Chapter 2: Literature Review
In this chapter, focus is on various relevant literatures that explain the concept of personality traits as well as the use of personality questionnaires.
One of the pioneering trait psychologists, Gordon Allport (1937), saw traits as organised mental structures, varying from person to person, which initiate and guide behaviour. There are two important qualifications to this general principle. First, as Hettema and Deary (1993) pointed out, the explaining of behaviour requires different levels of analysis, including genetics, physiology, learning and social factors. Allport's notion that all the various manifestations of traits can be explained at a single level of 'mental structure' is simplistic. Hence, causal models of trait action will vary depending on the level investigated, although the ultimate research aim is to develop a trait theory that will connect various levels. Second, the causal effects of traits on behaviour may be indirect. Traits interact with situational factors to produce transient internal conditions or states, which may sometimes be a more direct influence on behaviour than the trait. For example, trait anxiety may interact with an immediate situational threat to generate transient state anxiety, which in turn disrupts ongoing information processing and impairs performance (Spielberger, 1966).
Tendency to experience negative emotions are assumed by some to relate to some fundamental, core quality of the person, which might be influenced substantially by genetic factors (Eysenck, 1967; McCrae et al., 2000). Again, even within theories that are sympathetic to the traditional view of traits, there has been some modification of the basic view. For example, Cattell and Kline (1977) distinguished 'surface' traits, which are simply clusters of overt responses which tend to be associated, from 'source' traits, which are deeper properties of the person with causal effects on behaviour. Modern developments of traditional theory seek to identify and explain underlying sources of consistency in behaviour, whether these are conceived of as genetic, physiological or cognitive in nature. The process of relating operationally defined measures such as questionnaire scores to theory is often referred to as construct validation.
Both assumptions of traditional trait theory—their causal primacy and inner locus—have been challenged more radically. The alternative to causal primacy is the view that traits are a construction with no independent causal status. For example, Buss and Craik (1983) argued that traits are simply descriptions of natural categories of acts. Wright and Mischel (1987) characterised traits as conditional statements of situation—behaviour contingencies. Furthermore, traits may be jointly constructed by two or more people in social interaction, according to the social dynamics of the situation (Hampson, 1988). Social psychological approaches to traits tend also to abandon the inner locus assumption. Even if traits represent genuine psychological structures, these structures may be no more than the superficial mask the person presents to the outside world, in order to present a socially acceptable self-image to other people.
The upshot of these considerations is that there is no generally accepted scientific theory of traits. Some trait theorists have tended to take the relatively easy option of focusing on the dimensional structure and measurement of traits rather than investigating their underlying nature (Goldberg, 1993). However, it should be clear from the preceding discussion that trait descriptions cannot be accepted at face value, and that there may be various qualitatively different types of explanation for consistencies in self-reports and behaviours. In recent years, progress has been made in developing psychobiological information processing and social psychological trait theories, which are partly complementary and partly competing accounts.
Personality Questionnaire: Development and Quality Factors
Contemporary views of traits are intimately related to the processes of measurement and assessment necessary to identify basic personality dimensions. Typically, the trait researcher has some hypothesis about the number and nature of the principal dimensions, and designs a questionnaire to measure them. Subsequent work investigates how useful a measuring device the questionnaire actually is, and modifies the questionnaire items in response to any shortcomings detected.
The initial development of a satisfactory questionnaire for measuring traits is not easy. Care must be taken in the composition of items: they must be easily understood and unambiguous, applicable to all respondents, and unlikely to cause offence (Angleitner & Wiggins, 1986). There should also be some systematic sampling of the various expressions of the personality trait of interest. It is important also to check that items are not strongly contaminated by response sets or biases, such as social desirability, yea-saying or extreme responding. It is also necessary to assess its adequacy formally, by application of psychometrics, the science of psychological measurement. Psychometrics provides statistical techniques which determine the ability level of a particular questionnaire as a measuring tool. The sophistication of modern techniques and the number-crunching power afforded by computers provide the contemporary researcher with powers of data analysis far beyond those envisaged by the pioneering trait researchers.
In order to ensure the efficacy of a personality questionnaire, several factors must be considered. One of which is its reliability. This refers to the accuracy with which the questionnaire measures a given quality. Reliability may be assessed by administering two alternative measures of the trait to a sample of subjects and computing the correlation between them. If the correlation is high, the quality can be assessed consistently and the scale is reliable or internally consistent. Otherwise, if the two supposedly equivalent forms are not assessing the same quality, the scale is unreliable and the items must be revised. The Cronbach alpha statistic is a widely used measure of reliability calculated from a single set of test items. It is, in effect, the correlation of the test with itself. In general, alpha tends to increase both as inter-item correlation increases, and as the number of items on the test increases (Henson, 2001).
