Georgetown University
Decision Analysis in Healthcare


Modeling Preferences 


A total of 1030 people have rated this page an average of  2.58 out of five stars.  Rate this page.   See their suggestions and questions


Why model values?
Misleading numbers?


  1. Would it help?
  2. Select attributes
  3. Do it again
  4. Set attribute levels
  5. Assign values for single attributes
  6. Choose aggregation rule
  7. Estimate weights
  8. Evaluate model

Other methods for assessing:

Resulting severity index
Model-evaluation details
Preferential independence
Multi-attribute utility models
What do you know?

Biweekly project
Recent questions
Get answers to questions about decision analysis

Instructions on how to generate cases for testing a mode (SWF)

Calculate weights from ratio estimates (SWF)

Correlation between expert's rating & model score ( SWF)

Generate case scenarios

Recently asked:  The lecture talks about using hypothetical cases to validate the model. Are the hypothetical cases actually based on real cases but identified as hypothetical for privacy reasons? Using hypothetical outcomes to validate a model seems odd.  See answer to this and other questions.

This chapter introduces a flexible method for modeling decision makers' values and several means of validating such models. We assume that a model is developed jointly by one analyst and one decision maker.  If the development process involves a group, the analyst should use the integrative group process discussed in a later chapter.  While the model building effort focuses on the interaction between an analyst and a decision maker, it should be obvious that the same process can be used for self-analysis.  A decision maker can build a model of his/her own decisions without the help of an analyst.

Value models are based on Bernoulli's (1938) recognition that money's value did not always equal its amount. He postulated that increasing amount of income had decreasing value to the wage earner. Value models have received considerable attention from economists and psychologists (Savage .1972; Edwards 1974). A comprehensive and rather mathematical introduction to constructing value models is found in von Winterfeldt and Edwards (1986). This chapter focuses on instructions for making the models and ignores axiomatic/mathematical foundation of multi-attribute value models.

Value models quantify a person's preferences. By this, we mean that value models assign numbers to options so that higher numbers reflect more preferred options.  These models assume that the decision makers must select from several options and that the selection should depend on grading the preferences for the options. These preferences are quantified by examining the various attributes (characteristics, dimensions, or features) of the options. For example, if decision makers were choosing among different Electronic Medical Record (EMR) systems, the value of the different EMRs could be scored by examining such attributes as “compatibility with legacy systems”, “potential impact on practice patterns” and “cost.” First, the impact of each EMR on each attribute would be scored, this is often called single attribute value function. Second, scores would be weighted by the relative importance of each attribute. Third, the scores for all attributes would be aggregated, often by using a weighted sum. Fourth, the EMR with the highest weighted score would be chosen.

If each option was described in terms of n attributes A1, A2,  ... , An the option would be assigned a score on each attribute, V(A1), V(A2),  ... , V(An). The overall value of an option equals:

Value = Function [ V(A1), V(A2),  ... , V(An) ]

In words, the overall value of an option is a function of the value of the option on each attribute.

Why Model Values?

Values (attitudes, preferences) play major roles in making management decisions.  In organizations, decision making is often very complex and a product of collective action. Frequently, decisions must be made concerning issues on which few data exist, forcing managers to make decisions on the basis of opinions, not fact.  Often there is no correct resolution to a problem‑with all options having equally legitimate perspectives and values playing a major role in the final choice.

There are many everyday decisions that involve value tradeoffs.  Decision such as -- Which software to purchase? Which vendor to contract with? Whether to add a clinic? Whether to add nurses? Whether to shift to float staffing? Whether to buy a piece of equipment? Whether to add a program? Which person to hire? Whether to hire staff or pay overtime? How to balance missions (e.g. providing service with revenue generating activities)? What contractual relationships to enter into? Which quality improvement project to pursue?  In all these decisions, the manger has to tradeoff gains in something against other things.  Doing a quality improvement project in stroke unit means you do not have resources to do the same in trauma unit.  Hiring a technically savvy person may mean you have to put up with socially ineptness.  In business, difficult decisions almost always involve trading off various benefits against each other.

Most people acknowledge that a manager's decisions involves consideration of value tradeoffs.   This is not a revelation.  What is unusual is that decision analyst model these values.  Some may wonder why the analyst needs to model and quantify value tradeoffs.  The reason for modeling decisions maker's values includes the following:

  1. One reason to model values is to clarify and communicate how decision makers feel. Modeling values helps managers communicate their positions by explicitly showing their priorities. Value models divide multi-attribute problems into components that are systematically analyzed. A specific formula then quantifies the outcome of the analysis. These models clarify the basis of decisions so others can see the logic behind the decision and ideally agree with it.  For example, Cline and colleagues constructed a value model to determine the eligibility of nursing home residents for a higher level of reimbursement (the "super‑skilled" level of nursing care). This model showed which attributes of an applicant affected eligibility and how much weight each attribute deserved.  As a consequence of this effort, the regulator, the industry, the providers of care and the patients became more aware of how eligibility decisions were made (Cline et al. 1982).

  2. Another reason to model values is to aid decision making in complex situations.  In complicated decisions, decision makers face uncertain events as well as ill expressed values.  In these circumstances, modeling the values adds to the decision maker's understanding of the underlying problem.  It helps the decision maker break the problem into its parts and manage the decision more effectively.  In short, models help decision makers divide and conquer. 

  3. Still another reason to model the values of the decision maker is to repeatedly use the mathematical model instead of the decision maker.  Consider screening a large number of applicants. If the analyst models the decision maker's values, then one could go through thousands of applicants and select a few that the manager needs to interview.  Because the model reflects the manger's values, the analyst is reassured that he/she has not erroneously screened out applicants that the manger would have liked to interview.  

  4. Finally the last reason for modeling values is to quantify hard to measure concepts such as the severity of trauma (Detmer et al. 1977), the degree to which an area is medically underserved (Health Services Research Group 1975), or the quality of the remaining years of life (Pliskin et al. 1980). These hard‑to‑measure concepts are similar to preferences because they are subjective and open to disagreement.  Like complex preferences it is necessary to describe hard to measure concepts in terms of several attributes that may capture different dimensions of the concept.  Measuring a subjective construct may seem a contradiction in terms. After all, rational people look at the same construct and evaluate them very differently. For some broad distinctions, such as whether there is more of one thing than another, the task is easier. Thus, one might say that one patient is more ill than another, or that an area is more medically underserved than another.

Chatburn and Primiano (2001) used value models to examine large capital purchases such as the deicion to purchase a ventilator.  Value models have been used to model policymakers' priorities for evaluating standards for offshore oil discharges (von Winterfeldt 1980), energy alternatives (Keeney 1976), drug therapy options (Aschenbrenner and Kaubeck 1978), and family planning options (Beach et al. 1979). In a later chapter, we report on using value models to analyze conflicts among decision makers.

Misleading Numbers?

Though value models allow us to quantify subjective concepts, the resulting numbers are rough estimates that should not be mistaken for precise measurements. It is important that managers do not read more into the numbers than they mean. Analysts must stress that the numbers in value models are intended to offer a consistent method of tracking, comparing, and communicating rough, subjective concepts and not to claim a false sense of precision. 

An important distinction is whether the model is to be used for rank ordering (ordinal) or rating the extent of presence of a hard‑to measure concept (interval).  Some value models produce numbers that are only useful for rank ordering options. Thus, some severity indexes indicate whether one patient is sicker than another, not how much sicker. In these circumstances, a patient with a severity score of 4 may not be twice as ill as a patient with a severity of 2. Averaging such ordinal scores is meaningless. In contrast, value models that score on an interval scale show how much more preferable one option is than another. For example, a severity index can be created to show how much mare severe one patient's condition is than another’s. A patient scoring 4 can be considered twice as ill as one scoring 2. Further, averaging interval scores is meaningful. 

Numbers can also be used as a way of naming things (the so called Nominal scale). Nominal scales produce numbers that are neither ordinal or interval. The International Classification of Disease assigns numbers to diseases but these numbers are neither ordinal or interval.

In modeling decision makers values, single-attribute value functions must be interval scales. If single attributes are measured on an interval scale, then these numbers can be added or multiplied to produce the overall score.   If measured as an ordinal scale or nominal scale, one cannot calculate the overall severity from the single attribute values. In contrast, overall scores for options need only have an ordinal property. When it comes to choosing an option, all most decision makers care about is which option has the highest rating and not by how much is an option scored higher than others.

Keep in mind that the purpose of quantification is not to be precise in numerical assessment.  The analyst quantifies values of various attributes so that the calculus of mathematics can be used to keep track of them and produce an overall score that reflects the various attributes.  Quantification allows us to use the logic imbedded in numbers.  In the end, model scores are a rough approximation of preferences.  They are helpful not because they are precise but because they adequately track contributions of each attribute.

Examples & Discussion of Severity of AIDS Cases 

There are many occasions in which multi-attribute value models can be used to model a decision.  A common example is in hiring decisions.  In choosing among the candidates, the attributes in Table 1 might be used to screen applicants for subsequent interviews.

Attribute weight


Attribute level

Value of the level






Applicants education





No college degree


Bachelors of science or Bachelor of Arts


Master of Science in  health care field


Master of Science in health care related field


Ph.D. or higher degrees






Computer skills






Data entry


Experience with a database or a worksheet program


Experience with both databases & worksheet programs




Internal promotion









People skills



Not a strength of the applicant


Contributes to teams effectively


Organizes and leads teams


Table 1:  A model for hiring decisions

Each attribute has an assigned weight.  Each attribute levels has an assigned value score.  By convention, attribute levels are set to range from 0 to 100.  Attribute levels are set so that only one level can be assigned to each applicant.  Attribute weights are set so that all weights add up to 1.   The overall value of an applicant can be measured as the weighted sum of attribute level scores.  In this example, the model assigns to each applicant a score between 0 to 100, where 100 is the most preferred applicant.   Note that the way the decision maker has rated these attributes suggest that internal promotion is less important than appropriate educational degrees and computer experience.  The overall rating can be used to focus interviews on a handful of applicants.  

Consider for another example the organization of a health fair.   Let us assume that a choice needs to be made about what should be included in the fair:

  • Screening for blood pressure

  • Screening for peak air flow

  • Screening for lack of exercise

  • Screening for smoking habits

  • Screening for depression

  • Screening for poor food habits

  • Screening for knowledge of breast self examination

  • Screening for access to primary care clinician

  • Screening for blood sugar level

The decision maker is concerned about cost but would like to underwrite the cost of the fair if it leads to significant number of referrals.  Discussions with the decision maker led to the specification of the following attributes:

Attribute weight


Attribute level

Value of the level






Cost of providing the service

Interview cost


Interview and non-intrusive test costs


Interview and intrusive test costs






Need in target group






Less than 1% are likely to be positive


1% to 5% are likely to be positive


More than 5% likely to be positive




Generates a likely visit






Table 2:  A model for deciding on composition of a health fair

This simple model will score each screening based on three attributes, the cost of providing the service, the need in the target group and whether it may generate a visit to the clinic.  Once all screening options have been scored, then depending on funds available, the top scoring screening activities can be chosen and offered in the health fair.

Our third example, and one that we will describe at length though out this chapter, is concerned with constructing practice profiles using a severity index.  Practice profiles are helpful in order to hire, fire, discipline and pay physicians (Sessa 1992, Vibbert 1992, McNeil Pedersen, Gatsonis 1992).   A practice profile compares cost and outcomes of individual physicians to each other.  But since patients differ in severity of illness, it is important to adjust outcomes by the provider's mix of patients.  Only then can one compare apples to apples.  If a severity score existed, managers could examine patient outcomes and see if it was within expectations.  They could compare two clinicians and see which one had better outcomes for patients with same severity of illness.  Armed with a severity index, managers could compare cost of care for different clinicians and see which one was more efficient.  Almost 15 years ago, Alemi and colleagues used a value model to create a severity index for acquired immune deficiency syndrome (AIDS) (Alemi et al. 1990).  While much time has lapsed and care of AIDS patients has progressed, the method of developing the severity index is still relevant. 

The U.S. Centers for Disease Control defines an AIDS case as an HIV‑positive patient suffering from any of a list of opportunistic infections. After the diagnosis of HIV, patients often suffer a complex set of different diseases.  The cost of treatment for each patient is heavily dependent on the course of their illness.  For example, patients with disease of the skin cancer, Kaposi's sarcoma, have half the first‑year costs of patients with pneumocystic pneumonia, a lung infection (Pascal et al. 1989). Thus, if a manager wants to compare two clinicians in their ability to care for patients, it is important to measure the severity of AIDS among their patients.  We use the construction of severity index for AIDS to demonstrate the construction of multi-attribute value models.

Steps  in Modeling Values

Using the example of the AIDS severity index, this section shows how to examine the need for a value model and how to create such a model. 

Step 1. Would It Help?

The first, and most obvious, question is whether constructing a value model would help resolve the problem faced by the manager. Defining the problem is the most significant step of the analysis, yet surprisingly little literature is available for guidance (Volkema 1981).

To define a problem, the analyst must answer several related questions: Who is the decision maker? What objectives does this person wish to achieve? What role do subjective judgments play in these goals? Would it help to model the judgment? Should a value model be used? Finally, once a model is developed, how might it be used?

Who decides? In organizations, there are often many decision makers. No single person's viewpoint is sufficient, and the analyst needs a multidisciplinary consensus instead. A later chapter discusses how values of a group of people can be modeled. For simplicity, the following discussion assumes that only one person is involved in the decision‑making process.

The core of the problem here was that AIDS patients need different amounts of resources depending on the severity of their illness. The federal administrators of the Medicaid program wanted to measure severity of AIDS patients because the federal government paid for part of their care. The state administrators were likewise interested because state funds paid for another portion of their care.  Individual hospital administrators were interested in order to analyze a clinicians' practice patterns and recruit one who is more efficient. 

For the study, Alemi and colleagues assembled six experts known for clinical work with AIDS patients or for research on the survival of AIDS patients. Physicians came from several states, including New York and California (which at the time had the highest number of AIDS patients in the United States). California had more homosexual AIDS patients, while intravenous drug users were more prevalent in New York. The integrative group process was used to construct the severity index.

What must be done? Problem solving starts by recognizing a gap between the present situation and the desired outcome. Typically, at least one decision maker has noticed a difference between what is and what should be and begins to share this awareness with the relevant levels of the organization. Gradually a motivation is created to change, informal social ties are established to promote the change, and an individual or group receives a mandate to find a solution.

Often, a perceived problem must be reformulated to address the real issues. For example, a stated problem of developing a model of AIDS severity may indicate that the real issue is insufficient funds to provide for all AIDS treatment programs. When solutions are proposed prematurely, it is important to sit back and gain greater perspective on the problem. Occasionally, the decision makers have a solution in mind before fully understanding the problem, which shows the need for examining the decision makers' circumstances in greater depth. In these situations, it is the analyst's responsibility to redefine the problem to make it relevant to the real issues. VanGundy (1981) reviews 70 techniques used by analysts to redefine problems and creatively search for solutions, including structured techniques like brainstorming and less structured techniques like using analogies. Analysts can refer to VanGundy for guidance in restructuring problems.

