The Big Five Taxonomy

Based on a Qualifying Exam Answer

by Frank Fujita

The Question

1. What is the evidence on which the claim that there are five basic traits rests?
2. Summarize the major conceptual and empirical work relevant to the nature and usefulness of these dimensions?
3. Discuss whether this system fully captures what is meant by "personality." To what degree can individual differences be summarized by the Big Five? In other words, evaluate the strengths and limitations to this individual differences conception of personality. Might the Big Five be considered a complete description of personality?

My Answer

• Just like biology, personality psychology needs a taxonomy of its subject matter. All classification is inherently temporary and artificial, a structure serving a limited purpose. Holton (1978) said, "Concepts are primarily analytical instruments for the intellectual mastery of empirical data." A good taxonomy will point researchers in fruitful directions in their research and will allow a standard vocabulary by which their research results may be communicated and related to each other. Even a poor taxonomy, however, is better than no taxonomy at all. Without some common framework personality psychologists will have an impossible task integrating their results, or put the other way, with a good taxonomy of the domain of personality psychology, personality psychologists may achieve what many others areas of psychology have not: an integrated science.
• A taxonomy is structurally similar to a prefabricated building erected to house an aggregate of experience. The building is intended precisely to reflect its contents, and, therefore, must change to keep up with the changing nature of experience. The purpose of a taxonomy is to describe the universe of phenomena to be explained.
• In personality psychology all attempts to create a taxonomy have used factor analysis, probably because of the pioneering work done by Spearman (1923). There are many methods of taxonomy, and, of the empirically based, factor analysis ranks 13 of 18 different statistical methods used, based on external criterion validity values, 17 of 18 based on internal criterion validity values, 5 of 18 based on replicability correlation coefficients, and 16 of 18 based on an average of the three criteria listed above (Mezzich & Solomon, 1980). The winner of the methods is complete linkage or HICLUS cluster analysis. I wish I could say that I know of a replication of the Big Five factors using HICLUS cluster analysis, but I cannot.
• Personality psychology needs a good taxonomy, and I will argue that the Big Five is the best taxonomy that we currently have available. While there are some problems with the Big Five as a taxonomy, many psychologists unfairly criticize it because it is not a theory. I will argue that we should not expect more from any taxonomy of personality than biologists do from their Linnaeian taxonomy. When evaluated by this criteria, the Big Five is as good a taxonomy as we can expect until we achieve a more coherent physiology of personality processes.
• I will use McCrae's mnemonic convention for the five factors, Extraversion, Energy, and Enthusiasm; Agreeableness, Altruism, and Affection; Conscientiousness, Control, and Constraint; Neuroticism, Negative Affectivity, and Nervousness; and Openness, Originality, and Openmindedness. John (1990) proposes that these dimensions be considered the OCEAN of personality.

