Personality Assessment in Groups of Different Verbal Intelligence Levels
Funding: This research received a non-commercial University of Zagreb grant no. (11-933-1005) awarded to the second author.
ABSTRACT
The differentiation of personality by intelligence hypothesis, which has mixed support in the literature, predicts that personality is more variable for more intelligent individuals. This study aimed to test that hypothesis by comparing variances, reliability coefficients, and inter-scale correlations of personality as assessed by self-reports between groups of participants with different intelligence scores. We used two independent datasets (N1 = 655; N2 = 836; Ntotal = 1491) in which the same vocabulary test was used as a measure of verbal intelligence, but personality was measured as self-report by different inventories (NEO-FFI and HEXACO-100). As the verbal ability scores had a normal distribution, the combined mean was calculated, and empirical groups were generated within each sample to compare groups of participants who indicated a low-ability group (G1) and the high-ability group (G3). Results mostly support the differentiation hypothesis in the NEO-FFI dataset, where participants in G3 have higher variances and reliability coefficients than participants in G1, but do not show lower inter-scale correlation coefficients. However, the same trend was not found in the second sample where personality was assessed by the HEXACO inventory. In conclusion, the results of this study provide only partial support for the personality differentiation by intelligence hypothesis.
Summary
- Testing differentiation of personality by intelligence hypothesis in two samples.
- Low- and high-verbal ability groups were compared on NEO-FFI and HEXACO-PI-R.
- Testing differences in personality scale variances, reliability, and inter-scale correlations.
- Results partially support the differentiation hypothesis only in one sample where high-verbal ability group achieved more variable and more reliable results on the NEO-FFI than individuals with lower abilities.
1 Introduction
Personality and intelligence represent two core domains of individual differences. Personality reflects how individuals tend to think, feel, and behave across different situations, so its assessments should represent a measure of what a person typically and most commonly does (Larsen and Buss, 2021). However, in self-report measures of personality, the actual score also reflects the self-perception and self-presentation of the individual who is the object of measurement. On the other hand, intelligence is defined as the ability to solve problems, reason, make good decisions, learn from experience, adapt, etc. (Gottfredson 1997). Therefore, it is usually measured via tests that include tasks designed to assess what a person can do, that is, by maximum performance tests (Cronbach 1949). In this study, personality traits are self-report based, while intelligence was assessed using a maximum performance test.
There are different theoretical views about the relationship between personality and intelligence. One view states that personality and intelligence are essentially unrelated. For example, more than a century ago, Webb (1915) found that intelligence and personality items load on separate factors, and Eysenck (1994) supported the idea of two independent dimensions. However, the results of various studies are more in line with the view that a certain relationship between intelligence and personality does exist. The most consistent finding is that the personality domain of openness/intellect moderately correlates with intelligence (DeYoung 2020; Stanek 2014), although it seems that it is more related to Cattell's (1963) factor of crystallized and less with fluid intelligence (Ackerman and Heggestad 1997; Ashton et al. 2000; DeYoung et al. 2014; Stanek 2014). Moreover, both personality and intelligence may have an influence on each other's measurement. For instance, (i) personality may influence the measurement of intelligence. This is mostly pronounced in some stressful settings, like selection situations or in situations involving time pressure. Namely, individuals with high neuroticism are more prone to experience test anxiety when their intelligence is measured (Ackerman and Heggestad 1997), which in turn may increase the possibility of making errors and decrease their IQ result. A recent large-scale meta-analysis (Anglim et al. 2022) has shown that there is a statistically significant negative correlation between neuroticism and intelligence, but that the effect size is small (r = −0.08).
