Integr Psych Behav DOI 10.1007/s12124-015-9311-9 R E G U L A R A RT I C L E
The Big Five Factor Marker Adjectives Are Not Especially Popular Words. Are They Superior Descriptors? Eka Roivainen 1
# Springer Science+Business Media New York 2015
Abstract Vocabularies of natural languages evolve over time. Useful words become more popular and useless concepts disappear. In this study, the frequency of the use of 295 English, 100 German, and 114 French personality adjectives in book texts and Twitter messages as qualifiers of the words person, woman, homme, femme, and Person was studied. Word frequency data were compared to factor loadings from previous factor analytic studies on personality terms. The correlation between the popularity of an adjective and its highest primary loading in five- and six-factor models was low (−0.12 to 0.17). The Big five (six) marker adjectives were not more popular than Bblended^ adjectives that had moderate loadings on several factors. This finding implies that laymen consider Bblended^ adjectives as equally useful descriptors compared to adjectives that represent core features of the five (six) factors. These results are compatible with three hypotheses: 1) laymen are not good at describing personality, 2) the five (six) factors are artifacts of research methods, 3) the interaction of the five (six) factors is not well understood Keywords Personality theory . Personality descriptors . Word frequency
Introduction Travel guides often include a brief vocabulary of the language spoken in the destination country. For example, the words wine and water might be included in the restaurant vocabulary, and the words gasoline and mechanic in the motoring vocabulary. If a guidebook were to contain a short personality vocabulary, which words should be included? Personality can be defined as Bthose characteristics that account for a
* Eka Roivainen [email protected] 1
Verve Rehabilitation, PL 404, 90100 Oulu, Finland
Integr Psych Behav
person’s consistent patterns of feeling, thinking, and behaving^ (Pervin et al. 2005 p.6). The prevailing five-factor model of personality (Goldberg 1990; McCrae and Costa 1999) implies that the top five words of the personality vocabulary list should describe the five broad factors of personality that have emerged from factor analyses of the multitude of expressions used to describe personality. These five factors seem to account for most of the variation in human personality, and they have been labeled by adjectives with high primary loadings on the factors. A number of standardized tests (e.g., NEO-PI; Costa and McCrae 1995) have been devised for professional assessment and description of personality based on the five-factor model. Thus, the tourist vocabulary might include translations for the big five marker adjectives: open-minded, conscientious, extroverted, agreeable, and neurotic. The restaurant and motoring vocabularies are not based on factor analysis; rather, they are based on their assumed usefulness and frequency of the use. The label adjectives of the Big five personality scales are common words; however, they are not among the most popular personality adjectives. If the popularity criterion were used to choose five personality descriptors, they might include words such as nice, happy, friendly, smart, and confident, which are among the most prevalent concepts in openended personality descriptions in experimental settings (Ames and Bianchi 2008) and on the internet (Roivainen 2013). For example, during week 13/2014, the Big five marker adjectives listed above were used 167 times as qualifiers of the word person in Twitter tweets, while the frequency for friendly person was 1,700 (Topsy 2014). However, in Saucier and Goldberg’s (1996a, b) five-factor model, friendly does not have a very high loading on any of the Big five factors. In this framework, friendliness is a Bblended trait^ or a mixture of extraversion (a loading of 0.39 on this factor) and agreeableness (0.37). One of the great challenges for personality science has been the cross-situational variation of personality (Mischel and Shoda 1995; Uher 2014a). Textbook definitions of personality often refer to Brelatively enduring^ (Larsen and Buss 2005) or Bconsistent patterns^ (Pervin et al. 2005) of thought, emotion, and behavior. For example, a student may be shy in the classroom and extroverted during recess. To assess the general and typical ways of behaving and thinking, personality scientists have traditionally resorted to questionnaires (Bdo you/does your friend usually behave in a shy manner^). Obviously, this method has several drawbacks, including problems caused by the subjectivity of the data (Uher 2014c). Socially desirable answering is also a well-known problem in personality assessment (Backström and Björklund 2013) and, at a more theoretical level, concepts such as shy do not have a fixed, objective, and stable definitions (Toomela 2003). The ability to understand and predict the behavior of other people is a basic human skill. Languages that lack concepts such as smart, dumb, nice, and mean are unthinkable; societies with completely invalid models of human personality cannot be functional. While laymen may know little about neurochemistry or astronomy, they have a certain competence in assessing personality. We can assume that words that are useful in describing behavior and personality will progressively become more popular, while words that are less useful will disappear. The problem of generalizability of results from students or any other experimental samples to the general population may be overcome by studying the use of personality terms in large online corpora, such as the Google books database. Leising et al. (2014)
Integr Psych Behav
found that participants in a student sample and in a corpus of online communications used terms rated as important in describing person characteristics more often. It may be hypothesized that if the variation in personality is explained by five underlying factors that are well described by words appearing in natural languages, those words, Bfactor markers^ that have high loadings on only one factor should be more popular than blended concepts, such as friendly, which have moderate loadings on two or more factors. Words that represent the core features of the factors should be more powerful descriptors than words that fall between the factors. In a similar fashion, a subtest of an IQ test that has a high loading (e.g., perceptual organization factor [POI]) is a better predictor of scores in the other subtests of the POI than a subtest that has a lower loading. In a recent study, Roivainen (2013) found that the correlation between the frequency of the use of a personality adjective on the internet and its highest loading on one of the five factors in Saucier and Goldberg (1996a, b) model was fairly low (0.12). Roivainen examined bigrams composed of a personality adjective and the word person. The correlation between word frequency on the internet and in the Google books database was 0.58. In another study (Roivainen 2014), the correlation between usage frequency for bigrams composed of a personal adjective describing openness + person and personal adjective + woman was 0.60 in the Google books corpus. Thus, some variation exists in the usage frequency of personality adjectives across corpora and qualified nouns. In the present study, the correlation between the frequency of use and factor loadings of personality adjectives in five (Saucier and Goldberg 1996a, b; Ostendorf 1990) and six factor models (Lee and Ashton 2008; Boies et al. 2001) was examined using words from three different languages, English, French, and German, with the following qualified nouns: woman, person, homme, femme, Person. It was hypothesized that primary factor loading and communality are positively correlated with word popularity in book texts and Twitter messages.
