Bayesian Tutorial

a primer on
BAYESIAN STATISTICS
in Health Economics and
Outcomes Research
BS
CH
B AYESIAN I NITIATIVE IN H EALTH E CONOMICS E
& O UTCOMES R ESEARCH Centre for Bayesian Statistics
in Health Economics
A Primer on Bayesian Statistics Luce O’Hagan
a primer on
BAYESIAN STATISTICS
in Health Economics and
Outcomes Research
Anthony O’Hagan, Ph.D. Bryan R. Luce, Ph.D.
Centre for Bayesian Statistics MEDTAP ® International, Inc.
in Health Economics Bethesda, MD
Sheffield Leonard Davis Institute,
United Kingdom University of Pennsylvania
United States
With a Preface by
Dennis G. Fryback
Bayesian Initiative in Health Economics & Outcomes Research

Centre for Bayesian Statistics in Health Economics
Bayesian Initiative in Health Economics & Outcomes Research
(“The Bayesian Initiative”)
The objective of the Bayesian Initiative in Health Economics & Outcomes
Research (“The Bayesian Initiative”) is to explore the extent to which formal
Bayesian statistical analysis can and should be incorporated into the field
of health economics and outcomes research for the purpose of assisting rational
health care decision-making. The Bayesian Initiative is organized by scientific
staff at MEDTAP® International, Inc., a firm specializing in health and economics
outcomes research. www.bayesian-initiative.com.
The Centre for Bayesian Statistics in Health Economics (CHEBS)

The Centre for Bayesian Statistics in Health Economics (CHEBS) is a research
centre of the University of Sheffield. It was created in 2001 as a collaborative ini-
tiative of the Department of Probability and Statistics and the School of Health
and Related Research (ScHARR). It combines the outstanding strengths of these
two departments into a uniquely powerful research enterprise. The Department
of Probability and Statistics is internationally respected for its research in
Bayesian statistics, while ScHARR is one of the leading UK centers for economic
evaluation. CHEBS is supported by donations from Merck and AstraZeneca, and
by competitively-awarded research grants and contracts from NICE and research
funding agencies.
Copyright ® 2003 MEDTAP International, Inc.
All rights reserved. No part of this book may be reproduced in any form,
or by any electronic or mechanical means, without permission in
writing from the publisher.
ii A Primer on Bayesian Statistics in Health Economics and Outcomes Research

Table of Contents
Acknowledgements..............................................................iv
Preface and Brief History ......................................................1
Overview ..............................................................................9
Section 1: Inference ............................................................13
Section 2: The Bayesian Method ........................................19
Section 3: Prior Information ..............................................23
Section 4: Prior Specification..............................................27
Section 5: Computation ......................................................31
Section 6: Design and Analysis of Trials ..........................35
Section 7: Economic Models ..............................................39
Conclusions ........................................................................42
Bibliography and Further Reading ....................................43
Appendix ............................................................................47
A Primer on Bayesian Statistics in Health Economics and Outcomes R e s e a r c h iii

Acknowledgements
We would like to gratefully acknowledge the Health Economics Advisory

Group of the International Federation of Pharmaceutical Manufacturers
Associations (IFPMA), under the leadership of Adrian Towse, for their
members’ intellectual and financial support. In addition, we would like to
thank the following individuals for their helpful comments on early drafts
of the Primer: Lou Garrison, Chris Hollenbeak, Ya Chen (Tina) Shih,
Christopher McCabe, John Stevens and Dennis Fryback.
The project was sponsored by Amgen, Bayer, Aventis, GlaxoSmithKline,
Merck & Co., AstraZeneca, Pfizer, Johnson & Johnson, Novartis AG, and
Roche Pharmaceuticals.
iv A Primer on Bayesian Statistics in Health Economics and Outcomes Research

Preface and
Brief History
L et me begin by saying that I was trained as a Bayesian in the

1970s and drifted away because we could not do the computa-
tions that made so much sense to do. Two decades later, in the 1990s,
I found the Bayesians had made tremendous headway with Markov
chain Monte Carlo (MCMC) computational methods, and at long last
there was software available. Since then I’ve been excited about once
again picking up the Bayesian tools and joining a vibrant and growing
worldwide community of Bayesians making great headway on real life
problems.
In regard to the tone of the Primer, to certain readers it may sound
a bit strident – especially to those steeped in classical/frequentist sta-
tistics. This is the legacy of a very old debate and tends to surface when
advocates of Bayesian statistics once again have the opportunity to
present their views. Bayesians have felt for a very long time that the
mathematics of probability and inference are clearly in their favor,
only to be ignored by “mainstream” statistics. Naturally, this smarts a
bit. However, times are changing and today we observe the beginnings
of a convergence, with frequentists finding merit in the Bayesian goals
and methods and Bayesians finding computational techniques that
now allow us the opportunity to connect the methods with the
demands of practical science.
Communicating the Bayesian view can be a frustrating task since
A Primer on Bayesian Statistics in Health Economics and Outcomes R e s e a r c h 1

we believe that current practices are logically flawed, yet taught and taken
as gospel by many. In truth, there is equal frustration among some fre-
quentists who are convinced Bayesians are opening science to vagaries of
subjectivity. Curiously, although the debate rages, there is no dispute about
the correctness of the mathematics. The fundamental disagreement is
about a single definition from which everything else flows.
Why does the age-old debate evoke such passions? In 1925, writing in
the relatively new journal, Biometrika, Egon Pearson noted:
Both the supporters and detractors of what has been termed Bayes’
Theorem have relied almost entirely on the logic of their argument;
this has been so from the time when Price, communicating Bayes’
notes to the Royal Society [in 1763], first dwelt on the definite rule by
which a man fresh to this world ought to regulate his expectation of
succeeding sunrises, up to recent days when Keynes [A Treatise on
Probability, 1921] has argued that it is almost discreditable to base any
reliance on so foolish a theorem. [Pearson (1925), p. 388]
It is notable that Pearson, who is later identified mainly with the fre-
quentist school, particularly the Neyman-Pearson lemma, supports the
Bayesian method’s veracity in this paper.
An accessible overview of Bayesian philosophy and methods, often
cited as a classic, is the review by Edwards, Lindman, and Savage (1963).
It is worthwhile to quote their recounting of history:
Bayes’ theorem is a simple and fundamental fact about probability that
seems to have been clear to Thomas Bayes when he wrote his famous
article ... , though he did not state it there explicitly. Bayesian statistics is
so named for the rather inadequate reason that it has many more occa-
sions to apply Bayes’ theorem than classical statistics has. Thus from a
very broad point of view, Bayesian statistics date back to at least 1763.
From a stricter point of view, Bayesian statistics might properly be said

to have begun in 1959 with the publication of Probability and Statistics
2 A Primer on Bayesian Statistics in Health Economics and Outcomes Research

for Business Decisions, by Robert Schlaiffer. This introductory text pre-
sented for the first time practical implementation of the key ideas of
Bayesian statistics: that probability is orderly opinion, and that infer-
ence from data is nothing other than the revision of such opinion in
the light of relevant new information. [Edwards, Lindman, Savage
(1963) pp 519-520]
This passage has two important ideas. The first concerns the definition
of “probability”. The second is that although the ideas behind Bayesian sta-
tistics are in the foundations of statistics as a science, Bayesian statistics
came of age to facilitate decision-making.
Probability is the mathematics used to describe uncertainty. The dom-
inant view of statistics today, termed in this Primer the “frequentist” view,
defines the probability of an event as the limit of the relative frequency
with which it occurs in series of suitably relevant observations in which it
could occur; notably, this series may be entirely hypothetical. To the fre-
quentist, the locus of the uncertainty is in the events. Strictly speaking, a
frequentist only attempts to quantify “the probability of an event” as a
characteristic of a set of similar events, which are at least in principle
repeatable copies. A Bayesian regards each event as unique, one which
will or will not occur. The Bayesian says the probability of the event is a
number used to indicate the opinion of a relevant observer concerning
whether the event will or will not occur on a particular observation. To the
Bayesian, the locus of the uncertainty described by the probability is in the
observer. So a Bayesian is perfectly willing to talk about the probability of
a unique event. Serious readers can find a full mathematical and philo-
sophical treatment of the various conceptions of probability in Kyburg &
Smokler (1964).
It is unfortunate that these two definitions have come to be character-
ized by labels with surplus meaning. Frequentists talk about their proba-
bilities as being “objective”; Bayesian probabilities are termed “subjective”.
Because of the surplus meaning invested in these labels, they are perceived
to be polar opposites. Subjectivity is thought to be an undesirable proper-
Preface and Brief History 3

ty for a scientific process, and connotes arbitrariness and bias. The fre-
quentist methods are said to be objective, therefore, thought not to be con-
taminated by arbitrariness, and thus more suitable for scientific and arm’s-
length inquiries.
Neither of these extremes characterizes either view very well. Sadly,
the confusion brought by the labels has stirred unnecessary passions on
both sides for nearly a century.
In the Bayesian view, there may be as many different probabilities of
an event as there are observers. In a very fundamental sense this is why
we have horse races. This multiplicity is unsettling to the frequentist,
whose worldview dictates a unique probability tied to each event by (in
principle) longrun repeated sampling. But the subjective view of probabil-
ity does not mean that probability is arbitrary. Edwards, et al., have a very
important adjective modifying “opinion”: orderly. The subjective probabili-
ty of the Bayesian must be orderly in the specific sense that it follows all of
the mathematical laws of probability calculation, and in particular it must
be revised in light of new data in a very specific fashion dictated by Bayes’
theorem. The theorem, tying together the two views of probability, states
that in the circumstance that we have a long-run series of relevant obser-
vations of an event’s occurrences and non-occurrences, no matter how
spread out the opinions of multiple Bayesian observers are at the begin-
ning of the series, they will update their opinions as each new observation
is collected. After many observations their opinions will converge on near-
ly the same numerical value for the probability. Furthermore, since this is
an event for which we can define a long-run sequence of observations, a
lemma to the theorem says that the numerical value upon which they will
converge in the limit is exactly the long-run relative frequency!
Thus, where there are plentiful observations, the Bayesian and the fre-
quentist will tend to converge in the probabilities they assign to events. So
what is the problem?
There are two. First, there are events—one might even say that most
events of interest for real world decisions—for which we do not have

ample relevant data in just one experiment. In these cases, both Bayesians
and frequentists will have to make subjective judgments about which data
to pool and which not to pool. The Bayesian will tend to be inclusive, but
weight data in the pooled analysis according to its perceived relevance to
the estimate at hand. Different Bayesians may end at different probability
estimates because they start from quite different prior opinions and the
data do not outweigh the priors, and/or they may weight the pooled data
differently because they judge the relevance differently. Frequentists will
decide, subjectively since there are no purely objective criteria for “rele-
vance”, which data are considered relevant and which are not and pool
those deemed relevant with full weight given to included data.
Frequentists who disagree about relevance of different pre-existing
datasets will also disagree on the final probabilities they estimate for the
events of interest. An outstanding example of this happened in 2002 in the
high profile dispute over whether screening mammography decreases
breast cancer mortality. That dispute is still is not settled.
The second problem is that Bayesians and frequentists disagree to
what events it is appropriate and meaningful to assign probabilities.
Bayesians compute the probability of a specific hypothesis given the
observed data. Edwards, et al., start counting the Bayesian era from publi-
cation of a book about using statistics to make business decisions; the rea-
son for this is that the probability that a particular event will obtain (or
hypothesis is true), given the data, is exactly what is needed for making
decisions that depend on that event (or hypothesis). Unfortunately, with-
in the mathematics of probability this particular probability cannot be
computed without reference to some prior probability of the event before
the data were collected. And, including a prior probability brings in the
topic of subjectivity of probability.
To avoid this dilemma, frequentists—particularly RA Fisher, J Neyman
and E Pearson—worked to describe the strength of the evidence inde-
pendent of the prior probabilities of hypotheses. Fisher invented the P-
value, and Neyman and Pearson invented testing of the null hypothesis

using the P-value.
Goodman beautifully summarized the history and consequences of
this in an exceptionally clearly written paper a few years ago (Goodman,
1999). A statistician using the Neyman & Pearson method and P-values to
reject null hypotheses at the 5% level will, on average in the long run (say
over the career of that statistician), only make the mistake of rejecting a
true null hypothesis about 5% of the time. However, the computations say
nothing about a specific instance with a specific set of data and a specific
null hypothesis, which is a unique event and not a repeatable event. There
is no way, using the data alone, to say how likely it is that the null hypoth-
esis is true in a specific instance. At most the data can tell you how far you
should move away from your prior probability that the hypothesis is true.
A Bayesian can compute this probability because to a Bayesian it makes
sense to state a prior probability of a unique event.
Actually, as further recounted by Goodman, Neyman & Pearson were
smart and realized that hypothesis testing did not get them out of the bind
– as did many other intelligent statisticians. One response in the commu-
nity of frequentists was to move from hypothesis testing to interval esti-
mation – estimation of so-called confidence intervals likely to contain the
parameter value of interest upon which the hypothesis depends.
Unfortunately, this did not solve the problem but sufficiently regressed it
into deep mathematics as to obfuscate whether or not it was solved.
So what does all of this mean for someone who is trained in frequen-
tist statistics or for someone who is wondering what Bayesian methods
offer? Let us call this person “You”.
At the very least, it means You will discover a new way to compute
intervals very close to those you get in computing traditional confidence
intervals. Your only reward lies in the knowledge that the specific interval
has the stated probability of containing the parameter, which is not the
case with the nearly identical interval computed in the traditional manner.
Admittedly, this does not seem like much gain.
It also means that You will have to think differently about the statisti-