• The lexical approach was used by many psychologists as a starting point for scientific taxonomies (Allport, 1932; Baumgarten, 1933; Klages, 1926; Odbert, 1936). The lexical approach assumes that the socially-relevant personality characteristics have become encoded in our language. Examining the assumptions behind the lexical approach is an epistemological job. One of the consequences of the lexical approach is that our taxonomy will only describe those individual differences which are readily apparent. For example, although neuroticism seems to out predict extraversion in my non-random sample of empirical work, extraversion is considered the biggest factor of the Big Five, with neuroticism coming in a distant fourth. Probably the relative lack of observability of neuroticism accounts for its underepresentation in the English language. Thus neuroticism, strong in behavioral specification, is weak in comparison to extraversion on the lexical criteria.
• Why use a lexical approach? I think the decision can best be described as the least of a group of evils; that is, the lexical approach has the virtue of being atheoretical (It is interesting that early theorists chose the lexical approach because it was atheoretical, but that recent researchers find it lacking because it is atheoretical). Since factor analysis is basically a data reduction technique, the result of any factor analysis critically depends on the initial item pool. The lexical approach is best for the one reason that it doesn't depend on anyone's theory of personality. I can think of no reason for accepting the assumption that "the primary dimensions of personality will be reflected by a large number of trait descriptors which will covary" other than that without this assumption no further work can be done. I would feel more comfortable if I could design an empirical test of the "lexical hypothesis." Unfortunately, we only have candidates for the primary dimensions of personality if we accept the lexical hypothesis, and therefore no empirical test is possible. So, either we accept the lexical hypothesis or we take our ball and go home. Please note throughout the rest of this paper, that if the lexical hypothesis is false, the Big Five factors of personality become the Big Five factors of English personality descriptors.
• Allport and Odbert listed personality relevant terms from the 1925 edition of Webster's New International Dictionary. They came up with 18,000 terms. Terms were included if they possessed "The capacity to...distinguish the behavior of one human being from that of another." They categorized these words into four columns: traits, states, evaluation, and other. Cattell reduced these by discarding uncommon and derivative forms and by then applying cluster and factor analysis, along with a good measure of subjective judgment, to the reduced set. This produced 35 variables. John (1990) concludes, based on the subjective input at every step of the process that "Cattell's lists of variables and factors primarily represent those traits that he himself considered most important." From there, other major work was done by Fiske (1949), Tupes and Christal (1961), Norman (1963, 1967), Goldberg (1976), and most recently by McCrae and Costa (1965). These researchers have all discovered five major factors of personality. Although the original factor analyses all depended on Cattell's list of terms, many other conceptual replications have found five factors using different lexical criteria (Botwin & Buss, 1989; McCrae & Costa, 1985). Even critics of the Big Five have admitted that the five factor finding is robust.
• If the Big Five are basic personality traits, then they should be found in all cultures and in all languages. There has been one cross-language study that conceptually replicated the lexical criteria of the English Big Five. In that study, Hofstee used the Dutch language and found five replicable factors which were easily identified with the English Big Five (John, 1988). A German taxonomy project has also begun (Angleitner, Otsendorf, & John, 1990), with preliminary results indicating a clear replication of the Big Five factors. A bi-lingual study of German-American bilinguals shows Big Five factor correlations in the .72 to .84 range. Evidence for replicability in the Germanic languages seems clear.
• Using translations of English measures, White (1980) found replicability of Extraversion and Agreeableness in A'ara (Solomon islands) and Orissa (India). Japanese and Chinese translation replications also replicate the entire Big Five, (Bond & Forgas, 1984). The Filipino evidence is contradictory, with some researchers finding agreement (Church & Katigbak, 1989), and others not (Guthrie, & Bennett, 1971). Unfortunately, finding agreement of factors from translated scales is much less impressive than a replication using the lexical criterion from scratch. Thus, the evidence for non-western languages is much more muddy, and any conclusion about the universality of the Big Five would be premature.

• Before accepting the Big Five taxonomy of personality, one should look at the alternative taxonomies. Choosing a poor taxonomy when it is the best one around is defensible. Choosing a very good taxonomy when there is a better one is indefensible. So let's look at some competing taxonomies to see how the Big Five stacks up.
• Cattell's complex system has failed to independently replicate. No one (Banks, 1948; Digman & Takemoto-Chock, 1981; Fiske, 1949; Howarth, 1976) has been able to find more than seven factors in the original correlation of the rating studies that were the basis of the system. This failure is not a failure to replicate in a different language, or a different sample, but even with the same correlation matrix! This may be due to the primitive numerical analysis methods used on early computers, along with the eyeball methods of factor rotation used by Cattell. Thus, one could use the Cattellian system as an a priori theoretical system, depending on his clinical observation of the relevant dimensions of personality, but on purely empirical grounds, the Cattellian system is clearly inferior to the Big Five.
• Eysenck's Extraversion, Neuroticism, and Psychoticism have been routinely used in many studies of personality. Tellegen (1985) created the Multidimensional Personality Questionnaire (MPQ), which when factored gives three trait dimensions similar in content to Eysenck's. Block (1977) criticizes the labeling of Psychoticism, suggesting that it is an amalgam of Agreeableness and Conscientiousness. Adding intelligence would provide five factors; therefore, one can almost translate one into the other. Given the relative lack of empirical progress on the Psychoticism dimension, it seems that the Eysenckian system of personality can be subsumed without loss of information into the Big Five.
• Guilford also excluded intelligence from his personality system, and a second order factor analysis (Amelang & Borkenau, 1982) supports his claim (Guilford, 1975) that his concepts fit the four non-intellectual factors of the Big Five. Guilford then seems to have a middle-level of analysis theoretical framework from which one could begin to unpack the Big Five. Other personality systems can also be described in Big Five language. For example, Wiggins (1980) fit the interpersonal circle (Leary, 195 7) to the Extraversion and Agreeableness factors of the Big Five. Therefore, it seems possible to incorporate all seriously competing theoretical systems into the Big Five set.