What judgments must be made? After the problem has been defined, the analyst must examine the role subjective judgments can play in its resolution. One can do this by asking several "what if' questions: What plans would change if the judgment were different? What is being done now? If no one makes a judgment about the underlying concept, would it really matter, and who would complain? Would it be useful to tell how the judgment was made, or is it better to leave matters rather ambiguous? Must the decision maker choose among options, or should the decision maker let things unfold on their own? Is a subjective component critical to the judgment, or can it be based on objective standards?

In the example, the administrators needed to budget for the coming years, and they knew judgments of severity would help them anticipate utilization rates and overall costs. Programs caring for low severity patients would receive a smaller allocation than programs caring for high‑severity patients. But there were no objective measures of severity available, so clinician judgments was used instead.

How can the model of the judgment be used? In understanding what judgments must be made, it was crucial to attend to the limitations of circumstances in which these judgments are going to be made. The use of existing AIDS severity indexes was limited because they relied on physiological variables that were unavailable in automated data bases (Redfield and Burke 1988). Administrators asked us to predict prognoses from existing data. The only information widely available on AIDS patients was diagnoses, which were routinely collected after every encounter. Because these data did not include any known physiological predictors of survival (such as number of T4 cells), the manager had to find alternative ways to predict survival.

Experts seem to intuitively know the prognosis of the patient.  They easily recognize a patient that is very sick.  Would it make sense to model how experts make these judgments?  Although it was theoretically possible to have an expert panel review each case and estimate severity, from the outset it was clear that a model was needed because case‑by‑case review was extremely expensive. Moreover, the large number of cases would require the use of several expert panels, each judging a subset of cases, and the panels might disagree. Further, judgments within a panel can be quite inconsistent over time. In contrast, the model provided a quick way of rating the severity of patients. It also explained the rationale behind the ratings, which allowed skeptics to examine the fairness of judgments, thus increasing the acceptance of those judgments.

Step 2. Select Attributes

After defining the problem, the second step is to identify the attributes needed for making the judgment.  The analyst solicited the list of the attributes that may be needed in making a judgment of severity from invited experts. There are three steps to interviewing an expert. First, the analyst introduces himself or herself and describes the purpose of the meeting. Second, the analyst asks about the expert's relevant experience. Third, using the expert's terminology, the analyst asks about the attributes.

Introduce yourself and your purpose. Briefly say who you are, why you are talking to the expert, the model's purpose, and how it will be developed. Be as brief as possible. An interview is going well if the analyst is listening and the expert is talking. If it takes you five minutes just to describe your purpose, then something is amiss. Probably you haven't understood the problem well, or possibly the expert is not familiar with the problem.

Be assertive in setting the interview's pace and agenda. Because you are likely to receive a comment whenever you pause, be judicious about pausing. Thus, if you stop after saying, "Our purpose is to construct a severity index to work with existing databases," your expert will, likely discuss your purpose.  But if you immediately follow the previous sentence with a question about  the expert's experience in assessing severity, the expert is more likely to begin describing his or her background. The point is that, as the analyst, you set the agenda, and you should construct your questions to resolve your uncertainties, not to suit the expert's agenda.

Ask the experts to introduce themselves. It is important to establish that you are a record keeper and that the expert will provide the content. A good way of doing this is to ask the expert to introduce himself or herself by describing relevant experiences. This approach also stresses your respect for his or her expertise.

Start with tangible examples. Concrete examples help you understand which patient attributes should be used to predict severity and how they can be measured. Ask the expert to recall an actual situation and contrast it with other occasions to discern the key discriminators. For example, you might ask the expert to describe a severely ill patient in detail (to ensure that the expert is referring to a particular patient). Then ask for a description of a patient who was not severely ill and elicit the key differences between the two patients. These differences are attributes you can use to judge severity. Continue asking the expert to think about specific patients until you have identified a number of attributes. Here is a sample dialogue:


Can you recall a specific patient with a very poor prognosis?


I work in a referral center, and we see a lot of severely ill patients. They seem to have many illnesses and are unable to recover completely, so they continue to worsen.


Tell me about a recent patient who was severely ill.


A 28‑year‑old homosexual male patient deteriorated rapidly. He kept fighting recurrent influenza and died from gastrointestinal cancer. The real problem was that he couldn't tolerate AZT, so we couldn't help him much. Once a person has cancer, we can do little to maintain him.


Tell me about a patient with a good prognosis, say close to five years.


Well, let me think. A year ago we had a 32‑year‑old male patient diagnosed with AIDS who has not had serious disease since‑‑a few skin infections but nothing serious. His spirit is up, he continues working, and we have every reason to expect he will survive four or five years.


What key difference between the two patients made you realize that the first patient had a poorer prognosis than the second?


That's a difficult question.  Patients are so different from each other that it's tough to point at one characteristic. But if you really push me, I would say two characteristics: the history of illness and the ability to tolerate AZT.


What about the history is relevant?


If I must predict a prognosis, I want to know whether he has had serious illness in vital organs.


Which organs?


Brain, heart, and lungs are more important than, say, skin.

As you may have noticed in this dialogue, the analyst started with tangible examples and used the terminology and words introduced by the expert to become more concrete. There are two advantages to this process. First it helps the expert recall the details without analyst introducing unfamiliar words such as attributes.  Second, soliciting attributes by contrasting patients helps single out those attributes that truly affect prognosis. Thus, it does not produce a wish list of information that is loosely tied to survival ‑ an extravagance one cannot afford in model building.

After you have identified some attributes, you can ask directly for additional attributes that indicate prognosis. You might ask if there are other markers of prognosis, if the expert has used the word marker. If you have to, you might say: “In our terminology, we refer to the kinds of things you have mentioned as markers of prognosis. Are there other markers?”  Here is an example dialog:


Are there other markers for poor prognosis?


Co morbidities are important. Perhaps advanced age suggests poorer prognosis. Sex may matter.


Does the age or sex really matter in predicting prognosis?


Sex does not matter, but age does. But there are many exceptions. You cannot predict the prognosis of a patient based on age alone.


What are some other markers of poor prognosis?

As you can see in the dialogue, you might even express your own ideas without pushing them on the expert.  In general, analysts are not there to express their own ideas.  They are there to listen.  But they can ask questions to clarify things or even to mention things overlooked by the expert, as long as it does not change the nature of the relationship between the analyst and the expert.

Arrange the attributes in a hierarchy from broad to specific attributes (Keeney and Raiffa 1976). Some analysts suggest using a hierarchy to solicit and structure the attributes. For example, an expert may suggest that a patient's prognosis depends on medical history and demographics. Demographics include age and sex. Medical history involves the nature of the illness, co morbidities, and tolerance of AZT. The nature of illness breaks down into body systems involved (skin, nerves, blood, etc.). Within each body system, some diagnoses are minor and other diagnoses are more threatening. The expert then lists, within each system, a range of diseases. The hierarchical structure promotes completeness and simplifies tracking many attributes. 

Be careful about terminology. Always use the expert's terminology, even if you think a reformulation would help. Thus, if the expert refers to "sex," do not substitute "gender." Such new terminology may confuse the conversation and create an environment where the analyst acts more like an expert, which can undermine the expert's confidence that he or she is being heard. It is reasonable, however, to ask for clarification‑‑"sex" could refer to gender or to sex practices, and you must understand which meaning is intended. 

In general, the less esoteric prompts are more likely to produce the best responses, so formulate a few prompts and use the prompts that feel most natural for your task.  Avoid jargon, including the use of terminology from Decision Analysis (e.g. attribute, value function, aggregation rules, etc.). 

Take notes, and do not interrupt. Have paper and pencil available, and write down the important points. Not only does this help the expert's recall, but it also helps you review matters while the expert is still available. Experts tend to list a few attributes, then focus attention on one or two. Actively listen to these areas of focus. When the expert is finished, review your notes for items that need elaboration. If you don't understand certain points, ask for examples, which are an excellent means of clarification.  For instance, after the expert has described attributes of vital organ involvement, you may ask the expert to elaborate on something mentioned earlier, such as "acceptance of AZT." If the expert mentions other topics in the process, return to them after completing the discussion of AZT acceptance. This ensures that no loose ends are left when the interview is finished and reassures the expert that you are indeed listening.

Other approaches. Other, more statistical approaches to soliciting attributes are available, such as multidimensional scaling and factor analysis. However, we prefer the behavioral approach to soliciting attributes because it involves the expert more in the process and leads to greater acceptance of the model.

Step 3. Do It Again

After soliciting a set of attributes, it is important to examine and, if necessary, revise them.   Psychological research suggests that changing the framing of a question alters the response. Consider these two questions:

"What are the markers for survival?"

"What are the markers for poor prognosis?"

One question emphasizes survival, the other mortality. One would expect that patient attributes indicating survival would also indicate mortality, but researchers have found this to be untrue (for a review, see Nisbett and Ross i 980). Experts may identify entirely different attributes for survival and mortality. This research suggests that value‑laden prompts tap different parts of the memory and can evoke recall of different pieces of information.  Evidence about the impact of questions on recall and judgment is substantial (Hogarth 1975; Ericsson and Simon 1980). For example, in one study subjects used surprisingly different sets of attributes to judge whether a person was an introvert or an extrovert (Snyder and Cantor 1979). Studies like this suggest that analysts should ask their questions in two ways, once in positive terms and again in negative terms.

Several tests should be conducted to ensure that the solicitation process succeeded.  The first test ensures that the listed attributes are exhaustive by using them to describe several hypothetical patients and asking the expert to rate their prognosis. If the expert needs additional information for a judgment, solicit new attributes until you have sufficient information to judge severity.

A second test checks that the attributes are not redundant by examining whether knowledge of one attribute implies knowledge of another. For example, the expert may consider "inability to administer AZT" and "cancer of GI tract" redundant if no patient with GI cancer can accept AZT. In such cases, either the two attributes should be collapsed into one, or one must be dropped from the analysis.

A third test ensures that each attribute is important to the decision maker's judgment. You can test this by asking the decision makers to judge two hypothetical situations: one with the attribute at its lowest level and another with the attribute at peak level. If the judgments are similar, the attribute may be ignored. For example, gender may be unimportant if male and female AIDS patients with the same history of illness have identical prognoses.

Fourth, a series of tests examines whether the attributes are related or dependent (Keeney and Raiffa 1976; Keeney 1977). These words are much abused and variously defined. By independence we mean that in judging two different patients, the shared feature among these patients does not affect how other features are judged. This type of impendence is called preferential independence. There are many situations in which preferential independence does not hold. In predicting three‑year risks of hospitalization, age and lifestyle may be dependent (Alemi et al. 1987). Among young adults, drinking may be a more important concern than cholesterol risks, while among older adults, cholesterol is the more important risk factor. Thus, the relative importance of cholesterol and drinking risks depends on the age of the patients being compared.

In many circumstances, preferential independence holds. It often holds even when experts complain that the attributes are dependent in other senses. When preferential independence holds, it is reasonable to break a complex judgment into components. Or, to say it differently, with preferential independence, it is possible to find a formula that translates scores on several attributes into an overall severity score in such a manner as to resemble the expert's intuitions.  When preferential independence does not hold, it is often a sign that some underlying issue is poorly understood. In these circumstances, the analyst should query the expert further and revise the attributes to eliminate dependencies (Keeney 1980).

In the AIDS severity study, discussions with the expert and later revisions led to the following set of 18 patient attributes for judging severity of AIDS: 



Transmission mode

Defining diagnosis 

Time since defining diagnosis

Diseases of nervous system

Disseminated diseases

Gastrointestinal diseases

Skin diseases

Lung diseases

Heart diseases

 Recurrence of a disease

Functioning of the organs


Psychiatric co-morbidity

Nutritional status

Drug markers

Functional impairment

As the number of attributes in a model increases, the chances for preferential dependence also increase. The rule of thumb is that preferential dependencies are much more likely in value models with more than nine attributes.

Step 4. Set Attribute Levels

Now it is time to identify the possible levels of each attribute. The analyst starts by deciding if the attributes are discrete or continuous. Attributes such as age are continuous; attributes such as diseases of the nervous system are discrete. However, continuous attributes may be expressed in terms of a few discrete levels, so that age can be described in decades, not individual years. The four steps in identifying the levels of an attribute are to define the range, define the best and worst levels, define some intermediate levels, and fill in the other possible levels so that the listing of the levels is exhaustive (capable of covering all possible situations).

To define the range, the analyst must select a target population and ask the expert to describe the possible range of the variable in it. Thus, for the AIDS severity index, the analyst asked the experts to focus on adult AIDS patients and, for each attribute, suggest the possible ranges. To assess the range of nervous system diseases, the analyst asked:


In adult AIDS patients, what is a disease that suggests the most extensive involvement of the nervous sys­tem?

Next the analyst asked the expert to specify the best and the worst possible level of each attribute. In the AIDS index, one could easily identify the level with the best possible prognosis: the normal finding within each attribute‑in common language, the healthy condition. The analyst accomplished the more difficult task of identifying the level with the worst possible prognosis by asking the expert:


What would be the gravest disease of the central nervous system, in terms of prognosis?

A typical error in obtaining the best and the worst levels is failing to describe these levels in detail. For example, in assessing the value of nutritional status, it is not helpful to define the levels as:

Best nutritional status

Worst nutritional status

Nor does it help to define the worst level as "severely nutritionally deficient" because the adjective "severe" is not defined.  It is best to avoid using adjectives in describing levels, as experts perceive words like severely, or best in different ways. The levels must be defined in terms of the underlying physical process measured in each attribute, and the descriptions must be connected to the nature of the attribute. Thus, a good level for the worst nutritional status might be "patients on total parenteral treatment," and the best status might be "nutritional treatment not needed."

Next, ask the expert to define intermediate levels. These levels are often defined by asking for a level between the best and worst levels. In the example, this dialogue might occur:


I understand that patients on total parenteral treatment have the worst prognosis. Can you think of other relatively common conditions with a slightly better prognosis?


Well, a host of things can happen. Pick up any book . on nutritional diseases and you find all kinds of things.


Right, but can you give me three or four examples?


Sure. The patient may be on antiemetics or nutritional supplements.


Do these levels‑include a level with a moderately poor prognosis and one with a relatively good prognosis?


Not really. If you want a level indicative of moderately poor prognosis, then you should include whether the patient is receiving Lomotil or Imodium. 

It is not always possible to solicit all possible levels of an attribute from the expert interviews. In these circumstances, the analyst can fill in the gaps afterward by reading the literature or interviewing other experts. The levels specified by the first expert are used as markers for placing the remaining levels, so that the levels range from best to worst. In the example, a clinician on the project team reviewed the expert's suggestions and filled in a long list of intermediate levels.

Step 5. Assign Values to Single Attributes

The analysis proceeds with the evaluation of single attribute value function, i.e. a scoring procedure that assigns the relative value of each level in a single attribute.  A common method for doing so is to use the double-anchored estimation method (Kneppreth et al. 1974). This approach gets its name from the practice of selecting the best and worst levels first and rating the remaining levels according to these two "anchors." In this method, first the attribute levels are ranked, or, if the attribute is continuous, the most and least preferred levels are specified. Then the best and the worst levels are used as anchors for assessing the other levels. 

For example, skin infections have the following levels:

No skin disorder

Kaposi's sarcoma


 Herpes complex

Candida or mucus


The following interaction typifies the questioning for the double-anchored estimation method:


 Which among the skin disorders has the worst prognosis?


 None is really that serious.


Yes, I understand that, but which is, the most serious?


