Naturalistic Decision Makingand Selecting a Spouse

Some of my systems engineering graduate courses have been on decision analysis/expert systems. One course dealt with naturalistic decision making (how humans really make decisions) as opposed to classical theory (decision trees, probabilities, would you prefer a guarantee of \$3000 or an 80% chance at \$4000 and 20% chance of nothing). Without getting too bogged down in technical detail about all the proposed naturalistic theories, I've found lots of good parallels with how (at least in my opinion) people try and choose mates. With the divorce rate at 50% it appears that people are doing a lousy job at making this decision (yeah, I know there's a lot of assumptions and caveats but bear with me).

One primary difference between naturalistic and classical is that people tend to find a solution that's "good enough" (this is called satisficing) rather than optimal solution (in terms of expected value). This is typical, I think, in that people usually serial date/have relationships until they find a person who meets all of their qualifications. The prospective partner won't be guaranteed to be the "best possible fit" out of all the options (all eligible partners in the country, for instance), but will be good enough as a solution.

Here's another example. There's a Dominance Theory, which presupposes you can list a bunch of characteristics that you are looking for. First, you screen out any choices that don't meet the minimum criteria for any of the characteristics. After that, you look for one alternative that is dominant. That is, it's at least as good as any of the others for each characteristic, and better for at least one characteristic. This likely "solution" is often determined early on, even if all the data isn't in yet.

Then there's a key phase called Dominance Restructuring, part of which is that people tend to rationalize a prospective solution by bolstering the positive characteristics and possibly restructuring the problem so as to minimize or eliminate a negative characteristic. I think this happens a lot during the mate selection process, and it may not always be a good idea. "Oh, he's such a hunk, treats me well, is reliable, honest, etc. So what if he drinks a bit too much." Also, they might continue on with the relationship and even get married, hoping they can change the person.

Anyway, there's a whole lot more involved, but I think a lot of it could be used to improve the decision process. A lot of the self-help books talk about this more in "rule-based" terms and I think it would also help if people understood some of the rationale behind how people seek and interpret information, and actually make decisions.

Here's the slides from my class project, but first, a short description of each of the five theories I discuss:

Image Theory - This theory has three parts (images). The value image consists of the decision maker's principles; what's right or wrong, any organizational rules or principles that must be followed, etc. The second image is the trajectory image, the goals that the decision maker wants to achieve. The third image is the strategic image, which are the plans adopted to achieve the goals, including making decisions, evaluating, and modifying approaches based on results. Decisions can be made by screening out candidates because they don't pass a minimum level, or by doing some sort of combined comparison to rank the candidates in preference order.

Recognition Primed Decision Making (RPD) - This model describes how experts make decisions under stressful situations, perhaps due to time pressure or rapidly changing environments. The decision maker uses their expertise and experience to quickly asses the situation and to come up with an acceptable course of action. They then "play out" the course of action to see whether it is feasible or requires modification. If the first choice doesn't work, they will go back, select another option, and do the evaluation again. A good example is a firefighting captain who arrives on the scene of a burning building. He will quickly recognize what to do and act accordingly, but the situation may change rapidly and he will have to stay on top of the situation, perhaps changing priorities on the fly. One aspect of RPD is that the expert can quickly rule out unimportant information or unusable solutions, almost on a subconscious level, whereas a novice would need much more time to explicitly think through all possibilities.

Explanation Based Model - There are two parts to this model: The coherent story and the choices. The theory says that the decision maker will attempt to create a full story from some incomplete raw facts and then match this story against possible choice options to come up with a solution. For example, a jury will try to formulate a full explanation of a defendant's behavior from the evidence, general knowledge about similar events, and knowledge about story structures in general. With their completed story, they will then try to match that with the choices (verdict categories). If a match is found, they can make a decision, otherwise the process would have to be repeated with additional inputs.

Lens Model - The lens model is a part of Social Judgment Theory. It tries to analytically build a model of how well a person's judgments match up with the environment they are trying to predict. The interface between the two are the cues that represent the environment. An example is a trader trying to predict what the market will do so that they can pick good stocks. Some of the cues might be unemployment rate, price/earnings ratio of the stock, inflation rate, etc. The trader observes the cues and makes a judgment on how to interpret them, then selects stocks. The lens model takes a large number of these trial cases and comes up with equations for how well the trader does, plus other models for how well the cues are judged or how well they represent the environment. Even with perfect information, most task success rates are nowhere near 100%. This is due to many factors, including errors in judgment, insufficient or unrecognized cues, or important cue patterns that are hard to determine.

