Brian Donohue and J. Neil Otte Fall 2013 Ontological Engineering Prof. Barry Smith THE FIVE-FACTOR MODEL AND PERSONALITY ASSESSMENT ONTOLOGY 1. Introduction What factors lead to someone having a particular personality? If we know someone’s personality, what can we predict about their lifetime income, social success, and longevity? Is our personality fixed, or can we change it? The history of the study of personality in the twentieth century has been one of promise, disappointment, and rebirth. Initially, empirical psychologists proposed a number of testable components of personality. One psychologist might propose scores for thinking, feeling, sensing and intuiting; another might propose scores for extroversion and harm dependence. This led to a variety of competing models with no clear winner, and the impression that the study of personality was without a solid subject matter. As far back as 1958, Gordon Allport complained, “each assessor has his own pet units and uses a pet battery of diagnostic devices” (Allport 1958, p. 285).1 In the 1970s and 1980s, the study of personality declined due to a lack of consensus, and new studies emphasizing the role of circumstance and situation on individual action came to prominence. But now researchers again have grounds for optimism. The Five-Factor Model, also known as the Big Five Model of Personality, has brought new life into the study of personality. 1 This quote is reported by Nettle (2007). 2 1.1. What Is the Five-Factor Model (FFM)? The Five-Factor Model proposes that there is an exhaustive set of five personality dimensions that sufficiently capture a person’s unique personality, which we understand here to be a disposition of individual persons to respond similarly across a variety of situations. The dimensions of FFM include: Openness to Experience, Conscientiousness, Extroversion, Agreeableness, and Neuroticism (OCEAN, for short).2 Each of these dimensions represents a range on which individual subjects may be tested: an Extroversion score, for instance, will report an individual as falling somewhere between extreme extroversion and extreme introversion. Each dimension may also be thought of as tracking a unique set of traits and personality features that appear in descriptions of individuals strongly under each dimension. For example, Openness includes features like imagination and insight, and signifies a broad range interests. Conscientiousness commonly includes high levels of thoughtfulness, good organization skills, and excellent impulse control. Extraversion includes traits like edginess, amiability, confidence, and emotional expression. Agreeableness includes benevolence toward others and other prosocial behaviors. Neuroticism includes emotional distress, angst, irritability, and a proclivity to depression or sadness. Previous studies produced many testable characteristics, but lacked this core system of traits. To counter this, psychologists began running factorial analyses on the available data to see whether or not there were more primary characteristics underlying those that previous researchers had proposed. This allowed them to uncover a hierarchy 2 I will begin a convention here of capitalizing dimensions to more clearly distinguish them from other scores discussed. 3 of variables, of which many previous characteristics were merely synonymous or multicategorical measurements. What they found surprised them. Tupes and Christal (1961), who found five recurrent factors in analyses of personality ratings in eight different samples, wrote: “In many ways it seems remarkable that such stability should be found in an area which to date has granted anything but consistent results. Undoubtedly the consistency has always been there, but it has been hidden by inconsistency of factorial techniques and philosophies, the lack of replication using identical variables, and disagreement among analysts as to factor titles” (p. 12). As a result of these findings, FFM has become the dominant model in social psychology, and its broad ability to correlate with various features of our lives—e.g. marital success and longevity—make it a worthwhile model to pursue further research. In addition, contemporary personality tests based on the Five-Factor Model are much more reproducible, have greater universality, and offer a holistic picture of personality than previous tests. A good example of this is the Newcastle Personality Assessor (NPA), which asks twelve questions—a reduction of the longer and more accurate IPIP questionnaire.3 Because the FFM bears a complex mathematical relationship to other measurements of personality, its dominance also poses a particular project for linked data—and this is our subject. 2. Five-Factor Model and Our Project The discovery of these new surveys of personality leaves open questions about the original survey data using other instruments and its relationship to the new model. For 3 See http://ipip.ori.org 4 example, when a researcher uses the Adjective Check List, they may test a subject’s degree of self-ascriptions of terms such as anxious, self-pitying, tense, touchy, and unstable, according to a Likert scale. Factorial analyses suggest that such terms are tracking a real dimension of personality according to the Five-Factor Model: Neuroticism. This discovery suggests that a score for anxiety in the Adjective Check List can be counted as a score for Neuroticism according to the Five-Factor Model. McCrae and Oliver P. John (1992) offer the following chart, which displays a sample of the kinds of relationships we wish to capture. These include positive and negative correlations that hold between the scores of various instruments and the factors of the Five-Factor Model.4 McCrae, Robert R. and John P. Oliver. “An Introduction to the Five-Factor Model and Its Applications.” Journal of Personality. Vol. 60, Issue 2, pp. 175-215, June 1992 4 5 Figure 1: McCrae and John's depiction of personality instruments (inventories) and FFM These relationships have been well known to personality researchers for