Patients with thrush perhaps have a worse prognosis than patients with other skin infections.


 Let's rate the severity of thrush at 100 and place the severity of no skin disorder at 0. How would you rate shingles?


Shingles is almost as serious as thrush.


This tells me that you might rate the severity of shingles nearer 100 than 0. Where exactly would you rate it?


Maybe 90.


Can you now rate the remaining levels?

Several psychologists have questioned whether experts are systematically biased in assessing value. Yates and Jagacinski (1979) showed that using different anchors produced different value functions. For example, in assessing the value of money,  Kahneman and Tversky (1979) showed that values associated with gains or losses are different from values related to the amount of monetary return. They argued that the value of money is judged according to the decision maker's current assets. Because value may depend on the anchors used, it is important to use different anchors from just the best or worst levels. Thus, if the value of skin infections is assessed by anchoring on "shingles" and "no. skin infections," then it is important to verify the ratings relative to other levels. If the expert rated skin infections as follows:

No skin disorder


Kaposi's sarcoma




Herpes complex


Candida or mucus




The analyst might ask: 


You have rated herpes complex halfway between shingles and candida. Is this OK?


Not really. Prognosis of patients with herpes is closer to patients with candida.


How would you change the ratings?


Maybe we should rate herpes 98.

It is occasionally useful to change not only the anchors but also the assessment method.  A later section describes several alternative methods of assessing single‑attribute value functions. When a value is measured by two different methods, there would be inadvertent discrepancies; the analyst must ask the expert to resolve these differences.

By convention, the single attribute value function must range from 0 to 100.  Sometimes, experts and decision makers refuse to assign the 0 value.  In these circumstances, their estimated values should be revised to range from 0 to 100.  The following formula shows how to obtain standardized value functions from estimates that do not range between 0 to 100:

Standardized value for level X = 100*(Value assigned to level X - Value of least important level) /(Value of most important level - value of least important level)

For example, if the skin diseases attribute are rated as:

Attribute level Rating

No skin disorder


Kaposi's sarcoma




Then, the maximum value is 90 and the minimum value is 10 and standardized values can be assigned to each level using the formula above.  For example for Kaposi's sarcoma the value is:

Standardized value for Kaposi's Sarcoma = 100*(20 - 10) /(90-10)

Step 6. Choose an Aggregation Rule

In this step, the analysis proceeds by finding a way for aggregate single attribute functions into an overall scores evaluated across all attributes. Note that the scoring convention has produced a situation in which the value of each attribute is somewhere between 0 and 100. Thus, the prognosis of patients with skin infection and the prognosis of patients with various GI disease have the same range. Adding these scores will be misleading because skin infections are less serious than GI problems, so the analyst must find an aggregation rule that differentially weights the various attributes.

The most obvious rule is the additive model.  Assume that S represents the severity of AIDS. If a patient is described by a series of n attributes of {A1, A2, . . . , Ai, . . . , An}, then using the additive rule, the overall severity equals:

S = i Wi Vi(Aj)

where Vi(Aj) is the value of the jth level in the ith patient attribute, Wi is the weight associated with the ith attribute in predicting. prognosis, and iWi = 1.

Several other models are possible in addition to the additive model. The multiplicative model form is described in a later section.  

Step 7. Estimate Weights

The analyst can estimate the weights for an additive value model in a number of ways. This section presents the method of rating the ratio of importance of the attributes.  A later section presents other methods. It is often useful to mix several approaches.  Some analysts estimate weights by assessing the ratio of the importance of two attributes (Edwards 1977). The attributes are rank ordered, and the least important is assigned 10 points. Then the expert is asked to estimate the relative importance of the other attributes. There is no upper limit to the number of points other attributes can be assigned. For example, in estimating the weights for the three attributes (skin infections, lung infections, and GI diseases), the analyst and the expert might have the following discussion:


Which of the three attributes is most important?


Well, they are all important, but patients with either lung infections or GI diseases have worse prognoses than patients with skin infections.


Do lung infections have a worse prognosis than GI diseases?


That's more difficult to answer. No. I would say that for all practical purposes, they have the same prognosis. Well, now that I think about it, perhaps patients with GI diseases have a slightly worse prognosis.

Having obtained the rank ordering of the attributes, the analyst can proceed to estimating the importance weights.


Let's say that we arbitrarily rate the importance of skin infection in determining prognosis at 10 points. GI diseases are how many times more important than skin infections?


Quite a bit. Maybe three times.


That is, if we assign 10 points to skin infections, we should assign 30.points to the importance of GI diseases?


Yes, that sounds right. 


How about lung infections? How many more times important are they than GI diseases? 


I would say about the same.


(Checking for consistency in the subjective judgments.) Would you consider lung infections three times more serious than skin infections?


Yes, I think that should be about right.

In the dialogue above, the analyst first found the order of the attributes, then asked for the ratio of the weights of the attributes. Knowing the ratio of attributes allows us to estimate the attribute weights. If the model has only three attributes, the weights for the attributes can be obtained by solving the following three equations: 

W(GI diseases)  /  W(skin infection) = 3

W(lung diseases)  /  W(skin infection) = 3

W(lung diseases)  +  W(skin infection)  +  W(GI diseases) = 1

One characteristic of this estimation method is that its emphasis on the ratio of the importance of the attributes leads to relatively extreme weighting compared to other approaches. Thus, some attributes may be judged critical, and others rather trivial. The above method is refered to as SMART. Other approaches, especially the direct magnitude process, may judge all attributes as almost equally important.

In choosing a method to estimate weights, you should consider several trade‑offs. You can introduce errors by asking experts awkward and partially understood questions, but you can also cause error with an easier, but formally less justified, method. Our preference is to estimate weights in several ways and use the resulting differences to help experts think more carefully about their real beliefs. In doing so, the analysts usually start with a rank order technique, then move on to assess ratios, obtain a direct magnitude estimate, identify discrepancies, and finally ask the expert to resolve them.

One note of caution:  Some scientists have questioned whether experts can describe how they weight attributes. Nisbett and Wilson (1977) argued that directly assessed weight may not reflect an expert's true beliefs.  Yi (2004) found that patients decisions depended on the choice of methods of assessing their preferences.  Other investigators review the literature and find that directly assessing the relative importance of attributes is accurate (John and Edwards 1978, Naglie et al. 1997).  The debate is many years old.   The only way to decide if the directly assessed weights reflect the expert's opinions is to look at how the resulting models perform. In a number of applications, value models based on directly assessed weights correlated quite well with the subject's judgments (Fischer 1979). The typical correlation is actually in the upper .80s, which is high in comparison to most social science correlations. This success confirms the accuracy (perhaps one should say adequacy) of the subjective assessment techniques for constructing value models.

Step 8. Evaluate the Accuracy of the Model

While researchers know the importance of carefully evaluating value models, analysts often lack the time and resources to do this. Because of the importance of having confidence in the models and being able to defend the analytical methodology, we will present several ways of testing the adequacy of value models (Gustafson et al. 1980).

Most value models are devised to apply to a particular context, and they are not portable to other settings or uses. This is called "context dependence." In general, it is viewed as a liability, but this is not always the case. For example, the AIDS severity index maybe intended for evaluating clinicians' practice patterns, its use for evaluating treatment programs and prognosis of groups maybe inappropriate and possibly misleading.

The value model should require only available data for input. Relying on obscure data may increase the model's accuracy at the expense of practicality. Thus, the severity index should rely on reasonable sources of data, usually from existing data bases. A physiologically based database, for instance, would predict prognosis of AIDS patients quite accurately. However, such an index would be useless if physiological information is generally unavailable and routine collection of this information would take considerable time and money.  While the issue of data availability may seem obvious, it is a very common error in the development of value models. Experts used to working in organizations with superlative data systems may want data that are unavailable at average institutions, and they may produce a value model with limited usefulness. If there are no plans to compare scores across organizations, one can tailor indexes to each institution's capabilities and allow each institution to decide whether the cost of collecting new data are justified by the expected increase in accuracy. However, if scores will be used to compare institutions or allocate resources among institutions, then a single value model is needed that must be based on data available to all organizations.

The model should be simple to use. The index of medical under service (Health Services Research Group 19'75) is a good example of the importance of simplicity. This index, developed to help the federal government set priorities for funding HMOs, community health centers, and health facility development programs, originally had nine variables, but the director of the sponsoring federal agency rejected it because of the number of variables. Because he wanted to be able to "calculate the score on the back of an envelope," the index was reduced to four variables. The simplified version performed as well as one with nine variables; it was used for eight years to help set nationwide funding priorities. This example shows that simplicity does not always equal incompetence. Simplicity nearly always makes an index easy to understand and use.

When different people apply the value model to the same situation, they must arrive at the same scores, which is referred to as inter‑rater reliability. In the example, different registered record abstractors who use the model to rate the severity of a patient should produce the same score. If a model relies on hard‑to‑observe patient attributes, the abstractors will disagree about the condition of patients. If reasonable people using a value model reach different conclusions, then one loses confidence in the model's usefulness as a systematic method of evaluation. Inter-rater reliability is tested by having different abstractors rate the severity of randomly selected patients.

The value model should also seem reasonable to experts, something coined face validity. Thus, the severity index should seem reasonable to clinicians and managers. Otherwise, even if it accurate, one may experience problems with its acceptance. Clinicians who are unfamiliar with statistics will likely rely on their experience to judge the index, meaning that the variables, weights, and value scores must seem reasonable and practical to them. Face validity is tested by showing the model to a new set of experts and asking if they understand it and whether it is conceptually reasonable. 

One way to establish the validity of a model is to show that it simulates the judgment of the experts; then if one believes in the experts acumen then one should also consider the model as valid (Fryback 1976). In this approach the expert is asked to score several (perhaps 100) hypothetical case profiles described only by variables included in the model.  If the model accurately predicts the expert's judgments; confidence in the model increases, but this measure has the drawback of producing optimistic results. After all, if the expert who developed the model can't get the model to predict his or her judgments, who can? It is far better to ask a separate panel of experts to rate the patient profiles.   In the AIDS severity project, the analyst collected the expert's estimate of survival time for 97 hypothetical patients and examined whether the value model could predict these ratings. The correlation between the additive model and the rating of survival was ‑0.53. (The negative correlation means that high severity scores indicate shorter survival, and the magnitude of the correlation ranges between 0 and ‑1.0.) The ‑0.53 correlation suggests low to moderate agreement between the model and the expert's intuitions; correlations closer to 1.0 or  ‑1.0 imply greater agreement.  One can judge the adequacy of the correlations by comparing them with agreement among the experts. The correlation between several pairs of experts rating the same 97 hypothetical patients was also similar. The value model agreed with the average of the experts as much as the experts agreed with each other. Thus, the value model may be a reasonable approach to measuring severity of AIDS. ( A later section discusses an alternative value model with  better correlation.)

A model is considered valid if several different ways of measuring it lead to the same finding. This method of establishing validity is referred to as construct validity.  For example, the AIDS severity model should be correlated with other measures of AIDS severity. If the analyst has access to other severity indexes (such as physiologically based indexes), the predictions of the different approaches can be compared on a sample of patients. One such study was done for the index described in this section.  Alemi, Walker, Carey, and  Leggett  report that the proposed index was more accurate than physiological markers typically used for measuring severity of AIDS.  Construct validity is explored by comparing the model with a surrogate measure of severity. Because severely ill patients stay longer in the hospital, length of stay should be correlated with severity scores. The point is that convergence among several measures of severity increases confidence in the model.

In some situations, one can validate a value model by comparing the model's predictions against observable behavior. This method of establishing validity is referred to as predictive validity.  If a model is used to measure a subjective concept, its accuracy can be evaluated by comparing predictions to an observed and objective standard, which is often called the gold standard, to emphasize its status as being beyond debate.  In practice, gold standards are rarely available for judging the accuracy of subjective concepts (otherwise, one would not need the models in the first place). For example, the accuracy of a severity index can be examined by comparing it to observed outcomes of patients' care. When the severity index accurately predicts outcomes, then there is evidence favoring the model.  The model developed in this section was tested by comparing it to patients' survival rates.  Medical history of patients were analyzed using the model and the ability of the severity score in predicting patients' prognosis were examined. 

Other Methods for Assessing Single Attribute Value Functions

Single attribute value functions can be assessed in a number of different ways aside from the double‑anchored method (Johnson and Huber 1979).  The mid-value splitting technique sets the best and worst levels of the attribute at 100 and 0. Then the decision maker finds a level of the attribute that psychologically seems halfway between the best and the worst level.  This value for this level is set to 50. Using the best, worst, and mid-value points, the decision maker continues finding points that psychologically seem halfway between any two points. After several points are identified, the values of other points are assessed by linear extrapolation from existing points. The following conversation illustrates how the mid-value splitting technique could be used to assess the value of age in AIDS severity.


What is the age with the best prognosis?


A 20'-year-old has the best chance of survival.


What is the age with the worst prognosis?


AIDS patients over 70 are more susceptible to opportunistic infections and have the worst prognosis. Of course, infants with AIDS have an even worse prognosis, but I understand we are focusing on adults.


Which age has a prognosis half as bad as a 70-year­old?


I am going to say about 40, though I am not really sure.


I understand. We do not need exact answers. Perhaps it may help to ask the question differently. Do you think an increase in age from 40 to 70 causes as much of a deterioration in prognosis as an increase from 20 to 40 years?


If you are asking roughly, I agree.


If 20 years is rated as 0, 70 years as 100, do you think it would be reasonable to rate 40 years as 50?


I suppose my previous answers imply that I should say "yes."


Yes, but this is not binding‑you can revise your answers.


A rating of 50 for the age of 40 seems fine as a first approximation.


Can you tell me what age would have a prognosis halfway between 20 and 40 years old? 

Using the mid‑value splitting technique, the analyst chooses a value score, and the expert specifies the particular attribute level that matches it. This is opposite to the double‑anchored estimation, in which the analyst specifies an attribute level and asks for its value. The choice between the two methods should depend on whether the attribute is discrete or continuous. Often, with discrete attributes, there are no levels to correspond to a particular value scores, leading us to select the double‑anchored method.

A third method for assessing a value function is to draw a curve in the following fashion. The x axis is the various attribute levels. The y axis is the value associated with each attribute level. The best attribute level is assigned 100 and drawn on the curve. The worst attribute level is assigned 0. The expert is asked to draw a curve between these two points showing the value of remaining attribute levels (Pai et al, 1971). Once the graph is drawn, the analyst and the expert review its implications. For example, a graph can be constructed with age (20 to 70 years) on the x axis and value (0 to 100) on the y axis. Two points are marked on the graph (age 20 at 0 value and age 70 at 100 value). The analyst asks the expert to draw a line between these two points showing the prognosis for intermediate ages.

Finally, an extremely easy method, which requires no numerical assessment at all, is to assume a linear value function over the attribute. This arbitrary assumption introduces some errors, but they will be small if you are constructing an ordinal value scale, and the single attribute value function is monotonic (meaning that an increase in the attribute level will cause either no change or an increase in value) (Warner 1971).

For example, one cannot assume that increasing age will cause a proportionate decline in prognosis. In other words, the relationship between the variables is not monotonic: the prognosis for infants is especially poor, while 20‑year‑old patients have the best prognosis and 70‑year‑olds have a poor outlook. Because increasing age does not consistently lead to increasing severity‑‑and in fact it can also reduce severity‑‑an assumption of linear value is misleading.