Method The adjectives and their factor loadings were taken from four previous studies. Saucier and Goldberg (1996a, b) five-factor model is based on the factor analysis of correlations of 432 English adjectives selected from Norman’s (1967) list of 2,797 terms. Lee and Ashton’s (2008) six-factor Hexaco model is based on the analysis of 449 personality adjectives selected from Goldberg’s (1990) list of 1,711 adjectives. There were 300 adjectives, which were used in both the Goldberg and Saucier and Lee and Ashton studies, and these adjectives were examined in the present study. Ostendorf (1990) proposed a five-factor model based on the analysis of 430 rating scales composed of German personality adjectives. In the present study, the 50 highest loading German adjectives (mean loading = 0.66) and 50 adjectives with moderate loadings (mean loading = 0.40), ten for each factor, were compared. Boies and others (2001) analyzed the French personality lexicon using 388 popular personality adjectives and concluded that the data best fit a six-factor model. In the present study, the 10 highest loading French adjectives for five factors and seven highest loading adjectives for the openness factor, in all 57 adjectives (mean loading = 0.55), were compared to 57 adjectives with moderate loadings (mean loading = 0.36).
Integr Psych Behav
The frequency of the use of English personality adjectives was examined using bigrams composed of an adjective and the word person or woman: kind woman, kind person. The French bigrams were composed of an inflected adjective and the word homme or femme: homme doux, femme douce. The German bigrams were composed of a personality adjective and the word Person. The weak inflexion form of the adjective that follows both the masculine and feminine definite articles and that ends with -e for all cases was used (e.g., intelligente Person.) Based on preliminary analyses, the English adjectives natural, responsible, rational, reasonable and sexy were eliminated from final analyses because of poor correlations of usage frequency as qualifiers for person and woman and across corpora. The first four of these often appear in legal texts (natural person, responsible person) and do not necessarily refer to personality. The Google books database is based on over five million books and documents and is the largest existing text corpus (Michel et al. 2011). Usage frequency of the Ngrams for the year 2000 were searched with 3 years of smoothing. The American English, German, and French databases were searched. Topsy (2014) is a search engine for the social website Twitter. The Topsy index includes billions Twitter messages known as Tweets. The frequency of English and French bigrams described above was searched among Tweets from March 2014. The rationale for using two different databases, the Google books and the Twitter Topsy database was that they represent different types of texts. The Google books database includes all kinds of written texts, mainly books, and we may assume that the written language of the texts in the Google books database is more formal than everyday speech is. The Twitter tweets are more speech-like, being interactive communication, they are short and not necessarily grammatically correct.
Results Table 1 shows the factor loadings and usage frequency data for selected English adjectives. As Table 2 shows, the correlation between word popularity and factor loading was low for all conditions (−0.12 to 0.17). Table 3 shows that the six-factors of the Hexaco model combined account for 27–29 % of the total variance for the traits represented by the 100 most popular words and 23–24 % of the variance for the traits represented by the 95 least popular words. In comparison, the communalities for the marker adjectives sympathetic (Agreeableness), withdrawn (Extraversion) and organized (Conscientiousness) were 0.32, 0.45 and 0.44, respectively. Table 4 shows that correlations between word usage frequency across databases and qualified nouns ranged from 0.20 (frequency of adjectives qualifying homme in Twitter x frequency as qualifier of femme in Google books) to 0.74 (Google books person x Google books woman).
Discussion The results of the present study imply that the implicit layman theory of personality that underlies personality vocabularies seems to consider Bblended^ traits, such as
Integr Psych Behav Table 1 Primary factor loadings and the frequency of the use of selected English personality adjectives as qualifiers of person and woman Twitter
Twitter
Book
Book
a
b
messages
messages
texts
texts
factor
factor
…person
…woman
…person
…woman
loading
loading
Jealous
39000
6000
32
29
47 (N)
29 (H)
Smart
38000
8000
32
59
49 (O)
37 (O)
Loyal
31000
10000
11
6
43 (A)
43 (H)
Shy
20000
170
64
16
65 (Ex)
64 (Ex)
Simple
16000
3000
28
29
45 (O)
35 (O)
Sympathetic
561
14
18
10
62 (A)
45 (Em)
Withdrawn
6
0
5
2
67 (Ex)
63 (Ex)
Five
Six
Organized
849
25
24
6
65 (Cs)
64 (C)
Moody
4100
214
21
9
53 (N)
46 (A)
Intelligent
6500
3400
180
148
55 (O)
38 (O)
Wise
11000
9200
180
199
30 (Cs)
28 (Cs)
Independent
3300
21000
80
145
30 (O)
26 (H)
Sarcastic
4700
46
1.8
0.7
30 (A)
23 (Cs)
Religious
5100
442
170
64
31 (A)
20 (Em)
Cruel
2200
455
14
20
40 (A)
25 (Em)
Google books ngram frequency (%) × 10(−7) for year 2000 Twitter frequency for March 2014 a
Primary factor loading from Saucier and Goldberg (1996b)
b
Primary factor loading from Lee and Ashton (2008)
A Agreeableness, Cs Conscientiousness, Ex Extraversion, O Openness, H Honesty-humility, Em Emotionality, N Neuroticism
Table 2 Correlation between primary factor loading and usage frequency of English, German, and French personality adjectives in book texts and in Twitter tweets Five factor modela Qualified noun:
Woman
Person
Person
Google Books corpus
0.03
0.08
0.00
Twitter
0.02
0.17
Six factor modelb Qualified noun:
Woman
Person
Homme
Femme
Google books corpus
−0.02
0.03
−0.12
−0.02
Twitter
−0.03
0.01
−0.06
−0.02
a
Primary factor loading of 295 English adjectives from Saucier and Goldberg (1996b), 100 German adjectives from Ostendorf (1990) b
Primary factor loading of 295 English adjectives from Lee and Ashton (2008), 116 French adjectives from Boies et al. (2001) Google books Ngram frequency for year 2000 Twitter frequency for March 2014
Integr Psych Behav
Table 3 Word popularity and communality in the six-factor model, 295 English words
Mean communality calculated from data from Lee and Ashton (2008)
Word popularity ranking
Mean communality
1–50
0.274
51–100
0.294
101–150
0.251
151–200
0.220
201–250
0.242
251–295
0.230
friendliness (composed of extraversion and agreeableness), as being just as important as Bpure^ traits, such as agreeableness or extraversion because the frequency of the use of Bpure^ and Bmixed^ concepts is roughly the same. Several researchers, including Eysenck (1992) and Block (1995) have criticized the five factor model for its lack of theory (Uher 2014b). No consensus exists on the ontological nature of the five factors: are they basic ingredients or elements of personality or Bpolyglot generic arenas with fuzzy, overlapping boundaries?^ (McAdams 1992). While Goldberg saw the Big five model as a descriptive model that did not address issues of causality or mechanisms underlying behavior (Saucier and Goldberg 1996a, b), McCrae and Costa (1999) viewed the five factors as causal dispositions that were partly derived from biological factors, such as brain structure and neurochemistry. The six HEXACO factors Bcan be readily interpreted in terms of constructs from theoretical biology^ according to Ashton and Lee (2007). Other theories view the five factors from a more social psychological viewpoint and conceptualize them as relational constructs (Hogan 1996). Traits serve interpersonal functions,