Dominance Testing - There are four major steps to making a decision. First, the decision maker simply screens out alternatives that do not meet minimum standards. After that, if there is more than one choice left, the second step is to select a promising alternative. This can be a fairly subjective choice based on preferences or initial reaction. The third step is to test for dominance. An alternative is dominant if, for all the selection criteria, the alternative has no disadvantages and at least one advantage, it is selected. Often, this is not the case, and the fourth step is entered. This is where the decision maker tries to restructure or reinterpret the information in order to make the promising alternative dominant so it can be selected. This can be good or bad, since if overdone it can mean talking yourself into making a bad decision.

Naturalistic Decision Making (NDM) and Selecting a Spouse

• Introduction
• Background and assumptions
• Hypothesis
• General NDM relevance
• Specific decision theories: Image, RPD, Explanation Based, Lens Model, Dominance Search
• Applicability to the given problem
• Possible sources of error, strengths/weaknesses
• Decision support system (DSS) and testing the hypothesis

Notes - Other NDM theories could certainly apply but I had to limit the size of the presentation. Connolly's Decision Cycles could be used to represent the problem. There are the three domains; actual world (who the candidate really is), cognitive map (how the searcher perceives the candidate), and the values/purposes/goals (what the searcher wants and how they make the decision). The cycles represent feedback as the dating process becomes more serious, and also any modifications the searcher might make in terms of eventual goals.

There is probably some element of Noble's Situation Assessment relating to the reference problem concept. The more dating/relationship experience (reference problems) you have, the better you might be able to make a good decision. Hammond's cognitive continuum could also apply. Issues such as chemistry or emotional attachments would fall more towards the intuitive end of the spectrum, while other aspects of the problem (specific relationship goals) might be more suitable to an analytic, objective analysis.

Background and Assumptions

• Present task success rate is approximately 50% (divorce statistics, other measures could be used)
• Hypothesis - With proper partner selection, success rate ought to be in the 80-90% range (age/experience dependent)
• Typical steps in the process (not all may be used, iteration likely)
• Make decision, or at least be aware, that you want to be married
• Meet potential candidate(s)
• Casually date several candidates concurrently
• Exclusive relationship with prime candidate
• Live together
• Decide to get married, or else break up

Notes - Defining success is difficult. For the purposes of this project, I assumed it was loosely defined as a happy marriage for life. Also, this is a perfect example of the type of problem where you can make what you consider a very good decision and through totally unforeseen or impossible to predict circumstances the eventual outcome may be a failure. That's why the achievable success rate could never be close to 100% I believe the inherent achievable success rate to be dependent on age/experience. I would expect the inherent rate to be lower for a pair of 20 year olds than 30 year olds, for example. Not sure of the shape of the curve as age keeps increasing; it might level off or even decline. Most of this presentation deals with the process up to the point at which you make the decision on a final partner. There are many things that can happen after that point that could affect the final outcome, but for the most part these would be outside the scope of the original decision process (other than to consider their likelihood during forecasting or extrapolation).

Naturalistic Problem Characteristics That Make the Decision Difficult

• Searcher will often have to base decision on a rather incomplete set of information about the candidate
• High stakes - decision drastically affects whole life
• Multiple players - searcher is also playing candidate role at the same time; interpersonal dynamics can be complicated
• Uncertain, dynamic environment - the relationship is constantly evolving as the situation is being assessed. Also, the searcher and/or candidate may change over time
• Ill-defined goals - searcher may not have a clear picture of what they want
• Possible time stress (ticking biological clock, ultimatum given by partner, self-defined schedule)

General NDM Principles

• Satisficing - The searcher usually selects the first candidate that satisfies the nebulous "requirements" necessary rather than waiting for the "perfect person"
• Searcher uses various methods to quickly rule out non-solutions (compatibility testing)
• Biases, personal histories have huge impact
• No "clear-cut" algorithmic, classical approach. Problem contains an enormous number of interrelated variables and situations that the searcher must decode and interpret.
• Extensive use of mental imagery

Image Theory

• Values (Principles)
• Many of these will come from family upbringing, societal norms, and simply the experience of growing up. Examples:
• Being married is inherently better than being single
• Having children outside of marriage is wrong
• A woman is "incomplete" without a man
• Trajectory Image (Goals)
• Top level goal is getting married
• Underlying goals could include:
• Having children
• Retiring by age 55
• Relationship where most free time is spent together

Notes - The values shown are not implicitly right or wrong but are only given for example purposes. The trajectory image (particularly the underlying goal list) is similar to the relationship choice story part of the Explanation-Based model . This image somewhat aligns with the day-to-day behaviors of how the relationship operates.