Other Methods for Estimating Weights 

In the direct magnitude estimate, the expert is asked to rank order the attributes and then to rate their importance by assigning each a number between 0 and 100. Once the ratings are obtained, they are scaled to range between 0 and 1 by dividing each weight by the sum of the ratings. Subjects rarely rate the importance of an attribute near zero, so the direct magnitude estimation has the characteristic of producing weights that are close together, but the process has the advantage of simplicity and comprehensibility.

Weights can be estimated by having the expert distribute a fixed number of points, typically 100, among the attributes (Torgerson 1958). The main advantage of this method is simplicity, since it is only slightly more difficult than the ranking method. But if there are a large number of attributes, experts will have difficulty assigning numbers that total 100.

One approach to estimating weights is to ask the expert to rate "corner" cases. A corner case is a description of a patient with one attribute at its most extreme level and the remainder at normal levels. The expert's score for the corner case shows its relative importance. In multiplicative models (described later), the analyst can estimate other parameters by presenting corner cases with two or more attributes at peak levels. After the expert rates several cases, a set of parameters is estimated that optimizes the fit between model predictions and expert's ratings.

Another approach is to mix and match methods. Several empirical comparisons of assessment methods have shown that different weight estimation methods lead to similar assessments. A study that compared seven methods for obtaining subjective weights, including 100 point distribution, ranking, and ratio methods, found no differences in their results (Cook and Stewart 1975). Such insensitivity to assessment procedures is encouraging because it shows that the estimates are not by‑products of the method and thus are more likely to reflect the expert's true opinions. This allows us to substitute one method for another.

Other Aggregation Rules: 
Multiplicative Multiattribute Value Model

The additive value model assumes that single‑attribute value scores are weighted for importance and then added together. In essence, it calculates a weighted average of single‑attribute value functions.

The multiplicative model is another common aggregation rule. In the AIDS severity study, discussions with physicians suggested that a high score in any single attribute value function was sufficient ground for judging the patient severely ill. Using a multiplicative model, overall severity would be calculated thusly:

S = (- 1 + ∏i [ 1 + k ki U(Ai) ]) / k 

where Ki and K are constants chosen so that k = -1 + ∏i (1 + k ki )

In a multiplicative model, when the constant K is close to -1, high scores in one category are sufficient to produce an overall severe score even if other categories are normal. This model better resembled the expert's intuitions. To construct the multiplicative value model, the expert must estimate "n + 1" parameters: the n constants K; and one additional parameter, the constant K.

In the AIDS severity project, the analyst constructed a multiplicative value model. On 97 hypothetical patients, the severity ratings of the multiplicative and the additive models were compared to the expert's intuitive ratings.   The multiplicative model was more accurate (correlation between additive model and experts' judgment was 0.53, while the correlation between multiplicative model and expert judgment was 0.60). The difference in the accuracy of the two models was statistically significant. Therefore, the multiplicative severity model was chosen.

Resulting Multiplicative Severity Index

The following is an example of a multiplicative value model.  Experts on HIV/AIDS were interviewed and an index was built based on their judgment.  This index is intended for assessing the severity of the course of AIDS based on diagnosis and without access to physiological markers.  As such it is best suited for analysis of data from regions of the world where physiological markers are not readily available or for analysis of data from large administrative databases, where diagnoses are widely available.  Kinzbrunner and Pratt (1994) as well as Alemi, Walker, Carey, and  Leggett  (1999) provide evaluations of this index.  This index is in public domain and can be used without royalty payments.  Please note that advances in HIV/AIDS treatment may have changed the relative severity of various levels in the index. 

In the multiplicative multi-attribute value model used in the Severity of the Course of AIDS index the "k" value was set to -1 and all other parameters (single attribute value functions and Ki constants) were estimated by querying a panel of experts.  The scores presented in the index are the result of multiplying the single attribute value function by its Ki coefficient.    The index is scored by selecting a level within each attribute, finding the score associated with that level, multiplying all selected scores, and calculating the difference between 1 and the resulting multiplication.  



Step 1

Choose the lowest score that applies to the patient's characteristics.  If no exact match can be found, approximate the score by using the two markers most similar to the patient's characteristics.  


Less than 18 years, do not use this index
18 to 40 years, 1.0000
40 to 60 years, 0.9774
Over 60 years,  0.9436


White, 1.0000
Black, 0.9525
Hispanic, 0.9525
Other, 1.0000

Defining AIDS diagnosis

Kaposi's sarcoma, 1.0000
Candida esophagitis, 0.8093
Pneumocystis carinii pneumonia, 0.8014
Toxoplasmosis, 0.7537
Cryptococcosis, 0.7338
Cytomegalovirus retinitis, 0.7259
Cryptosporidiosis, 0.7179
Dementia, 0.7140
Cytomegalovirus colitis, 0.6981
Lymphoma, 0.6981
Progressive multi-focal leukoencephalopathy, 0.6941

Mode of transmission

Blood transfusion for non-trauma, 0.9316
Drug abuse, 0.8792
Other, 1.0000

Skin disorders

No skin disorder, 1.0000
Kaposi's sarcoma, 1.0000
Shingles, 0.9036
Herpes simplex, 0.8735
Cutaneous candidiasis, 0.8555
Thrush, 0.8555

Heart disorders

No heart disorders, 1.0000
HIV cardiomyopathy, 0.7337

Gastrointestinal diseases

No gastrointestinal disease, 1.0000
Isosporidiasis, 0.8091
Candida esophagitis, 0.8058
Salmonella, 0.7905
Tuberculosis, 0.7897
Non-specific diarrhea, 0.7803
Herpes esophagitis, 0.7536
Mycobacterium aviumintercellulare, 0.7494
Cryptosporidiosis, 0.7369
Kaposi's sarcoma, 0.7324
Cytomegalovirus colitis, 0.7086
Gastrointestinal cancer, 0.7060

Time since AIDS


Less than 3 months


 More than 18 months


More than 3 months


 More than 21 months


More than 6 months


 More than 24 months


More than 9 months


 More than 36 months


More than 12 months


 More than 48 months


More than 15 months


 More than 60 months


Lung disorders

No lung disorders, 1.0000
Pneumonia unspecified, 0.9208
Bacterial pneumonia, 0.8960
Tuberculosis, 0.8911
Mild Pneumocystis carinii pneumonia, 0.8664
Cryptococcosis, 0.8161
Herpes simplex, 0.8115
Histoplasmosis, 0.8135
Pneumocystis carinii pneumonia with respiratory failure, 0.8100
Mycobacterium aviumintercellulare, 0.8020
Kaposi's sarcoma, 0.7772

Nervous system diseases

No nervous system involvement, 1.0000
Neurosyphilis, 0.9975
Tuberculosis meningitis, 0.7776
Cryptoccoccal meningitis, 0.7616
Seizure, 0.7611
Myelopathy, 0.7511
Cytomegalovirus retinitis, 0.7454
Norcardiosis, 0.7454
Meningitis encephalitis unspecified, 0.7368
Histoplasmosis, 0.7264
Progressive multi-focal leukoencephalopathy, 0.7213
Encephalopathy/HIV dementia, 0.7213
Coccidiomycosis, 0.7189
Lymphoma, 0.7139

Disseminated disease

No disseminated illness, 1.0000
Idiopathic thrombocytopenic pupura, 0.9237
Kaposi's sarcoma, 0.9067
Non salmonella sepsis, 0.8163
Salmonella sepsis, 0.8043
Other drug induced anemia, 0.7918
Varicella zoster, 0.7912
Tuberculosis, 0.7910
Norcardiosis, 0.7842
Non-tuberculosis mycobacterial disease, 0.7705
Transfusion, 0.7611
Toxoplasmosis, 0.7591
AZT drug induced anemia, 0.7576
Cryptococcosis, 0.7555
Histoplasmosis, 0.7405
Hodgkin's disease, 0.7340
Coccidiodomycosis, 0.7310
Cytomegalovirus, 0.7239
Non-hodgkin's lymphoma, 0.7164
Thrombotic thrombocytopenia, 0.7139

RECURRING acute illness

No, 1.0000              Yes, 0.8357

Functional impairment

No marker, 1.0000
Boarding home care, 0.7933
Home health care, 0.7655
Nursing home care, 0.7535
Hospice care, 0.7416

Psychiatric co-morbidity

None, 1.0000
Psychiatric problem in psychiatric hospital, 0.8872
Psychiatric problem in medical setting, 0.8268
Severe depression, 0.8268

Drug markers

None, 1.0000
Lack of prophylaxis, 0.8756
Starting AZT on 1 gram, 0.7954
Starting and stopping of AZT, 0.7963
Dropping AZT by 1 gram, 0.7673
Incomplete treatment in herpes simplex virus, varicella zoster
virus, mycobacterium avuimintercellulare, or cytomegalovirus retinitis, 0.7593
Prescribed oral narcotics, 0.7512
Prescribed parenteral narcotics, 0.7192
Incomplete treatment of Pneumocystis carinii pneumonia, 0.7111
Incomplete treatment in Toxoplasmosis, 0.7031
Incomplete treatment in Cryptococcal infection, 0.6951

Organ involvement

None, 1.0000































None, 1.0000
Hypertension, 1.0000
Influenza, 0.9203
Legionella, 0.9402
Alcoholism, 0.8406

Nutritional status

No markers, 1.0000
Antiemetic, 0.9282
Nutritional supplement, 0.7687
Payment for nutritionist, 0.7607
Lomotil/imodium, 0.7447
Total parenteral treatment, 0.7248

Step two

Multiply all selected scores and enter here:

Step three

Subtract one from above entry and enter here:

Step four

Divide by -0.99 and enter here:

The last entry indicates the severity of the course of illness.  The higher the score the worse the prognosis.  Maximum score is one, minimum score is zero.  This score is a rough estimate of prognosis and should not be used to guide treatment of individual patients.

Model Evaluation

In evaluating multi-attribute value models, it is sometimes necessary to compare model scores against experts' ratings of cases.  For example, the analyst might want to see if a model makes similar prediction on applicants for a job as a decision maker.  Or the analyst might want to test if a model's score is similar to a clinician rating of severity of illness.  This section describes how a model can be validated by comparing it to the expert or decision maker's judgments.

Clearly, models should be evaluated against objective data; but objective data does not always exist.  In these circumstances, one can evaluate a model by comparing it against consensus among experts.  A model is consider valid, if it replicates the average rating of the experts and if there is consensus among experts about the ratings.

The steps in testing the ability of a model in predicting an expert are the following:

  1. Generate or identify cases that will be used to test the model.

  2. Ask the experts to rate each case individually, discuss their differences and rate the case again

  3. Compare the experts to each other and establish that there is consensus in ratings

  4. Compare the model scores against the experts ratings.  If there is more agreement between the model and the average of the experts, than among the experts, consider the model as effective in simulating the experts' consensus

The first step in comparing a model to experts' rating is to have access to a number of cases, at a minimum about 30 cases.  A case is defined as a collection of the attributes in the model: for each attribute one level is chosen.  A case is the combination of the chosen levels.  For example, a case can be constructed for judging the severity of AIDS patients by selecting a particular level for each attribute in the severity index.  There are two ways for constructing cases.  The first is to rely on real cases.  A real case is organized by using the model to abstract patients or situations.  The second approach is to create a hypothetical case from combination of attributes. 

We prefer to rely on hypothetical rather than real cases for two reasons.  First, often the analyst does not have time or resources to pull together a minimum of 30 real cases.  Second, in real cases attributes are positively correlated and in these circumstance, any model, even models with wrong attribute weights, will produce ratings similar to the experts.  In generating hypothetical cases, a combination of attributes, called orthogonal design, is used that generates cases more likely to detect differences between the model and the expert.  In an orthogonal design, the best and worst of each attribute are combined in a manner that there is no correlation between the attributes. 

The test of the accuracy of a model depends in part on what cases are used.  If the cases are constructed in a way that all of the attributes point to the same judgment, the test will not be very sensitive and any model, even models with improper attribute weights, will end up predicting the cases accurately.    For example, if a hypothetical applicant is described so that they have all of the desired features, then both the model and the decision maker will not have a difficult time accurately rating the overall value associated with the case.  A stricter test of the model occurs only when there are conflicting attributes, one suggesting one direction and the other the opposite.  When cases are constructed to resemble real situations, attributes are often correlated and they point to the same conclusions.  In contrast when orthogonal design is used, attributes have zero correlation, and it is more likely to find differences between the model score and expert's judgments. 

The steps for constructing orthogonal cases, also called scenario generation, are the following:

  1. Select two extreme levels for each attribute (best and worst)

  2. Start with 2 to the power of number of attribute cases.  For example, if there are 4 attributes, you would need 16 cases.

  3. Divide the cases into half and assign each half the levels of first attribute

  4. Divide the cases into quartiles and assign each quartile the levels of the second attribute.

  5. Continue this process until each other case is assigned the level of the last attribute.   

  6. Review the cases to drop those that are not possible, e.g. pregnant males.

  7. If there are too many cases, ask the expert or decision maker to review a randomly chosen sample of cases.

  8. Summarize each case in a separate piece of paper so that the decision maker/ expert can rate the case without being overwhelmed with information from other cases.

The following table shows an orthogonal design of cases for a three attribute model:

Scenario/Case Attribute 1 Attribute 2 Attribute 3
1 Best Best Best
2 Best Best Worst
3 Best Worst Best
4 Best Worst Worst
5 Worst Best Best
6 Worst Best Worst
7 Worst Worst Best
8 Worst Worst Worst
Table 3:  Orthogonal Design for Three Attributes

To generate your own scenarios, for a decision with less than 15 attributes, click here.  The second step in comparing model scores to expert's judgments is to ask the expert or decision maker to review each case and rate it on a scale from 0 to 100, where 100 is the best (defined in terms of the task at hand) and 0 is the worst (again defined in terms of task at hand).  If multiple experts are available, experts can discuss the cases in which they differ and rate again.   This process is known as estimate-talk-estimate and is an efficient method of getting experts to come to agreement on their numerical ratings.  In this fashion, a behavioral consensus and not just a mathematical average can emerge. 

When asking the expert to rate a case, present the each case in a separate page, so that information from other cases would not interfere.  For example, Table 4 shows an orthogonal design for cases needed to judge severity of HIV/AIDS based on three attributes:

Scenario/Case Skin Disease Lung diseases GI disease
1 No skin disorder No lung disorders No gastrointestinal disease
2 No skin disorder No lung disorders Gastrointestinal cancer
3 No skin disorder Kaposi's sarcoma No gastrointestinal disease
4 No skin disorder Kaposi's sarcoma Gastrointestinal cancer
5 Thrush No lung disorders No gastrointestinal disease
6 Thrush No lung disorders Gastrointestinal cancer
7 Thrush Kaposi's sarcoma No gastrointestinal disease
8 Thrush Kaposi's sarcoma Gastrointestinal cancer
Table 4:  Orthogonal Design for Three Attributes
in Judging Severity of AIDS

These cases are presented one at a time.  For example, Table 5 shows an example case and the question asked from the expert:

Case number 4:
Rated by expert:  XXXX

Patient has the following conditions:
      Skin disorders: None
      Lung disorders:  Kaposi's sarcoma
      Gastrointestinal disorders:  Gastrointestinal cancer

On a scale from 0 to 100, where 100 is the worst prognosis, i.e. a person with less than 6 months to live and 0 is the best, i.e. a person with no disorders, where would you rate this case:
       First rating before consultations:   ______
       Second rating after consultations: ______

Table 5:  An example of a case constructed from three attributes

In step three, if there are multiple experts, their judgments are compared to each other by looking at pair wise correlations.  Two experts are said to be in excellent agreement if the correlation between their ratings are relatively high, at least more than 0.75.    For correlation between 0.50 to 0.65 experts are said to be in moderate agreement.  For correlations lower than 0.5, the experts are said to be in low agreement.  If experts are in low agreement, it is important to explore the reason why.  If there is one decision maker or one expert, this step is skipped.