• One problem with the Big Five is its specificity, or lack thereof. Briggs (1989) finds fault with the Big Five because each researcher comes up with a different Big Five. Waller & Ben-Porath (1987) ask "Which Big Five?" For example, Openness has been called Culture and Intellect. Are these three factors or just one? John (1990) attempted to answer this question. Using the 300 items in the Adjective Check List, he had 10 judges assign each item into one of the Big Five, or into a sixth, unassigned category. Of the 300 items, 112 were assigned with 90% agreement or better to one of the Big Five. When the ACL was factor analyzed (on a different data set) 111 of the 112 items loaded on the hypothesized factor in the hypothesized direction. These results would suggest that there is basic agreement about what is represented by each of the Big Five dimensions.
• Unfortunately, this points to a second difficulty since the same data can be rephrased as 188 of 300 items were not assigned with 90% agreement or better to one of the Big Five. This suggests that there is a large portion of the personality sphere left to describe. Thus, the Big Five is not inclusive enough, for example, John (1990) finds that the Big Five fail to capture dimensions of individuation/autonomy, maturity, physical characteristics, traditional values, and gender terms. I do not think that personality psychology is willing to declare these issues non-problems. Thus, at some point in time, probably after the Big Five factors have been incorporated into a theory of personality, we will need to return to these personality dimensions to see if our theory is truly general or only capable of describing the domain on which it was based.
• Related to the problem of lack of inclusiveness, one might object to the Big Five on the a priori grounds that five dimensions cannot account for all of personality. The same objection could be made that the three categories "animal", "vegetable", and "mineral" cannot possibly account for all material objects. The Big Five are very broad dimensions, and the disadvantage of their large bandwidth is their low fidelity. Thus, if one is interested in predicting specific behaviors, one should not use Big Five scales, but rather, more behavior-specific scales. However, one should not confuse predictive specificity with theoretical value. Because of their large bandwidth, Big Five dimensions will tend to have some predictive utility on almost any personality-relevant behavior. And, of course, to the extent that behaviors are determined by more than one Big Five dimension, the largest correlation that one could expect is .30 (Ahadi & Diener, 1990).
• A third difficulty with the Big Five is that, unfortunately, there seems to be little agreement as to the basic nature of the fifth factor--openness. McCrae doesn't consider openness to include intelligence and has some empirical evidence to back up his conclusion (McCrae, 1990). On the other hand, if Openness doesn't include intelligence, then the mapping of Guilford's and Eysenck's personality systems onto the Big Five becomes problematic. Thus, we may have six big factors, with factors 5 and 6 alternating in various factor solutions.
• A fourth problem is the identification of the referent of the Big Five dimensions. Thus, Briggs (1990) raises objections to the Big Five because it may only be the structure of the personality lexicon, not the structure of personality. He suggests that it is premature to close off alternate research programs involved with discovering the underlying psychophysical processes involved in the production of visible differences in behavior. I think that this is a viable reservation to keep in mind; however, until an alternate taxonomy of personality comes along, I would encourage physiological psychologists to use the five factor model. Indeed, research by Strelau & Eysenck (1987) and Gray (1987) show very promising avenues of research into the physiological bases of Big Five dimensions. Note, however, that Gray (1987) finds that based on lateralization studies, the extraversion/neuroticism dimensions should be rotated to an arousal/valence solution. Thus, we have some preliminary evidence that the rotation process of the lexical strategy may have been misconceived.
• Fifth, and last, is parsimony. One thing that would be desirable in a dimensional taxonomy would be independence of dimensions. This would allow one to consider any dimension without confounding it with any of the others. Unfortunately, the Big Five are significantly correlated. Extraversion and Neuroticism correlate about -.30 in observed data. In my master's thesis, the latent variables of Extraversion and Neuroticism correlated over -.50 across two studies using three methods. In addition, Extraversion and Agreeableness correlate between .30 and .40 in observed data. Thus, the Big Five do not provide independent dimensions. In summary, there are some empirical and conceptual difficulties with the Big Five, but these difficulties do not include poor predictive validity, the impossibility of condensing personality to five dimensions, or a better competing taxonomy.