Table 4 Correlation between word usage frequency across corpora and qualified nouns Google books person
Google books Twitter woman person
Google books homme
Google books femme
Twitter homme
Google books person Google books woman
0.74
Twitter person
0.38
0.36
Twitter woman
0.40
0.68
0.52
Google books homme Google books femme
0.61
Twitter homme
0.72
0.39
Twitter femme
0.20
0.52
Note. 295 English adjectives from Saucier and Goldberg (1996b) 114 French adjectives from Boies et al. (2001) Google books Ngram frequency for year 2000 Twitter frequency for March 2014
0.28
Integr Psych Behav
and we perceive individual differences that are important in interaction with others. Thus, the Big five is essentially a model of what people want to know about one another (Srivastava 2010). The results of the present study are compatible with three hypotheses: 1) laymen are not very good at describing personality, 2) the five (or six) factors are psycho-lexical artefacts associated with the factor analytic method, and 3) the five (or six) factors are causal and functional but their interaction is not well understood. The history of science is full of examples where firmly established popular concepts turn out to be erroneous. In personality psychology, popular ideas concerning the correlation between physique and temperament, or that between personality and birth month have been proved wrong. Scientific progress can be based on new data not accessible to laymen and everyday observation, or otherwise on data analysis that is more complex than are layman methods. In the case of the five factor theory, the superiority of the scientific model, compared to layman descriptions of personality, is supposedly based on data analysis methods. Concerning basic variables, traits are well known to laymen. However, the factor marker adjectives have not been identified as core concepts of the personality lexicon or as better descriptors than other adjectives. In a way, popular vote and scientific analysis seem to be at odds on the question of the most useful personality descriptors. The fact that the correlation between factor loading and usage frequency of personality terms seems low across nations suggests that the problem lies with the theory. A contrasting example comes from information technology concepts that quickly cross borders and become part of everyday language because of their usefulness. It is also difficult to imagine that names for broad categories of any important everyday phenomena were missing from our vocabulary and factor analysis was needed to show that, for example, cats, dogs, and hamsters have something in common, as do cakes, biscuits, and pies. The fact that we do not have five highly popular words to describe the five factors of personality like we do for animals and pastry may reflect that the five factor taxonomy is not very descriptive or practical. In a critical review of quantitative methods in psychology, Toomela (2010, p.2) suggested that in much of modern psychological research, research questions are adjusted to the methods and not vice versa: BFor instance, a researcher may ask, how many factors emerge in the analysis of personality or intelligence test results. But why to look for the number of factors if personality or intelligence is studied ?^ According to Toomela, theory that looks for mechanisms should be clearly distinguished from pure descriptions of regularities in superficial observations. Lykken (1971) presented a critique of the factor analytic method in a paper on automobiles. He selected several dozen variables from a motoring magazine (e.g., acceleration, braking distance, and fuel consumption). Some of the resulting factors, such as Bbrake size and price^ made little sense and seemed unrelated to how cars actually work. Empirical studies that have questioned the functional significance of the five factors include a recent study by Franic et al. (2013) who used common and independent pathway model comparisons to test whether the five personality factors mediated the genetic and environmental effects on responses on test items of the popular five-factor tests. They found that, while the genetic and environmental covariance components displayed a 5-factor structure, the pathway modeling showed that they do not comply with the collinearity constraints entailed in the common pathway model. Another problematic issue comes from
Integr Psych Behav
Paunonen and Ashton (2001, 2013) who found that, in many cases, test scores of a few narrow facet scales are superior in predicting behavior compared to test scores of all Big Five factor scales combined. The five-factor theory is also challenged by its poor performance in studies in which confirmatory factor analysis has been applied to validate the factor model based on exploratory analyses (McCrae et al. 1996). The results of the present study are compatible with a model of five broad and causal factors only in the case that the interaction of the factors is more complicated than existing theories suggest. For example, personality tests based on the five factor model usually have x facets for each of the five factors, and the score for each factor is simply the sum of the facet scores. The maximum score on each scale is the same. First, it is improbable that the effects of 25 or 30 facet variables on behavior are equal. Second, the combined effect of two, three, or more factors may not simply be the sum of these factors. According to the Gestalt school of thought Bthere are wholes, the behavior of which is not determined by that of their individual elements, but where the part-processes are themselves determined by the intrinsic nature of the whole^ (Wertheimer 1938, p. 2). For example, age, sex, and height are three important variables in describing people. However, a person who is 170 cm tall may be described as a tall woman but a man of average height. The correlation between age and height is also complicated with extreme shortness seen in individuals of very young age but (supposedly) no correlation between these two variables in samples of psychology undergraduates. Understanding the nature of these interacting variables is necessary to explain such anomalies. If friendliness is simply a blend of agreeableness and extraversion and we have a personality test with valid scales for these constructs (F, A, and E), then is the F score equal to A+E/2, or xA+yE+c; does a nonlinear relationship exist? In practical terms, is a highly extroverted and moderately agreeable person equally as friendly as a very agreeable and moderately extroverted person? Finally, it can be concluded that personality tests developed to specifically measure the Big five factors of personality may not be optimal for the purpose of general personality assessment. It can be hypothesized that personality tests should include scales that measure traits represented by popular terms in everyday language independent of their standing in the five (or six) factor frame.