In step four, the average of experts (in case where there are multiple experts) or the experts' ratings (in case where there is a single expert) are compared to the model scores.  For each case, model is used to score the case.  The correlation between the model score and the expert is used to establish the validity of the model.  This correlation should be at least as high as agreement between the experts on the same cases. 

Preferential Independence

Independence has many meanings.  The Merriam Webster Dictionary provides the following meanings:

  1. Not subject to control by others

  2. Not affiliated with a larger controlling unit

  3. Not requiring or relying on something else : not contingent,  for example as in "an independent conclusion"

  4. Not looking to others for one's opinions or for guidance in conduct

  5. Not bound by or committed to a political party

  6. Not requiring or relying on others (as for care or livelihood), for example as in "independent of her parents"

  7. Being enough to free one from the necessity of working for a living, for example "a man of independent means"

  8. Showing a desire for freedom, for example "an independent manner"

  9. Not determined by or capable of being deduced or derived from or expressed in terms of members (as axioms or equations) of the set under consideration; especially : having linear independence

  10. Having the property that the joint probability (as of events or samples) or the joint probability density function (as of random variables) equals the product of the probabilities or probability density functions of separate occurrence

  11. Neither deducible from nor incompatible with another statement, for example, "independent postulates."

All of these are various meanings of independence.  To these we want to add yet another meaning known as "Preferential Independence."  Preferential independence is defined as:

  • One attribute is preferentially independent from another if changes in shared aspect of the attribute does not affect preferences in the other.

  • Two attributes are mutually preferentially independent from each other if each is preferentially independent of the other.

For example, prognosis of  patients with high cholesterol level is always worst than patients with low cholesterol level independent of shared levels of age.  To test this, present two patients to the expert and ask which one has worst prognosis:

Analyst: Lets look at two patients.  Both of these patients are young.  One has high cholesterol level and the other low levels.  Which one has worst prognosis?
Expert: This is obvious the person with high cholesterol level.
Analyst: Yes, I agree it is relatively obvious but I need to check for it.  Let me now repeat the question, but this time both patients are frail elderly.  Who has worst prognosis, the one with high cholesterol or the one with low cholesterol?
Expert: If both are elderly, then my answer is the same, the one with high cholesterol.
Analyst: Great, this tells me in my terminology that cholesterol levels are preferentially independent of age.

Please note that in testing the preferential independence we change the shared feature and not the actual items that the client is comparing:  we change from young to old for both patients and not the cholesterol level. 

Many experts may say that two attributes are dependent (because they have other meanings in mind) but when the analyst checks, they remain preferentially independent.  In many circumstances preferential independence holds despite appearances to the contrary.  But occasionally there are situations where it does not hold.  Lets take the previous example and now combine more facts in one of the attributes so that preferential independence does not hold: 

Analyst: Lets look at two patients.  Both of these patients are young.  One has high cholesterol level and low alcohol use.  The other has high alcohol use and low cholesterol level.  Which one has worst prognosis?
Expert: Well for a young person, alcohol abuse is a worse indicator than cholesterol levels.
Analyst: Ok, now lets repeat the question, but this time both patients are frail elderly.  First patient has high cholesterol and low alcohol use.  Second patient has low cholesterol and high alcohol use.
Expert: If both are elderly, I think the one with high cholesterol is at more risk.  You see for young people I am more concerned with alcohol use but for older people I am more concerned with cholesterol levels.
Analyst: Great, this tells me that the combination of alcohol and cholesterol levels is not preferentially independent of age.

Obviously to assess preferential independence a large number of comparisons need to be made, as any pair of attributes must be compared to any other attribute.  Keeney and Raiffa (1976) show that if any two consecutive pairs are mutually preferentially independent from each other, then all possible pairs are mutually preferentially independent.  This reduces the number of assessment necessary to only comparison of consecutive pairs, as arranged by the analyst or the decision maker.

When preferential independence does not hold, the analyst should take this as a signal that the underlying attributes have not been full y explored.  Perhaps a single attribute can be broken down into multiple attributes. 

A multi-attribute value model assumes that any pair of attributes is mutually preferentially independent of a third attribute.  When this assumption is not met, as in the above dialogue, there is no mathematical formula that can combine single attribute functions into an overall score that reflects the decision maker's preferences.  In these circumstance, one has to build different models for each level of the attribute.  For example, the analyst would be forced to build one model for young people and another for older people and still another for frail elderly. 

When the analyst identifies preferential independence, several different courses of actions could be followed.  If the preferential dependence is not systematic or large, it could be ignored as a method of simplifying the model.  On the other hand, if preferential independence is violated systematically for a few attributes, then a different model can be built for each value of the attributes.  For example, in assessing risk of hospitalization a model can be built for young people and a different model can be built for older people.  Finally, one can search for a different formulation of attributes so that they are preferentially independent. 

Multi-attribute Utility Models

Utility models are value models that reflect the decision maker's risk preferences.  Instead of assessing the decision maker's values directly, utility models reflect the decision maker's preferences among uncertain outcomes.  Single attribute Utility functions are constructed by asking the decision maker to choose among a sure return and a gamble.  If we a continuous variable, e.g. return on investment, the decision maker is asked to find a return that will make him indifferent to a gamble at 50% chance of maximum return and 50% chance of worst possible return.  The decision maker's sure return is assigned a utility of 50.  The process is continued by posing gambles involving the mid-point and the best and worst points.  For example, suppose you want to estimate the utility associated with returns ranging from 0 to $1000.   The decision maker is asked how much of a return he/she is willing to take for sure to give up a 50% chance of making $1,000 and a 50% chance of making $0. 

What return ~ 50% of $1000 or 50% of $0

If the decision maker gives a response that is less than midway, i.e. less than $500, then the decision maker is a risk seeker.  He/she assigns a utility to the midway point that is higher than expected value of returns.  If the decision maker gives a response above the midway point, then he/she under values a gamble less and prefers more of the sure return.  He is risk averse.  The utility he/she assigns to gambles is less than the expected value of the gamble as risk itself is something this decision maker is trying to avoid.  If the decision maker responds with the mid-point, then he/she is considered to be risk neutral.  A risk neutral person values each additional dollar in the same fashion, whether it is the first dollar or the last dollar made.  A risk neutral person is indifferent between a gamble for various returns and the expected monetary value of the gamble.

Suppose the decision maker in our case has responded with a value of $400.  Then, we assign 50 utilities to the return of $400.  The mid-point of the scale is $500.  The decision maker is a risk seeker because he assigns to gambles a utility more than its expected value.  Of course, one point does not establish risk preferences and several points need to be estimated before one has a reasonable picture of the utility function.  The analyst continues the interview in order to assess utility of additional gambles.  The analyst can ask for a gamble involving the mid-point and the best return.  The question would be:  "How much do you need to get for sure in order to give up a 50% chance of making $400 and a 50% chance of making $0."  Suppose the response is $175.  The return is assigned a utility of 25.  Similarly, the analyst can ask "How much do you need to get for sure in order to give up a 50% chance of making $400 and a 50% chance of making $1000."  Suppose the response is $675.  The response is assigned a utility of 75.  After the utility of a few points have been estimated, it is possible to fit the points to a polynomial curve, so that a utility score for all returns can be estimated.  For our example, Figure 1 shows the resulting utility curve:

Figure 1:  A risk seeking utility function

Sometimes, we have to estimate a utility function over an attribute that is not continuous or does not have a natural physical scale.  For example, we might want to measure a utility scale (or in better words a disutility scale) over different types of skin diseases.  In this approach, the worst and the best levels are fixed at 0 and 100 utilities.  The decision maker is asked to come up with a probability that would make him indifferent between a new level in the attribute and a gamble involving the worst and best possible levels in the attribute.  For example, suppose we want to estimate the utility associated with the following five skin diseases (listed in order of severity):  No skin disorder, Kaposi's sarcoma, Shingles, Herpes complex, Candida or mucus, and Thrush.

The analyst assigns the best possible level a utility of 0.  The worst possible level, Thrush, is assigned a utility of 100. The decision maker is asked to think if they prefer to have Kaposi's Sarcoma or a 90% chance of Thrush and 10% chance of no skin disorders.  No matter what the response, the decision maker is asked the same question again but now with probabilities reversed:  "Do you prefer to have Kaposi's Sarcoma or a 10% chance of Thrush and 90% chance of no skin disorders."  The analyst points to the decision maker that the choice between the sure disease and the risky situation was reversed when the probabilities were changed.  Because the choice is reversed, there must exist a probability at which point the decision maker is indifferent between the sure thing and the gamble.  The probabilities are changed until a point is found where the decision maker is indifferent between having Kaposi's Sarcoma and probability p of having Thrush and probability (1-p) of having no skin disorders.     The utility associated with Kaposi's Sarcoma is 100 times the estimated probability, p.  A utility function assessed in this fashion will reflect not only the values associated with different diseases but decision maker's risk taking attitude.  Some decision makers may consider a sure disease radically worst than a gamble involving a chance, even though remote, of having no disease at all.  These estimates thus clearly reflect not only their values but their willingness to take risks.  Value functions do not reflect risk attitudes.  Therefore, one would expect single attribute value and utility functions to be different.

If preferential independence is met, several single attribute utility functions are aggregated into an overall score using additive or multiplicative summation rules.    An additive multi-attribute utility function is commonly used in the literature.  When the decision maker is assumed to be risk neutral, an additive multi-attribute utility function is mathematically the same as calculating an expected value. 

Hierarchical Modeling of Attributes

A Summary Prepared by Jennifer A. Sinkule Based on Chatburn RL, Primiano FP Jr. Decision analysis for large capital purchases: how to buy a ventilator. Respir Care. 2001 Oct;46(10):1038-53.

It is sometimes helpful to introduce a hierarchical structure among the attributes, where broad categories are considered first and then within these broad categories, weights are assigned to attributes.  By convention the weights for the broad categories add up to one and the weight for the attributes within each category also add up to one.  The final weight for an attribute is the product of the weight for its category and the weight of the attribute within the category.  The following example shows the use of hierarch in setting weights for attribute. 

Chatburn & Primiano (2001) employed an additive, compensatory, multi-attribute utility model to assist the University Hospitals of Cleveland in their purchase of new ventilators for use in the hospitals’ intensive care units. A decision making model was useful in this instance because ventilators are expensive and complicated machines, and the administration and staff needed an efficient way to analyze the costs and benefits of the various purchase options, as even long-term equipment trials can be ineffective at identifying all the relevant strengths and weaknesses of a particular machine. The decision making model utilized input from relevant stakeholders regarding factors important to the decision to weigh the various purchase options. Each possible ventilator purchase was weighed according to the identified factors and ranked in terms of how well each option met each key decision factor. The model also provides an overall score which identifies the best purchase option, taken as a whole. The decision process began with an analysis of the hospitals’ current ventilator situation. Many factors suggested that the purchase of new ventilators would be advantageous. First, all of the ventilators owned by the hospitals were between 12 and 16 years old, while the depreciable lifespan of a ventilator is only 10 years. Thus, the age of the equipment put the hospitals at greater risk to experience equipment failures. Because ventilators are used primarily for life support, should this equipment fail the hospitals would be highly liable. Second, the costs to maintain the older equipment were beginning to outweigh the initial capital investment. Third, the current fleet of ventilators varied in age and model. Some ventilators could be used only for adults while others could only be used for infants or children, and different generations of machines ran under different operating systems. The result was that not all members of the staff were facile with every model of ventilator, yet it seemed impractical to invest in the type of extensive staff training which would be required to correct this problem. Therefore, the goals for the ventilator purchase were to advance patient care capabilities and increase staff competence, to reduce maintenance costs by acquiring ventilators which could be maintained via technical support updates and upgrades versus replacement of parts, and to reduce the capital inventory in order to lower both maintenance and staff training costs. To begin the selection process, the consultants wanted to limit the analysis to only the most relevant choices, those machines that were designed for use in intensive care units with an ability to ventilate multiple types of patients. In addition, it was important to select a company with good customer support and the availability of software upgrades. Also, as well as cost analysis, the analysis involved both a clinical and technical evaluation of each ventilator model. Each possible ventilator was used in the hospitals units on a trial basis for 18 months so that staff could familiarize themselves with each model. The technical evaluation utilized previously published guidelines for ventilators as well as vendor-assisted simulations of various ventilator situations so that administrators and staff could compare the functionality of the different models. A checklist was used in this instance to evaluate each ventilator in three major areas: control scheme, operator interface and alarms. Figure 1 depicts the attributes, their levels and relative weights used in the final decision model.

Figure 1: A Hierarchy for Assessment of Attribute Weights

Note that weights were first assessed across broad categories (cost, technical features and customer service). Two of these broad categories were broken into additional attributes. Weights for broad categories were assessed (note that these weights add up to one).  In addition, the weights for each attribute within the categories were also assessed (note that these weights also add up to one within the category).  In the end, the model had eight attributes in total and the weight for attributes was calculated as the product of the weight for broad category and the weight of the attribute within that category.

So how did this model perform in the real world? According to personal communication from the authors, there were two major problems experienced by the hospital staff. The first was reaching consensus on the weight of the cost factor. Ultimately, a more practical approach was to remove cost as a weighted factor and increase the weights of the remaining factors proportionately (i.e., technical features = 0.6 and customer service = 0.4.) Then, the score for each ventilator model was divided by its cost to give a more intuitively satisfying “benefit/cost” ratio. The second problem was more political than practical. Any rational decision can be based on only one of three strategies: (1) minimize cost, (2) maximize benefit, or (3) optimize the ratio of benefit/cost. In a hospital, physicians often have a large influence on purchase decisions when they argue from the standpoint of “what is best for the patient”, which usually ignores cost. Senior hospital administration, on the other hand, is usually more likely to base a decision on cost because they have to choose among many proposals competing for limited capital funds. Middle managers, like the authors of this paper, naturally sought a balance and took the benefit/cost approach. In the end, the physicians won out and the most expensive ventilator was purchased based on the perception that it provided the highest quality. As a result, the cost had to be spread out over 3 years with equal purchases of ventilators each year. Unfortunately, by the third year, funding for this project was no longer available. This shows how varied and unpredictable real-world factors can degrade an otherwise very logical decision process.

What Do You Know?

What are two methods for assessing values that should be assigned to single attributes?

What are two methods for aggregating  values assigned to different attributes into one overall score?

Make a numbered list of what to do and what not to do in selecting attributes?

Describe how attribute levels are solicited?  This question is asking you to describe the process of soliciting attribute levels and not any specific list of attributes or attribute levels.

Send your response by email to your instructor.  Include both the question and the answers in your response.  Include your contact information.