cal problem You are solving, which will mean additional work. In particu-
lar, You may have to put real effort into specifying a prior probability that
You can defend to others. While this may be uncomfortable Bayesians are
working on ways to help You with both the process of understanding and
specifying the prior probabilities as well as the arguments to defend them.
Here is what You will get in return. First, in any specific analysis for a
specific dataset and specific hypothesis (not just the null hypothesis) You
will be able to compute the probability that the hypothesis is true. Or, often
more useful, You will be able to specify the probability that the true value
of the parameter is within any given interval. This is what is needed for
quantitative decision-making and for weighing the costs and benefits of
decisions depending on these estimates.
Second, You will get an easy way to revise Your estimate in an order-
ly and defensible fashion as you collect new data relevant to Your problem.
The first two gains give You a third: this way of thinking and comput-
ing frees You from some of the concerns about peeking at Your data before
the planned end of the trial. In fact, it gives You a whole new set of tools
to dynamically optimize trial sizes with optional stopping rules. This is a
very advanced topic in Bayesian methods – far beyond this Primer –but for
which there is growing literature.
Yet another gain is that others who depend on Your published results
to compute such things as a cost-effectiveness ratio can now directly incor-
porate uncertainty in a meaningful way to specify precision of their results.
While this may be an indirect gain to You it gives added value to Your
analyses.
A fifth gain, stemming from advances in computation methods stimu-
lated by Bayesians’ needs, is that you can naturally and easily estimate dis-
tributions for functions of parameters estimated in turn in quite compli-
cated statistical models to represent the data generating processes. You will
be freed from reliance on simplistic formulations of the data likelihood
solely for the purpose of being able to use standard tests. In many ways this
is analogous to the immense advances in our capability to estimate quite

sophisticated regression models over the simple linear models of yester-
year.
Finally, You will not get left behind. There is a beginning sea change
taking place in statistics and the ability to understand, apply and criticize a
Bayesian analysis will be important to researchers and practitioners in the
near future.
I hope You will find all these gains accruing to You as time marches
forward. It will require investment in relearning some of the fundamentals
with little apparent benefit at first. But if You persist, my probability is high
that You will succeed.
Dennis G. Fryback
Professor, Population Health Sciences
University of Wisconsin-Madison
References
Edwards W, Lindman H, Savage LJ. Bayesian statistical inference for psycho-
logical research. Psychological Review, 1963; 70:193-242.
Goodman, SN. Toward Evidence-Based Medical Statistics. 1: The P Value

Fallacy. Annals of Internal Medicine, 1999; 130(12): 995-1004.
Kyburg HE, Smokler, HE [Eds.] Studies in Subjective Probability, New York: John
Wiley & Sons, Inc. 1964.
Pearson ES. Bayes’ theorem, examined in the light of experimental sampling.

Biometrika 1925; 17:388-442.

Overview
T his Primer is for health economists, outcomes research practi-

tioners and biostatisticians who wish to understand the basics
of Bayesian statistics, and how Bayesian methods may be applied in
the economic evaluation of health care technologies. It requires no
previous knowledge of Bayesian statistics. The reader is assumed only
to have a basic understanding of traditional non-Bayesian techniques,
such as unbiased estimation, confidence intervals and significance
tests; that traditional approach to statistics is called ‘frequentist’.
The Primer has been produced in response to the rapidly growing
interest in, and acceptance of, Bayesian methods within the field of
health economics. For instance, in the United Kingdom the National
Institute for Clinical Excellence (NICE) specifically accepts Bayesian
approaches in its guidance to sponsors on making submissions. In the
United States the Food and Drug Administration (FDA) is also open to
Bayesian submissions, particularly in the area of medical devices. This
upsurge of interest in the Bayesian approach is far from unique to this
field, though; we are seeing at the start of the 21st century an explo-
sion of Bayesian methods throughout science, technology, social sci-
ences, management and commerce. The reasons are not hard to find,
and are similar in all areas of application. They are based on the fol-
lowing key benefits of the Bayesian approach:

(B1) Bayesian methods provide more natural and useful inferences
than frequentist methods.
(B2) Bayesian methods can make use of more available informa-
tion, and so typically produce stronger results than frequentist
methods.
(B3) Bayesian methods can address more complex problems than
frequentist methods.
(B4) Bayesian methods are ideal for problems of decision making,
whereas frequentist methods are limited to statistical analyses
that inform decisions only indirectly.
(B5) Bayesian methods are more transparent than frequentist
methods about all the judgements necessary to make infer-
ences.
We shall see how these benefits arise, and their implications for health
economics and outcomes research, in the remainder of this Primer.
However, even a cursory look at the benefits may make the reader won-
der why frequentist methods are still used at all. The answer is that there
are also widely perceived drawbacks to the Bayesian approach:
(D1) Bayesian methods involve an element of subjectivity that is
not overtly present in frequentist methods.
(D2) In practice, the extra information that Bayesian methods uti-
lize is difficult to specify reliably.
(D3) Bayesian methods are more complex than frequentist meth-
ods, and software to implement them is scarce or non-exis-
tent.
The authors of this Primer are firmly committed to the Bayesian

approach, and believe that the drawbacks can be, are being and will be
overcome. We will explain why we believe this, but will strive to be hon-
est about the competing arguments and the current state of the art.

This Primer begins with a general discussion of the benefits and draw-
backs of Bayesian methods versus the frequentist approach, including an
explanation of the basic concepts and tools of Bayesian statistics. This part
comprises five sections, entitled Inference, The Bayesian Method, Prior
Information, Prior Specification and Computation, which present all of the key
facts and arguments regarding the use of Bayesian statistics in a simple,
non-technical way.
The level of detail given in these sections will, hopefully, meet the
needs of many readers, but deeper understanding and justification of the
claims made in the main text can also be found in the Appendix. We stress
that the Appendix is still addressed to the general reader, and is intended
to be non-technical.
The last two sections, entitled Design and Analysis of Trials and Economic
Models, provide illustrations of how Bayesian statistics is already contribut-
ing to the practice of health economics and outcomes research. We should
emphasize that this is a fast-moving research area, and these sections may
go out of date quickly. We hope that readers will be stimulated to play their
part in these exciting developments, either by devising new techniques or
by employing existing ones in their own applications.
Finally, the Conclusions section summarizes the arguments in this
Primer, and a Further Reading list provides some general suggestions for fur-
ther study of Bayesian methods and their application in health economics.
Overview 11
SECTION 1 Inference
I n order to obtain a clear understanding of the benefits and draw-

backs to the Bayesian approach, we first need to understand the
basic differences between Bayesian and frequentist inference. This sec-
tion addresses the nature of probability, parameters and inferences
under the two approaches.
Frequentist and Bayesian methods are founded on different
notions of probability. According to frequentist theory, only repeatable
events have probabilities. In the Bayesian framework, probability sim-
ply describes uncertainty. The term “uncertainty” is to be interpreted
in its widest sense. An event can be uncertain by virtue of being intrin-
sically unpredictable, because it is subject to random variability, for
example the response of a randomly selected patient to a drug. It can
also be uncertain simply because we have imperfect knowledge of it,
for example the mean response to the drug across all patients in the
population. Only the first kind of uncertainty is acknowledged in fre-
quentist statistics, whereas the Bayesian approach encompasses both
kinds of uncertainty equally well.
Example.
Suppose that Mary has tossed a coin and knows the outcome,
Heads or Tails, but has not revealed it to Jamal. What probability
should Jamal give to it being Head? When asked this question,

most people say that the chances are 50-50, i.e. that the probability is
one-half. This accords with the Bayesian view of probability, in which
the outcome of the toss is uncertain for Jamal so he can legitimately
express that uncertainty by a probability. From the frequentist per-
spective, however, the coin is either Head or Tail and is not a random
event. For the frequentist it is no more meaningful for Jamal to give
the event a probability than for Mary, who knows the outcome and is
not uncertain. The Bayesian approach clearly distinguishes between
Mary’s and Jamal’s knowledge.
Statistical methods are generally formulated as making inferences about

unknown parameters. The parameters represent things that are unknown,
and can usually be thought of as properties of the population from which
the data arise. Any question of interest can then be expressed as a ques-
tion about the unknown values of these parameters. The reason why the
difference between the frequentist and Bayesian notions of probability is
so important is that it has a fundamental implication for how we think
about parameters. Parameters are specific to the problem, and are not gen-
erally subject to random variability. Therefore, frequentist statistics does
not recognize parameters as being random and so does not regard proba-
bility statements about them as meaningful. In contrast, from the Bayesian
perspective it is perfectly legitimate to make probability statements about
parameters, simply because they are unknown.
Note that in Bayesian statistics, as a matter of convenient terminology,
we refer to any uncertain quantity as a random variable, even when its
uncertainty is not due to randomness but to imperfect knowledge.
Example.
Consider the proposition that treatment 2 will be more cost-effective
than treatment 1 for a health care provider. This proposition concerns
unknown parameters, such as each treatment’s mean cost and mean
efficacy across all patients in the population for which the health care

provider is responsible. From the Bayesian perspective, since we are
uncertain about whether this proposition is true, the uncertainty is
described by a probability. Indeed, the result of a Bayesian analysis of
the question can be simply to calculate the probability that treatment 2
is more cost-effective than treatment 1 for this health care provider.
From the frequentist perspective, however, whether treatment 2 is
more cost-effective is a one-off proposition referring to two specific
treatments in a specific context. It is not repeatable and so we cannot talk
about its probability.
In this last example, the frequentist can conduct a significance test of

the null hypothesis that treatment 2 is not more cost-effective, and there-
by obtain a P-value. At this point, the reader should examine carefully the
statements in the box “Interpreting a P-value” below, and decide which
ones are correct.
Interpreting a P-value
The null hypothesis that treatment 2 is not more cost-effective than treatment 1
is rejected at the 5% level, i.e. P = 0.05. What does this mean?
1. Only 5% of patients would be more cost-effectively treated by treatment 1.
2. If we were to repeat the analysis many times, using new data each time, and
if the null hypothesis were really true, then on only 5% of those occasions
would we (falsely) reject it.
3. There is only a 5% chance that the null hypothesis is true.
Statement 3 is how a P-value is commonly interpreted; yet this inter-

pretation is not correct because it makes a probability statement about the
hypothesis, which is a Bayesian, not a frequentist, concept. The correct
interpretation of the P-value is much more tortuous and is given by
Statement 2. (Statement 1 is another fairly common misinterpretation.
Since the hypothesis is about mean cost and mean efficacies, it says noth-
Section 1: Inference 15
ing about individual patients.)
The primary reason why we cannot interpret a P-value in this way is
because it does not take account of how plausible the null hypothesis was a
priori.
Example.
An experiment is conducted to see whether thoughts can be transmit-
ted from one subject to another. Subject A is presented with a shuffled
deck of cards and tries to communicate to Subject B whether each card
is red or black by thought alone. In the experiment, Subject B correct-
ly gives the color of 33 cards. The null hypothesis is that no thought-
transference takes place and Subject B is randomly guessing. The
observation of 33 correct is significant with a (one-sided) P-value of
3.5%. Should we now believe that it is 96.5% certain that Subject A
can transmit her thoughts to Subject B?
Most scientists would regard thought-transference as highly implausi-

ble and in no way would be persuaded by a single, rather small, experi-
ment of this kind. After seeing this experimental result, most would still
strongly believe in the null hypothesis, regarding the outcome as due to
chance.
In practice, frequentist statisticians recognize that much stronger evi-
dence would be required to reject a highly plausible null hypothesis, such
as in the above example, than to reject a more doubtful null hypothesis.
This makes it clear that the P-value cannot mean the same thing in all sit-
uations and to interpret it as the probability of the null hypothesis is not
only wrong but could be seriously wrong when the hypothesis is a priori
highly plausible (or highly implausible).
To many users of statistics and even to many practicing statisticians, it
is perplexing that one cannot interpret a P-value as the probability that the
null hypothesis is true. Similarly, it is perplexing that one cannot interpret

a 95% confidence interval for a treatment difference as saying that the true
difference has a 95% chance of lying in this interval. Nevertheless, these
are wrong interpretations…and can be seriously wrong. The correct inter-
pretations are far more indirect and unintuitive. (See the Appendix for
more examples.)
Bayesian inferences have exactly the desired interpretations. A
Bayesian analysis of a hypothesis results precisely in the probability that it
is true. In addition, a Bayesian 95% interval for a parameter means pre-
cisely that there is a 95% probability that the parameter lies in that inter-
val. This is the essence of the key benefit (B1) – “more natural and inter-
pretable inferences” – offered by Bayesian methods.
Section 1: Inference 17
TABLE 1. Summary of Key Differences Between
Frequentist and Bayesian Approaches
FREQUENTIST BAYESIAN
Nature of probability
Probability is a limiting, long-run Probability measures a personal degree

frequency. of belief.
It only applies to events that are (at least It applies to any event or proposition
in principle) repeatable. about which we are uncertain.
Nature of parameters
Parameters are not repeatable or Parameters are unknown.

random.
They are therefore not random variables, They are therefore random variables.
but fixed (unknown) quantities.
Nature of inference
Does not (although it appears to) make Makes direct probability statements
statements about parameters. about parameters.
Interpreted in terms of long-run repeti- Interpreted in terms of evidence from the

tion. observed data.
Example
“We reject this hypothesis at the 5% “The probability that this hypothesis is
level of significance.” true is 0.05.”
In 5% of samples where the hypothesis The statement applies on the basis of

is true it will be rejected (but nothing is this sample (as a degree of belief).
stated about this sample).

