Decision-Making Theory: Implications for Academic Advising

Introduction

Everyone wants to make good decisions, but often the decisions that are made do not achieve their intended goals. This is particularly true for college students deciding on a major. They initially select majors that, according to their personal rationale, will enable them to accomplish certain personal goals. Often their logic is unexamined and their decision-making skills undeveloped, and as a result, their goals go unmet. When this occurs, students change their rationale and goals and eventually change majors. Nationally, 50 percent of college students change their educational plans at least once, with many doing so two and three times (Ronan, 2005).

The National Academic Advising Association states that one student-learning outcome of academic advising should be “students will use complex information from various sources to set goals, reach decisions, and achieve those goals” (National Academic Advising Association, 2006, p. 2). Therefore, advising should teach students the skills necessary for processing information, making personally beneficial decisions, and achieving personal goals. Essentially, advisers teach and students learn how to become effective decision makers. One likely outcome of teaching and learning effective decision making might be an increase in students selecting appropriate majors with a subsequent decrease in students switching majors.

An effective decision maker makes rational decisions that offer the greatest benefit or utility to the decision maker (Brown, 2005, p.54). Decision makers can maximize benefits when they consider: (1) their own assets, including abilities and attitudes; (2) the consequences of their decisions; and (3) the value or satisfaction associated with each consequence.

Problems Encountered with Qualitative Decision Making

Decision-making models are qualitative, quantitative, or an integration of both. Research has indicated that quantitative models are less prone to error than qualitative methods in determining the decision that offers the highest utility to the decision maker (Hastie & Dawes, 2001). Additionally, the degree of effectiveness of quantitative methods is directly proportional to the complexity of the decision (Shafir & LeBoeuf, 2004).

It is theorized that the errors associated with qualitative models are a function of decision makers' overconfidence in their abilities to predict outcomes. Three reasons are given for overconfidence. First is an “illusion of control” or people's belief that they have the ability to control future events. Second is “risk perception” or people's ability to accurately predict and consider future risks. Third is “distortion of information” or people's tendency to use heuristics (Schoemaker, 2004).

Heurisitcs

When making decisions without working through options quantitatively, people use what researchers call heuristics. Heuristics are considered to be automatic processes that selectively review decision options (Payne & Bettman, 2004). Because of the automatic nature of cognitive processing associated with heuristics, information is not thoroughly and completely analyzed, thus making the process inherently error prone (Payne & Bettman, 2004). The errors are a result of three biases that are related to one's reliance upon reasoning, which is bounded by the decision maker's working memories (Hastie & Dawes, 2001). These biases are representation, availability, and anchoring and adjustment (Keren & Teigen, 2004).

Representation Bias

Representation bias occurs through memory recall. When a decision maker has to analyze the options of a decision, she tends to recall a memory or experience that is similar to the present decision-making situation. For example, an incoming first-year student having difficulty choosing a major may recall certain secondary education experiences. Suppose the student chose to take honors biology in high school and did very well. Thus, the student may decide to major in biology in college because of the success experienced in high school. However, there exists a high probability that the high school representation is inaccurate. In high school, the student could have had a very motivating biology teacher, the course may have been a year long, and, most importantly, it probably did not have chemistry as a prerequisite. In college, the first biology course may be taught by a teacher who is not very engaging; the course meets only for a semester and covers more breadth and depth of subject matter; and, probably most significantly, it has chemistry as a prerequisite. These factors can significantly affect the student's success but may not have had a major influence on the student's choice. Therefore, examining all of these factors through a quantitative analysis would assist in making them more apparent to the student (Keren & Teigen, 2004).

Availability Bias

Availability bias is related to vivid memories of the recent past, or the primacy effect. For example, if a person is deciding between seeing two movies and has just watched an intense preview for one, she is more likely to choose the previewed one. Availability bias also occurs when the decision maker predicts “easily conceivable outcomes” (Keren & Teigen, 2004). That is, if we believe one outcome to be more likely, we predict that outcome to occur without seriously considering other possible outcomes. For example, a student may have always imagined medical school in her future and recently very much enjoyed shadowing a surgeon. For this student, it may be nearly impossible to predict a postgraduate future other than being accepted to medical school.

Anchor and Adjustment Bias

The third heuristic bias is anchor and adjustment. The anchor represents an idea that is familiar and conceivable to the decision maker. In a given situation, students often start with the anchor but fail to realize how much adjustment may be necessary for reaching a goal. For example, a first-year college student who achieved a 3.9 grade-point average (GPA) in high school finds a 3.9 GPA in college just as conceivable. The high school GPA represents the anchor. However, the student does not take into account the many differences that exist between high school and college and thus does not adequately adjust the predicted college GPA. Without thoroughly processing all the factors that are related to earning a high GPA in college, the student is poorly prepared to adequately predict or impact her college GPA.

Quantitative Models

Research has demonstrated that the more complex the decision, the more error prone qualitative models become (Shafir & LeBoeuf, 2004). By using quantitative models, we can help reduce the errors associated with qualitative decision-making models.