Another important factor to consider in personality questionnaires is their stability. Reliability should be distinguished from stability, which is the test— retest correlation of the scale over a given time interval. Personality is expected to change slowly as the person grows older, but it is expected that stabilities of trait measures will be fairly high over periods of a year or more. If the assessor has a scale that is reliable but has a low test—retest correlation, the assessor may be evaluating a mood or some other transient quality of the person, rather than a genuine trait (Neville et al., 2001).
The third essential factor for a personality questionnaire is validity, which pertains to the tool's ability to assess what it purports to assess. A scale may be reliable but not valid. For example, a fortune teller might use a highly consistent method for inferring a person's future from the lines on their palm, but the consistency of the technique would be no guarantee that the fortune teller's predictions were accurate. The most straightforward and convincing method for assessing validity is referred to as criterion or predictive validity. The trait measure is correlated with some independent index of a quality associated with the trait. Other external criteria frequently used in personality research include measures of job performance and behaviour, psychophysiological functioning and clinical abnormality (Bartram, 1995). Establishing predictive validity is indeed important. The essence of integrating validity is that correlations between the trait and external criteria are predicted in advance from an adequate scientific theory, rather than from common sense or a superficial analysis of trait characteristics. For example, the psychobiological theory of personality can be used to predict how a particular trait should correlate with measures of autonomic functioning, such as heart rate.
Another form of validity is called construct validity. This arises out of the total web of empirical data and theoretical analysis, which builds up around a trait, sometimes referred to as its nomological network (Eysenck, 1981). The difficulties of construct validity are those of establishing scientific truth. Even 'good' theories are never fully satisfactory, and require periodic modification of hypotheses and concepts as new research findings are obtained (Lakatos, 1976). Hence, construct validity is always somewhat provisional, and may be reduced or enhanced by fresh research.
Personality Questionnaire: Empirical Studies
There had been a number of studies related to personality questionnaires and its ability to detect various personal attributes. In one research, Epstein (1977) asked subjects to rate and describe their positive and negative emotions, impulses, behaviour and situations for over two weeks. Although the correlation between single days was as low as suggested by the work of Mischel (1968) and Bem (1972) suggested, the reliability of measures in each of these categories ranged from 0.40 to 0.88, with a median of 0.72 when odd and even days were correlated for data collected for between twenty-four and thirty-four days. Another message of this study was that, in all of the above categories, a certain minimum frequency of occurrence and variance was required to achieve high reliability, whether it was between behaviour and emotions. Epstein reckoned that, given the frequent assertion that there is a 0.30 barrier for cross-situational reliability coefficients, the findings of this study are no less than dramatic.
Personality, behaviour, and even situations as scored by judges independent of the subjects, were all highly reliable when aggregated over several days; the low predictive validity coefficients claimed by the situationists for personality variables were imposed by error of measurement as the result of single observations. Therefore, the procedure that others have employed all but guarantees reliability coefficients to be low. It may be concluded that those who have argued that personality is unstable have simply not used procedures that can establish its stability. As Eysenck (1981) pointed out, aggregation of data actually provides quite good evidence for cross-situational consistency in studies such as that of Hartshorne and May (1928) which purport to show situation specificity of behaviour. Similarly, when personality is assessed through judges' ratings, large numbers of behavioural observations may be needed for the behavioural consistency and predictive validity of traits to appear (Moskowitz & Schwartz, 1982).
Moskowitz (1988) studied the reliability of ratings and behaviour counts of friendliness and dominance in forty-three subjects who visited a laboratory on six occasions in order to conduct a problem-solving exercise with one partner. Correlating ratings (inferred traits) of friendliness and dominance made in one situation with only one other situation gave coefficients of 0.26 and 0.12, respectively; both were non-significant, but of the order expected from the criticisms of Mischel. The same analyses performed on behaviour counts gave coefficients of 0.37 (p < 0.05) for friendliness and 0.06 for dominance. However, when generalisability (using coefficient alpha) was calculated using the six situations the ratings values for friendliness and dominance were 0.68 (p < 0.001) and 0.44 (p < 0.01), respectively. The value for behaviour counts for friendliness was 0.78 (p < 0.001) and for dominance, 0.28 (ns). She concluded that there were high levels of cross-situational generality for behaviour count and ratings measures of friendliness (aggregated over six laboratory situations), and moderate levels of generality for ratings of dominance.