2
SECTION 1 The Bayesian Method
T he fundamentals of Bayesian statistics are very simple. The

Bayesian paradigm is one of learning from data.
The role of data is to add to our knowledge and so to update what
we can say about the parameters and relevant hypotheses. As such,
whenever we wish to learn from a new set of data, we need to iden-
tify what is known prior to observing those data. This is known as prior
information. It is through the incorporation of prior information that
the Bayesian approach utilizes more information than the frequentist
approach. A discussion of precisely what the prior information repre-
sents and where it comes from can be found in the next section: Prior
Information. For purposes of exposition of how the Bayesian paradigm
works, we simply suppose that the prior information has been identi-
fied and is expressed in the form of a prior distribution for the
unknown parameters of the statistical model. The prior distribution
expresses what is known (or believed to be true) before seeing the
new data. This information is then synthesized with the information
in the data to produce the posterior distribution, which expresses
what we now know about the parameters after seeing the data. (We
often refer to these distributions as ‘the prior’ and ‘the posterior’.)
The mathematical mechanism for this synthesis is Bayes’ theo-
rem, and this is why this approach to statistics is called “Bayesian”.
From a historical perspective, the name originated from the Reverend

Thomas Bayes, an 18th century minister who first showed the use of the
theorem in this way and gave rise to Bayesian statistics.
The process is simply illustrated in the box “Example of Bayes’
theorem”.
Figure 1 is called a triplot and is a way of seeing how Bayesian meth-
ods combine the two information sources. The strength of each source of
information is indicated by the narrowness of its curve – a narrower curve
rules out more parameter values and so represents stronger information.
In Figure 1, we see that the new data (red curve) are a little more inform-
ative than the prior (grey curve). Since Bayes’ theorem recognizes the
strength of each source, the posterior (black dotted curve) in Figure 1 is
influenced a little more by the data than by the prior. For instance, the pos-
terior peaks at 1.33, a little closer to the peak of the data curve than to the
prior peak. Notice that the posterior is narrower than either the prior or
the data curve, reflecting the way that the posterior has drawn strength
from both information sources.
The data curve is technically called the likelihood and is also important
in frequentist inference. Its role in both inference paradigms is to describe
the strength of support from the data for the various possible values of the
parameter. The most obvious difference between frequentist and Bayesian
methods is that frequentist statistics uses only the likelihood, whereas
Bayesian statistics uses both the likelihood and the prior information.
In Figure 1, the Bayesian analysis produces different inferences from
the frequentist approach because it uses the prior information as well as
the data. The frequentist estimate, using the data alone, is around 1.5. The
Bayesian analysis uses the fact that it is unlikely, on the basis of the prior
information, that the true parameter value is 2 or more. As a result, the
Bayesian estimate is around 1. The Bayesian analysis combines the prior
information and data information in a similar way to how a meta-analysis
combines information from several reported trials. The posterior estimate
is a compromise between prior and data estimates and is a more precise
estimate (as seen in the posterior density being a narrower curve) than

Example of Bayes’ Theorem
0.4
0.3
0.2
0.1
-4 -2 0 2 4
Figure 1. The prior distribution (grey) and information from the new data (red)
are synthesized to produce the posterior distribution (black dotted).
In this example, the prior information (grey curve) tells us that the parameter is
almost certain to lie between – 4 and + 4, that it is most likely to be between – 2
and + 2, and that our best estimate of it would be 0.
The data (red curve) favor values of the parameter between 0 and 3, and strongly
argue against any value below – 2.
The posterior (black dotted curve) puts these two sources of information together.
So, for values below – 2 the posterior density is tiny because the data are saying
that these values are highly implausible. Values above + 4 are ruled out by the
prior; again, the posterior agrees. The data favors values around 1.5, while the
prior prefers values around 0. The posterior listens to both and the synthesis is a
compromise. After seeing the data, we now think the parameter is most likely to be
around 1.
either information source separately. This is the key benefit (B2) – “ability
to make use of more information and to obtain stronger results” – that the
Bayesian approach offers.
According to the Bayesian paradigm, any inference we desire is
derived from the posterior distribution. One estimate of a parameter might
be the mode of this distribution (i.e. the point where it reaches its maxi-
mum). Another common choice of estimate is the posterior expectation. If
we have a hypothesis, then the probability that the hypothesis is true is
also derived from the posterior distribution. For instance, in Figure 1 the
Section 2: The Bayesian Method 21

probability that the parameter is positive is the area under the black dot-
ted curve to the right of the origin, which is 0.89.
In contrast to frequentist inference, which must phrase all questions in
terms of significance tests, confidence intervals and unbiased estimators,
Bayesian inference can use the posterior distribution very flexibly to pro-
vide relevant and direct answers to all kinds of questions. One example is
the natural link between Bayesian statistics and decision theory. By com-
bining the posterior distribution with a utility function (which measures
the consequences of different decisions), we can identify the optimal deci-
sion as that which maximizes the expected utility. In economic evaluation,
this could reduce to minimizing expected cost or to maximizing expected
efficacy, depending on the utility function. However, from the perspective
of cost-effectiveness, the most appropriate utility measure is net benefit
(defined as the mean efficacy times willingness to pay, minus expected
cost).
For example, consider a health care provider that has to choose which
of two procedures to reimburse. The optimal decision is to choose the one
that has the higher expected net benefit. A Bayesian analysis readily pro-
vides this answer, but there is no analogous frequentist analysis. To test the
hypothesis that one net benefit is higher than the other simply does not
address the question properly (in the same way that to compute the prob-
ability that the net benefit of procedure 2 is higher than that of procedure
1 is not the appropriate Bayesian answer). More details of this example
and of Bayes’ theorem can be found in the Appendix.
This serves to illustrate another key benefit of Bayesian statistics, (B4)
– “Bayesian methods are ideal for decision making”.

3
SECTION 1 Prior Information
T he prior information is both a strength and a potential weak-

ness of the Bayesian approach. We have seen how it allows
Bayesian methods to access more information and so to produce
stronger inferences. As such it is one of the key benefits of the
Bayesian approach. On the other hand, most of the criticism of
Bayesian analysis focuses on the prior information.
The most fundamental criticism is that prior information is sub-
jective: your prior information is different from mine, and so my prior
distribution is different from yours. This makes the posterior distribu-
tion, and all inferences derived from it, subjective. In this sense, it is
claimed that the whole Bayesian approach is subjective. Indeed,
Bayesian methods are based on a subjective interpretation of proba-
bility, which is described in Table 1 as a “personal degree of belief”. This
formulation is necessary (see the Appendix for details) if we are to give
probabilities to parameters and hypotheses, since the frequentist inter-
pretation of probability is too narrow. Yet for many scientists trained
to reject subjectivity whenever possible, this is too high a price to pay
for the benefits of Bayesian methods. To its critics, (D1) “subjectivity”
is the key drawback of the Bayesian approach.
We believe that this objection is unwarranted both in principle
and in practice. It is unwarranted in principle because science cannot
be truly objective. In practice it is unwarranted because the Bayesian

method actually very closely reflects the real nature of the scientific
method, in the following respects:
Subjectivity in the prior distribution is minimized through basing prior
information on defensible evidence and reasoning.
Through the accumulation of data, differences in prior positions are
resolved and consensus is reached.
Taking the second of these points first, Bayes’ theorem weights the prior
information and data according to their relative strengths in order to derive
the posterior distribution. If prior information is vague and insubstantial
then it will get negligible weight in the synthesis with the data, and the pos-
terior will in effect be based entirely on data information (as expressed in the
likelihood function). Similarly, as we acquire more and more data, the
weight that Bayes’ theorem attaches to the newly acquired data relative to
the prior increases. Again, the posterior is effectively based entirely on the
information in the data. This feature of Bayes’ theorem mirrors the process
of science, where the accumulation of objective evidence is the primary
process whereby differences of opinion are resolved. Once the data provide
conclusive evidence, there is essentially no room left for subjective opinion.
Returning to the first point above, it is stated that where genuine, sub-
stantial prior information exists it needs to be based on defensible evidence
and reasoning. This is clearly important when the new data are not so
extensive as to overwhelm the prior information, so that Bayes’ theorem
will give the prior a non-negligible weight in its synthesis with the data.
Prior information of this kind exists routinely in medical applications, and
in particular in economic evaluation of competing technologies.
Two examples are presented in the Appendix. One concerns the analy-
sis of subgroup differences, where prior skepticism about the existence of
such effects without a plausible biological mechanism is naturally accom-
modated in the Bayesian analysis.
The other example concerns a case where a decision on the cost-effec-
tiveness of a new drug versus standard treatment depends in large part on
evidence about hospitalizations. A small trial produces an apparently large

(and, in frequentist terms, significant) reduction in mean days in hospital.
However, an earlier and much larger trial produced a much less favorable
estimate of mean hospital days for a similar drug. There are two possible
responses that a frequentist analysis can have to the earlier trial:
1. Take the view that there is no reason why the hospitalization rate
under the old drug should be the same as under the new one, in which
case the earlier trial is ignored because it contributes no information
about the new drug.
2. Take the view that the two drugs should have essentially identical
hospitalization rates – and so we pool the data from the two trials.
The second option will lead to the new data being swamped by the
much larger earlier trial, which seems unreasonable, but the first option
entails throwing away potentially useful information. In practice, a fre-
quentist would probably take the first option, but with a caveat that the
earlier trial suggests this may underestimate the true rate.
It would usually be more realistic to take the view that the two hospital-
ization rates will be different but similar. The Appendix demonstrates how a
Bayesian analysis can accommodate the earlier trial as prior information
although it necessitates a judgement about similarity of the drugs. How dif-
ferent might we have believed their hospitalization rates to be before con-
ducting the new trial?
The Bayesian analysis produces a definite and quantitative synthesis of
the two sources of information rather than just the vague “an earlier trial on
a similar drug produced a higher mean days in hospital, and so I am skepti-
cal about the reduction seen in this trial”. This synthesis results from making
a clear, reasoned and transparent interpretation of the prior information.
This is part of the key benefit (B5) – “more transparent judgements” – of the
Bayesian approach. Without the Bayesian analysis it would be natural to
moderate the claims of the new trial. The extent of such moderation would
still be judgmental, but the judgement would not be so open and the result
would not be transparently derived from the judgement by Bayes’ theorem.
Section 3: Prior Information 25

This leads to another important way in which Bayesian methods are
transparent. Once the prior distribution and likelihood have been formu-
lated (and openly laid on the table), the computation of the posterior dis-
tribution and the derivation of appropriate posterior inferences or deci-
sions are uniquely determined. In contrast, once the likelihood has been
determined in a frequentist analysis there is still the freedom to choose
which of many inference rules to apply. For instance, although in simple
problems it is possible to identify optimal estimators, in general, there are
likely to be many unbiased estimators – none of which dominates any of
the others in the sense of having uniformly smaller variance. The practi-
tioner is then free to use any of these or to dream up others on an “ad hoc”
basis. This feature of frequentism leads to a lack of transparency because
the respective choices are, in essence, arbitrary.
So what of the criticism (D1), that Bayesian methods are inherently sub-
jective? It is true that one could carry out a Bayesian analysis with a prior
distribution based on mere guesswork, prejudice or wishful thinking. Bayes’
theorem technically admits all of these unfortunate practices, but Bayesian
statistics does not in any sense condone them. Also, recall that in a proper
Bayesian analysis, prior information is not only transparent but is also based
on both defensible evidence and reasoning which, if followed, will lead any
above-mentioned abuses to become transparent, and so to be rejected.
A compact statement of what should constitute prior information is
provided in the box ‘The Evidence’.
The ‘Evidence’
Prior information should be based on sound evidence and reasoned judgements.
A good way to think of this is to parody a familiar quotation: the prior distribution
should be ‘the evidence, the whole evidence and nothing but the evidence’:
• ‘the evidence’ – genuine information legitimately interpreted;
• ‘the whole evidence’ – not omitting relevant information (preferably a
consensus that pools the knowledge of a range of experts);
• ‘nothing but the evidence’ – not contaminated by bias or prejudice.
26 A Primer on Bayesian Statistics in Health Economics and Outcomes R e s e a r c h

4
SECTION 1 Prior Specification
W e hope that the preceding sections convince the reader that

prior information exists and should be used, in as rea-
soned, objective and fully transparent a way as possible. Here we
address the question of how to formulate a prior probability distribu-
tion, the grey curve in Figure 1.
Refer to the example in the previous section where prior infor-
mation consists of information about hospitalization in a trial of a sim-
ilar drug. In the Appendix this is formulated as a prior distribution
with mean 0.21 (average days in hospital per patient) and standard
deviation 0.08. This is justified by reference to the trial in question,
where the average days in hospital under the different but similar drug
was estimated to be 0.21 with a standard error of 0.03. But how is the
stated prior distribution obtained from the given prior information?
Judgement inevitably intervenes in the process of specifying the
prior distribution. As in the above case, it typically arises through the
need to interpret the prior information and its relevance to the new
data. How different might the hospitalization rates be under the two
drugs? Different experts may interpret the prior information different-
ly. As well, a given expert may interpret the information differently at
a later time, such as in the example of deciding on a prior standard
deviation of 0.75 rather than 0.8.