References Ames, D. R., & Bianchi, E. C. (2008). The agreeableness asymmetry in first impressions: perceivers’ impulse to (Mis)judge agreeableness and how it is moderated by power. Personality and Social Psychology Bulletin, 34(12), 1719–1736. Ashton, M. C., & Lee, K. (2007). Empirical, theoretical, and practical advantages of the HEXACO model of personality structure. Personality and Social Psychology Review, 11, 150–166. Backström, M., & Björklund, F. (2013). Social desirability in personality inventories: symptoms, diagnosis and prescribed cure. Scandinavian Journal of Psychology, 54(2), 152–159. Block, J. (1995). A contrarian view of the five-factor approach to personality description. Psychological Bulletin, 117, 187–229. Boies, K., Lee, K., Ashton, M., Pascal, S., & Nicol, A. M. (2001). The structure of the french personality lexicon. European Journal of Personality, 15, 277–295. Costa, P. T., & McCrae, R. R. (1995). Domains and facets: hierarchical personality assessment using the revised NEO personality inventory. Journal of Personality Assessment, 64, 21–50.
Integr Psych Behav Eysenck, H. J. (1992). Four ways five factors are not basic. Personality and Individual Differences, 13, 667– 673. Franic, S., Borsboom, D., Dolan, C. V., & Boomsma, D. I. (2013). The big five personality traits: psychological entities or statistical constructs? Behavior Genetics. doi:10.1007/s10519-013-9625-7. Goldberg, L. R. (1990). An alternative Bdescription of personality^: the big five factor structure. Journal of Personality and Social Psychology, 59, 1216–1229. Hogan, R. (1996). A socioanalytic perspective on the five-factor model. In J. S. Wiggins (Ed.), The five-factor model of personality: Theoretical perspectives (pp. 180–207). New York: Guilford Press. Larsen, R. J., & Buss, D. M. (2005). Personality psychology: Domains of knowledge about human nature (2nd ed.). New York: McGraw Hill. Lee, K., & Ashton, M. (2008). The HEXACO personality factors in the indigenous personality lexicons of English and 11 other languages. Journal of Personality, 76(5), 1001–1054. Leising, D., Scharloth, J., Lohse, O., & Wood, D. (2014). What types of terms do people use when describing an individual’s personality? Psychological Science. doi:10.1177/0956797614541285. Lykken, D. (1971). Multiple factor analysis and personality research. Journal of Experimental Personality Research, 5, 161–170. McAdams, D. P. (1992). The five factor model in personality: a critical appraisal. Journal of Personality, 60, 329–361. McCrae, R. R., Zonderman, A. B., Costa, P. T., Bond, M. H., & Paurnonen, S. (1996). Evaluating replicability of factors in the revised NEO personality inventory: confirmatory factor analysis versus procrustes rotation. Journal of Personality and Social Psychology, 70, 552–566. McCrae, R. R., & Costa, P. T. (1999). A five–factor theory of personality. In L. A. Pervin & O. P. John (Eds.), Handbook of personality (pp. 139–153). New York: Guilford. Michel, J. B., et al. (2011). Quantitative analysis of culture using millions of digitized books. Science, 331, 176–182. Mischel, W., & Shoda, Y. (1995). A cognitive-affective system theory of personality: reconceptualizing situations, dispositions, dynamics, and invariance in personality structure. Psychological Review, 102, 246–268. Norman, W.T. (1967). 2800 personality trait descriptors: normative operating characteristics for a university population. University of Michigan, Department of psychology. Ostendorf, F. (1990). Sprache und Persöhnlichkeitsstruktur. Regensburg: Roderer Verlag. Paunonen, S. V., & Ashton, M. C. (2001). Big five factors and facets and the prediction of behavior. Journal of Personality and Social Psychology, 81(3), 524–539. Paunonen, S. V., & Ashton, M. C. (2013). On the prediction of academic performance with personality traits: a replication study. Journal of Research in Personality, 47(6), 778–781. Pervin, L. A., Cervone, D., & John, O. P. (2005). Personality: Theory and research (9th ed.). Hoboken: Wiley. Roivainen, E. (2014). Personality adjectives in twitter tweets and in the Google books corpus. An analysis of the openness factor of personality. Current Psychology. doi:10.1007/s12144-014-9274-x. Roivainen, E. (2013). Frequency of the use of english personality adjectives: implications for personality theory. Journal of Research in Personality, 47(4), 417–420. Saucier, G., & Goldberg, L. R. (1996a). The language of personality: Lexical perspectives on the five factor model. In J. S. Wiggins (Ed.), The five-factor model of personality: Theoretical perspectives (pp. 21–50). New York: Guilford Press. Saucier, G., & Goldberg, L. (1996b). Evidence for the big five in analyses of familiar english personality adjectives. European Journal of Personality, 10, 61–77. Srivastava, S. (2010). The five-factor model describes the structure of social perceptions. Psychological Inquiry, 21, 69–75. Toomela, A. (2003). Relationships between personality structure, structure of word meaning, and cognitive ability: a study of cultural mechanisms of personality. Journal of Personality and Social Psychology, 85(4), 723–735. Toomela, A. (2010). Quantitative methods in psychology: inevitable and useless. Frontiers in Psychology, 1(29), 1–14. Topsy (2014). Topsy.com. Twitter search, monitoring & analytics.www.Topsy.com [access 03.04.2014]. Uher, J. (2014). Conceiving Bpersonality^: psychologists’ challenges and basic fundamentals of the transdisciplinary philosophy-of-science paradigm for research on individuals. Integrative Psychological and Behavioral Science, 49. Uher, J. (2014). Developing Bpersonality^ taxonomies: metatheoretical and methodological rationales underlying selection approaches, methods of data generation and reduction principles. Integrative Psychological and Behavioral Science, 49.
Integr Psych Behav Uher, J. (2014). Interpreting Bpersonality^ taxonomies: why previous models cannot capture individualspecific experiencing, behaviour, functioning and development. Major taxonomic tasks still lay ahead. Integrative Psychological and Behavioral Science, 49. Wertheimer, M. (1938). Gestalt theory. In W. D. Ellis (Ed.), A source book of Gestalt psychology. London: Routledge & Kegan Paul. Eka Roivainen received his Lic.A.Psych degree from the University of Tampere, Finland in 1995. He works at Verve Rehabilitation, Oulu, Finland, as an assessment psychologist. His research interests include personality theory, intelligence theory, and the validity of personality and intelligence tests.