Biweekly Project

Construct a value function for a decision at work.  Select a decision that does not involve predicting uncertain outcomes (e.g. see examples).  Select an expert that will help you construct the model.  Make an appointment with the expert and construct the model.  Prepare a report that answers the following questions:

  1. Who decides?  What is the problem to be addressed? What judgment must be made and how can the model of the judgment be useful?
  2. What options were considered?  What attributes influence the desirability of various outcomes Search Medline to report if similar studies have been done using multi-attribute value or utility models.   
  3. What values were assigned by the expert to each attribute and its levels?
  4. How were single attribute values aggregated to produce one overall score?  What were the estimated weights?
  5. What is the evidence that the model is valid?
    • Is the model based on available data?
    • Did the expert consider the model simple to use?
    • Did the expert consider the model "Face valid"?
    • Does the model correspond with other measures of the same concept (Construct validity)?
    • Does the model simulate the experts' judgment on at least 15 cases?
    • Does the model predict any objective "gold standard."

Prepare your report using narrated Power Point slides (SWF).  Keep the following rules in mind in narrating a slide presentation: 

  1. Use a microphone mounted on your head and not set up on a table so that the distance between your mouth and microphone stays the same
  2. Put the recorder to the side of your mouth and not in front of it so that the noise of the air flowing out of your mouth is not recorded.
  3. Use embedded files in your narration so that you do not need to assemble the slide show at end.
  4. Write out all slides as fully as possible, narrate the slides and then cut as much of the text out as possible after the narration so as to provide the listener a visual guide. 
  5. Do not read the titles of the slides
  6. Be brief.
  7. Do not stop narration and start over from the middle.  Instead do the entire narration and correct the errors by narrating single slides (using ESC after narration) afterwards.

Send your report to your instructor and at least one other student in class.  Mail your report, do not email narrated slides (narrated slides take considerable space).  Keep a copy for yourself till end of semester.  Your instructor will ask for your entire portfolio of assignments at end of semester again.

The slide show should contain following slides:

  1. Title slide:  Include your initials and acknowledgement of the initials of the expert involved.
  2. Privacy slide:  Indicate if you give permission for posting of the slide show to the web.  Do not include any names of organization or persons anywhere on the slides
  3. Problem definition slide:  Describe who decides, what is the problem and why modeling could be of help.
  4. Attribute slide:  Describe what attributes affected desirability of various options and how the single attribute value functions were assessed.
  5. Weights slide:  Describe how the weights associated with all attributes were assessed
  6. Evidence of validity slide:  present data that the model constructed accurately reflects the experts opinions or can predict some objective gold standard 

Do not exceed 7 slides, do not exceed 2 minutes per slide.  To see how the slide show will be graded, click here

Here are example of work done by students (please note that these are not necessarily model projects):



To assist you in reviewing the material in this lecture, please see the following resources:

  1. Listen to lecture on modeling preferences (SWF file)

  2. See the slides for the lecture 

  3. See a video on how to create an X-Y scatter chart and calculate correlations (SWF file) in Excel.

  4. See a video on how to calculate correlations in Excel using Data Analysis Tool Pack

  5. Generate case scenarios

Narrated lectures require use of Flash.

More & References


Alemi F, Stokes J 3rd, Rice J, Karim E, LaCorte W, Saligman L, Nau R. 1987. "Appraisal of Modifiable Hospitalization Risk." Medical Care 25 (7): 582‑91.

Alemi, F., B. Turner, L. Markson, and T. Maccaron, 1990. "Severity of the Course of AIDS." Interfaces 21: 105‑6.

Alemi F, Walker L, Carey J, Leggett J. Validity of three measures of severity of AIDS for use in health services research studies. Health Serv Manage Res. 1999 Feb;12(1):45-50.

Aschenbrenner, K., and T. Kaubeck. 1978. "Challenging the Cushing Syndrome: Multi‑Attribute Evaluation of Cortisone Drugs." Organizational Behavior and Human Performance 22 (2): 215‑34.

Beach, L. R., F. L. Campbell, and B. D. Townes. 1979. "Subjective Expected Utility and the Prediction of Birth‑Planning Decisions." Organizational Behavior and Human Performance 24: 18‑28.  See also Wood RJ, Campbell FL, Townes BD, Beach LR.  Birth planning decisions.  Am J Public Health. 1977 Jun; 67(6): 563-5.

Bernoulli, D. 1938. "Spearman theoria novai de mensura sortus." Comettariii Academiae Saentiarum Imperialses Petropolitica 5: 175‑92. Translated by L. Somner in Econometrica 22 (1954): 23‑36.

Chatburn RL, Primiano FP Jr. Decision analysis for large capital purchases: how to buy a ventilator. Respir Care. 2001 Oct;46(10):1038-53.

Cline, B., F. Alemi, and K. Bosworth 1982. "Intensive Skilled Nursing Care: A Multi‑Attribute Utility Model for Level of Care Decision Making." Journal of American Health Care Association.

Cook, R. L., and T. R, Stewart. 1975. "A Comparison of Seven Methods for Obtaining Subjective Description of Judgmental Policy." Organizational Behavior and Human Performance 12: 31‑45.

Detmer DE, Moylan JA, Rose J, Schulz R, Wallace R, Daly R.. 1977. "Regional Categorization and Quality of Care in Major Trauma." Journal of Trauma 17 (8): 592.

Edwards, W. 1974. "The Theory of Decision Making." Psychology Bulletin 51 (4): 320.

Edwards, W. 1977. "How to Use Multi‑Attribute Utility Measurement for Social Decision‑Making." IEEE Transactions on Systems, Man and Cybernetics SMC7: 326‑40.

Ericsson, K. A., and H. A. Simon, 1980. "Verbal Reports as Data." Psychological Review 87: 215‑51.  See citations of this work.

Fischer, G. W. 1979. "Utility Models for Multiple Objective Decisions: Do They Accurately Represent Human Preferences?" Decision Science 10: 451.

Fryback DG, Gustafson DH, Detmer DE.   Local priorities for allocation of resources: comparison with the IMU.  Inquiry. 1978 Sep; 15(3): 265-74.

Gustafson, D. H., M. E. Hiles, and C. Taylor. 1980. "Report on the Trauma Severity Index Conference." Center for Health Systems Research and Analysis, University of Wisconsin‑Madison.

Health Services Research Group, Center for Health Systems Research and Analysis, University of Wisconsin. 1975, "Development of the Index for Medical Under‑service." Health Services Research 10 (2): 168‑80.

Hogarth, R. M. 1975. "Cognitive Processes and the Assessment of Subjective Probabilities." Journal of American Statistical Association 70: 271‑94,

Humphrey, S. P., and A. Humphrey. 1975. "An Investigation of Subjective Preference Orderings for Multi‑Attribute Alternatives." In Utility, Probability and Human Decision‑Making, edited by D. Wendt and C. Vlek, pp. 119‑23. The Netherlands: D. Reidel Publishing Co.

John, R. S., and W. Edwards. 1978. "Subjective versus Statistical Importance Weights: a Criterion Validation." Research report, Social Science Research Institute, SSRI 78‑7, University of Southern California.

Johnson, E. M., and G. P. Huber. 1979. "The Technology of Utility Assessments: Issues and Problems." IEEE Transactions in Systems, Man and Cybernetics. Kahneman, D., and A. Tversky. 1979. "Prospect Theory: An Analysis of Decisions Under Risk." Econometrica 47: 263‑91.

Kahneman, D., and A. Tversky. 1979. "Prospect Theory: An Analysis of Decisions Under Risk." Econometrica 47: 263‑91.

Kao, E. P. C. 1972. "A Semi Markov Model to Predict Recovery Progress of Coronary Patients." Health Services Research 7: 191‑208.

Keeney, R. 1980. "Analysis of Preference Dependencies among Objectives." Woodward and Clyde Consultants (July).

Keeney, R. 1977. "The Art of Assessing Multi‑Attribute Utility Functions." Organizational Behavior and Human Performance 19: 267‑310.  See citations of this work. 

Keeney, R. L., and H. Raiffa, 1976. Decisions and Multiple Objectives: Preferences and Value Tradeoffs. New York: John Wiley.

Kinzbrunner B, Pratt MM. Severity index scores correlate with survival of AIDS patients. Am J Hosp Palliat Care. 1994 May-Jun;11(3):4-9.

Kneppreth, N. P., D. H. Gustafson, and R. P. Leifer. 1974. "Techniques for Assessment of Worth." Technical Paper 254, U.S. Army Research Institute for the Behavioral and Social Sciences, Arlington, MD.

McNeil BJ, Pedersen SH, Gatsonis C. Current issues in profiles: potentials and limitations. In: Physician Payment Review Commission Conference on Profiling. No. 92-2. Washington, D.C.: Physician Payment Review Commission, 1992:46-70.

Naglie G, Krahn MD, Naimark D, Redelmeier DA, Detsky AS. Primer on medical decision analysis: Part 3--Estimating probabilities and utilities. Med Decis Making. 1997 Apr-Jun;17(2):136-41.

Nisbett, R., and L. Ross. 1980. Human Inferences. Englewood Cliffs, NJ: Prentice‑Hall.

Nisbett, R. E., and T. D. Wilson. 1977. "Telling More Than We Can Know: Verbal Reports on Mental Processes." Psychological Review 84 (3): 231‑59.

Pai, G. K., D. H. Gustafson, and G. W. Kiner. 1971. "Comparison of Three Non‑Risk Methods for Determining a Preference Function." Center for Health Systems Research and Analysis, University of Wisconsin‑Madison.  See also Gustafson DH, Fryback DG, Rose JH, Prokop CT, Detmer DE, Rossmeissl JC, Taylor CM, Alemi F, Carnazzo AJ.  An evaluation of multiple trauma severity indices created by different index development strategies.  Med Care. 1983 Jul; 21(7): 674-91.

Pascal, A., C. L. Bennett, and M. C. Bennett. 1989. "The Cost of Financing Care for AIDS Patients: Results of a Cohort Study in Los Angeles." National Center for Health Services Research Conference Proceedings on New Perspectives on HIV‑Related Illness, NCHSR publications, Rockville, MD.

Pliskin, J. S., D. S. Shepard, and M. C. Weinstein. 1980. "Utility Functions for Life Years and Health Status." Operation Research 28 (1): 206.  See also related articles

Redfield, R. R., and D. S. Burke. 1988. "HIV Infection: The Clinical Picture." Scientific American 259 (4): 90‑98.

Rose W, Maffetone M, Suarez E, Whisler K, Bielitzki LD. A model for selecting a clinical laboratory information system. A four-phase process. Clin Lab Manage Rev. 1992 Jan-Feb;6(1):18-9, 22-4, 26-9.

Savage, L. D. 1972. Foundations of Measurement. New York: John Wiley.

Scitovsky, A. A. 1989. "Past Lessons and Future Directions: The Economics of Health Services Delivery for HIV‑Related Illness." National Center for Health Services Research Conference Proceedings on New Perspectives on HIV‑Related Illness, NCHSR publications, Rockville, MD.

Sessa EJ. Information is power: the Pennsylvania experiment. J Health Care Benefits 1992;Jan-Feb:44-8.

Snyder, M., and N. Cantor. 1979. "Testing Theories about Other People: Remembering All the History that Fits." Unpublished manuscript, University of Minnesota.

Torgerson, W. S. 1958. Theory and Methods of Scaling. New York: John Wiley.

VanGundy, A. B. 1981. Techniques of Structured Problem Solving. New York: Van Nostrand Reinhold.

Vibbert S. Illinois Blues target doctors. Medical Utilization Review. April 2, 1992.

Volkema, R. 1981. "Different Approaches to Problem Solving." Ph.D. thesis, University of Wisconsin.

Von Winterfeldt, D., and W. Edwards. 1986. Decision Analysis and Behavioral Research. New York: Cambridge University Press.

Warner, H. 1971. "Estimating Coefficients in Linear Models: It Don't Make No Never Mind." Psychological Bulletin 83 (2): 213‑14.

Yates, J. F., and C. M. Jagacinski, 1979. "Reference Effects in Multi‑Attribute Evaluations:" Organizational Behavior and Human Performance 24 (3): 400‑410.

Hawthorne G.  The effect of different methods of collecting data: mail, telephone and filter data collection issues in utility measurement. Qual Life Res. 2003 Dec; 12(8): 1081-8.

Additional Reading (logoff after each article, requires library membership)

Other sites

  • Links to online databases of Quality of Life analyses can be see at
  • Examples of use of multi-attribute utility models in health care can be found through Medline    
  • For a Google scholar search of Multiattribute Utility models in health care click  here
  • For James L. Corner and Craig W. Kirkwood, ``Decision Analysis Applications in the Operations Research Literature, 1970--1989,'' Operations Research, Vol. 39, No. 2, pp. 206--219 (March-April 1991) click here

Recently Asked Questions

In this section, you will find answers to questions asked by you or others.

Question: The lecture talks about using hypothetical cases to validate the model. Are the hypothetical cases actually based on real cases but identified as hypothetical for privacy reasons? Using hypothetical outcomes to validate a model seems odd.  Answer: Test cases should make sense, so for example, when building model for hiring decision test cases should correspond to candidates that may apply. If data about real cases is available it can be used, but it is important to test models on a diverse set of cases, to make sure it performs well in different situations.   Asked on 1/30/2010 7:35:59 PM and answered on 2/17/2010 12:11:55 PM.

Question: Can you explain the multiplicative model formula in more detail? I wasn't able to get a thorough understanding of the variables and how they inter-relate.  Answer: This formula looks complex because of complicated normalization of the score into [0, 1] rage. You can find its many properties in different papers when you, for example, search google. I will explain intuition and simple mathematics behind the formula. A way to combine single attribute value scores is to multiply them, i.e. calculate V(A1) V(A2) V(A3)...V(An). The problem with multiplication is that even if a single value V(Ai) equals 0, the entire product equals 0. To solve this problem we can multiply (1 + V(A1)) (1 + V(A2)) ... (1 + V(An)). Now, we may want to add weights to indicate importance of attributes, we have (1 + k1 V(A1)) (1 + k2 V(A2)) ... (1 + kn V(An)). Be cause we want the final score to be in the range zero to one, we need to things: (1) minimum possible value of the score should be when all V(Ai) = 0. When we insert all zeros in the product above we get value 1 (this is the smallest value we can get from the product). We subtract 1 and the minimal value for the score is 0. We have -1 + (1 + k1 V(A1)) (1 + k2 V(A2)) ... (1 + kn V(An)). Similarly (2), we can calculate maximum possible value when all V(Ai) = 1 and get that that the formula needs to be divided by -1 + (1 + k1) (1 + k2) ... (1 + kn) to have maximum value 1. By using k ki instead of only weights ki we add extra smoothness to the formula.   Asked on 6/1/2008 10:50:47 PM and answered on 6/2/2008 10:27:45 PM.

Question: The reading mentions the usefullness of constructing practice profiles using severty indexes. Could this method be applied to other professions outside of healthcare?   Answer: Multi-attribute utility model has been applied to many situations outside of health care. A search within Google for the terms MAU model will show many examples in different industries.    Asked on 2/10/2008 8:48:18 PM and answered on 2/11/2008 7:52:37 AM.