Even though our prior information might be genuine evidence with a
clear relation to the new data, we cannot convert this into a prior distri-
bution with perfect precision and reliability. This is the drawback (D2) –
“prior specification is unreliable”.
Nevertheless, in practice we only need to specify the prior distribution
with sufficient reliability and accuracy. We can explore the range of plau-
sible prior specifications based on reasonable interpretations of the evi-
dence and allowing for imprecision in the necessary judgements. If the
posterior inferences or decisions are essentially insensitive to those varia-
tions, then the inherent unreliability of the prior specification process does
not matter. This practice of sensitivity analysis with respect to the prior
specification is a basic feature of practical Bayesian methodology as it is in
all decision analysis applications.
Types and definitions of prior distribution

Informative (or genuine) priors: represent genuine prior information and best
judgement of its strength and relation to the new data.
Noninformative (or default, reference, improper, weak, ignorance) priors:

represent complete lack of credible prior information.
Skeptical priors: supposed to represent a position that a null hypothesis is likely

to be true.
Structural (or hierarchical) priors: incorporate genuine prior information about

relationships between parameters.
The precision needed in the prior specification to achieve robust infer-

ences and decisions depends on the strength of the new data. As we have
seen, given strong enough data, the prior information matters little or not
at all and differences of judgement in interpreting the data will be unim-
portant. When the new data are not so strong, and prior information is
appreciable, then sensitivity analysis is essential. It is also important to note
that, despite obvious drawbacks, expert opinion is sometimes quite a use-

ful component of prior information. The procedures to elicit expert judge-
ments are an active topic of research by both statisticians and psychologists.
Up until now, we have been considering genuine informative prior
distributions. Some other ways to specify the prior distribution in a
Bayesian analysis are set out in the box, “Types and definitions of prior
distribution”.
In response to the difficulty of accurately and reliably eliciting prior
distributions, some have proposed conventional solutions that are sup-
posed to represent either no prior beliefs or a skeptical prior position.
The argument in favor of representing no prior information is that this
avoids any criticism about subjectivity. There have been numerous
attempts to find a formula for representing prior ignorance, but without
any consensus. Indeed, it is almost certainly an impossible quest.
Nevertheless, the various representations that have been derived can be
useful – at least for representing relatively weak prior information.
When the new data are strong (relative to the prior information), the
prior information is not expected to make any appreciable contribution to
the posterior. In this situation, it is pointless (and not cost-effective) to
spend much effort on carefully eliciting the available prior information.
Instead, it is common in such a case to apply some conventional ‘nonin-
formative’, ‘default’, ‘reference’, ‘improper’, ‘vague’, ‘weak’ or ‘ignorance’
prior (although the last of these is really a misnomer). These terms are
used more or less interchangeably in Bayesian statistics to denote a prior
distribution representing very weak prior information. The term ‘improp-
er’ is used because technically most of these distributions do not actually
exist in the sense that a normal distribution with an infinite variance does
not exist.
The idea of using so-called ‘skeptical’ priors is that if a skeptic can be
persuaded by the data then anyone with a less skeptical prior position
would also be persuaded. Thus, if one begins with a skeptical prior position
with regard to some hypothesis and is nevertheless persuaded by the data,
so that their posterior probability for that hypothesis is high, then some-
Section 4: Prior Specification 29

one else with a less skeptical prior position would end up giving that
hypothesis an even higher posterior probability. In that case, the data are
strong enough to reach a firm conclusion. If, on the other hand, when we
use a skeptical prior the data are not strong enough to yield a high poste-
rior probability for that hypothesis, then we should not yet claim any def-
inite inference about it. Although this is another tempting idea, there is
even less agreement or understanding about what a skeptical prior should
look like.
The rather more complex ideas of structural or hierarchical priors (the
last category in the box “Types and definitions of prior distribution”) are
discussed in the Appendix.

5
SECTION 1 Computation
S oftware is essential for any but the simplest of statistical tech-

niques, and Bayesian methods are no exception. In Bayesian
statistics, the key operations are to implement Bayes’ theorem and
then to derive relevant inferences or decisions from the posterior dis-
tribution. In very simple problems these tasks can be done algebraical-
ly, but this is not possible in even moderately complex problems.
Until the 1990s, Bayesian methods were interesting, but they found
little practical application because the necessary computational tools and
software had not been developed. Anyone who wanted to do serious
statistical analysis had no alternative but to use frequentist methods. In
little over a decade that position has been dramatically turned around.
Computing tools were developed specifically for Bayesian analysis that
are more powerful than anything available for frequentist methods in
the sense that Bayesians can now tackle enormously intricate problems
that frequentist methods cannot begin to address. It is still true that
Bayesian methods are more complex and that, although the computa-
tional techniques are well understood in academic circles, there is still a
lack of user-friendly software for the general practitioner.
The transformation is continuing, and computational develop-
ments are shifting the balance between the drawback (D3) – “com-
plexity and lack of software” – and the benefit (B3) – “ability to tack-
le more complex problems”. The main tool is a simulation technique

called Markov chain Monte Carlo (MCMC). The idea of MCMC is in a sense
to bypass the mathematical operations rather than to implement them.
Bayesian inference is solved by randomly drawing a very large simulated
sample from the posterior distribution. The point is that if we have a suf-
ficiently large sample from any distribution then we effectively have that
whole distribution in front of us. Anything we want to know about the dis-
tribution we can calculate from the sample. For instance, if we wish to
know the posterior mean we just calculate the mean of this ‘inferential
sample’. If the sample is big enough, the sample mean is an extremely
accurate approximation to the true distribution mean, such that we can
ignore any discrepancy between the two.
The availability of computational techniques like MCMC makes exact
Bayesian inferences possible even in very complex models. Generalized
linear models, for example, can be analyzed exactly by Bayesian methods,
whereas frequentist methods rely on approximations. In fact, Bayesian
modelling in seriously complex problems freely combines components of
different sorts of modelling approaches with structural prior information,
unconstrained by whether such model combinations have ever been stud-
ied or analyzed before. The statistician is free to model the data and other
available information in whatever way seems most realistic. No matter
how messy the resulting model, the posterior inferences can be computed
(in principle, at least) by MCMC.
Bayesian methods have become the only feasible tools in several
fields such as image analysis, spatial epidemiology and genetic pedigree
analysis.
Although there is a growing range of software available to assist with
Bayesian analysis, much of it is still quite specialized and not very useful
for the average analyst. Unfortunately, there is nothing available yet that
is both powerful and user-friendly in the way that most people expect sta-
tistical packages to be. Two software packages that are in general use, freely

available and worth mentioning are First Bayes and WinBUGS.
First Bayes is a very simple program that is aimed at helping the begin-
ner learn and understand how Bayesian methods work. It is not intend-
ed for serious analysis of data, nor does it claim to teach Bayesian sta-
tistics, but it is in use in several universities worldwide to support cours-
es in Bayesian statistics. First Bayes can be very useful in conjunction
with a textbook – such as those recommended in the Further Reading
section of this Primer – and can be freely downloaded from
http://www.shef.ac.uk/~st1ao/.
WinBUGS is a powerful program for carrying out MCMC computations
and is in widespread use for serious Bayesian analysis. WinBUGS has been
a major contributing factor to the growth of Bayesian applications and can
be freely downloaded from http://www.mrc-bsu.cam.ac.uk/bugs/. Please
note, however, that WinBUGS is currently not very user-friendly and
sometimes crashes with inexplicable error messages. Given the growing
popularity of Bayesian methods, it is likely that more robust, user-friendly
commercial software will emerge in the coming years.
The Appendix provides more detail on these two sides of the Bayesian
computing coin: the drawback (D3) – “complexity and lack of software” –
and the benefit (B3) – “ability to tackle more complex problems”.
Section 5: Computation 33
6
SECTION 1 Design and Analysis
of Trials
B ayesian techniques are inherently useful for designing clinical

trials because trials tend to be sequential, each designed based
in large part on prior trial evidence. The substantial literature that is
available regarding clinical trial design using Bayesian techniques is, of
course, applicable to design of cost-effectiveness trials.
By their nature, cost-effectiveness trials always have prior clinical
and probably some form of economic information, which in the fre-
quentist approach would be used to set the power requirements for
the trial, and hence to identify the sample size. Since the prior infor-
mation is explicitly stated in Bayesian design techniques, the depend-
ence of the chosen design on prior information is fully transparent. A
Bayesian analysis would formulate prior knowledge about how large
an effect might be achieved. For instance, in planning a Phase III trial
there will be information from Phase II studies on which to base a
prior distribution for the effect. This permits an informative approach
to setting sample size.
For a given sample size, the Bayesian calculation computes the
probability that the trial will successfully demonstrate a positive effect
(see O’Hagan and Stevens, 2001b). This can then be directly linked to
a decision about whether a trial of a certain size (and hence cost), with
this assurance of success (and consequent financial return), is worth-

while. This contrasts with the frequentist power calculations, which only
provide a probability of demonstrating an effect conditional on the
unknown true effect taking some specific value.
An important simplifying feature of Bayesian design is that interim
analyses can be introduced without affecting the final conclusions, and
they do not need to be planned in advance. This is because Bayesian analy-
sis does not suffer from the paradox of frequentist interim analysis, that
two sponsors running identical trials and obtaining identical results may
reach different conclusions if one performs an interim analysis (but does
not stop the trial then) and the other does not. A Bayesian trial can be
stopped early or extended for any appropriate reason without needing to
compensate for such actions in subsequent analysis.
Aside from designing trials, a Bayesian approach is also useful for ana-
lyzing trial results. Today we see a growing interest in economic evaluation
that has led to inclusion of cost-effectiveness as a secondary objective in
traditional clinical trials. This may simply mean the collection of some
resource use data alongside conventional efficacy trials, but may extend to
more comprehensive economic data, more pragmatic enrollment, more
relevant outcome measures and/or utilities. Methods of statistical analysis
have begun to be developed for such trials. A useful review of Bayesian
work in this area is O’Hagan and Stevens (2002).
Early statistical work concentrated on deriving inference for the incre-
mental cost-effectiveness ratio, but the peculiar properties of ratios result-
ed in less than optimal solutions for various reasons. More recently, inter-
est has focused on inference for the (incremental) net benefit, which is
more straightforward statistically. Bayesian analyses have almost exclu-
sively adopted the net benefit approach. In fact, when using net benefits
the most natural expression of the relative cost-effectiveness of two treat-
ments is the cost-effectiveness acceptability curve (van Hout et al, 1994);
the essentially Bayesian nature of this measure is discussed in the
Appendix.

Costs in trials, as everywhere, are invariably highly skewed. Bayesian
methods accommodate this feature easily. O’Hagan and Stevens (2001a)
provide a good example where the efficacy outcome is binary. They model
costs as lognormally distributed (so explicitly accommodating skewness)
and allow different lognormal distributions for patients who have positive
or negative efficacy outcomes in each treatment group. They also illustrate
how even simple structural prior information can help provide more real-
istic posterior inferences in a dataset where two very high cost patients
arise in one patient group. Such a model is straightforwardly analyzed by
MCMC. Stevens et al (2003) provide full details of WinBUGS code to com-
pute posterior inferences. Another good example is Sutton et al (2003).
Section 6: Design and Analysis of Trials 37

7
SECTION 1 Economic Models
E conomic evaluation is widely practiced by building economic

models (Chilcott et al, 2003). Even when cost-related data have
been collected alongside efficacy in a clinical trial, they will rarely be
adequate for a proper economic evaluation. This is because, as is wide-
ly understood, practice patterns within clinical trials differ radically
from practice patterns in community medicine (the latter being the
context of practical interest). More realistically, such data will inform
some of the inputs (e.g. clinical efficacy data) to the cost-effectiveness
model, while other input values (e.g. resource use, prices) will be
derived from other sources.
Inputs to economic models can at best only be estimates of the
unknown true values of these parameters – a fact that is recognized in
the practice of performing sensitivity analysis. Often, this consists of a
perfunctory one-way sensitivity analysis in which one input at a time
is varied to some ad hoc alternative value and the model rerun to see
if the cost-effectiveness conclusion changes. Even when none of these
changes in single parameter values is enough to change the conclusion
as to which treatment is more cost-effective, the analysis gives no quan-
tification of the confidence we can attach to this being the correct infer-
ence. The true values of individual inputs might be outside the ranges
explored. Furthermore, if two or more inputs are varied together with-
in those ranges they might change the conclusion.

The statistically sound way to assess the uncertainty in the model out-
put that arises from uncertainty in its inputs is probabilistic sensitivity
analysis (PSA). This is the approach that is recommended by NICE, other
statutory agencies and many academic texts. It consists of assigning proba-
bility distributions to the inputs, so as to represent the uncertainty we have
in their true values, and then propagating this uncertainty through the
model. There is increasing awareness of the benefits of PSA; two examples
are Briggs et al (2002) and Parmigiani (2002).
It is important to appreciate that in PSA we are putting probability dis-
tributions on unknown parameters, which makes it unequivocally a
Bayesian analysis. In effect, Bayesian methods have been widely used in
health economics for years. The recognition of the Bayesian nature of these
probability distributions has important consequences. The distributions
should be specified using the ideas discussed in Section 4, Prior Specification.
In particular, the evidence sought to populate economic models rarely relates
directly to the parameters that the model actually requires in any applica-
tion. Trial data will be from a different population (possibly in a different
country) and with different compliance, registry data are potentially biased
and so forth. Just as we considered with the use of prior information in gen-
eral Bayesian analyses, the relationship between the data used to populate
the model and the parameters that define the use we wish to make of the
model is a matter for judgement. It is common to ignore these differences.
However, using the estimates and standard errors reported in the literature
as defining the input distributions will under-represent the true uncertainty.
The usual technique for PSA is Monte Carlo simulation, in which random
sets of input values are drawn and the model run for each set. This gives a sam-
ple from the output distribution (which is very similar to MCMC sampling
from the posterior distribution). This is feasible when the model is simple
enough to run almost instantaneously on a computer but for more complex
models it may be impractical to obtain a sufficiently large sample of runs. For
such situations, Stephenson et al (2002) describe an alternative technique,
based on Bayesian statistics, for computing the output distribution using far
fewer model runs.