The Big Five Factor Marker Adjectives Are Not Especially Popular Words. Are They Superior Descriptors? Eka Roivainen 1
# Springer Science+Business Media New York 2015
Abstract Vocabularies of natural languages evolve over time. Useful words become more popular and useless concepts disappear. In this study, the frequency of the use of 295 English, 100 German, and 114 French personality adjectives in book texts and Twitter messages as qualifiers of the words person, woman, homme, femme, and Person was studied. Word frequency data were compared to factor loadings from previous factor analytic studies on personality terms. The correlation between the popularity of an adjective and its highest primary loading in five- and six-factor models was low (−0.12 to 0.17). The Big five (six) marker adjectives were not more popular than Bblended^ adjectives that had moderate loadings on several factors. This finding implies that laymen consider Bblended^ adjectives as equally useful descriptors compared to adjectives that represent core features of the five (six) factors. These results are compatible with three hypotheses: 1) laymen are not good at describing personality, 2) the five (six) factors are artifacts of research methods, 3) the interaction of the five (six) factors is not well understood Keywords Personality theory . Personality descriptors . Word frequency
Introduction Travel guides often include a brief vocabulary of the language spoken in the destination country. For example, the words wine and water might be included in the restaurant vocabulary, and the words gasoline and mechanic in the motoring vocabulary. If a guidebook were to contain a short personality vocabulary, which words should be included? Personality can be defined as Bthose characteristics that account for a
* Eka Roivainen [email protected] 1
Verve Rehabilitation, PL 404, 90100 Oulu, Finland
Integr Psych Behav
person’s consistent patterns of feeling, thinking, and behaving^ (Pervin et al. 2005 p.6). The prevailing five-factor model of personality (Goldberg 1990; McCrae and Costa 1999) implies that the top five words of the personality vocabulary list should describe the five broad factors of personality that have emerged from factor analyses of the multitude of expressions used to describe personality. These five factors seem to account for most of the variation in human personality, and they have been labeled by adjectives with high primary loadings on the factors. A number of standardized tests (e.g., NEO-PI; Costa and McCrae 1995) have been devised for professional assessment and description of personality based on the five-factor model. Thus, the tourist vocabulary might include translations for the big five marker adjectives: open-minded, conscientious, extroverted, agreeable, and neurotic. The restaurant and motoring vocabularies are not based on factor analysis; rather, they are based on their assumed usefulness and frequency of the use. The label adjectives of the Big five personality scales are common words; however, they are not among the most popular personality adjectives. If the popularity criterion were used to choose five personality descriptors, they might include words such as nice, happy, friendly, smart, and confident, which are among the most prevalent concepts in openended personality descriptions in experimental settings (Ames and Bianchi 2008) and on the internet (Roivainen 2013). For example, during week 13/2014, the Big five marker adjectives listed above were used 167 times as qualifiers of the word person in Twitter tweets, while the frequency for friendly person was 1,700 (Topsy 2014). However, in Saucier and Goldberg’s (1996a, b) five-factor model, friendly does not have a very high loading on any of the Big five factors. In this framework, friendliness is a Bblended trait^ or a mixture of extraversion (a loading of 0.39 on this factor) and agreeableness (0.37). One of the great challenges for personality science has been the cross-situational variation of personality (Mischel and Shoda 1995; Uher 2014a). Textbook definitions of personality often refer to Brelatively enduring^ (Larsen and Buss 2005) or Bconsistent patterns^ (Pervin et al. 2005) of thought, emotion, and behavior. For example, a student may be shy in the classroom and extroverted during recess. To assess the general and typical ways of behaving and thinking, personality scientists have traditionally resorted to questionnaires (Bdo you/does your friend usually behave in a shy manner^). Obviously, this method has several drawbacks, including problems caused by the subjectivity of the data (Uher 2014c). Socially desirable answering is also a well-known problem in personality assessment (Backström and Björklund 2013) and, at a more theoretical level, concepts such as shy do not have a fixed, objective, and stable definitions (Toomela 2003). The ability to understand and predict the behavior of other people is a basic human skill. Languages that lack concepts such as smart, dumb, nice, and mean are unthinkable; societies with completely invalid models of human personality cannot be functional. While laymen may know little about neurochemistry or astronomy, they have a certain competence in assessing personality. We can assume that words that are useful in describing behavior and personality will progressively become more popular, while words that are less useful will disappear. The problem of generalizability of results from students or any other experimental samples to the general population may be overcome by studying the use of personality terms in large online corpora, such as the Google books database. Leising et al. (2014)
Integr Psych Behav
found that participants in a student sample and in a corpus of online communications used terms rated as important in describing person characteristics more often. It may be hypothesized that if the variation in personality is explained by five underlying factors that are well described by words appearing in natural languages, those words, Bfactor markers^ that have high loadings on only one factor should be more popular than blended concepts, such as friendly, which have moderate loadings on two or more factors. Words that represent the core features of the factors should be more powerful descriptors than words that fall between the factors. In a similar fashion, a subtest of an IQ test that has a high loading (e.g., perceptual organization factor [POI]) is a better predictor of scores in the other subtests of the POI than a subtest that has a lower loading. In a recent study, Roivainen (2013) found that the correlation between the frequency of the use of a personality adjective on the internet and its highest loading on one of the five factors in Saucier and Goldberg (1996a, b) model was fairly low (0.12). Roivainen examined bigrams composed of a personality adjective and the word person. The correlation between word frequency on the internet and in the Google books database was 0.58. In another study (Roivainen 2014), the correlation between usage frequency for bigrams composed of a personal adjective describing openness + person and personal adjective + woman was 0.60 in the Google books corpus. Thus, some variation exists in the usage frequency of personality adjectives across corpora and qualified nouns. In the present study, the correlation between the frequency of use and factor loadings of personality adjectives in five (Saucier and Goldberg 1996a, b; Ostendorf 1990) and six factor models (Lee and Ashton 2008; Boies et al. 2001) was examined using words from three different languages, English, French, and German, with the following qualified nouns: woman, person, homme, femme, Person. It was hypothesized that primary factor loading and communality are positively correlated with word popularity in book texts and Twitter messages.