• Given the hierarchical nature of trait concepts and the hegemony of the Big Five at the top level of personality description, many current personality psychologists are trying to advance their conception of the "middle level" of personality description. Norman (1982) and Goldberg (1989) have developed 40 to 75 middle-level categories. Unfortunately, these categories have not been systematically investigated or operationalized. Costa and McCrae (1989) have created "facets" of their NEO, and Hogan ( 1985) has conceptualized "homogeneous item composites." These four middle levels of analysis do not converge with each other in the same way as the Big Five do from researcher to researcher. But unfortunately, the conceptual confusion gets worse.
• Somewhat different approaches in staking out the middle-level units include Little's (1990) personal projects, Emmons (1986) personal strivings, Norem's (1990) cognitive strategies, Cantor & Kihlstrom's (1987) life tasks, Buss & Craik's (1983) act trends, and more fundamentally Kelly's (1963) personal constructs, and Murray's (1938) needs. It seems evident that most of the individual differences in these middle-level units of analyses overlap, both with each other, and under the Big Five. Of course , each of these different-named, different-operationalized concepts focuses on a different aspect of the human experience. When a process theory of individual differences is capable of distinguishing one of these middle level concepts from the other, then the field will be ready for these different foci. Until then, the propagation of these different concepts will only confuse the field. Costa & McCrae (1988) have begun the work of integrating Murray's needs as operationalized in the PRF (Jackson, 1984 ) into the five factor model. McCrae (1990) shows that many of the omnibus personality scales (NEO-PI, MBTI, Goldberg adjective scale, CQS, GZTS) are measuring the same (five) things. It would be useful to be able to reduce the (rhetorically) infinite number of personality scales to five dimensions. I hope that by consistently placing the dimensions of these "new" conceptual units of psychology under the umbrella of the Big Five, the incessant relabeling of basic individual difference phenomena can be muted, if not halted altogether.

• There are problems with the Big Five. Conceptually, we must assume the lexical hypothesis. This is especially problematic given recent physiological investigations of individual differences processes of arousal and valence. The Big Five do not encompass over 50% of the 300 ACL items; they leave independence, maturity, gender, and attractiveness, among others, outside the personality domain. The conceptual evidence suggests that Openness may not be the same factor from analysis to analysis. All of this suggests that the Big Five leaves much to be desired in a taxonomic system.
• However, every research decision must be a choice between alternatives. Compared to any of the competing taxonomies, the Big Five is a progressive research program, as described by Lakatos (1968). McDougal (1938) posited that personality is the study of both species-wide tendencies and tendencies specific to individuals. One feature of the Big Five is that it is only a taxonomy of the second type of tendencies. Thus, even if the Big Five were a perfect taxonomy of individual differences, it would not be a complete description of personality. This is not a fault, per se. Thus, it is difficult to imagine a taxonomic scheme that would cross the species-wide versus individual differences domain of behavior. Another non-problem with the Big Five is that it is atheoretical. That is, with the exception of the assumption of the lexical strategy, the Big Five does not tell us anything about the workings of personality. What the Big Five does, and does well, is delineate a domain and announce to the world of personality psychologists "I am a problem to be solved!" Thus, the Big Five is instrumental to the research strategy of personality psychology because it makes salient one domain of data to be addressed by any personality theory: namely, the origin and function of the Big Five factors of personality.