Once the uncertainty in the model output has been quantified in PSA
by its probability distribution, the natural way to express uncertainty about
cost-effectiveness is again through the cost-effectiveness acceptability curve.
As mentioned already, this is another intrinsically Bayesian construction.
A natural response to uncertainty about cost-effectiveness is to ask
whether obtaining further data might reduce uncertainty. In the United
Kingdom, for instance, one of the decisions that NICE might make when asked
to decide on cost-effectiveness of a drug is to say that there is insufficient evi-
dence at present, and defer approving the drug for reimbursement by the
National Health Service until more data have been obtained. Bayesian decision
theory provides a conceptually straightforward way to inform such a deci-
sion, through the computation of the expected value of sample information.
Expected value of information calculations have been advocated by Felli
and Hazen (1998), Claxton and Posnett (1996), Brennan et al (2003) and a
Bayesian calculation for complex models developed by Oakley (2002). There
is a strong link between such analyses and design of trials, since balancing
the expected value of sample information against sampling costs is a stan-
dard Bayesian technique for identifying an optimal sample size.
There is another important link between the analysis of economic mod-
els and the analysis of cost-effectiveness trials. Where the evidence for indi-
vidual parameters in an economic model comes from a trial or other statisti-
cal data, the natural distribution to assign to those parameters is their poste-
rior distribution from a fully Bayesian analysis of the raw data. This assumes
that the data are directly relevant to the parameter required in the model,
rather than relating strictly to a similar, but different, parameter. In the lat-
ter case, it is simple to link the posterior distribution from the data analysis
to the parameters needed for the model, using structural prior information.
This linking of statistical analysis of trial data to economic model inputs
is a form of evidence synthesis and illustrates the holistic nature of the
Bayesian approach. Examples are given by Ades and Lu (2002) and Cooper
et al (2002). Ades et al synthesize evidence from a range of overlapping data
sources within a single Bayesian analysis. Synthesizing evidence is exactly
what Bayes’ theorem does.
Section 7: Economic Models 41

Conclusions
In this section we will briefly summarize the main messages being con-
veyed in this Primer.
• Bayesian methods are different from and, we posit, have certain
advantages over conventional frequentist methods, as set out in
benefits (B1) to (B5) of the Overview. These benefits are explored and
illustrated in various ways throughout subsequent sections of the
Primer.
• There are some perceived disadvantages of Bayesian methods, as set
out in the drawbacks (D1) to (D3) in the Overview. These are also
discussed in subsequent sections and we describe how they are
being addressed. It is up to the reader to judge the degree to which
the benefits may outweigh the drawbacks in practice.
• Bayesian technologies have already been developed in many of the
key methodologies of health economics. Already we see clear
advantages in the design and analysis of cost-effectiveness trials,
quantification of uncertainty in economic models, expression of
uncertainty about cost-effectiveness, assessment of the value of
potential new evidence, and synthesis of information going into
and through an economic evaluation.
• There is enormous scope for the development of new and more
sophisticated Bayesian techniques in health economics and out-
comes research. We are confident that Bayesian analysis will
increasingly become the approach of choice for the development
and evaluation of submissions on cost-effectiveness of medical
technologies, as well as for pure cost or utility studies.

Bibliography and
Further Reading
Articles and books cited in the text (including those in the

Appendix) are listed below. This is by no means an exhaustive list of
Bayesian work in the field. Suggestions for further reading are given at
the end.
Ades, A. E. and Lu, G. (2002). Correlations between parameters in risk

models: estimation and propagation of uncertainty by Markov Chain
Monte Carlo. Technical report, MRC Health Services Research
Collaboration, University of Bristol.
Brennan, A., Chilcott, J., Kharroubi, S. A. and O’Hagan, A. (2003).

Calculating expected value of perfect information – resolution of the uncer-
tainty in methods and a two level Monte Carlo approach. Technical report,
School of Health and Related Research, University of Sheffield.
Briggs, A. H., Goeree, R., Blackhouse, G. and O’Brien, B. J. (2002).

Probabilistic analysis of cost-effectiveness models: choosing between treat-
ment strategies for gastroesophageal reflux disease. Medical Decision Making
22, 290-308.
Chilcott, J. Brennan, A., Booth, A., Karnon, J. and Tappenden, P. (2003).

The role of modelling in the planning and prioritisation of clinical trials. To
appear in Health Technology Assessment.
Claxton, K. (1999). The irrelevance of inference: a decision-making

approach to the stochastic evaluation of health care technologies. Journal of
Health Economics 18, 341-364.

Claxton, K. and Posnett, J. (1996). An economic approach to clinical trial design
and research priority-setting. Health Economics 5, 513-524.
Cooke, R. M. (1991). Experts in Uncertainty: Opinion and Subjective Probability in

Science. Oxford: Oxford University Press.
Cooper, N. J., Sutton, A. J., Abrams, K. R. (2002). Decision analytical econom-

ic modelling within a Bayesian framework: application to prophylactic antibi-
otics use for caesarean section. Statistical Methods in Medical Research 11, 491-512.
Felli, C. and Hazen, G. B. (1998). Sensitivity analysis and the expected value of
perfect information. Medical Decision Making 18, 95-109.
Kadane, J. B. and Wolfson, L. J. (1998). Experiences in elicitation. The

Statistician 47, 1-20.
Lichtenstein, S., Fischhoff, B. and Phillips, L. D. (1982). Calibration of probabil-

ities: the state of the art to 1980. In Judgement Under Uncertainty: Heuristics and
Biases, D. Kahneman, P. Slovic. and A. Tversky (Eds.) Cambridge University
Press: Cambridge, pp. 306-334.
Löthgren, M. and Zethraeus, N. (2000). Definition, interpretation and calcula-

tion of cost-effectiveness acceptability curves. Health Economics 9, 623-630.
Meyer, M. and Booker, J. (1981). Eliciting and Analyzing Expert Judgement: A

Practical Guide, volume 5 of “Knowledge-Based Expert Systems”. Academic
Press.
Oakley J. (2002). Value of information for complex cost-effectiveness models.

Research Report No. 533/02, Department of Probability and Statistics,
University of Sheffield.
O’Hagan, A. and Stevens, J. W. (2001a). A framework for cost-effectiveness

analysis from clinical trial data. Health Economics 10, 302-315.
O’Hagan, A. and Stevens, J. W. (2001b). Bayesian assessment of sample size for

clinical trials of cost-effectiveness. Medical Decision Making 21, 219-230.
O’Hagan, A. and Stevens, J. W. (2002). Bayesian methods for design and analy-
sis of cost-effectiveness trials in the evaluation of health care technologies.
Statistical Methods in Medical Research 11, 469-490.

O’Hagan, A., Stevens, J. W. and Montmartin, J. (2000). Inference for the cost-
effectiveness acceptability curve and cost-effectiveness ratio. PharmacoEconomics
17, 339-349.
Parmigiani, G. (2002). Measuring uncertainty in complex decision analysis

models. Statistical Methods in Medical Research 11, 513-537.
Stevens, J. W., O’Hagan, A. and Miller, P. (2003). Case study in the Bayesian
analysis of a cost-effectiveness trial in the evaluation of health care technolo-
gies: Depression. Pharmaceutical Statistics 2, 51-68.
Stevenson, M. D., Oakley, J. and Chilcott, J. B. (2002). Gaussian process mod-

elling in conjunction with individual patient simulation modelling. A case study
describing the calculation of cost-effectiveness ratios for the treatment of osteo-
porosis. Technical report, School of Health and Related Research, University of
Sheffield.
Sutton, A. J., Lambert, P. C., Billingham, L., Cooper, N. J. and Abrams, K. R.

(2003). Establishing cost-effectiveness with partially missing cost components.
Technical report, Department of Epidemiology and Public Health, University of
Leicester.
Thisted. R. A. (1988). Elements of Statistical Computing. Chapman and Hall; New

York.
Van Hout, B. A., Al, M. J., Gordon, G. S. and Rutten, F. (1994). Costs, effects
and C/E ratios alongside a clinical trial. Health Economics 3, 309-319.
Further reading on Bayesian statistics
Preparatory books.
These books cover basic ideas of decision-theory and personal probability.
Neither book assumes knowledge of mathematics above a very elementary
level.
Lindley, D.V. (1980). Making Decisions, 2nd ed. Wiley, New York.
O’Hagan, A. (1988). Probability: Methods and Measurement. Chapman and Hall,

London.
Bibliography and Further Reading 45

Introductory books.
Berry’s book is completely elementary. Lee’s book is aimed at undergraduate
mathematics level.
Berry, D.A. (1996). Statistics: A Bayesian Perspective. Duxbury, London.
Lee, P.M. (1997). Bayesian Statistics: An Introduction, 2nd ed., Arnold, London.
More advanced texts.

At an intermediate level, Migon and Gamerman is a nice, readable but concise
book. Congdon concentrates on how to develop models and computations for
the practical application of Bayesian methods. The last two books, by Bernardo
& Smith, and by O’Hagan, are the most advanced, for people wishing to learn
Bayesian statistics in depth.
Migon, H. S. and Gamerman, D. (1999). Statistical Inference: An Integrated

Approach. Arnold, London.
Congdon, P. (2001). Bayesian Statistical Modelling. John Wiley.
Bernardo, J. M. and Smith, A. F. M. (1994). Bayesian Theory. Wiley, New York.
O’Hagan, A. (1994). Bayesian Inference, volume 2B of “Kendall’s Advanced

Theory of Statistics”. Arnold, London.
Philosophy
Finally, the following presents the case for Bayesian inference from the per-
spective of philosophers of science.
Howson, C. and Urbach, P. (1993). Scientific Reasoning, 2nd edition. Open Court;
Peru, Illinois.

Appendix
This Appendix offers more detailed discussion of the issues raised in the
first four sections of this Primer.
Inference – Details
The following subsections give details of the arguments presented in the
“Inference” section and of the key differences between Bayesian and frequentist
statistics identified in Table 1.
The nature of parameters

The most fundamental distinction between the Bayesian and frequentist
approaches is that in the Bayesian approach the unknown parameters in a sta-
tistical model are random variables. Accordingly, in a Bayesian analysis the
parameters have probability distributions, whereas in frequentist analysis they
are fixed but unknown quantities, and it is not permitted to make probability
statements about them.
This can be puzzling to the layman because a standard frequentist state-
ment like a confidence interval certainly seems to be making probability state-
ments about the unknown parameter. If we see the statement that [3.5, 11.6]
is a 95% confidence interval for a parameter µ, surely this is saying that there
is a 95% chance that µ lies between 3.5 and 11.6. No, it cannot mean this
because µ is not a random quantity in frequentist inference. We shall see what
it does mean when we discuss The nature of inferences.
In Bayesian statistics, the parameters do have probability distributions, and
if a Bayesian analysis produces a 95% interval then it does have exactly the
interpretation that is usually put on a confidence interval.

The nature of probability
Underlying this distinction between the two approaches to statistics over the
nature of parameters is a difference in how they interpret probability itself. In fre-
quentist statistics, a probability can only be a long-run, limiting, relative frequen-
cy. This is the familiar definition used in elementary courses, often motivated by
ideas like tossing coins a very large number of times and looking at the long-run,
limiting frequency of ‘heads’. It is because it is based firmly on this frequency def-
inition of probability that we call those traditional methods ‘frequentist’. Bayesian
statistics, in contrast, rest on an interpretation of probability as a personal degree
of belief. Although to some this may seem ‘woolly’ and unscientific, it is impor-
tant to recognize that Bayesian statisticians have widely and successfully devel-
oped analyses based on this interpretation. As explained in Prior Information, it
does not lead to unbridled subjectivity and unscientific practices.
The necessity of such an interpretation becomes clearer when we appreci-
ate how many events that we would be willing to consider having probabilities
could never have a frequentist probability. The probability of a hypothesis is an
obvious example, since a particular hypothesis either is or is not true, and we
cannot consider any experimental repetitions of it like we could for tossing a
coin. Some other examples are the probability of rain tomorrow or the proba-
bility that you will experience a myocardial infarction (MI) during your life-
time. Tomorrow is a unique day, with meteorological conditions preceding it
today that have not existed in precisely the same form before, and never will
again, and you are a unique person whose genetic makeup and lifestyle make
you more or less disposed to MI at some point in your life in a way not precisely
matched by anyone else. Despite their uniqueness, we generally have no diffi-
culty in thinking of these events as having probabilities. Most of the uncertain
events and variables of real interest to scientists and practitioners are one-off
things, and the frequency interpretation of probability is completely unable to
accommodate our wish to describe them by probabilities.
The nature of inferences