decades, but they are hidden in silos of data that exist in various forms: online databases, journal archives, and individually published papers. Such large caches of personality studies using various methods could be combined under the interpretation of the Five-Factor Model to create a massively universal scoring system for all personality tests. This would allow for inferences to be drawn across all studies under the ontology, alongside their individual correlative variables, negating much of the need to reproduce previous work under the same FFM instrument. It is to this end that our project proposes an ontological model for rendering a single set of results, defined by the Five-Factor Model, for all instruments of personality, 6 which in our ontology we refer to as “inventories.” Our model uses the familiar, widely used Basic Formal Ontology (BFO) 1.0 as an upper-level ontology, and connects instruments and the scores they generate to particular factors in FFM of particular persons. The resulting ontology is missing necessary components to render reliable determinations of evidence between other instruments and the FFM. But it is our hope that this ontology is a proof of principle, which with further development could be used to organize multiple databases into a single, searchable resource using an RDF data model and the SPARQL query language. 3. Utilized and Related Ontologies Our project, Personality Assessment Ontology (PAO), either utilizes or relates in potentially significant ways to a few other ontologies. Below, we explicate how PAO utilizes the content and structure of Basic Formal Ontology and Neuropsychological Testing Ontology (NPT) in order to represent the factorial analyses of personality dimensions. Also, we discuss the ways in which PAO relates to Mental Functioning Ontology (MF) and Emotion Ontology (EMO). Thus, with respect to PAO, BFO and NPT count as utilized ontologies, whereas MF and EMO count only as related ontologies. These latter ontologies show similar concern for representing the information of interest to the psychological sciences, but their entities, classes, etc., are not directly incorporated into the architecture of PAO. 3.1 Basic Formal Ontology 7 PAO was built from the top down, as it were, according to the framework set up by the Basic Formal Ontology. Of particular importance were the basic classes of Process, Person, Role, Disposition, and Generically Dependent Continuant. Since PAO aims at capturing and consolidating the process of personality assessment, it utilizes BFO’s distinction between continuant and occurrent entities. Furthermore, two of BFO’s realizable entities, Role and Disposition, provided a few crucial distinctions for some of the foundational entities within PAO: subjects of assessment are persons in the role of evaluant, and personalities are dispositions belonging to and realized by persons (or persons’ behavior). Finally, under the BFO class of Generically Dependent Continuant, PAO is capable of representing the data produced by personality assays as subclasses of Information Content Entity. As we describe below, each assay of personality represents a distinct data item with a definite numeric value as determined by the directives given to it by the assay’s developers. 3.2 Mental Functioning Ontology and Emotion Ontology In part, PAO was inspired by the work accomplished already in the Mental Functioning Ontology (MF) and the closely related Emotion Ontology (EMO). These ontologies seek to encapsulate the functional, characteristic, and responsive aspects of human psychology. This links the broader projects of MF and EMO with the narrower one of PAO, since much of the assessment portion of PAO will constitute, in effect, an assessment of the aspects of psychologies depicted within MF and EMO. Nonetheless, many of the other facets of MF and EMO are largely irrelevant to the basic structure of PAO. There is no need, at this stage in PAO’s development, to represent discrete classes 8 of psychological acts performed by a subject (e.g. cognitive appraisal or affective representation), even if the elements of these classes ultimately form the psychological mosaic that constitutes a personality type. It is by no means inconceivable that further ontological work could forge a firmer link between MF and EMO on the one hand and PAO on the other. That being said, far more extensive research would be required to realize this, especially in characterizing the transition from isolated psychological acts to a settled, though impermanent, psychological disposition. 3.3 Neuropsychological Testing Ontology Among all the related and utilized ontologies, PAO draws foremost from the structure and resources of the Neuropsychological Testing Ontology (NPT).5 NPT represents a more general range of neuropsychological assessments, including memory, learning, attention, and expression. Along similar lines, PAO includes a list of personality scales that test for habitual behavioral and emotional patterns. In NPT, these assays provide measurement data about cognitive functions. Likewise, in PAO, the personality assessments provide measurement data about personality dimensions. Borrowing from NPT, the assays in PAO are planned processes that realize a plan that concretizes specific directive information. 4. Personality Assessment Ontology (PAO) In the interest of representing the full process of personality assessment, we divided our project into three main tasks: (4.1) the representation of roles involved in the 5 We are indebted to Barry Smith for directing us to this ontology. 9 assessment process, (4.2) the representation of particular processes of assessment, and (4.3) the representation of data accumulated from that process. 