Method The adjectives and their factor loadings were taken from four previous studies. Saucier and Goldberg (1996a, b) five-factor model is based on the factor analysis of correlations of 432 English adjectives selected from Norman’s (1967) list of 2,797 terms. Lee and Ashton’s (2008) six-factor Hexaco model is based on the analysis of 449 personality adjectives selected from Goldberg’s (1990) list of 1,711 adjectives. There were 300 adjectives, which were used in both the Goldberg and Saucier and Lee and Ashton studies, and these adjectives were examined in the present study. Ostendorf (1990) proposed a five-factor model based on the analysis of 430 rating scales composed of German personality adjectives. In the present study, the 50 highest loading German adjectives (mean loading = 0.66) and 50 adjectives with moderate loadings (mean loading = 0.40), ten for each factor, were compared. Boies and others (2001) analyzed the French personality lexicon using 388 popular personality adjectives and concluded that the data best fit a six-factor model. In the present study, the 10 highest loading French adjectives for five factors and seven highest loading adjectives for the openness factor, in all 57 adjectives (mean loading = 0.55), were compared to 57 adjectives with moderate loadings (mean loading = 0.36).
Integr Psych Behav
The frequency of the use of English personality adjectives was examined using bigrams composed of an adjective and the word person or woman: kind woman, kind person. The French bigrams were composed of an inflected adjective and the word homme or femme: homme doux, femme douce. The German bigrams were composed of a personality adjective and the word Person. The weak inflexion form of the adjective that follows both the masculine and feminine definite articles and that ends with -e for all cases was used (e.g., intelligente Person.) Based on preliminary analyses, the English adjectives natural, responsible, rational, reasonable and sexy were eliminated from final analyses because of poor correlations of usage frequency as qualifiers for person and woman and across corpora. The first four of these often appear in legal texts (natural person, responsible person) and do not necessarily refer to personality. The Google books database is based on over five million books and documents and is the largest existing text corpus (Michel et al. 2011). Usage frequency of the Ngrams for the year 2000 were searched with 3 years of smoothing. The American English, German, and French databases were searched. Topsy (2014) is a search engine for the social website Twitter. The Topsy index includes billions Twitter messages known as Tweets. The frequency of English and French bigrams described above was searched among Tweets from March 2014. The rationale for using two different databases, the Google books and the Twitter Topsy database was that they represent different types of texts. The Google books database includes all kinds of written texts, mainly books, and we may assume that the written language of the texts in the Google books database is more formal than everyday speech is. The Twitter tweets are more speech-like, being interactive communication, they are short and not necessarily grammatically correct.
Results Table 1 shows the factor loadings and usage frequency data for selected English adjectives. As Table 2 shows, the correlation between word popularity and factor loading was low for all conditions (−0.12 to 0.17). Table 3 shows that the six-factors of the Hexaco model combined account for 27–29 % of the total variance for the traits represented by the 100 most popular words and 23–24 % of the variance for the traits represented by the 95 least popular words. In comparison, the communalities for the marker adjectives sympathetic (Agreeableness), withdrawn (Extraversion) and organized (Conscientiousness) were 0.32, 0.45 and 0.44, respectively. Table 4 shows that correlations between word usage frequency across databases and qualified nouns ranged from 0.20 (frequency of adjectives qualifying homme in Twitter x frequency as qualifier of femme in Google books) to 0.74 (Google books person x Google books woman).
Discussion The results of the present study imply that the implicit layman theory of personality that underlies personality vocabularies seems to consider Bblended^ traits, such as
Integr Psych Behav Table 1 Primary factor loadings and the frequency of the use of selected English personality adjectives as qualifiers of person and woman Twitter
Book
Book
a
b
messages
messages
texts
texts
factor
factor
…person
…woman
…person
…woman
loading
loading
Jealous
39000
6000
32
29
47 (N)
29 (H)
Smart
38000
8000
32
59
49 (O)
37 (O)
Loyal
31000
10000
11
6
43 (A)
43 (H)
Shy
20000
170
64
16
65 (Ex)
64 (Ex)
Simple
16000
3000
28
29
45 (O)
35 (O)
Sympathetic
561
14
18
10
62 (A)
45 (Em)
Withdrawn
6
0
5
2
67 (Ex)
63 (Ex)
Five
Six
Organized
849
25
24
6
65 (Cs)
64 (C)
Moody
4100
214
21
9
53 (N)
46 (A)
Intelligent
6500
3400
180
148
55 (O)
38 (O)
Wise
11000
9200
180
199
30 (Cs)
28 (Cs)
Independent
3300
21000
80
145
30 (O)
26 (H)
Sarcastic
4700
46
1.8
0.7
30 (A)
23 (Cs)
Religious
5100
442
170
64
31 (A)
20 (Em)
Cruel
2200
455
14
20
40 (A)
25 (Em)
Google books ngram frequency (%) × 10(−7) for year 2000 Twitter frequency for March 2014 a
Primary factor loading from Saucier and Goldberg (1996b)
b
Primary factor loading from Lee and Ashton (2008)
A Agreeableness, Cs Conscientiousness, Ex Extraversion, O Openness, H Honesty-humility, Em Emotionality, N Neuroticism
Table 2 Correlation between primary factor loading and usage frequency of English, German, and French personality adjectives in book texts and in Twitter tweets Five factor modela Qualified noun:
Woman
Person
Person
Google Books corpus
0.03
0.08
0.00
0.02
0.17
Six factor modelb Qualified noun:
Woman
Person
Homme
Femme
Google books corpus
−0.02
0.03
−0.12
−0.02
−0.03
0.01
−0.06
−0.02
a
Primary factor loading of 295 English adjectives from Saucier and Goldberg (1996b), 100 German adjectives from Ostendorf (1990) b
Primary factor loading of 295 English adjectives from Lee and Ashton (2008), 116 French adjectives from Boies et al. (2001) Google books Ngram frequency for year 2000 Twitter frequency for March 2014
Integr Psych Behav
Table 3 Word popularity and communality in the six-factor model, 295 English words
Mean communality calculated from data from Lee and Ashton (2008)
Word popularity ranking
Mean communality
1–50
0.274
51–100
0.294
101–150
0.251
151–200
0.220
201–250
0.242
251–295
0.230
friendliness (composed of extraversion and agreeableness), as being just as important as Bpure^ traits, such as agreeableness or extraversion because the frequency of the use of Bpure^ and Bmixed^ concepts is roughly the same. Several researchers, including Eysenck (1992) and Block (1995) have criticized the five factor model for its lack of theory (Uher 2014b). No consensus exists on the ontological nature of the five factors: are they basic ingredients or elements of personality or Bpolyglot generic arenas with fuzzy, overlapping boundaries?^ (McAdams 1992). While Goldberg saw the Big five model as a descriptive model that did not address issues of causality or mechanisms underlying behavior (Saucier and Goldberg 1996a, b), McCrae and Costa (1999) viewed the five factors as causal dispositions that were partly derived from biological factors, such as brain structure and neurochemistry. The six HEXACO factors Bcan be readily interpreted in terms of constructs from theoretical biology^ according to Ashton and Lee (2007). Other theories view the five factors from a more social psychological viewpoint and conceptualize them as relational constructs (Hogan 1996). Traits serve interpersonal functions,