Probabilities in the frequentist approach must be based on repetition. The
statement that [3.5, 11.6] is a 95% confidence interval for a parameter µ says

that if we repeated this experiment a great many times, and if we calculated an
interval each time using the rule we used this time to get the interval [3.5,
11.6], then 95% of those intervals would contain µ. The 95% probability is a
property of the rule that was used to create the interval, not of the interval
itself. It is simply not allowed, and would be wrong to attribute that probabili-
ty to the actual interval [3.5, 11.6]. This is a very unintuitive argument.
Frequentist statements such as these are widely misinterpreted, even by trained
statisticians. The erroneous interpretation of the confidence interval, that there
is a 95% chance of the parameter µ lying in the particular interval [3.5, 11.6],
is almost universally made by statistician and non-statistician alike.
Nevertheless, it is incorrect. A Bayesian interval, however, does have that inter-
pretation.
The Bayesian approach uses different terminology from the familiar fre-
quentist terms. Bayesian intervals are generally called credible intervals to make it
clear that they are different from confidence intervals. The highest density interval
is a credible interval with the property of being shortest among all available cred-
ible intervals with a given probability of containing the parameter’s true value.
An entirely similar argument can be made about frequentist significance
tests. If a hypothesis is rejected with a P-value of 1%, this does not mean that
the hypothesis has only a 1% probability of being true. Hypotheses do not have
probabilities in a frequentist analysis any more than parameters do. The P-value
must again be based on repetition of a rule. In this case it is a rule which says
that when the data satisfy some condition we will formally reject the hypothe-
sis. The P-value’s proper interpretation is that if we repeated the experiment
many times, and if the hypothesis really were true every time, then on only 1%
of such experiments would the rule lead us to (wrongly) reject that hypothesis.
This is a tricky and convoluted idea. It is not surprising that practitioners regu-
larly misinterpret a P-value as the probability that the hypothesis is true.
To interpret a P-value in this way is not only wrong but also dangerously
wrong. The danger arises because this interpretation ignores how plausible the
hypothesis might have been in the first place. Here are three examples.
Appendix 49
Examples.
1) Screening. Consider a screening test for a rare disease. The test is very
accurate, with false-positive and false-negative rates of 0.1% (i.e. only
one person in a thousand who does not have the disease will give a
positive result, and only one person in a thousand with the disease will
give a negative result). You take the screen and your result is positive.
What should you think? Since the screen only makes one mistake in
a thousand, doesn’t this mean you are 99.9% certain to have the dis-
ease? In hypothesis testing terms, the positive result would allow you
to reject the null hypothesis that you don’t have the disease at the
0.1% level of significance, a highly significant result agreeing with that
99.9% diagnosis. But the disease is rare, and in practice we know that
most positives reporting for further tests will be false positives. If only
one person in 50,000 has this disease, your probability of having it
after a positive screening test is less than 1 in 50.
Although this example may not be obviously concerned with hypoth-

esis testing, in fact there is a direct analogy. We can consider using the
observation of a positive screening outcome as data with which to test
the null hypothesis that you do not have the disease. If the null
hypothesis is true, then the observation is extremely unlikely, and we
could formally reject the null hypothesis with a P-value of 0.001. Yet,
the actual probability of the null hypothesis is more than 0.98. This is
a dramatic example of the probability of the hypothesis (> 0.98) being
completely different from the P-value (0.001). The difference clearly
arises because the null hypothesis of you having the disease begins
with such a low prior probability. Nobody who is familiar with the
nature of screening tests would be likely to make the mistake of inter-
preting the false positive rate as the probability of having the disease
(but it is important to make the distinction clear to patients!) By the
same token, it is wrong to interpret a P-value as the probability of the
null hypothesis, because this fails to take account of the prior proba-
bility in exactly the same way.

2) Subgroup analysis. Statisticians are continually warned against trawl-
ing through the data for significant subgroup effects. In clinical trials,
subgroup analyses are generally only permitted if they were prespeci-
fied and have a plausible biological mechanism. This is a clear case
where it is recognized that the interpretation of significance depends
on how plausible the hypothesis was in the first place.
3) Drug development. Pharmaceutical companies synthesize huge num-

bers of compounds looking for clinical effects. By analogy with the
screening example, even after a trial has produced an effect that is
highly significant, the probability that this effect is real may not be
large; we expect false positives. Despite this, and despite the experience
of many drugs going into expensive Phase III trials only to prove inef-
fective, companies continue to wrongly and over-optimistically inter-
pret significant P-values.
In contrast to the frequentist approach, a Bayesian analysis can give a prob-

ability that a hypothesis is true or false. A Bayesian hypothesis test does just that
– reports the probability that the hypothesis in question is true. A proper
Bayesian analysis would correctly identify the probabilities in all the above
examples.
Frequentist point estimates are also based on repetition, and although they
are less often misinterpreted in Bayesian ways, there are still important differ-
ences. Suppose that a frequentist analysis reports that an unbiased estimate of
µ is 9.1. Unbiasedness is a property of the rule of estimation, not of the estimate
itself, and in this case means that if we repeated the experiment many times
and applied the same rule each time to produce an estimate of µ, then the aver-
age value of these estimates would be µ itself. On average, the estimates will be
neither too high nor too low, but nothing is or can be said about whether 9.1 is
expected to be too high or too low. A Bayesian analysis might report that 9.1 is
the expected value of µ and has precisely the interpretation that 9.1 is not
expected to be too high or too low as an estimate of µ.
Appendix 51
Example.
Consider the rate of side effects from a drug. In a trial with 50 patients,
we observe no side effects. The standard unbiased estimator of the side
effect rate per patient is now zero (0/50). In what sense can we believe
that this is “on average neither too high nor too low”? It obviously can-
not be too high and is almost certain to be too low. It is true that the
estimation rule (which is to take the number of patients with side
effects and divide by 50) will produce estimates that on average are
neither too high nor too low if we keep repeating the rule with new
sets of data. It is also clear, though, that we cannot apply this interpre-
tation to the individual estimate. To do so is like interpreting a P-value
as the probability that the null hypothesis is true; it is simply incorrect.
In any Bayesian analysis, given no side effects among 50 patients, the
expected side effect rate would be positive. Furthermore, the posterior
expectation has the desired interpretation that this estimate is
expected to be neither too high nor too low.
More natural and useful inferences

In frequentist inferences the words ‘confidence’, ‘significance’ and ‘unbi-
asedness’ are technical terms, and it is important to interpret them according
to their definitions.
The fact that practitioners invariably, but incorrectly, wish to interpret a
confidence interval as making a probability statement about the parameter is
evidence that the Bayesian approach is more intuitive and natural and gives
more direct answers to the client’s questions. It is similarly tempting to interpret
a P-value as the probability that a hypothesis is true, because this is exactly what
the practitioner wants to hear. Again, the Bayesian inference naturally and
directly addresses the practitioner’s needs.
This is the key benefit (B1) – “more natural and useful inferences” – of the
Overview section. It is easy to see how it becomes a very real benefit for the
health economist. For instance, one widely used way of presenting a cost-effec-
tiveness analysis is through the Cost-Effectiveness Acceptability Curve (CEAC),

introduced by van Hout et al (1994). An example is shown in Figure A1. For
each value of the threshold willingness to pay λ, the CEAC plots the probabili-
ty that one treatment is more cost-effective than another.
It should already be clear to the reader that this probability can only be mean-
ingful in a Bayesian framework. It refers to the probability of a one-off event (the
relative cost-effectiveness of these two particular treatments is one-off, and not
repeatable), and that event is expressed in terms of the unknown parameters of
the statistical model used to analyze the available evidence. We note that it is pos-
sible to construct a frequentist alternative CEAC, defined in terms of P-values
(O’Hagan et al, 2000; Löthgren and Zethraeus, 2000), and which plots for each
value of λ, the probability that the data would have fallen into a set bounded by
the observed data, assuming the truth of the hypothesis that the two treatments
are equally cost-effective. However, it would seem rather perverse to adopt that
frequentist approach when a Bayesian analysis yields the far more direct and use-
ful CEAC that plots the probability that treatment 2 is more cost-effective than
treatment 1.
FIGURE A1. CEAC plot.

Probability that new treatment is more cost-effective than placebo,
plotted against willingness to pay (in units of $10,000/QALY).
1
0.8
0.6
0.4
0.2
0 0.5 1 1.5 2 2.5 3
Appendix 53
The Bayesian Method – Details
The following subsections give more details of the Bayesian method.
Bayes’ theorem
The simplest way to express Bayes’ theorem without using mathematical
notation is this:
The posterior is proportional to the prior times the likelihood.
Several terms in this statement need to be defined and explained. It will be

useful to refer to Figure 1 in the box ‘Example of Bayes’ theorem’. ‘The poste-
rior’ means the posterior distribution of the unknown parameter(s). Strictly, it
is the posterior probability density function, which is shown as the black dotted
curve in Figure 1. Similarly, ‘the prior’ means the prior distribution of the
unknown parameter(s), also in the form of the prior probability density func-
tion and shown as the grey line in Figure 1. The ‘likelihood’ is the common fac-
tor in both frequentist and Bayesian theory. It represents the information in the
data and is shown as the red curve in Figure 1. Formally, for any given value of
the unknown parameter, the likelihood plots the probability of observing the
data that were actually observed. Bayes’ theorem states that we should multi-
ply the two curves. Since the area under any probability density curve must be
equal to one we scale the product to satisfy this condition, which Bayes’
Theorem expresses by saying that the posterior is ‘proportional to’ the product.
Thus, the black dotted curve in Figure 1 results from multiplying the grey and
red curves and scaling the result so that the area under it is equal to one.
This mechanism of multiplying the two curves also makes it clear that
Bayes’ theorem weights each source of information according to its strength.
Consider the situation in which the prior information is very weak. This would
be represented by a very flat grey curve, giving more or less equal prior proba-
bility to a wide range of parameter values. When we apply Bayes’ theorem,
then the posterior becomes almost a constant times the likelihood, and because

it must be scaled to integrate to one, the posterior is in effect the same as the
red curve. This is shown in Figure A2, where we have weakened the prior
information relative to Figure 1.
When the prior information is very weak, relative to the data information,
the prior distribution gets so little weight in Bayes’ theorem that the posterior
distribution is effectively just the likelihood. In this situation we might expect,
and in simple problems often find, that Bayesian methods lead to similar infer-
ences to conventional frequentist methods. Bayesian methods make use of
more information than frequentist methods, but give each source of informa-
tion its due weight; weak information is naturally downweighted.
FIGURE A2. Triplot with weaker prior information.
0.4
0.3
0.2
0.1
-4 -2 0 2 4
Another useful feature of the Bayesian paradigm that is worth mentioning

and nicely captured in a simple phrase is:
Today’s posterior is tomorrow’s prior.
The paradigm is about learning, and we can always learn more. When we
acquire more data, Bayes’ theorem tells us how to update our knowledge to
synthesize the new data. The old posterior contains all that we know before see-
Appendix 55
ing the new data, and so becomes the new prior distribution. Bayes’ theorem
synthesizes this with the new data to give the new posterior. And on it
goes…Bayesian methods are ideal for sequential trials!
Bayes’ theorem also makes it more clear as to why the common misinter-
pretation of frequentist inferences is wrong. The likelihood expresses the prob-
ability of obtaining the actual data, given any particular value of the parameter.
In simple terms,
Likelihood = P (data | parameters).
This distribution is the basis of frequentist inference. On the other hand,

the basis of Bayesian inference is the posterior distribution, which is the proba-
bility distribution of the parameters, given the actual data,
Posterior = P (parameters | data).
It is the unjustified switching around of parameters and data that leads to

misinterpretations. For instance, a P-value is P (data | hypothesis), whereas
what a decision-maker wants, and what Bayesian inference provides, is P
(hypothesis | data). It is clear that the two are quite different things, and Bayes’
theorem shows the relationship between them: we can only derive the poste-
rior probability we want by combining the P-value with the prior distribution.
Bayesian inference
In the Bayesian approach, all inferences are derived from the posterior dis-
tribution. When a Bayesian analysis reports a probability interval (a credible
interval) for a parameter, this is a posterior interval, derived from the para-
meter’s posterior distribution, based not only on the data but also on whatever
other information or knowledge the investigator possesses. The probability that
a hypothesis is true is a posterior probability and a typical example of an esti-
mate of a parameter would be the posterior mean (the ‘expected’ value).
These are the Bayesian analogues of the three kinds of inferences that are
available in the frequentist framework. However, Bayesian inference is much
more flexible than this.

Often the real question of interest does not fit one of these frequentist
inference modes. For instance, the investigator frequently wants to know
“What do we now know about this parameter, after seeing the data?” There is
no straightforward frequentist answer to that very natural question. The
Bayesian answer is simplicity itself – we plot the posterior density. Thus, the
black dotted curve in Figure 1 fully expresses what is known about that param-
eter after synthesizing all of the available evidence.
Decision theory provides another good example of the flexibility of
Bayesian inference. In this theory we have a set of possible decisions and a util-
ity function that specifies how good it would be to make a particular decision if
the parameters turned out to have particular values. For instance, for a hypoth-
esis test we could define a utility function that said it would be good (high util-
ity) to accept the hypothesis if it turned out to be true, or to reject it if it turned
out to be false, but otherwise the utility would be low.
If we knew the parameters it would be easy to arrive at a decision – we
would just choose the decision with the largest utility for those values of the
parameters. However, the parameters are generally unknown. Decision theory
says we should choose the decision with the highest (posterior) expected util-
ity. This expectation is the average value of the utility, averaged with respect to
the posterior distribution of the parameters. This is a technical statement, but
the point is that there is no frequentist way to find that optimal decision. It is
essentially a Bayesian construct and is yet another way that the posterior dis-
tribution allows us to answer the real question of interest.
For a health care provider (HCP) choosing between two alternative drugs,
the net benefit is an appropriate utility function. The net benefit (strictly, the
net monetary benefit) of a given drug is obtained by taking the drug’s mean effi-
cacy, multiplying it by the price that the HCP is willing to pay for a unit increase
in efficacy, and then subtracting the mean cost to the HCP of using this drug.
The rule of maximizing expected utility then implies that the HCP should
choose the drug with the larger expected net benefit. Equivalently, it should
choose drug 2 if the expected incremental net benefit over drug 1 is positive
(Claxton, 1999).
Appendix 57
Prior Information – Details
The following subsections give more details of the discussion presented in
the Prior Information section of the main text.
Subjectivity
The fact that Bayesian methods are based on a subjective interpretation of
probability was introduced in the subsection The Nature of Probability in this
Appendix. We explained there that this formulation is necessary if we are to
give probabilities to parameters and hypotheses since the frequentist interpre-
tation of probability is too narrow. Yet this leaves Bayesian methods open to
the charge of subjectivity, which is thought by many to be unacceptable and
unscientific.
Yet science cannot be truly objective. Schools of thought and contention
abound in most disciplines. Science attempts to minimize subjectivity through
the use of objective data and reason, but where the data are not conclusive we
have to use our judgement and expertise.
Bayesian methods naturally accommodate this approach. Figure 1 demon-
strates how Bayes’ theorem naturally weights each information source accord-
ing to its strength. In that example, the data were only slightly more informa-
tive than the prior, and so the posterior is quite strongly influenced by the prior
as well as by the likelihood.
We also saw in Figure A2 how if the prior information is weakened then
Bayes’ theorem effectively places all of the weight on the likelihood. Often we
are in the more fortunate position of having strong data. Then the situation will
be more like the triplot in Figure A3. As new data accumulate the prior distri-
bution again becomes less influential.