Thus, the most attractive feature of the Big Five is not that it is a comprehensive theory of personality, for the Big Five is not a personality theory at all. It is a datum, which any theory of personality must explain. Thus, its empirical atheoretical nature is not a handicap. Rather, by reducing the plethora of psychological constructs to a manageable five it allows a personality theory the possibility of parsimoniously explaining of the brute facts of individual differences.

 

The Illusory Nature of Personality Traits

Here are some ways in which people may appear more trait-like than they really are.
1. Impoverished sampling of environments

 

We do N0T take into account that we generally see a particular person in the same situations, so people seem more consistent than they really are. This creates the impression that traits must underlie behavior, not situations.
 

2. Incomplete adjustment for environmental restraints

 

We fail to sufficiently adjust for the situational forces present in a particular environment. This error used to be called the `fundamental attribution error` in the scientific literature, but it is now called the `correspondence bias.`
 

3. The effects of our own presence

Our perceptions of others generally require our physical presence, and we fail to realize that the target person (whom we are observing) is probably reacting to our presence; this is a type of situational constraint.
 

4. Behavioral confirmation and self-fulfilling prophecies

We interact with others according to a set of preconceptions or expectations. These expectations may cause the person to behave in ways that confirm our expectations. Or, we might misinterpret the other`s behavior to be consistent with expectation.
 
 
 
 

Berscheid`s Ugly Duckling study: men can make women beautiful merely by expected them to be beautiful.
 
 
 
 

Other studies have shown these impressions can, under some circumstances, become internalized by the target person.
 

Because of these confirmation effects, trait impressions have a self-perpetuating nature to themselves, even when false.

5. Biased sampling of other people`s perceptions.

What others (third parties) tell us about a person can be biased:

1. We associate with others who are similar to us and share our prejudices and beliefs.

 
 
 
 
2. Linguistic bias--Our language is laden with trait terms so it is very difficult to explain someone`s behavior without implying traits.

 
 
 
 
6. Perceptual bias of physical appearance and mannerisms

 

Because people look and act pretty much the same, their constant physical appearance deludes us into thinking that their internal traits are constant.

7. The need to control and predict the behavior of others

 
 
 
 

We have to acknowledge that trait impressions serve our need to control or predict the course of interaction with others.

For further reading, see:

Goldberg, LR. (1992). The development of markers for the big-five factor structure. Psychol. Assessment, 4, 26-42.

Goldberg, LR (1993a). The structure of phenotypic personality traits. Am. Psychol., 48, 26-34.

Goldberg, LR. (1993b). The structure of personality traits: vertical and horizontal aspects. In DC Funder, RD Parke, C Tomlinson-Keasey, & K Widaman (Eds.), Studying lives through time: personality and development (pp. 169-88). Washington, D.C.: American Psychological Association.

John, OP. (1990). The "Big Five" factor taxonomy: Dimensions of personality in the natural language and in questionnaires. In LA Pervin (Ed.), Handbook of personality: Theory and research . New York: Guilford.

Revelle, W. (1995). Personality Processes, Annual Review of Psychology.

Zuckerman, M. (1991). Psychobiology of personality. Cambridge: Cambridge University Press.

Zuckerman, M. (1994). Impulsive unsocialized sensation seeking: The biological foundations of a basic dimension of personality. In JE Bates & TD Wachs (Eds.), Temperament: Individual differences at the interface of biology and behavior (pp. 219-55). Washington, D.C.: American Psychological Association.