Goals, Options, and Outcomes Model

This very simple model, although qualitative in nature, can be used to begin to develop quantitative approaches. According to this model, the decision maker first answers three qualitative questions: (1) What do I want? (goals), (2) What can I do? (options), and (3) What might happen? (outcomes) (Brown, 2005). This exercise makes the decision-making process more explicit and available to analysis and scrutiny. Goals can be listed as criteria for the decision, and choices are the options available; however, it is often very difficult to predict outcomes.

The Personalist Approach

As previously noted, research has converged to conclude that quantitative methods are less prone to errors for predicting outcomes. What is needed is a way to integrate quantification with qualitative data in order to better examine outcomes (Hastie & Dawes, 2001). The most straightforward approach to quantitative predictive methods is the Personalist Approach, which is an enhancement of the Goals, Options, and Outcomes model. In addition to listing the outcomes of a decision, the Personalist model quantifies the outcomes (Brown, 2005). To use this approach, the decision maker lists the outcomes vertically next to the options, as illustrated in Figure 1. The options are compared using pluses and minuses for each outcome. The pluses and minuses are assigned from an arbitrary scale determined by the decision maker, for example +5 to -5. The algebraic totals are then used to find the better option (Brown, 2005). Quantitatively examining options in this manner can greatly aid a student in making academic decisions, such as choosing a major. Figure 1 illustrates the analysis of two options with three possible outcomes.

Figure 1 - Pluses and minuses assigned to an outcome using an arbitrary scale (in this case, -5 to +5) chosen by the decision maker

Lens Model

Algebraic probability models have evolved from the Personalist approach (Hastie & Dawes, 2001). The Lens Model creates an algebraic model of probability that measures and scales the importance of each piece of information available to the decision maker. Theoretically, the Lens Model engages the decision maker in trying to see a “distal” true state of outcomes through a “proximal lens” of cues. These cues represent information or characteristics that the decision maker uses to choose a final decision. In the Lens Model, the relationship between the cues and the importance of the cues is used in helping one make a decision.

In academic advising, conversing with a student may help to identify what variables seem to be important in predicting success in a major(s), but an empirically derived statistical model may improve the accuracy of the prediction. Levin and Wyckoff (1988) developed a Lens-type model that used both academic and on-academic student variables to predict success in baccalaureate engineering majors.

The purpose of the Levin and Wyckoff study was to “develop models that predict freshman persistence and success in baccalaureate engineering programs by analyzing five intellective and nine non-intellective variables in relation to these criteria” (Levin & Wyckoff, 1988, p. 178). Using logistic regression analyses, they first determined the weight or importance of each variable in predicting persistence and success in engineering. Next, the predictor variables were used to calculate an estimate of the probability of an individual student's persistence and success in an engineering program. The overall predicted success was stated in terms of percentage chance of graduating in an engineering major.

Figure 2 shows a Lens-type model that indicates the significant predictor variables for persistence and success in bachelor of science (B.S.) degrees in engineering. By entering individual student data, the solution of the equation is the natural log odds of persisting and succeeding in undergraduate engineering. This model can be used in academic advising to identify the significant student predictor variables. By discussing the variables the student has control over, advisers can help the student to improve the probability of successful persistence or in some cases begin exploring more appropriate majors.

Figure 2a Lens-type model that identifies significant predictor variables and calculates an estimated probability of persistence and success in B.S. engineering

Natural log odds (ln) (persisting & succeeding/not persisting & succeeding) = -3.8 - .02(interests in non-science majors) - .01(SAT-V) + .69(high school GPA) + .83(algebra entrance test score) + .69(chemistry entrance test score) + .38(intrinsic motivation) - .38(extrinsic motivation) + .20(male) - .20(female)

Figure 2b Substituting sample student data into the equation

ln (persisting & succeeding/not persisting & succeeding) = - 3.8 - .02(10) - .01(520) + .69(3.00) + .83(25) + .69(12) + .38(1) - .38(0) + .20(1) - .20(0) = .236

odds ln (persisting & succeeding/not persisting & succeeding) = e^.236 = 2.72^.236 = 1.27/1

Therefore, the probability of this sample student persisting successfully in engineering = 1.27/2.27 = 56%.

From the Lens Model came several experimental studies testing the accuracy of a prediction of a quantitative statistical model compared to the qualitative prediction of a human expert. It was concluded that statistical models are always equal to or more accurate than purely qualitative psychological judgments (Hastie & Dawes, 2001). Furthermore, Hastie and Dawes concluded that “experts correctly select variables that are important in making predictions, but that a linear model combines these variables in a way that is superior to the global judgments of these very same experts” (Hastie & Dawes, 2001, p. 58).

Simple Utility Equation Model

Another statistical linear model is the “simple utility equation,” which measures the utility of a decision. A decision tree is used in which each option represents a major branch, and from each branch stems the possible outcomes. In Figure 3, there are two options, engineering and physics. For each option there are two grade-point outcomes: < 3.0 and > 3.0. For each of these outcomes a specific quantitative probability is assigned so that the sum of the outcomes stemming from each option adds up to 1, or 100 percent.