4.1 Persons, Roles, and Dispositions The aim of PAO is to represent the data produced through the process of assessing various dimensions of personality. However, personalities are not free-floating entities; they inhere in persons, who are the subjects of assessment. Therefore, the first entities that we sought to represent were persons, roles, and the personality belonging to a person. Along with ‘evaluant,’ i.e. a person in the role of undergoing some assessment, we included the role of ‘evaluator,’ i.e., the role of the person who implements the plan of assessment. As for classifying Personality, we mimicked the practice of NPT: where NPT counted various cognitive functions as subclasses of Disposition, we counted Personality as a subclass of Disposition. Alongside Personality, we identified the class, Personality Dimension. A Personality Dimension is not a subclass of Personality (since, by the “true path” rule, this would render each Personality Dimension as some Personality, which is false), but rather each Personality Dimension is a part of Personality. The subclasses of Personality Dimension we identified according to FFM: Openness to Experience, Conscientiousness, Extraversion, Agreeableness, and Neuroticism. Each of the five is a part or facet that helps make up the total Personality, which is a Disposition of the Person toward certain behaviors and beliefs. 4.2 Assays 10 Our second task was to represent the process of personality assessment such that disparate inventories and instruments could be interpreted as assessments of the five subclasses of Personality Dimension. We developed the class Assay as a subclass of Process and its subclass Planned Process (the latter borrowed from NPT). Under Assay, we created a subclass of assays (Personality Scale) specified to assess Personality, and then included subclasses of entities that were themselves specified assays, e.g. Likability Assay, Judgment Assay, and Well-Being Assay. Each subclass of Assay is drawn from the sample of personality inventories included within PAO. Both the full inventories (e.g. Myers-Briggs Type Indicator and Hogan Personality Inventory), and the individual parts of those inventories (e.g. Likability Assay is part of Hogan Personality Inventory), are included here as subclasses of Personality Scale, since both the full inventories and their parts assess Personality. Finally, each specified personality assay has as its output some score that represents the degree of presence or absence of Personality factor (see 4.3 below). Following NPT, we counted the assays as realizations of a Plan Specification, which is a subclass of Directive Information Entity, a subclass of Information Content Entity (BFO). This included as parts Action Specification (i.e. actions to be performed by the evaluator and evaluant) and Assay Objective Specification. (Recall that Personality Scale is a subclass of Planned Process.) 11 Figure 2: A personality scale and its parts 4.3 Score Data and Values The final task was to represent in PAO data values as the specified outputs of personality assessments. We needed to be able to represent, for example, whether the output of a Personality Scale recorded that an evaluant’s personality is in part characterized by high or low Conscientiousness. It became evident that PAO required the creation of some subclasses of information or data entities. To this end, we created a class of Scalar Personality Measurement Datum, which is a subclass of Information Content Entity, which is in turn a subclass of Generically Dependent Continuant (BFO). Under Scalar Personality Measurement Datum, we created only five subclasses of data entities to match FFM. This was a significant step, since PAO is claiming (perhaps not uncontroversially) that each measurement produced by the wide variety of personality inventories just is measuring the dimensions identified by FFM. In building the ontology, the question arose of whether we should create a class of data entities to match each assay (e.g. Likability Assay has specified output Likability Score), and then stipulate that the Likability Score should undergo data transformation to become an Agreeableness Score. We opted against this, since (a) we wanted to avoid introducing data transformation in a case in which the pools of data were already largely amenable, and (b) we wanted to stand behind the strong claim, in accordance with McCrae & John (1992), that a measurement of Likability produced by the Likability 12 Assay just is a measurement of Agreeableness on FFM. Thus, in PAO, a Likability Assay, which is part of the Hogan Personality Inventory, produces a Likability Score, but this score is a subclass of Agreeableness Score, and both are treated as measuring the Agreeableness of some Personality. At the same time, we excluded separate personality dimension entities corresponding to each personality inventory (e.g. Likability as a subclass of Personality Dimension). After identifying the output of personality assessments as data entities (scores), the problem of representing the specified values of these scores remained. Thus, we created the data property: ‘has score of.’ That is to say, the data entity (Generically Dependent Continuant), the score that is the output of the assessment, has the data property ‘has score of’, which is a definite numeric figure. This figure, we stipulated, has a maximum cardinality of 100 and a minimum cardinality of 1. In turn, the wide variety of inventories and questionnaires can be translated into roughly a percentage scale simply through dividing the total positive answers in the numerator by the total possible answers in the denominator: 𝑥= 𝑛 𝑠 For instance, in a questionnaire consisting of fifteen questions assessing likability, an evaluant who answers thirteen of these questions positively receives a Likability Score, and thus an Agreeableness Score, of 87%. In theory, this translation is possible regardless of the psychometric mechanism that the questionnaire employs, for instance, whether the assay uses a Likert scale, or self-ascriptions from an adjective list. However, it should be noted that the current version of PAO does not contain any representation or 13 annotation for precisely how transformations of variegated kinds of scales would take place. Furthermore, there is no indication of the significance