Table 4 Correlation between word usage frequency across corpora and qualified nouns Google books person
Google books Twitter woman person
Google books homme
Google books femme
Twitter homme
Google books person Google books woman
0.74
Twitter person
0.38
0.36
Twitter woman
0.40
0.68
0.52
Google books homme Google books femme
0.61
Twitter homme
0.72
0.39
Twitter femme
0.20
0.52
Note. 295 English adjectives from Saucier and Goldberg (1996b) 114 French adjectives from Boies et al. (2001) Google books Ngram frequency for year 2000 Twitter frequency for March 2014
0.28
Integr Psych Behav
and we perceive individual differences that are important in interaction with others. Thus, the Big five is essentially a model of what people want to know about one another (Srivastava 2010). The results of the present study are compatible with three hypotheses: 1) laymen are not very good at describing personality, 2) the five (or six) factors are psycho-lexical artefacts associated with the factor analytic method, and 3) the five (or six) factors are causal and functional but their interaction is not well understood. The history of science is full of examples where firmly established popular concepts turn out to be erroneous. In personality psychology, popular ideas concerning the correlation between physique and temperament, or that between personality and birth month have been proved wrong. Scientific progress can be based on new data not accessible to laymen and everyday observation, or otherwise on data analysis that is more complex than are layman methods. In the case of the five factor theory, the superiority of the scientific model, compared to layman descriptions of personality, is supposedly based on data analysis methods. Concerning basic variables, traits are well known to laymen. However, the factor marker adjectives have not been identified as core concepts of the personality lexicon or as better descriptors than other adjectives. In a way, popular vote and scientific analysis seem to be at odds on the question of the most useful personality descriptors. The fact that the correlation between factor loading and usage frequency of personality terms seems low across nations suggests that the problem lies with the theory. A contrasting example comes from information technology concepts that quickly cross borders and become part of everyday language because of their usefulness. It is also difficult to imagine that names for broad categories of any important everyday phenomena were missing from our vocabulary and factor analysis was needed to show that, for example, cats, dogs, and hamsters have something in common, as do cakes, biscuits, and pies. The fact that we do not have five highly popular words to describe the five factors of personality like we do for animals and pastry may reflect that the five factor taxonomy is not very descriptive or practical. In a critical review of quantitative methods in psychology, Toomela (2010, p.2) suggested that in much of modern psychological research, research questions are adjusted to the methods and not vice versa: BFor instance, a researcher may ask, how many factors emerge in the analysis of personality or intelligence test results. But why to look for the number of factors if personality or intelligence is studied ?^ According to Toomela, theory that looks for mechanisms should be clearly distinguished from pure descriptions of regularities in superficial observations. Lykken (1971) presented a critique of the factor analytic method in a paper on automobiles. He selected several dozen variables from a motoring magazine (e.g., acceleration, braking distance, and fuel consumption). Some of the resulting factors, such as Bbrake size and price^ made little sense and seemed unrelated to how cars actually work. Empirical studies that have questioned the functional significance of the five factors include a recent study by Franic et al. (2013) who used common and independent pathway model comparisons to test whether the five personality factors mediated the genetic and environmental effects on responses on test items of the popular five-factor tests. They found that, while the genetic and environmental covariance components displayed a 5-factor structure, the pathway modeling showed that they do not comply with the collinearity constraints entailed in the common pathway model. Another problematic issue comes from
Integr Psych Behav
Paunonen and Ashton (2001, 2013) who found that, in many cases, test scores of a few narrow facet scales are superior in predicting behavior compared to test scores of all Big Five factor scales combined. The five-factor theory is also challenged by its poor performance in studies in which confirmatory factor analysis has been applied to validate the factor model based on exploratory analyses (McCrae et al. 1996). The results of the present study are compatible with a model of five broad and causal factors only in the case that the interaction of the factors is more complicated than existing theories suggest. For example, personality tests based on the five factor model usually have x facets for each of the five factors, and the score for each factor is simply the sum of the facet scores. The maximum score on each scale is the same. First, it is improbable that the effects of 25 or 30 facet variables on behavior are equal. Second, the combined effect of two, three, or more factors may not simply be the sum of these factors. According to the Gestalt school of thought Bthere are wholes, the behavior of which is not determined by that of their individual elements, but where the part-processes are themselves determined by the intrinsic nature of the whole^ (Wertheimer 1938, p. 2). For example, age, sex, and height are three important variables in describing people. However, a person who is 170 cm tall may be described as a tall woman but a man of average height. The correlation between age and height is also complicated with extreme shortness seen in individuals of very young age but (supposedly) no correlation between these two variables in samples of psychology undergraduates. Understanding the nature of these interacting variables is necessary to explain such anomalies. If friendliness is simply a blend of agreeableness and extraversion and we have a personality test with valid scales for these constructs (F, A, and E), then is the F score equal to A+E/2, or xA+yE+c; does a nonlinear relationship exist? In practical terms, is a highly extroverted and moderately agreeable person equally as friendly as a very agreeable and moderately extroverted person? Finally, it can be concluded that personality tests developed to specifically measure the Big five factors of personality may not be optimal for the purpose of general personality assessment. It can be hypothesized that personality tests should include scales that measure traits represented by popular terms in everyday language independent of their standing in the five (or six) factor frame.