The probability for each outcome is multiplied by an assigned number that represents how the decision maker would feel about that outcome. For example, if option A yields possible outcomes 1 and 2 and the decision maker would be most satisfied with outcome 1 and least satisfied with outcome 2, we could label outcome 1 with a value of 100 and outcome 2 with a value of -100, assuming an arbitrary scale of -100 to 100. Each outcome value is then multiplied by the probability of that outcome occurring. The values of each option's outcomes are then summed, as in Figure 3, to give the utility value {Σ (probability_i x value_i)}, of that option (Hastie & Dawes, 2001).

Figure 3 The two options are engineering and physics. In this case, the utility of option ENGR has a much higher utility to the decision maker than does PHYS. Utility = Σ (probability_outcome x value_outcome)

Additive Linear Multi Attribute Utility Theory (MAUT)

Building on the utility equation, Hastie and Dawes (2001) give excellent descriptions of other models, including the Additive Linear Multi Attribute Utility Theory (MAUT). MAUT is the method that all reviewed research considered to be the most successful predictor of the highest utility option.

When using MAUT, the interactions of options and attributes are analyzed. Each option is assigned a scaled value that represents the importance of that option to the decision maker. Each attribute is assigned a statistically derived weight that indicates its relationship to the option. Summing the products of the option-scaled values and the weighted attributes results in a utility value for each option. Although MAUT is very effective in predicting the optimal choice in decision making, Hastie and Dawes (2001) consider it to demand the most effort in analyzing a decision.

Table 1 demonstrates a complex MAUT-like model taken from in-progress research (Levin & Cross, unpublished). Using stepwise regression modeling, Levin determined the weight of each attribute for predicting persistence and success in the two considered options, the life science majors and the physical science majors in Penn State's Eberly College of Science. The attributes were both quantitative (such as SAT math score) and non-quantitative (such as certainty about majors, certainty). The non-quantified attributes (remedial math, certainty, chem, math, and biol attitude) were converted to quantified variables by using dummy variables. A negative sign indicates an attribute that has a significant (p < .10) negative impact on students' likelihood of success, while a positive sign indicates a significant positive impact. Empty cells indicate the predictor variable was not significant at the .10 level.

Next, each attribute is measured for every Penn State first-year student using the University's mathematics and chemistry placement exams (Division of Undergraduate Studies, 2008a) and the Educational Planning Survey (Division of Undergraduate Studies, 2008b). Table 1 lists the attribute measures for an individual student.

For advising purposes, the individual student's attribute measures are multiplied by the weights for each attribute. These products are summed for each option, resulting in a utility value for each option as indicated in Table 1. The utility value for the option of life sciences is 2.60 and for the physical sciences 5.95. Although obvious to any academic adviser, such an analysis should not be the sole determinant for a student to decide upon a major in the physical sciences. Discussions with an academic adviser are critical and should include, among other things, the relationship of mathematics to the physical sciences, the student's willingness to study pre-calculus before calculus, how non-science interests can be accommodated in the physical sciences, and the need for more study hours than the student anticipates.

Table 1 - Predictors of success for life science and physical science majors and sample student data

Summary

The National Academic Advising Association states that one student-learning outcome of academic advising should be “students will use complex information from various sources to set goals, reach decisions, and achieve those goals” (National Academic Advising Association, 2006, p. 2). Essentially, advisers teach and students learn how to become effective decision makers.

A decision maker who is effective makes rational decisions that offer the greatest benefit or utility to the decision maker (Brown, 2005). Decision-making models are either qualitative, quantitative, or an integration of both. When making decisions qualitatively, or without working through options quantitatively, people use what researchers call heuristics. Heuristics are considered to be automatic processes that selectively review decision options (Payne & Bettman, 2004). However, heuristics are error prone and plagued by the biases of representation, availability, and anchor and adjustment.

Research demonstrates that the more complex the decision, the more error prone qualitative models become (Shafir & LeBoeuf, 2004). By using quantitative models, the errors associated with qualitative decision-making models can be reduced.

The quantitative models reviewed were Goals, Options, and Outcomes, which is mostly qualitative but serves as the foundation for future quantitative models; the Personalist model, which is the simplest quantitative approach; the Lens Model, which enables the decision maker to view the importance of each attribute's impact on multiple options; the Simple Utility Equation Model that introduces the idea of the utility or benefit of each option being considered; and the Additive Linear Multi Attribute Utility Theory (MAUT), which enables the decision maker to analyze statistically derived weighted attributes. The MAUT model is considered to be the most effective but demands the most effort in analyzing a decision.

Quantitative models make explicit the relevant weighted variables in academic decisions. Quantitative models engage students in reflection and analysis, which improve academic advising and students' decision-making skills. One goal of academic advising is to assist students in making informed educational decisions. Decision-making models that identify variables that predict educational outcomes reduce the likelihood of students making risky decisions. It is widely accepted that students are more likely to function well academically and make rational, informed educational decisions when they understand how their interests and abilities relate to the likelihood of success in their chosen fields of study (Levin & Hussey, 1993). Without quantitative models, students and advisers operate at an intuitive level, which increases the likelihood of risky decisions (Levin & Wyckoff, 1995).