Question: How do you ensure that you do not offend the expert, when you ask questions to make the expert answer more specifically?  Answer: First, you make sure that you use the same vocabulary as the expert. Second, you do not introduce an idea of your own but ask for clarification and third you acknowledge that you do not have expertise in the area.   Asked on 2/10/2008 3:23:48 PM and answered on 2/11/2008 7:43:53 AM.

Question: When creating attributes values, what is the affect of scale size and direction? For example, in the examples given, the direction of Blood Pressure implicitly means higher values tend to be worse than lower values. In contrast, the direction of other attributes’ measures -- such as T-cell scores -- might move in a different direction. This raises two questions: (1.) Does direction of attribute values need to be uniformly scaled (e.g., best to worse or good to bad or no condition to severe manifestation, etc.)? (2.) Do the attribute value scales of a given model require comparable scaling (e.g., five-point scale versus three or seven intervals)?  Answer: The assignment of numbers to attribute levels should be consistent across attributes. For example, in measuring severity higher numbers in all attributes should always mean more severe cases. The functional form of the attribute scores within each attribute is, of course, variable. Meaning that scores can go up and down depending on attribute level being considered. But higher scores across attributes always mean the same thing. There is no need for all attributes to be defined in the same fashion, e.g. five point scale. Attributes should be defined based on what makes sense to the decision maker.    Asked on 2/11/2007 11:36:57 AM and answered on 2/23/2007 9:38:19 AM.

Question: I really understood the lecture for the most part, however, I'm still unclear on evaluating validity. Please explain as simple as you can, what R2 represents and how one should interpret it. For instance, my results were: R2 = 0.5895. I'm not sure if I fully understand the results.  Answer: R2 represents the percent of variance of the experts' opinions explained by your model. You may be familiar with R which is the correlation between the experts' opnions and the model.   Asked on 2/9/2007 12:23:24 PM and answered on 2/10/2007 10:04:38 AM.

Question: Can the expert we interview be a non-decision maker? a Team lead for example?   Answer: Yes, as long as this person has expertise in the topic you want to model   Asked on 2/4/2007 1:40:53 PM and answered on 2/6/2007 5:39:29 PM.

Question: For the biweekly projects, it calls for the model evaluation of at least 15 cases. In the absence of real data, are we to use random sceranios, or must we generate cases using orthogonal design. In many cases, this would result in more than 15 cases, and might be burdensome for our experts to review and evaluate.  Answer: The best approach is to generate the full set of scenarios and choose 15 of them randomly.    Asked on 2/2/2007 2:21:35 PM and answered on 2/6/2007 5:37:12 PM.

Question: What if we cannot find an expert to interview a health-related topic? (especially for those of us who are full-time graduate students and not working). Could the modeling preferences bi-weekly project be non-health related?  Answer: Yes, if you do not want to stay in a career within health care. In either case it needs to be management related and not clinical. In the past, full time students have found a friend or a classmate that has a specific expertise. Many faculty can also be interviewed. This is a good opportunity for you to get inside a health care issue.   Asked on 1/31/2007 3:59:51 PM and answered on 2/2/2007 7:35:38 AM.

Question: The chapter mentioned an easy method, which requires no numerical assessment at all for assessing single value functions. However, the example provided was not an ideal example as stated in the text. Thus, can you provide an example in which this method would be ideal. Thanks.  Answer: A good example for using linear extrapolation for single attribute values arises in measuring severity of patients on multiple highly correlated attributes. In these circumstances, each attribute is arranged so that it increases the overall severity of the patient. In addition, the best is typically "Normal" and the worst is typically the highest level of the attribute. For example, Normal blood pressure versus blood pressure exceeding 140. If the best and the worst are assigned 0 and 100. All remaining levels can be assigned by linear extrapolation.   Asked on 6/1/2006 4:09:06 PM and answered on 6/6/2006 4:15:27 PM.

Question: For students that don't work at the moment, who can be the expert?   Answer: A family member or a friend who is working can be the expert.   Asked on 5/23/2006 9:59:51 PM and answered on 5/26/2006 11:48:26 AM.

Question: Dr Alemi, after developing the case senario to be rated by the expert. The expert will rate the accuracy for each developed hypotheical case between (0-100). To develop the correlation, I will be ploting these rating against the weight of each attribute? I am confused n this part  Answer: Once you have obtained the exprt's ratings of each scenario, then you use your model to score each scenario. Thus if you have $ attributes, then you use the value of these 4 attributes to calculate an overall score for the case. Then you correlate the case scores against the experts' ratings   Asked on 2/15/2006 2:03:33 PM and answered on 2/15/2006 10:23:43 PM.

Question: I had trouble distinguishing between attributes being dependent and not redundant. If the attributes are dependent or related does that mean that I have to reconsider them?  Answer: The concept of dependence and redundance are related concepts. Preferential independence is a mathematical relationship that suggests the form of the multi-attribute value model. Redundance is used with no special mathematical relationship in mind. When ever dependence and redundance are discovered, the best approach is to reformulate the list of attributes so that this problem goes away.   Asked on 2/12/2006 12:38:42 AM and answered on 2/12/2006 12:12:50 PM.

Question: I would like to confirm my understading. "The expert would rate scores following value of each attribute, and then the expert would rank the scale from 0 to 100. When we have 2 values we will calculate correlations through excel program". 1 Is this idea right or wrong? 2 Is the Modeling Preference suitable for only clinical problems? (because of most your examples focusing on clinical situations)Generally, the diseases are evaluated by standard of diagnosis or knowledge of medical care. This strategy is very clear to differentiate each health status of the patient. Why we use the model to compare? Is it appropriate to use for administration or other problems?   Answer: A scenario is a collection of attribute levels. The experts are provided scenarios and asked to rate them on a scale from 0 to 100. The expert's rating is correlated against the model score for the same scenario. The procedures are useful for any modeling effort, whether or not it is clinical. Examples of appropriate application of multiattribute value model are provided in the begining of the lecture and including activities such as budget allocation, defining underserved health care areas, evaluating applicants, evaluating land purchases, etc.   Asked on 2/11/2006 2:09:16 PM and answered on 2/12/2006 12:03:34 PM.

Question: How long should our powerpoint presentation be (minutes and slides)?  Answer: Not longer than 15 minutes, the shorter the better. The various slides that should be included are provided in the lecture assignments, but generally the same as the student work already posted   Asked on 2/7/2006 11:48:57 AM and answered on 2/7/2006 10:14:11 PM.

Question: I think convincing non-academic healthcare business experts to sit through long meetings where they are grilled on their area of expertise will take some doing. The analyst must have to practice going over the questions and establish goals and objectives for these meetings to make them as short as possible. Most high level people I know about have no time for this. So, do you think it is necessary in our assignments to identify a serious problem or issue area they face as an incentive for them to grant us imterviews designed to help them solve their problems?   Answer: Yes, the more the problem is important for them the more likely they are willing to give of their time.    Asked on 2/5/2006 3:27:29 PM and answered on 2/6/2006 9:45:54 PM.

Question: For Question #3, are we supposed to answer it based on the AIDS severity index example provided in the text.  Answer: I am not sure to what you are referring-In the what do you know section- this comes straight from the lecture notes on what you do and do not do when attaining the attributes- for example- you do want to use the experts own terminology. The biweekly project requires you to work a preferential driven decision from your work. Please let me know is this does not answer your question.    Asked on 6/5/2005 10:43:33 PM and answered on 6/7/2005 10:33:19 PM.

Question: I think different examples of calculations would be beneficial  Answer: Thank you- your suggestions will be noted.   Asked on 6/2/2005 6:41:30 PM and answered on 6/3/2005 11:14:46 PM.

Question: Where can you hear the biweekly projects done by the student?  Answer: You will be able to hear some slected bi weekly projects online at the place that where the biweekly project is announced. It usually takens several weeks for these projects to be posted   Asked on 2/16/2005 9:04:04 PM and answered on 2/16/2005 9:52:07 PM.

Question: It was seem hard at the beginning but after practicing more reading, following the examples, and benefits from other questions & answers, it looks fine and easy, just it took more time to get the model ready, thanks for your effort.  Answer: Well I am glad to hear this. I was quite worried about this first assignment, glad to see you and others are doing ok.   Asked on 2/14/2005 8:38:05 PM and answered on 2/14/2005 8:46:10 PM.


Suggested Changes


Following comments were made recently about this page:

Suggestion: Will I have to work on Saturdays? <a href=" ">clomid how long to get pregnant r kelly</a> She does, however, pay touching tribute to her son, saying that she overcame her instinctive reluctance to write the book in the hope that it might one day be read by any children he might have. She wants them to learn first what a fine grandfather they would have had, and secondly “what a mentally strong and resolute man is their father”. If future Tebbutts ever do read grandma’s memoirs, they will think she was made of pretty strong stuff, too.   Left on 7/13/2015 11:27:00 PM.

Suggestion: We've got a joint account <a href=" ">celexa antidepressant side effects hallucinations</a> Terry Holt, a longtime Republican strategist, said Mr. Obama’s strategy rests on a cold-eyed calculation that Republicans are the ones with the most to lose. “As long as the president thinks his poll numbers are going to be good, I don’t expect the government to reopen,” he said. <a href=" ">increasing citalopram dosage side effects me</a> "Bed occupancy has now reached the potentially worrying level of 85% nationally. It is generally accepted that this is the maximum level at which hospitals can operate safely and that occupancy above 85% can cause increased risks and critical problems for patient care." <a href=" ">celexa medicine side effects erectile dysfunction</a> WFAN radio host Mike Francesa discussed the failed negotiations on Tuesday during an interview with Daily News baseball columnist Bill Madden. The News reported on Aug. 3 that Rodriguez’s claims that MLB and the Yankees were conspiring to keep him off the field in order to void his contract ended any settlement talks with baseball officials and led to the 211-game ban imposed by commissioner Bud Selig. <a href=" ">citalopram increased dose side effects erectile</a> "This is a dream come true," Kipsang told The Associated Press. "Ten years ago, I watched (Kenya's) Paul Tergat break the world record in Berlin (in 2:04.55) and now I have achieved the dream. I felt strong, so I attacked at 35K, because the pace had become a little too slow."   Left on 7/10/2015 12:55:50 PM.

Suggestion: Thanks funny site <a href=" ">ventoline 0 4 mg dosering</a> "I say to them, 'Let's play Westgate.' And they grab a stick and say it's a gun, or they hide behind something and say they are the terrorist. It can bring on discussion. It's a way to get it out of them. It becomes less painful. I talk to them, and say, 'OK so the terrorist went there? Where could you have gone to get away?'" <a href=" ">skelaxin not working</a> Unfortunately, the same cannot be said of the eurozone. Like the US, the pace of fiscal consolidation is set to slow considerably next year, but barely a start has been made on household and banking sector repair. <a href=" http://WWW.PATH-23.ORG/nexium-or-prilosec-otc/ ">nexium mups 20 mg english</a> "I was hoping to go back to Goldman Sachs and in themeantime make the SEC to understand this transaction," Tourre,34, told the court. Instead, he ended up on trial in federalcourt in New York on securities fraud charges.   Left on 7/2/2015 9:22:08 AM.

Suggestion: I quite like cooking <a href=" ">is metoprolol tartrate the same as toprol xl</a> China has spoken clearly: its government wants to orchestrate a "soft landing" for its economy as it tries to slow the rate of inflation. Policy makers are seeking to constrain the money supply to suppress what they perceive as possible price bubbles in areas like real estate. They would prefer to have Chinese growth settle into the target 7.5 percent range, down from the 9 to 10 percent growth rates seen from 2008-2011. <a href=" http://www.xn--80aaghfnqt9b8g5a.xn--p1ai/buy-leather-tricorn-hat/#bellamy ">lipanthyl 300 mg fenofibrate</a> “Given the absence of a timely review of those practices with corresponding explanations and commitment to cease the interception activities, the conditions for a visit on the previously agreed-upon date are not present,” she said. <a href=" ">thuoc flagyl 250mg chua benh gi</a> Analysts had expected the session to be subdued as there waslittle market-moving news and investors are waiting for anupdate on the outlook for U.S. monetary stimulus after the Fed'smeeting ends on Wednesday.   Left on 7/2/2015 9:11:23 AM.

Suggestion: Could I make an appointment to see ? <a href=" ">fluticasone furoate nasal spray long term safety</a> Senate Majority Leader Harry Reid, a Democrat, and hisRepublican counterpart, Mitch McConnell, ended a day of constanttalks with optimistic proclamations, as details leaked out ofthe pact they were negotiating. <a href=" ">how much does trileptal cost</a> It will be heard — loud and clear — on YES, but only when Francesa yaks. As long as the A-Rod circus is in town, Francesa will be the ringmaster blabbering on about A-Rod’s raw deal, or the Yankees’ money-grubbing ways, or how Rodriguez is the world’s most picked-upon abuser of performance-enhancing drugs. <a href=" ">can i buy terbinafine pills over counter</a> After giving it a quick filter through a fine sieve and some clean cheesecloth, I started the evaporation process, running my oven at its lowest setting for three days and turning it off overnight. Little by little, tiny fleur de sel started growing along the sides of the three bowls in my oven as the Sound water disappeared. By the time the water was gone, there was a layer of chunkier crystals growing from the bottom of the bowl as well.   Left on 7/1/2015 7:06:46 AM.

Suggestion: I'd like a phonecard, please <a href=" ">is it legal to order domperidone online</a> “He has already had many job offers. This issue is being considered now. I’m absolutely confident that Edward will find a job, and will find a very well-paid job as he’s very skilled,” Kucherena said. <a href=" ">500mg amoxicillin for tooth infection</a> The idea of fostering young companies at a time of sluggisheconomic growth had strong appeal, and those critics concernedabout the erosion of investor protections were only able to makemodest changes to the legislation. <a href=" ">buy ibuprofen online uk</a> He had almost ended it in the 11th, leading off that inning with a single off the glove of shortstop Adeiny Hechavarria. Ruben Tejada advanced him to second with a sacrifice bunt and he became the first Mets runner to reach third on Josh Satin's ground out. He was stranded when Eric Young, Jr. struck out.   Left on 7/1/2015 12:51:27 AM.

Suggestion: I want to report a <a href=" ">cost buspar</a> "It's a big relief," said Jenifer Bakki, who works at the Grand Canyon Squire Inn a couple of miles from the park's entrance, which went from full capacity to less than half full following the shutdown. "Ever since the park closed we have slowed down," she said. <a href=" ">eskalith cr generic</a> The aid package is a far cry from the $80 billion infinancing extended to the U.S. auto industry during the2008-2009 financial crisis that saved General Motors Co and Chrysler Group LLC from collapse. <a href=" ">flagyl forte 500 mg untuk</a> On Peru’s Iquitos-Nauta Highway, I was photographing under a waning moon along a new asphalt road that bored into recently lacerated ancient forests when I was swarmed by these miniature Draculas, desperate to suck through my skin. Amazingly, few penetrated my completely buttoned shirt. Even more fortuitous, those that did bite did not carry the proper protozoa, which is carried only by some female mosquitoes. Male mosquitoes have little passion for blood, while their mates depend on it.   Left on 6/30/2015 11:07:51 PM.