Further, using data from only five situations to predict friendliness ratings or behaviour counts in a single situation, multiple R values of 0.50 and 0.57 were obtained for ratings and behaviour counts, respectively (both p < 0.01). For dominance, the expression of relevant behaviour was affected by whether the subject knew the partner they were with in the situation. The use of abstract qualities such as friendliness also seems to raise behavioural consistency. Funder and Colvin (1991) showed cross-situational consistencies typically of 0.4–0.6 for behaviour coded by meaning, but substantially smaller consistencies for specific instances of behaviour. For example, 'humour' is more consistent than 'joke-telling'.
Other later works on personality tests made use of trait constructs to predict behaviours with remarkable success. Researchers used behavioural dispositions in a particular way—one that takes the context into account and may be seen as a form of interactionism (Wright & Mischel, 1987). As an alternative to theories that see traits as causal agents or as mere summaries of observed behaviours (e.g. Buss & Craik, 1983), Mischel sees a trait statement as the 'conditional probability of a category of behaviours in a category of contexts'. It is hard to imagine any trait theorist taking exception to this definition, and the present authors consider it to be a good, mainstream definition of a trait, stripped of beliefs about the origin of the trait. In particular, the point that traits most reliably express themselves in situations that are suited to their expression is accepted by most, if not all, personality trait theorists. That is, it is difficult to express extraversion whilst marching with other soldiers in a parade, but much easier to express it at a party. What is remarkable about Mischel and colleagues' research is the care with which it is formulated and executed, and the high level of predictive validity that it provides for personality traits from this once champion of situationism.
In a study, Wright and Mischel (1987) asked raters to assess children on the traits of 'aggression' and 'withdrawal'. Several different observers watched the children's actual behaviours over a period of time. The raters were also asked to judge how demanding the situation was for the child, in comparison to the child's competence. The hypotheses were complex: that children with high levels of a trait would show more behaviours that were central to that trait (feature-centrality), and that correlations between traits and behaviours would be especially high if the situation was a demanding one for the child. Feature-centrality needs explanation: with regard to aggression, one 'feature-central' behaviour would be a threat issued to another child. The feature-central threatening behaviour would be expected to show higher correlations with aggression than would a non-feature-central trait such as distractibility.
As hypothesised, children with given levels of a trait showed more trait-relevant behaviours. The correlations are especially strong when the demand level of the situation is high, and when the rated behaviour is central to the trait concept, although correlations are substantial for feature-central behaviours even in low-demand situations. Ratings of traits made by others do predict objectively observed behaviours. Wright and Mischel's study is a success for trait theory, situationism and interactionism all at once: traits were highly predictive of behaviours, the relevance of the situation made a difference to the behavioural scores, and there was also a significant trait—situation interaction. Thus, highly aggressive children displayed more overall feature-central behaviours such as pushing and shoving, which further increased as the demands of the situation rose.
This model of interactionism has continued to develop, and Mischel and colleagues have conceptualised personality as a dynamical system (Mischel and Shoda, 1995; Shoda, LeeTiernan, & Mischel, 2002). The authors' Cognitive Affective Personality System (CAPS) describes affects, goals, expectancies, beliefs, competencies, and self-regulatory plans and strategies as the basic units of personality. The outcome of these interacting units is typically of an if…then… form: e.g., if you encounter someone you know, then behave in a friendly manner. The individual's repertoire of if—then connections provides a unique behavioural signature or profile for that person. Typically, these outcomes are then highly contextually dependent: e.g., showing friendly behaviour towards acquaintances, but not to strangers or work colleagues. Nevertheless, the model assumes some personality stability that produces consistency in how the individual behaves in specific situations. As with trait models, it assumes personality develops from both biological and cognitive-social influences, a point to be elaborated in subsequent chapters.
Mendoza-Denton and associates (2001) studied person by situation interactions by asking subjects to describe themselves in 'if—then' terms ('I am…. when….') after they had performed a task for which they were given either positive or negative feedback. In doing so, the subjects were less likely to put themselves on extreme ends of dimensions (as they might using standard personality inventories), and less likely to misattribute—or overgeneralise—success or failure to themselves, rather than to the specific situation. In addition, the 'if—then' framework also reduced the likelihood that subjects would attribute reasons for others' behaviour to stereotypes. This 'dynamical system' has also been modeled using computer simulations of in the same terms, this idea is borne out in applied fields of research, too. For example, while behaviours in certain crime situations are consistent across individuals, people's traits alone do not predict criminal involvement (Alison et al., 2002). Other studies have also shown that 'driver stress' is predicted from situational factors such as traffic congestion and time pressure of the journey (Hennessy, Wiesenthal & Kohn, 2000), together with dispositional stress vulnerabilities that are specific to driving (Matthews, 2002).