Image Theory (continued)

• Strategic Image (Plans)
• Meeting candidates
• Candidates must pass compatibility test to be dateable (initial screening out of obvious losers)
• Dating several candidates
• Use profitability testing to decide among best alternative
• Serious relationship with single candidate
• Further round of screening (with higher minimum standards). Also, forecasting to determine how well the goals will be met if this candidate is selected.
• Make final decision, yes or no

Image Theory - Possible Sources of Error

• Unclear underlying goals
• Could lead to screening errors (ruling out good candidates or including poor choices)
• Outcome could be choosing a candidate with the wrong characteristics/criteria
• Poor planning or implementation of strategy (example - not putting yourself in a position to meet the proper "pool" of candidates)
• Values not sufficiently or wrongly defined

Notes - People's values are sometimes unduly influenced by external sources, such as family or society. To make the best decision, the searcher should have a stable set of internally defined values that are not likely to change much, if at all, during the process or after they get married. A good example of this is a woman who is conditioned to believe that having children is a positive value, and lets this external bias override her own desire to be childfree, skewing one of her subgoal requirements. This can result in future disaster if she marries a partner who wants children, and she later realizes she doesn't want that after all. I've also talked with a number of women who got married when they were young because the idea of being married was so ingrained in them that it overrode their knowledge of potential mismatches. Many of these people said that if they had it to do over again they wouldn't have married that person. Another example of how values can strongly influence the decision process.

Recognition Primed Decision Model

• Similar to the General Hod scenario - most of the time is spent in situation assessment until candidate type is known (either a yes or no for a possible match)
• Very expertise based, since it is a personal values decision and each person applies their unique expertise
• Use knowledge from previous relationships or interactions with people to infer behaviors or attributes from the observed cues
• Take actions to test hypotheses
• schedule a vacation together
• meet each other's families (how candidate behaves under stress)
• Possible time pressures (ticking biological clock, ultimatum given by partner, self-defined schedule)
• Probably the best model for evaluating the intangibles such as love, attraction, and chemistry

Notes - RPD mostly applied for the situation assessment part of the process, not so much if the searcher had to decide between two or more promising early candidates. One of the main differences of RPD is that most of the time there will not be undue pressure to come to a decision, and that the searcher can spend a lot of time thinking about the problem or gathering information for situation assessment.

Notes - The type of relationship is not really a choice among alternatives, but rather a single, preferred goal. This is different than the normal model, which presents various choice options for this box. However, there is a definite matching process, where the searcher tries to evaluate how well the candidate will fit in to the desired relationship story. If the match is perceived as good enough, the candidate will be selected, otherwise discarded. There may be some feedback if the searcher decides to modify their relationship details in order to successfully match up with a desirable candidate. This could produce a good or bad end result, depending on many factors.

Explanation-Based Model - Possible Sources of Error

• Searcher's constructed story of candidate does not match reality
• Previous biases or experiences produce improper judgments
• Searcher does not have good general knowledge of human behavior
• Unclear or underdeveloped relationship choice story
• Searcher simply does not spend enough time thinking about the problem
• Insufficient relationship experience to create a full set of criteria
• Searcher decides to modify relationship details in order to produce a match for a perceived desirable candidate (could be good or bad, depending on many factors)

Notes - The cue examples given here are of the measurable, objective type. I think the lens model is more suited to these types of cues, rather than trying to use cues that already contain a measure of the searcher's judgment (such as sense of humor or how attractive). This is difficult to avoid; even in the example given here, constantly late, late for one person might not be for another.

The number of cues to characterize a whole person would be enormous. One of the textbooks seemed to imply that something on the order of 50 cues is already a large number. To use the lens model you might have to limit the candidate's criteria to a very specific subset of behaviors.

Lens Model - Possible Sources of Error and Strengths/Weaknesses

• Missing cues
• Searcher didn't know candidate long enough
• Searcher didn't observe candidate in situations that might be common after marriage
• Judgment errors - searcher misinterprets cues
• Selective perception bias: judgments affected by searchers previous experiences and life history
• Confirmation bias: searcher sees what they want to see and may ignore or discount undesirable information
• Searcher too life inexperienced (young?) to accurately judge
• Extrapolation errors
• Searcher assumes judgments that hold for present scenarios will be the same for future (different but similar) scenarios

Lens Model - Sources of Error, Strengths/Weaknesses (continued)

• Cue validity
• Candidate may deliberately provide misleading cues
• Candidate may change later on (original cues no longer valid)
• Incorrect or insufficient pattern recognition
• Searcher makes a judgment about candidate's behavior that might be different if they knew underlying causes or explanations (that could be inferred from additional related cues)
• Good model for interpreting measurable cues that correspond to objective criteria such as reliability, honesty, introvert/extrovert, etc.
• Not all cues are purely objective; many will already be tainted by the searcher's own biases.
• Problem as a whole may not be solvable - too many variables. Might be possible to model a specific criteria subset.