Suggestion: I'd like to send this to <a href=" ">generic pantoprazole sod dr 40 mg</a> Don’t expect to see the feature go live any time soon, though. Despite their coaxing, the 9to5 team wasn’t able to get the feature running on current generation Apple hardware, which they’re taking as a sign that Mogul is actually meant for a forthcoming version of the iPhone. <a href=" ">best place to get nolvadex</a> "This is really bad luck after (tropical storm) Manuel,"said Anaberta Lopez, 26, who works on the city's beachesbraiding tourists' hair, surveying a vacant stretch of sand."There's no work now and people here live off tourism." <a href=" ">chlorpromazine equivalents table</a> In some cases firms will claim current regulatory assets under management (RAUM). This practice may be scrutinized but is not prohibited. The figures should be accurate and from a recent period. It would be especially misleading if a firm had a large change in RAUM and did not reflect the change on its public website.   Left on 6/30/2015 7:28:42 PM.

Suggestion: Through friends <a href=" http://www.xn--80aaghfnqt9b8g5a.xn--p1ai/bisacodyl-5-mg-dose/ ">dulcolax overnight relief laxative 50 reviews</a> Barbara Young, chief executive of Diabetes UK, said that restricting access was not only causing distress, but would also create a problem for the NHS in the long term, as complications caused by lack of monitoring mounted up. Diabetes already costs the NHS around £10bn annually. <a href=" ">cephalexin 250 mg/5ml susp</a> Robert Sims, 35, of Brooklyn, was arraigned Saturday in Manhattan on charges of first-degree gang assault, first-degree assault and third-degree criminal possession of a weapon. Bail was set at $100,000. <a href=" ">cheap cipralex</a> For what? First, for a bit of ogling of the boys as they surf. “They’re like young gods!” coos Roz, as she watches Lil’s now 20-year-old son Ian (Xavier Samuel) and her own twentyish son Tom (James Frecheville) stride ashore. The bland, affectless “young gods” she’s referring to spend long days lounging with their mums on a raft. Then, when he’s alone with Lil, Tom abruptly kisses her.   Left on 6/29/2015 10:00:40 PM.

Suggestion: Your account's overdrawn <a href=" ">topamax 150 mg</a> NEW YORK - Small business owners plan for growth this year but are closely tracking recent interest rate rises and any impact they might have on their business and customers, according to a spot survey of firms in the New York region. <a href=" ">buy maxalt cheap</a> Former Adams County, Colo., prosecutor Bob Grant believes Holmes’ lawyers are looking for any issue they can use to launch an eventual appeal. The defense believes having Holmes appear in restraints may also prejudice the jury, Grant said. <a href=" ">ipratropium bromide nasal spray price</a> The dollar lost momentum after rising initially inanticipation of an end to the fiscal impasse, and fell to its intraday lows versus the yen after the U.S. House ofRepresentatives approved a deal that had already been passed bythe Senate.   Left on 6/29/2015 5:48:12 PM.

Suggestion: It's serious <a href=" ">hydroxyzine hcl (atarax) 25mg tab</a> “We look forward to restoring the building, creating dozens of good paying health-care jobs, and helping address the medical needs of the community,” said Kaufman, whose Harmony Group has a track record of transforming upstate properties into modern housing and commercial space. <a href=" ">finasteride online uk</a> PepsiCo Global Investment Holdings Ltd, which providesfinancing to other companies in the drinks group and is one of14 Irish subsidiaries, made a profit of almost $6 million in2011 and paid tax of $215 to Curacao, giving it a rate of 0.004percent, Irish company records show. <a href=" ">best treatment for scabies permethrin</a> It is inappropriate for Obama to be commenting on a case, in which he displays obvious bias and pre-judgment. Mr. Zimmerman tutored black children, took a black girl to his prom, and had some blacks stay in his home. He doesn’t have a prejudicial bone in his body.   Left on 6/29/2015 5:20:10 PM.

Suggestion: Until August <a href=" ">artane 2 mg 50 tb</a> Josh, who previously reviewed the PureView 800, called that phone one of the best, if not the best, smartphone cameras out there. And in his most recent post, he explains why he thinks the Lumia 1020 will likely be equally as impressive. <a href=" ">2000 mg amoxicillin h pylori</a> Look party perfect in a pastel lace playsuit and show off your effortless style this summer. A great wardrobe staple that will take you from day-to-night dressing with complete ease, a playsuit is a stylish alternative to a party dress. <a href=" ">order liquid amoxicillin online</a> “If you’re gonna have a quarterback that has those skills, like a (Colin) Kaepernick or somebody like that, then obviously it becomes more of a weapon, because you’re defending, you’re not just knowing that Tebow’s going in there and he’s just going to run the ball,” the coach said. “And that’s kind of where we were last year it seemed like.”   Left on 6/29/2015 8:13:34 AM.

Suggestion: I'm on business <a href=" ">buy elimite cream</a> WASHINGTON, Sept 26 (Reuters) - Nine companies based inJapan and two executives have agreed to plead guilty and to payalmost $745 million in fines for their roles in long-runningconspiracies to fix the prices of auto parts sold to U.S. carmanufacturers, the Department of Justice said on Thursday. <a href=" ">canesten clotrimazole thrush treatment 3 day</a> He disappeared from public view in the summer of 2012 and speculation swirled for weeks about his condition. At first Jackson Jr. said he was being treated for exhaustion, and his doctor said in July 2012 said he was being treated for a "mood disorder." <a href=" ">order finasteride canada</a> About 90 investor orders took part in the deal, consistentwith Ford's prior two transactions this year. Orders were saidto be in the US$30m to US$40m-size range from investors, asopposed to larger US$100 m requests.   Left on 6/28/2015 11:36:04 AM.

Suggestion: We need someone with qualifications <a href=" ">price of valtrex</a> In the study, researchers took a group of eight adults averaging around the age of 30 to Colorado's Rocky Mountains for a week while wearing a wristband device that measures light exposure. Campers could use only natural sources of light, such as sunlight and a campfire, and abstained from using computers, flashlights, and mobile phones. <a href=" ">buy zanprol 10mg tablets omeprazole</a> According to BART officials, that makes it difficult to make technological changes like having station agents file reports by e-mail instead of writing them out longhand, using e-mail instead of fax machines to send documents and sending paycheck stubs to each work location electronically instead of hand-delivering them. <a href=" ">albendazole 400 mg tablets formulation</a> Analysts say South Sudan might collapse without oil, themain source for the budget apart from foreign grants. They pointto recent looting of aid agencies by soldiers as a sign thatJuba is struggling to pay salaries.   Left on 6/28/2015 10:22:19 AM.

Suggestion: How do you know each other? <a href=" ">generic paxil blue pill</a> Lincoln may have grown a beard for another reason. He knew he wasn't a particularly attractive man, and often cracked jokes about his homeliness. During one of his famous debates with Stephen Douglas, Douglas called him two-faced. Lincoln turned and appealed to the crowd. "If I had another face," he asked, "do you think I'd wear this one?" <a href=" ">tamoxifen induced endometrial cancer</a> I believe that the writer is pointing to the pillars of industry, just as the news people did at the beginning of the great depression. I had two good teachers, my dad and granddad that went through it. To quote them, “they said the pillars of industry would stand and not fall, but they fell”. We are at that point again and greed is still king in industry and government. Hang on tight, here we go. <a href=" ">clomid over the counter uk</a> Meanwhile, last year Pizza Hut pulled off a similar publicity stunt which went terribly awry. The original marketing pitch offered a lifetime supply of pizza to the person who asked US presidential candidates their pizza topping preference during a live broadcast of the debate.   Left on 6/28/2015 10:22:18 AM.

Suggestion: Do you know what extension he's on? <a href=" ">methotrexate mg/kg</a> However, I also see that perhaps half of those transgressing are listening to their iPods. I don’t think that any helmet will save them from being crushed by a lorry that they do not hear coming. <a href=" ">hw much des zyban cst in canada</a> Sen. Lisa Murkowski, the committee's ranking Republican,ended the hearing by saying that she "reluctantly" could notsupport the nomination. The Alaskan said she was "not convinced"that Binz's views were compatible with FERC's mission. <a href=" ">phenytoin first order</a> "Chariot's share price in (the first half of 2013) wasweighed down by disappointing results from HRT ParticipaçõesSA's drilling campaign offshore Namibia ... however,management has today embarked on its own initiative to createvalue and generate new interest in the stock."   Left on 6/28/2015 10:22:13 AM.

Suggestion: I work for a publishers <a href=" ">buscopan equine dose</a> During the same period, total wine consumption rose from 5 billion to 8.8 billion bottles.   Left on 6/25/2015 5:09:40 PM.

Suggestion: In a meeting <a href=" ">buy septilin online</a> Stocks have slumped since the start of the month on speculation that the Federal Reserve could raise its benchmark interest rate in June as hiring continues to improve <a href=" ">rebetol</a> Struggling apparel retailer Wet Seal Inc, which last month warned that it could file for bankruptcy protection if it failed to meet immediate liquidity needs, said on Wednesday that it laid off about 3,700 full- and part-time employees as it closed 338 stores.   Left on 6/24/2015 11:23:25 AM.

Suggestion: this post is fantastic <a href=" ">buy starlix</a> C John Jaso will be placed on the 7-day concussion list before the A's play Houston. <a href=" ">purchase arava </a> They had been chanting slogans against Mubarak and against Sisi and Mursi, the two men who have served as president since him   Left on 6/24/2015 9:03:30 AM.

Suggestion: Punk not dead <a href=" ">where to buy dulcolax suppositories</a> The diversified group is also exploring more "risk-sharing"deals under which the company would receive a fee linked to thesuccess of some of its products including in-vitro fertilitytreatments, the newspaper said on Monday, citing StefanOschmann, executive board member. <a href=" ">venlafaxine hydrochloride overdose</a> They contend that keeping private the initial discussionswith regulators spares companies the heartburn of putting outfinancials and then being prevented from elaborating on them dueto "quiet period" rules. Those rules prevent company executivesfrom making public statements related to information in thefiling. <a href=" ">lisinopril dosage 40 mg</a> Low interest rates in advanced economies drove many investors overseas in search of higher returns. Now some developing economies are witnessing rising borrowing costs and weakening exchange rates as those investors pull out, Mr. Bean said.   Left on 6/23/2015 5:13:57 PM.

Suggestion: I'm afraid that number's ex-directory <a href=" ">buy cyproheptadine hcl</a> Andrew Parker, the new head of MI5, used his first speech to launch a scathing attack on the leaks warning they had “gifted” the terrorists the ability to attack “at will”. <a href=" ">betnovate n skin cream india</a> Goerner’s ability to turn sounds into coiled expressive gestures was revealed time and again in Chopin’s 24 Preludes. Sometimes it was too much, as in the G major Prelude, where gesture swallowed up the music entirely. In the final D minor Prelude it was a marvel, turning a small piece into a mighty expression of steely tragedy. In these little masterworks something only glimpsed before really stood out: Goerner’s ability to set a melody in relief by slightly anticipating the beat. It was a wonderful evening in many ways, but it would be good to see how Goerner handles music that relies less on colour and gesture. At his next Wigmore Hall date (on September 11) we’ll have a chance to find out. <a href=" ">diamox costo</a> Hannah Bishop, Infiniti UK Fleet Manager, said: “A car of this size, power and standard specification for under £300 a month (plus VAT) is a very attractive and competitive proposition for the business user.   Left on 6/23/2015 11:42:20 AM.

Suggestion: Whereabouts in are you from? <a href=" ">glycomet 500 sr tablet purpose</a> Of course, the Washington Post Co. still owns, and is trying to sell, its building. It's still holding fast to its profitable enterprises like Kaplan and TV and cable channels that bring in needed funds. But my guess is that the $80 million that is listed by the D.C. government for that office building is way on the low side. So bricks and mortar and baseball are worth more than storied newspapers. <a href=" ">does seroquel 50 mg cause weight gain</a> However, on being asked by Nomura Equity Research analyst Romit Shah in a post-earnings conference call, Chief Executive John Daane said "we would expect that there will be more ramp transitions over time that will benefit Altera from a market share perspective, but there is nothing specifically that happened in the third calendar quarter or in the fourth calendar quarter." <a href=" ">weaning off 25 mg amitriptyline</a> On Friday, Egyptian security officials told the AP on Friday that a drone firing from the Israeli side of the border had killed five suspected militants preparing to fire rockets at Israel. The site of the strike was about five kilometers (three miles) inside Egypt. The officials spoke on condition of anonymity as they were not authorized to brief journalists.   Left on 6/23/2015 9:36:44 AM.

Suggestion: Hello good day <a href=" ">cheap tricor 145 mg</a> Even if it successfully defends itself from class actionlawsuits by aluminium manufacturers, the LME may have to acceptgreater external oversight into a trade that until nowflourished with little external supervision. <a href=" ">contain miconazole clotrimazole or tolnaftate</a> According to the insider, the couple pulled the plug on their romance after realizing they were in two different places, with Dobrev reportedly not ready to settle down as much as her 10-years-older beau. <a href=" ">hard money financing</a> "It is akin to saying we won't have an London Metal Exchange, you'll have to go to Chile to get your copper. It is blatantly inefficient - though in terms of politics and development, if I were president I'd do the same."   Left on 6/23/2015 9:35:38 AM.

Suggestion: I'm on work experience <a href=" ">cheap finasteride canada</a> He has hosted a debate on his blog, after receiving an e-mail from a reader, who suggested that the much-maligned McDouble actually provided far more nutritional bang for its buck than had conventionally been thought. <a href=" ">1 clotrimazole cream</a> More games, of course, would start with one. The Jayhawks' last conference victory came against Colorado in 2010, and Weis is still drawing more attention for his blunt assessments of Kansas football. <a href=" ">imitrex canada</a> Both schemes were allegedly orchestrated by Sandy Winick,55, a Canadian who has lived in China, Thailand, Vietnam and theUnited States, according to an indictment. Winick remains atlarge and is presumably hiding in Thailand, according to theU.S. attorney's office in Brooklyn.   Left on 6/23/2015 6:00:49 AM.

Suggestion: Where's the nearest cash machine? <a href=" ">starting allopurinol treatment</a> The program was started in 2006 by Guy Rancourt, a Queens native who played for Jack Curran at Archbishop Molloy. Rancourt wanted to give American kids a chance to broaden their minds and their games. <a href=" ">small cash loans in pennsylvania</a> "China is likely to overtake the United States (as an oilimporter), and Russia has to stake its claim in China," saidAndrey Polishchuk, analyst at Raiffeisenbank in Moscow, sayingsome oil flows to Europe would be redirected eastward. <a href=" ">metronidazole flagyl 250 mg</a> While “Jasmine” is spiced with light comedy, and splinters some of Williams’ characters into two, Blanchett has the toughest task: making Blanche Dubois into a modern Woody Allen heroine. The Australian Oscar-winner knows the terrain — she’s led a “Streetcar” production Down Under — and as her twitchiness turns to anger and panic, what seems like a sketch becomes a mosaic. The way she anchors this superb dramedy is a thing of beauty.   Left on 6/23/2015 1:28:13 AM.



This page is part of the course on Decision Analysis, the lecture on Preferences.  It was last edited on 05/12/2003 by Farrokh Alemi, Ph.D© Copyright protected.  This page is based on a chapter with the same name published by Gustafson DH, Cats-Baril WL, Alemi F. in the book Systems to Support Health Policy Analysis:  Theory, Model and Uses, Health Administration Press: Ann Arbor, Michigan, 1992.