FIGURE A3. Triplot with more informative data
1.2 1.2
1 1
0.8 0.8
0.6 0.6
0.4 0.4
0.2 0.2
-4 -2 0 2 4 0 0.5 1 1.5 2 2.5 3
(a) on the same scale as Figure A2 (b) with scale changed to show
difference between likelihood
and posterior.
In this case, the data are based on 10 times as much information as in

Figure 1. The red likelihood curve is much narrower than the grey prior densi-
ty. The prior contributes very little information to the synthesis, and the poste-
rior density (black dotted curve) is almost identical to the likelihood. When the
data are sufficiently strong they will outweigh any subjective prior information.
Although different experts in the field might bring different prior knowledge
and opinions, the data will outweigh their priors, and they will all agree close-
ly on the posterior distribution. This is an excellent model for science.
What if the data are not conclusive in this way? Then different experts in
the field will have materially different posterior distributions. We do not have
consensus, although their differences will generally have been lessened by the
data. It is actually a strength of the Bayesian approach that by considering the
consequences of using different prior distributions we can see whether the data
are adequate to outweigh those differences. If the data are not strong enough,
then it would be misleading to present any analysis as if it were definitive.
Appendix 59
Whose prior?
Suppose that a sponsor of some medical technology (e.g. a pharmaceutical
company or device maker) wishes to present a Bayesian analysis in support of
a case for cost-effectiveness of that technology. What would be acceptable in the
form of prior information? One way to approach this question is to ask whose
prior should be used.
As shown above, it may not matter. Consensus can be reached if the data
are strong enough to overrule the differences in the prior opinions and knowl-
edge of all interested parties. However, the argument does not work if one per-
son has sufficiently strong prior information or opinions. It only takes one
extremely opinionated person to prevent agreement. We are familiar with the
person whose views on some matters are so prejudiced that they will not listen
to any facts or arguments to the contrary – Bayes’ theorem explains these peo-
ple, too!
This clarifies some aspects of subjectivity. While we should accept that dif-
ferent people might legitimately have different background knowledge or might
legitimately interpret the same information differently, there is no place for
prejudice or perverse misinterpretation of the available facts in health econom-
ics (or anywhere else). An important aspect of Bayesian analysis is that the prior
distribution is set out openly. If it is not based on reasonable use of information
and experience, the resulting analysis will not convince anyone. This is the key
benefit (B5) – “more open judgements” – of Bayesian analysis. All cards should
be on the table and nothing hidden.
Returning to the question of whose prior a sponsor might use, it is likely
that the sponsor’s own prior distribution would be unacceptable. In principle,
their opinions might be defensible on the basis of the company’s own substan-
tial experiences in development and testing of the product, but there is the risk
of selective use of information unless full disclosure can be enforced. To quote
the box “The Evidence” in the main text, the sponsor would need to be able to
show that its prior not only represented the evidence but also the whole evi-
dence.
The most natural choice of prior distribution might be the considered and
defensible prior of an expert in the field. The agency to which the cost-effec-

tiveness case is being made would probably be interested in whether the infer-
ences might change if the views of alternative experts were considered. Ideally,
some kind of consensus view of the profession would be a good choice.
Examples
These ideas are illustrated in the following two examples, which are dis-
cussed briefly in the main text.
Subset analysis is a notoriously tricky question. The risk of dredging the
TABLE 2. Mean costs in hypothetical trial
NAMES A-D NAMES E-Z

Treatment 1 $800 $800
Treatment 2 $450 $850
data to find subgroups of patients that respond differently is real. For example,
suppose that a cost study found the following mean costs table (Table 2), split
by treatment and the initial letter of the patient’s surname.
It looks like treatment 2 is cheaper for patients whose names begin with the
letters A to D. There is highly unlikely to be any plausible reason for such a sub-
group effect. To avoid the risk of declaring spurious subgroup effects, standard
clinical trials guidance requires that the analysis of possible subgroups must be
specified before a trial begins, and there must be a plausible mechanism for the
proposed subgroup effects.
From a Bayesian perspective, the absence of a plausible mechanism simply
constitutes prior information – the subgroup effect would have a very small
prior probability. Combining the prior information with the data would result
in a small posterior probability, regardless of how convincing the data appeared
to be. The prior information is strong enough to override the data. The standard
guidance, therefore, applies equally to Bayesian analyses; subgroup analyses
must be prespecified, so that prior information about their plausibility can be
quantified.
Appendix 61
Bayesian methods will then automatically moderate the data and prevent
us from claiming implausible effects that arise by chance in the data. The prior
information is clearly important. To an extent, the existing procedures for fre-
quentist subgroup analysis incorporate the prior information (and so, we would
argue, are unconsciously Bayesian). However, the frequentist analysis simply
splits subgroup hypotheses into those that are plausible a priori and those that
are not, whereas the Bayesian assigns a prior probability that can take any value
from 0 to 1, and thereby allows a far more subtle gradation.
Hospitalization. Suppose that, after evaluating good evidence on efficacy of

a new drug relative to the standard treatment, a decision on whether it is more
cost-effective rests on whether it reduces hospitalizations. Here the data come
from a trial in which the number of days in hospital was recorded for each of
100 patients in each treatment group. A total of 25 hospital days were record-
ed in the standard treatment group and only 5 in the group receiving the new
drug. In frequentist terms, the difference is found to be significant at the 5%
level (one-sided). (We will not give technical details of calculations, but all of
this, and subsequent analysis, may be reconstructed using the additional infor-
mation that the sample variances were 1.2 in the standard treatment group and
0.248 in the new drug group.) The drug company could then, in conventional
frequentist terms, claim an effect on hospitalization, estimate the mean number
of days per patient as 0.25 under standard treatment and 0.05 under the new
drug, perhaps with the result that the new drug is now found to be more cost-
effective than standard treatment.
This is, however, a rather small trial and the data are far from conclusive.
If other evidence were available it would be prudent to incorporate it in the
analysis. Suppose that a much larger trial (in comparable conditions) of a simi-
lar drug produced a mean number of days in hospital per patient of 0.21, and
that the standard error of this estimate is only about 0.03. This extra informa-
tion suggests that the observed rate of 0.05 for the new drug is optimistic and
casts doubt on the magnitude of the real difference between it and standard
treatment. However, the interpretation of this evidence is necessarily judgmen-
tal. Nobody would claim that the two drugs should necessarily yield the same

hospitalization rates, but it is reasonable to suppose that they would not be
markedly different. Because they cannot be treated as completely comparable
with the trial data on the new drug, we cannot treat this other, larger trial as
part of the data and just merge it with the new data. There is no apparent way
to use the evidence on the related drug in a frequentist analysis. Yet any clini-
cian or health care provider who was aware of this external evidence would be
disinclined to take the new trial evidence at face value.
A Bayesian analysis resolves the question by treating the earlier trial as pro-
viding prior information but entails an element of subjectivity. Suppose that
your interpretation of the prior information is that your prior expectation of the
mean days in hospital for the new drug should be 0.21 but with a standard devi-
ation of 0.08 to reflect the fact that the two drugs are not the same. Bayes’ the-
orem now yields for you a posterior estimate of 0.095 for the mean hospital
days using the new drug. There is still a reasonably strong probability (90%)
that the new drug reduces this hospitalization rate, but now the estimated dif-
ference may not be large enough to provide the same assurance that it is more
cost-effective than standard treatment.
The subjectivity in this analysis arises in the necessary judgement about
how different the hospitalization rates might be for the two drugs. Another cli-
nician or decision-maker might interpret the prior information differently and
employ a different prior distribution. In particular, they may have a different
prior standard deviation. In fact this answer is fairly robust to reasonable
changes in the prior distribution. On that basis we might conclude that an
‘objective’ interpretation of the combined data is that there is a strong proba-
bility (perhaps around 90% but not as high as 95%) that the new drug reduces
mean days in hospital but that it achieves a mean number of days nearer to 0.1
than to 0.05.
This example has shown how a Bayesian analysis, making use of genuine
prior information and considering a range of reasonable interpretations of that
information, can produce a scientifically sound conclusion. The answer differs
substantially from the frequentist analysis which has no technical way of mak-
ing use of the extra information. The Bayesian answer is also sound because it
formalizes the natural intuitive reaction that anyone would have to the fre-
Appendix 63
quentist analysis, knowing the result of the other trial.
In this and previous examples, we have stressed the power of the Bayesian
approach to temper overly optimistic interpretations of P-values, but it is impor-
tant to recognize that the reverse situation is equally common and important.
Pharmaceutical company executives and biostatisticians will be very familiar
with occasions where a Phase III trial of a drug has just failed to produce a sig-
nificant effect, yet there is plenty of evidence (from related drugs, from a Phase
II trial that was restricted to acute cases, etc.) to suggest that the drug really is
effective. A properly conducted Bayesian analysis would allow the responsible
incorporation of this additional evidence to demonstrate the drug’s true efficacy.
Both situations are of enormous importance to the developers and users of
health care technologies –the first in avoiding costly mistakes due to being over-
ly optimistic and the second in allowing beneficial products to be brought to mar-
ket that otherwise would have to be abandoned or delayed for more testing.
Prior Specification – Details

The following subsections give details of three aspects of prior specification;
the elicitation of expert priors, conjugate priors and the construction of struc-
tural priors.
Elicitation
We will consider the process of eliciting a prior distribution for an expert
without reference to the actual nature of the underlying prior information. In
practice, of course, the expert will base her analysis on that information, but in
this subsection we will not try to deal with the specifics of the underlying
information.
Suppose that we decide to formulate a prior distribution for a particular
parameter (such as the mean utility gain arising from some treatment), repre-
senting the knowledge of a single expert about that parameter. The first diffi-
culty we will face is that the expert will almost certainly not be an expert in
probability or statistics. That means it will not be easy for this person to express
her beliefs in the kind of probabilistic form demanded by Bayes’ theorem.
Our expert might be willing to give us an estimate of the parameter, but

how do we interpret this? Should we treat it as the mean (or expectation) of the
prior distribution, or as the median of that distribution, as its mode, or something
else? In statistics, the mean, median and mode might all be regarded as sound
point estimates of a quantity, but they are different things. The mean is the
‘expected value’, the median is the ‘central value’ and the mode is the ‘most like-
ly value’. In principle, we might explore with the expert these nuances of mean-
ing, but for someone not trained in statistics it will not be easy for her to appre-
ciate the differences and give us a reliable interpretation for her estimate.
We could go on to elicit from the expert some more features of her distri-
bution, such as some measure of spread to indicate her general level of uncer-
tainty about the true value of the parameter. (Remember, the strength of infor-
mation is indicated by the narrowness of the distribution representing that
information.)
The next difficulty is that no matter how much information of this type we
extract from the expert it will not be enough to identify her distribution exact-
ly. This is easy to see when we recognize that to uniquely identify the grey curve
in Figure 1 representing the prior distribution, we must specify its height at
every single point – potentially an infinite number of facts must be obtained
from the expert. To compound this problem, the more detail we ask for, the
more difficult it is for the expert. In practice, the best we can do is to elicit a few
simple expressions of her knowledge, in the form of things like median and
quartiles and then fit some sensible distribution to those statements.
Even the judgements that we elicit from the expert cannot be treated as
precise. Suppose that the expert gives us the estimate of 0.85 for the parame-
ter, and we interpret this as the mean of her prior distribution. Even if we are
right to interpret it in that way, we cannot realistically treat it as a precise num-
ber. Our expert almost certainly gave us a round figure and couldn’t say
whether 0.86 might be a more accurate reflection of her prior mean. The read-
er should now amply appreciate the criticism (D2), that “prior specification is
unreliable”.
Nevertheless, there is a growing body of research into how to elicit experts’
knowledge accurately and reliably. The difficulties that people face in assessing
probabilities have been extensively studied, particularly by cognitive psycholo-
Appendix 65
gists; some useful reviews can be found in Lichtenstein et al (1980), Meyer and
Booker (1981), Cooke (1991), Kadane and Wolfson (1998). Although the psy-
chologists have tended to emphasize the tasks that people conceptualize poor-
ly, the practical significance of this work is that we know a great deal about how
to avoid the problems. This and ongoing research seeks to identify the kinds of
questions that are most likely to yield good answers, avoiding pitfalls that have
already been identified by psychologists and statisticians, and the kind of feed-
back mechanisms that help to ensure good communication between statistician
and expert.
The second answer is that, fortunately, the imprecision of distributions
elicited from experts may not matter much. As discussed earlier, we can think
of a range of prior distributions that are consistent with the statements we have
elicited from the expert, and if the data are sufficiently strong then all these dif-
ferent specifications of the prior distribution will lead to essentially the same
posterior distribution. The box “Example of elicitation” explores these ideas.
Example of Elicitation
An expert estimates a relative risk 1.2
(RR) parameter to be about 50%, but 1
has considerable uncertainty about its
0.8
true value. She says that it is unlikely
to be less than 0.2 or greater than 0.6
1.5. The distribution to the right fits 0.4
those statements, but so would many 0.2