Dominance Search Model

• Pre-editing used to screen out obvious mismatches
• Promising alternative/Dominance testing
• Used early in the process if searcher needs to choose from a number of possible candidates
• Dominance Structuring
• Could be used to select from a number of casual candidates
• After single candidate is selected, is used against the searcher's absolute standards
• Deemphasizing: She isn't as attractive as I'd hoped for but at least guys won't be hitting on her all the time (cancellation)
• Bolstering: The chemistry between us is unbelievable. Even though some other things aren't ideal, I can live with that
• Collapsing: Emotions such as love or chemistry could be considered combinations of lower level attributes

Notes - While there can be dominance testing in the early stages of partner selection (where you are deciding among several candidates), I felt that most of the relevance was during the �serious� evaluation phase when you are considering a single candidate. Here, the processes such as bolstering and de-emphasizing would come into play when the searcher was trying to justify their promising alternative against ideal standards. The candidate is never going to fully satisfy ALL of the searcher's criteria, and some compromises may have to be made. Some of the concepts of the dominance structuring phase appear to match up with how people typically seem to resolve these issues.

Dominance Search Model - Possible Sources of Error

• Pre-editing screen too rigid
• Could be eliminating a promising alternative
• Dominance Structuring errors
• Deemphasizing and bolstering
• Latching on to a promising alternative and trying to force fit into a solution
• Modification of original requirements in order to accept a promising alternative
• Assuming candidate will change in the future

Conceptual Functionality of a Spouse Selection Decision Support System

• Part of the training would be an overview of NDM theory
• Assist in determining relationship goals/requirements to flesh out explanation based relationship choice description
• Provide lens model analysis of candidate's cues to assist in situation assessment
• Reveal cue patterns or relationships between cues that might be overlooked by searcher
• Give an objective, unbiased assessment that searcher can compare against own judgment
• Match candidate's characteristics against searcher's goals or requirements (forecasting or predicting success probability) for specific criteria as well as overall assessment
• Results presented in cognitive-friendly manner (graphs vs. text, similar concepts grouped together, etc.)

Notes - Concentrate the aid where it will do the most good - what the searcher has the most control over (internal vs external items) One way this could be done is in helping to structure an ill-defined problem by helping the searcher clarify their requirements or fill in details about the type of relationship they want (the relationship choice in the explanation model). This might be done via some sort of questionnaire or text based elicitation, then the DSS provides options about choices that the user could refine. The DSS should be adaptable to the user's methodology - some people might be top-down processors, starting with a high level goal and working downward, while others might prefer a bottom-up approach by providing very detailed behaviors and then working to combine those into more easily understood (for the purposes of decision making) goals.

What I had in mind for the lens model was something similar to police profiles. By having a very large database, you might be able to somewhat accurately predict the candidates characteristics based on the cues supplied by the searcher. In theory this ought to be possible (although the inherent success rate would still be nowhere near 100%) but in practice you would have to limit the number of cues to something manageable. And even if you had a proven model, the searcher might still refuse to use the results if it presented disconfirming information, since the choice is such a personal, emotional one.

Testing the Hypothesis

• Design and build the DSS
• Large amount of data needed to formulate environmental model of human behavior and characteristics
• Perform experiment with one control group and one group using the DSS
• Need wide variety of age groups (and other differentiators)
• Experiment would take years to complete
• Much of the success/failure data would be subjective and likely questionnaire based
• Analyze results (compare success rates)
• How to define success
• Might be difficult to isolate DSS influence from other variables

Notes - A successful choice is a hard thing to define. You might have to limit it for the purposes of a study to something like 5 or 10 years of a stable, happy marriage (and you would have to rely on the participants for feedback, which would likely be biased anyway) instead of a whole lifetime. Maybe lifetime is too rigid a definition in any event. These days, perhaps a 10 year marriage that ends in a friendly manner might be classified as a success.

 Send email (larryk@alum.mit.edu) Back to my home page Sign Guestbook View Guestbook

people are either still trying to decide or else regret their choice.