of each score (see below). Figure 3: An assay outputs a score At this stage in its development the ontology faces a sizable difficulty. Although the majority of the personality assays correlate positively with the assessed personality dimensions (e.g. some Likability Assay with an output of 70% represents that the subject has 70% Agreeableness), a handful of the assays correlate negatively with the personality dimensions (e.g. some Well-Being Assay with an output of 30% represents that the subject has 70% Neuroticism), namely: Aggression, Critical Parent, Paranoid, Social Introversion, Adjustment, Ideal Self, Objectivity, and Well-Being. Within Protégé 5.1, we could find no tool by which to automate the inversion of the scores generated by the assays. We considered forcing each output of each assay, whether positive or negative, to undergo data transformation, but this seemed either unnecessary or misleading in the case of positively correlated data. Thus, we decided to create a separate data property, ‘has negatively correlated score,’ and annotate the inverse data property to indicate the sort of transformation the assay score (n) must undergo in order to count as a measurement of some Personality Dimension (x): 14 𝑥 = 100 − 𝑛 This is a patchwork fix to the architecture of our ontology to be sure. But we cannot at present conjure up a better solution. Figure 4: Neuroticism Score and its subclasses 5. Conclusions and Further Research At present, our model captures the relation of evidential support between instrumental measurements and the FFM, but treats the nature of that support rather crudely as equivalent percentages recorded as measurements of dimensions. This is clearly inadequate. Not all positive correlations between a score and a personality dimension will maintain an equivalent score. This is also true of negative correlations. And we have presently no means by which to distinguish the statistical significance of one type of instrument versus another, let alone specific applications of an instrument. To make this ontology useful, information concerning the individual instruments would need to be accounted for in the ontology. Such information would need to include the number of questions and the kind of questions asked (e.g. true-false, Likert scales, self-descriptions), along with the nature of the correlations to FFM. There is also information that would be required from individual applications of each measurement. This information would include basic statistical information about each study such as population size and standard deviation, as well as significance and power, but also 15 information such as geographical location, the name of the evaluator who conducted the study, and a timestamp. This kind of information would be especially important for determining the plasticity of personality over time, and the effects of socio-economic status on displays of typical behaviors associated with personality types. 5.1. Auto-Diagnostic Inference Auto-diagnostic inference would be possible if other indicators of personality types and thresholds could be incorporated into the ontology. Fascinating studies correlating personality types and lifespan, success at work, and marital bliss could be employed to infer individual diagnostic profiles for persons based on their responses to individual tests. This kind of information would not be particular to our ontology, but to any ontological mapping of the FFM. What would be unique to our approach would be the granularity achieved by our ontology, which would be able to draw inferences across test types with distinctive purposes and from times in the past before the FFM was available. The promise of our model is its ability to extend the potential of the available data to avoid reinventing the wheel. 5.2 fMRI, Ontologies, and the Need for Further Research Part of the resurgence of interest in personality stems from significance advances in neurosciences and the ability to track blood flow through the brain using fMRI. As Nettle (2007) accurately observes, “a new science is emerging of individual differences in brain structure and functioning and the results of this science can be mapped back to 16 the big five personality dimensions.”6 The mission to incorporate emerging neuroscience, social psychological research, and genetic discovery is a massive task. If it is to be successful, the ontologies describing neurological and cognitive functioning that have been characterized here, as well as the one we propose here, will need to be incorporated to capture new results. This makes the use of a proper upper-level ontology like BFO especially important. If ontologies built for distinct purposes are to be successfully merged, they must be interoperable. PAO raises multifarious questions as it answers them. It provides only a preliminary solution to the problem of amassing pools of variegated personality assessment data. Nonetheless, this step is an indispensable one. The advances within personality psychology procured by factorial-analytic method, the fruit of which is the Five-Factor Model, clear the way for further research, e.g. the correlations of health and mental disorder with shifts in personality dimensions. Personality Assessment Ontology, insofar as it supplies a model of a panoply of personality assessments, and insofar as it does so successfully, opens one avenue toward an extended, more comprehensive system of ontologies of mental functioning and mental health. 6 p.11 17 Works Cited Allport, G. W. (1937). Personality: a psychological interpretation Holt. New York, NY. McCrae, Robert R. and John P. Oliver. (June 1992) “An introduction to the five-factor model and its applications.” Journal of Personality. Vol. 60, Issue 2, pp. 175-215 Nettle, Daniel. (2007) Personality: what makes you the way you are. Oxford University Press. New York, NY. Tupes, E. C , & Christal, R. E. (1961). Recurrent personality factors based on trait ratings (USAF ASD Tech. Rep. No. 61-97). Lackland Air Force Base, TX: U.S. Air Force.