaining alliances of several small firmsThe model assesses the impact of limiting and
or by banning premium differentials based on firms’ then abolishing the tax free status of insurance
specific characteristics. Rand then assessed the premiums that employers pay for their
respective advantages of State by State reformsemployees. It simulates the employer’s
and reform at federal level. decision to provide group policies for its
employees using two distinct processes
Other proposed reforms aim to curb health care depending on the size of the firm. Large firms
expenditure in America. The two models bargain with trade unions, which defend their
described above were linked with a model members’ interests. In the end, the decision on
simulating firms’ decisions to provide health whether or not to offer health coverage depends
coverage and then used to assess the impact of ona the preferences of the average union
reduction of tax subsidies for insurance member. On the other hand, small firms’
(Marquis and Buchanan, 1994). Employers’ decisions on whether or not to offer health
Box 4
The Health Episode Simulation Model was about future expenditure and the probability of
developed largely from the data collected in the exceeding their deductible. It is postulated that
Health Insurance Experiment. The model each individual has a propensity to experience each
distinguishes between several types of health type of episode. The propensity depends on the
episodes: dental care, hospital care, ambulatory known family characteristics (socio demographic
care for acute diseases, ambulatory for chronic structure, income, health) and hidden
diseases and "well care" (Keesey et al., 1985). characteristics (individual or family effect). The
Health episodes are defined using information succession of potential episodes is then simulated
supplied by practitioners, who fill in forms indicating on the basis of earlier propensities using Pissono
which diseases and which episodes are related to processes. The implicit costs of each episode are
their acts. Where this is not possible, the health drawn from a log normal distribution where the
episode is defined using an algorithm. For example, average varies depending on the type of episode
in the case of acute care, doctor’s visits are less and the characteristics of the individual concerned.
than 16 days apart and either deal with the same In the first stage, the simulated episodes do not
diagnosis, or else involve the same doctor. In take account of the quality of insurance coverage.
Rand’s approach, one visit can correspond to The decision to actually use health care services on
several health episodes, in which case the cost is an episode by episode basis is only simulated after
spread between the various episodes. All of the a succession of potential episodes for the whole
health care expenditure relating to an episode is year has been generated. The model derives usage
counted on the date at which it first becomes under partial insurance coverage from usage under
foreseeable. For example, maternity expenditure is full insurance coverage, which makes it possible to
all counted on the date of the first antenatal visit. reduce simulation variance by not introducing a
This enables the model to analyse changes in random variable stemming from the simulation of
individuals’ behaviour according to the expectations morbidity.
10 INSEE Studies no. 36, June 1999