other distributions. The question is 0
whether, if we tried those other distri- 0.5 1 1.5 2 2.5 3
butions in a Bayesian analysis of

some data, they would give materially different posterior inferences.
Conjugate priors
As explained in the preceding discussion, the usual approach to specifying
a prior distribution for some parameter consists of first specifying (or eliciting)
a few features of the distribution, such as a prior expectation and some meas-

ure of prior uncertainty (e.g. the prior variance), then choosing a suitable dis-
tribution to fit these features. In the box “Example of elicitation”, for instance,
a particular form of distribution known as a gamma distribution has been fitted
to the expert’s two specific statements.
This may seem rather cavalier and arbitrary, but in practice the prior distri-
bution is often quite well determined even if only a few actual features have
been specified or elicited from the expert. Although the actual distribution cho-
sen is arbitrary, any other reasonable prior distribution that fits the specifica-
tions is likely to be very similar, and hence to lead to very similar inferences or
decisions.
It is then sensible to make the choice of distributions on grounds of sim-
plicity and convenience. Mathematically, in some simple statistical problems
there exist classes of priors known as conjugate priors that are particularly con-
venient. This is because of two features. First, if the prior distribution is a mem-
ber of the relevant conjugate class then the posterior distribution will also be a
member of that class. Second, the conjugate distributions are sufficiently simple
for us to be able to derive a great many inferences from them without resort to
computational methods. Whenever the statistical model is such that a conjugate
family exists and a member of that family fits the prior specification, then it is
particularly convenient to choose that distribution. Very simple posterior analy-
sis then follows.
Indeed, in the early days of modern Bayesian statistics, in the 1960s and
1970s, Bayesian analysis was essentially restricted to the use of conjugate pri-
ors, since computational tools did not exist to tackle more complex situations.
They are now much less used because statisticians are building models that do
not have corresponding conjugate priors, and the desire for more realistic for-
mulation of prior information also means that conjugate priors may not fit even
when they are available.
Structural priors
Structural prior distributions express information about relationships
between parameters, usually without saying anything about the specific values
of individual parameters. For instance, we might specify a prior distribution that
Appendix 67
represents effective ignorance about the mean cost under two different treat-
ments, but says that we expect the ratio of these means to be in the range 0.2
to 5. The mean cost under any given treatment could be anything at all, but
whatever value it actually takes we expect the mean cost under the other treat-
ment will be within a factor of 5 of this.
A simple example is the prior distribution in the example of hospitaliza-
tion. This can be dissected into two parts. We have substantial prior informa-
tion about mean days in hospital under the related drug, and we have struc-
tural prior information about how it might differ from the corresponding mean
hospitalization under the new drug. Indeed, if the raw data of the earlier trial
are available we might formally analyze these as data with the results of the
new trial. In that case, the prior information is purely structural. The frame-
work now looks a little like a meta-analysis, and indeed Bayesian meta-analy-
sis is based strongly on structural prior information.
In a Bayesian meta-analysis, we have several datasets, each of which
addresses the efficacy of a treatment in slightly different conditions. So we
have a separate parameter for mean efficacy in each trial, but we formulate a
structural prior representing the prior expectation that these parameters
should not be too different. This is usually done in a hierarchical model,
where a common ‘underlying’ mean efficacy parameter is postulated, and each
of the trial mean efficacies is considered to be independently distributed
around this common parameter.
The hierarchical structure, introducing one or more common parameters,
is often used to link several related parameters and to express a belief that they
should be similar via the fact that they should all be similar to the common
parameter. Another example is to formulate structural prior information that
cost data arising in different arms of a trial should not be markedly different in
their degrees of skewness. This has the benefit of moderating the influence of
a very small number of patients with unusually high costs.
Computation – Details
The following subsections give details of Bayesian computation and its
ability to address very complex models.

Complexity
The source of the extra complexity in Bayesian analysis is again the prior
distribution. Suppose that the problem is the simple, canonical, statistical prob-
lem of estimating the mean of a normal distribution (with known variance),
given a sample from that distribution. To the frequentist, this is a complete spec-
ification of the problem, and in principle there can be a single answer (and in
fact the sample mean is universally accepted as the best estimate). To the
Bayesian, the specification is incomplete because we also need to state the prior
distribution. Then the answer will synthesize the two information sources and
will depend on the prior. Given this simple problem, the Bayesian approach can
give a very wide range of answers.
This makes the construction of Bayesian software very difficult. The software
itself must be more complex. The reason for this is because it must allow for spec-
ification of the prior (which could be any distribution at all) and must be able to
compute the desired inferences no matter what that prior distribution may be.
Mathematically, Bayesian inferences usually require us to be able to inte-
grate the product of the prior and the likelihood. We have to do this, for
instance, just to find the area under the curve, so as to know how to scale it to
make that area 1. Just applying Bayes’ theorem demands integration. More
integration is then needed, for instance, to find the posterior mean or the prob-
ability that a hypothesis is true. Now even if the prior and the likelihood are
individually quite nice functions whose integrals are well known, the product
will almost invariably be sufficiently complex for us to be unable to derive the
necessary integrals by mathematical principles. So these integrals need to be
done numerically (for an introduction to the ideas and methods of numerical
integration, see Thisted, 1988). The main reason why Bayesian methods were
impractical until the 1990s was that we did not have effective integration algo-
rithms. The reason for the subsequent explosion of Bayesian applications is that
a very powerful, general solution was developed.
MCMC
The technique that has revolutionized Bayesian computation is Markov
chain Monte Carlo (MCMC). The idea of MCMC has been outlined in the main
Appendix 69
text: Bayesian inference is solved by randomly drawing a very large sample
from the posterior distribution. Any inference we want to obtain from the pos-
terior distribution we can calculate from the sample.
At this stage, it may be helpful to emphasize that we are not talking about
the sample data. (Usually, we have little control over how many data we can
get, and we don’t expect to have such an enormous sample that we can calcu-
late anything we like from the sample in this simple way.) Instead, we are talk-
ing about artificially generating a sample of parameter values, by some random
simulation method that is somehow constructed to deliver a sample from the
posterior distribution of those parameters.
Strictly speaking, the idea of drawing a large sample from the posterior dis-
tribution is called Monte Carlo. Monte Carlo simulation is used very widely in
science as a powerful indirect way of calculating things that direct mathemati-
cal analysis cannot solve. What makes MCMC different is the way the sample
is drawn. Simple Monte Carlo can be visualized as playing darts – ‘throwing‘
points randomly into the space of all possible values of the parameters, with
each point independent of the others. This approach is impractical for Bayesian
analysis because in a model with many parameters it is extremely difficult to
construct an efficient algorithm for randomly ‘throwing‘ the points according to
the desired posterior distribution. MCMC operates by having a point wander-
ing around the space of possible parameter values. See Figure A4.
FIGURE A4. Sampling from the posterior distribution
1
3
2
5 6
6 4 5
3
4
2
Monte Carlo MCMC

Technically, a probability model called a Markov chain generates this wan-
dering point. Successive steps in this Markov chain produce the sample. It turns
out that it is really quite simple to construct a Markov chain such that the sam-
ple is drawn from any desired posterior distribution.
The power of Bayesian computations derives from the availability of
MCMC solutions to compute posterior inferences from almost arbitrarily
complex statistical models with almost arbitrarily huge numbers of parame-
ters. However, this simple statement hides some important complications.
It is clear that these successive values are connected and not independent
in the way that simple Monte Carlo points are. If there is too much depend-
ence, then the points move very slowly around the space of possible param-
eter values. Therefore, an extremely large sample will be needed to cover the
space properly to represent the posterior distribution adequately. So the effi-
ciency of MCMC methods depend critically on getting a Markov chain that
moves rapidly around the space (a property that is referred to as ‘good mix-
ing’). Unfortunately, although it is generally very easy to devise an MCMC
algorithm that works in principle, it often requires considerable skill and
experience to construct one that mixes well.
Another complication is that the algorithms require a ‘burn-in’ period,
for the randomly moving point to settle into the part of the parameter space
supported by the posterior distribution. It is by no means simple to diagnose
when the Markov chain has run long enough.
In a very wide range of moderately complex models (such as those that
can be implemented successfully in the software WinBUGS described in the
main text), these problems are minimal. For large and complex problems,
however, MCMC remains something of an arcane art. Nevertheless, there is
growing familiarity with the technique based on its widespread use in
Bayesian statistics – and a growing literature on MCMC algorithms that is
gradually advancing the frontier of problems that can be tackled routinely.
Appendix 71
Tackling hard problems
There is a frequentist parallel to the computational problems of Bayesian
methods, which is that it is extremely difficult to obtain exact frequentist tests,
confidence intervals and unbiased estimators except in really simple models. It
is often overlooked that the great majority of frequentist techniques in general
use are, in fact, only approximate. This includes all methods based on general-
ized linear models, generalized likelihood ratio tests, bootstrapping and many
more. The honorable exception is inference in the standard normal linear
model. Even here, it is not straightforward to compare non-nested models using
frequentist methods.
As described in the main text, the availability of computational techniques
like MCMC makes exact Bayesian inferences possible even in very complex
models. As statisticians strive to address larger, more complex data structures
(micro-array data, data mining, etc), the benefit (B3) – “ability to tackle more
complex problems” – of Bayesian methods becomes increasingly important.

Bayesian Tutorial

Uploaded by

Copyright:

Available Formats

Bayesian Tutorial

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Bayesian Tutorial

Uploaded by

Copyright:

Available Formats

What is the objective of the Bayesian Initiative in Health Economics & Outcomes Research?

What is the objective of the Bayesian Initiative in Health Economics & Outcomes Research?

What is the Centre for Bayesian Statistics in Health Economics (CHEBS) and what are its goals?

What is the Centre for Bayesian Statistics in Health Economics (CHEBS) and what are its goals?

a primer on

Bayesian Initiative in Health Economics & Outcomes Research

The Centre for Bayesian Statistics in Health Economics (CHEBS)

Copyright ® 2003 MEDTAP International, Inc.

ii A Primer on Bayesian Statistics in Health Economics and Outcomes Research

Preface and Brief History ......................................................1

Section 1: Inference ............................................................13

Section 2: The Bayesian Method ........................................19

Section 3: Prior Information ..............................................23

Section 4: Prior Specification..............................................27

Section 5: Computation ......................................................31

Section 6: Design and Analysis of Trials ..........................35

Section 7: Economic Models ..............................................39

Bibliography and Further Reading ....................................43

A Primer on Bayesian Statistics in Health Economics and Outcomes R e s e a r c h iii

We would like to gratefully acknowledge the Health Economics Advisory

iv A Primer on Bayesian Statistics in Health Economics and Outcomes Research

L et me begin by saying that I was trained as a Bayesian in the

A Primer on Bayesian Statistics in Health Economics and Outcomes R e s e a r c h 1

From a stricter point of view, Bayesian statistics might properly be said

2 A Primer on Bayesian Statistics in Health Economics and Outcomes Research

Preface and Brief History 3

4 A Primer on Bayesian Statistics in Health Economics and Outcomes Research

Preface and Brief History 5

6 A Primer on Bayesian Statistics in Health Economics and Outcomes Research

Preface and Brief History 7

Goodman, SN. Toward Evidence-Based Medical Statistics. 1: The P Value

Pearson ES. Bayes’ theorem, examined in the light of experimental sampling.

8 A Primer on Bayesian Statistics in Health Economics and Outcomes Research

T his Primer is for health economists, outcomes research practi-

A Primer on Bayesian Statistics in Health Economics and Outcomes R e s e a r c h 9

The authors of this Primer are firmly committed to the Bayesian

10 A Primer on Bayesian Statistics in Health Economics and Outcomes Research

I n order to obtain a clear understanding of the benefits and draw-

A Primer on Bayesian Statistics in Health Economics and Outcomes R e s e a r c h 13

Statistical methods are generally formulated as making inferences about

14 A Primer on Bayesian Statistics in Health Economics and Outcomes Research

In this last example, the frequentist can conduct a significance test of

1. Only 5% of patients would be more cost-effectively treated by treatment 1.

3. There is only a 5% chance that the null hypothesis is true.

Statement 3 is how a P-value is commonly interpreted; yet this inter-

Most scientists would regard thought-transference as highly implausi-

16 A Primer on Bayesian Statistics in Health Economics and Outcomes Research

Probability is a limiting, long-run Probability measures a personal degree

Parameters are not repeatable or Parameters are unknown.

Interpreted in terms of long-run repeti- Interpreted in terms of evidence from the

In 5% of samples where the hypothesis The statement applies on the basis of

18 A Primer on Bayesian Statistics in Health Economics and Outcomes Research

T he fundamentals of Bayesian statistics are very simple. The

A Primer on Bayesian Statistics in Health Economics and Outcomes R e s e a r c h 19

20 A Primer on Bayesian Statistics in Health Economics and Outcomes Research

Section 2: The Bayesian Method 21

22 A Primer on Bayesian Statistics in Health Economics and Outcomes Research

T he prior information is both a strength and a potential weak-

A Primer on Bayesian Statistics in Health Economics and Outcomes R e s e a r c h 23

24 A Primer on Bayesian Statistics in Health Economics and Outcomes Research