References Ames, D. R., & Bianchi, E. C. (2008). The agreeableness asymmetry in first impressions: perceivers’ impulse to (Mis)judge agreeableness and how it is moderated by power. Personality and Social Psychology Bulletin, 34(12), 1719–1736. Ashton, M. C., & Lee, K. (2007). Empirical, theoretical, and practical advantages of the HEXACO model of personality structure. Personality and Social Psychology Review, 11, 150–166. Backström, M., & Björklund, F. (2013). Social desirability in personality inventories: symptoms, diagnosis and prescribed cure. Scandinavian Journal of Psychology, 54(2), 152–159. Block, J. (1995). A contrarian view of the five-factor approach to personality description. Psychological Bulletin, 117, 187–229. Boies, K., Lee, K., Ashton, M., Pascal, S., & Nicol, A. M. (2001). The structure of the french personality lexicon. European Journal of Personality, 15, 277–295. Costa, P. T., & McCrae, R. R. (1995). Domains and facets: hierarchical personality assessment using the revised NEO personality inventory. Journal of Personality Assessment, 64, 21–50.
Integr Psych Behav Eysenck, H. J. (1992). Four ways five factors are not basic. Personality and Individual Differences, 13, 667– 673. Franic, S., Borsboom, D., Dolan, C. V., & Boomsma, D. I. (2013). The big five personality traits: psychological entities or statistical constructs? Behavior Genetics. doi:10.1007/s10519-013-9625-7. Goldberg, L. R. (1990). An alternative Bdescription of personality^: the big five factor structure. Journal of Personality and Social Psychology, 59, 1216–1229. Hogan, R. (1996). A socioanalytic perspective on the five-factor model. In J. S. Wiggins (Ed.), The five-factor model of personality: Theoretical perspectives (pp. 180–207). New York: Guilford Press. Larsen, R. J., & Buss, D. M. (2005). Personality psychology: Domains of knowledge about human nature (2nd ed.). New York: McGraw Hill. Lee, K., & Ashton, M. (2008). The HEXACO personality factors in the indigenous personality lexicons of English and 11 other languages. Journal of Personality, 76(5), 1001–1054. Leising, D., Scharloth, J., Lohse, O., & Wood, D. (2014). What types of terms do people use when describing an individual’s personality? Psychological Science. doi:10.1177/0956797614541285. Lykken, D. (1971). Multiple factor analysis and personality research. Journal of Experimental Personality Research, 5, 161–170. McAdams, D. P. (1992). The five factor model in personality: a critical appraisal. Journal of Personality, 60, 329–361. McCrae, R. R., Zonderman, A. B., Costa, P. T., Bond, M. H., & Paurnonen, S. (1996). Evaluating replicability of factors in the revised NEO personality inventory: confirmatory factor analysis versus procrustes rotation. Journal of Personality and Social Psychology, 70, 552–566. McCrae, R. R., & Costa, P. T. (1999). A five–factor theory of personality. In L. A. Pervin & O. P. John (Eds.), Handbook of personality (pp. 139–153). New York: Guilford. Michel, J. B., et al. (2011). Quantitative analysis of culture using millions of digitized books. Science, 331, 176–182. Mischel, W., & Shoda, Y. (1995). A cognitive-affective system theory of personality: reconceptualizing situations, dispositions, dynamics, and invariance in personality structure. Psychological Review, 102, 246–268. Norman, W.T. (1967). 2800 personality trait descriptors: normative operating characteristics for a university population. University of Michigan, Department of psychology. Ostendorf, F. (1990). Sprache und Persöhnlichkeitsstruktur. Regensburg: Roderer Verlag. Paunonen, S. V., & Ashton, M. C. (2001). Big five factors and facets and the prediction of behavior. Journal of Personality and Social Psychology, 81(3), 524–539. Paunonen, S. V., & Ashton, M. C. (2013). On the prediction of academic performance with personality traits: a replication study. Journal of Research in Personality, 47(6), 778–781. Pervin, L. A., Cervone, D., & John, O. P. (2005). Personality: Theory and research (9th ed.). Hoboken: Wiley. Roivainen, E. (2014). Personality adjectives in twitter tweets and in the Google books corpus. An analysis of the openness factor of personality. Current Psychology. doi:10.1007/s12144-014-9274-x. Roivainen, E. (2013). Frequency of the use of english personality adjectives: implications for personality theory. Journal of Research in Personality, 47(4), 417–420. Saucier, G., & Goldberg, L. R. (1996a). The language of personality: Lexical perspectives on the five factor model. In J. S. Wiggins (Ed.), The five-factor model of personality: Theoretical perspectives (pp. 21–50). New York: Guilford Press. Saucier, G., & Goldberg, L. (1996b). Evidence for the big five in analyses of familiar english personality adjectives. European Journal of Personality, 10, 61–77. Srivastava, S. (2010). The five-factor model describes the structure of social perceptions. Psychological Inquiry, 21, 69–75. Toomela, A. (2003). Relationships between personality structure, structure of word meaning, and cognitive ability: a study of cultural mechanisms of personality. Journal of Personality and Social Psychology, 85(4), 723–735. Toomela, A. (2010). Quantitative methods in psychology: inevitable and useless. Frontiers in Psychology, 1(29), 1–14. Topsy (2014). Topsy.com. Twitter search, monitoring & analytics.www.Topsy.com [access 03.04.2014]. Uher, J. (2014). Conceiving Bpersonality^: psychologists’ challenges and basic fundamentals of the transdisciplinary philosophy-of-science paradigm for research on individuals. Integrative Psychological and Behavioral Science, 49. Uher, J. (2014). Developing Bpersonality^ taxonomies: metatheoretical and methodological rationales underlying selection approaches, methods of data generation and reduction principles. Integrative Psychological and Behavioral Science, 49.
Integr Psych Behav Uher, J. (2014). Interpreting Bpersonality^ taxonomies: why previous models cannot capture individualspecific experiencing, behaviour, functioning and development. Major taxonomic tasks still lay ahead. Integrative Psychological and Behavioral Science, 49. Wertheimer, M. (1938). Gestalt theory. In W. D. Ellis (Ed.), A source book of Gestalt psychology. London: Routledge & Kegan Paul. Eka Roivainen received his Lic.A.Psych degree from the University of Tampere, Finland in 1995. He works at Verve Rehabilitation, Oulu, Finland, as an assessment psychologist. His research interests include personality theory, intelligence theory, and the validity of personality and intelligence tests.