On the other hand, (ii) it is also possible that intelligence moderates results of personality assessment. Although potential effects in this direction had been much less studied in the literature, there are some potential causal mechanisms which can explain why it is plausible. Firstly, more intelligent individuals may have better understanding of the item content in personality questionnaires (Austin et al. 2000). Personality questionnaires sometimes have complex sentences with terms that people do not often use in everyday life and/or are not understandable to everyone. Secondly, it is possible that more intelligent individuals may have a better insight into their typical behavior, feelings, or experiences. Thirdly, more intelligent individuals may have more accurate estimates of the typical points of the Likert scale, that is, they may better understand what “average” means and what is above or below the average, so they have better ability to compare their own behavior with individual differences in the population, which is a standard task in personality assessment. Lastly, one possible effect of intelligence on personality assessment is represented by the personality differentiation by intelligence hypothesis (Brand et al. 1994). This hypothesis states that personality is more differentiated within more intelligent individuals, that is, that there is a greater personality variability in groups with a higher level of intelligence compared to those with a below average level. Austin et al. (2000, p. 407) suggested that this effect may occur because “higher g allows an individual more choice or freedom in personality development, leading to the existence of more or more well-defined personality dimensions in higher ability groups”. Moreover, it is possible that personality items better match with personality of highly intelligent individuals, so they can find more meaning in the items and perceive them more coherently, which can lead to more extreme responding, larger standard deviations, and higher scale reliability (Austin et al. 2000).
This hypothesis has been of interest and has had mixed support in the literature (e.g., Schermer, Bratko and Bojić, 2020; Schermer, Krammer et al. 2020; Schermer and Furnham 2020). Some studies reported higher variances with the increase in intelligence for some (or most) personality scales that were measured. For example, Austin et al. (1997) found increases in variance with increased ability for the NEO-FFI neuroticism and openness scales in the sample of adult Scottish farmers. Harris et al. (2005) used Personality Research Form (PRF; Jackson 1989) as a personality marker and found higher variances in more intelligent individuals on 15 out of 20 personality scales. In a study with a large job applicant sample of more than 20,000 individuals, significant differences in variability were confirmed for three out of five NEO-PI-R scales, that is, openness to experience, neuroticism, and extraversion when more extreme groups were compared (De Fruyt et al. 2006).
On the other hand, some studies did not support the hypothesis. The most recent example is Schermer et al. (2024) study based on more than 6000 participants who applied for teaching programs at 12 university-level programs in Austria. In this study, personality was assessed using the German version of the Big Five Inventory (BFI; Lang et al. 2001), and the authors compared low versus high intelligence groups splitting at the median and the lowest versus highest 10th percentile. The results supported the hypothesis only for agreeableness and only for the median split, but not for the extreme groups split (where the results showed statistically significant difference, but in the opposite direction). Andersson et al. (2022) in the Sweden sample of older adults also reported that in three out of the five Mini IPIP's personality scales, the low-intelligence groups were somewhat more variable than high-intelligence groups, which is the opposite direction from expected. Similarly, in a German study with two different samples (Harris et al. 2006), although the differentiation hypothesis was mostly confirmed when employed adults were assessed, results using a student sample were in the opposite direction. They assessed personality with NEO-FFI and five PRF subscales, and in 8 out of 10 measured scales, slightly greater variability was found in the low-intelligence group.
As stated earlier, Austin et al. (2000) proposed that differences in variability can be confounded by the scale reliability, that is, more intelligent groups tend to have more reliable results than less intelligent groups because they are more consistent in responding. Moreover, assessments of individuals with lower ability are probably more under the influence of the error variance and “that is reflected in (a) less variability in the true scores of the trait measured, and (b) lower inter-item correlations, which in turn impact the factorial structure of the measures.” (Escorial et al. 2019, p. 332).
Findings of different studies support the idea about differentiation in reliability coefficients. For example, Mottus et al. (2007) reported higher alpha coefficients across the facet scales of EPIP-NEO (Estonian Personality Item Pool-NEO; Mottus et al. 2006), a linguistically simpler version of NEO-PI-R. Average alpha values in the high-ability group were 0.80, and 0.73 in the low-ability group. Results of Andersson et al. (2022) study also showed that the alpha coefficients for the Mini IPIP scales are consistently larger for the more intelligent group, with the biggest difference found for conscientiousness. Moreover, one study demonstrated a positive relationship between “person reliability” and general intelligence, and authors concluded that their results are in line with the idea that traits in more intelligent individuals are represented with greater strength and clarity (Navarro-Gonzalez et al. 2018).
Since the personality differentiation by intelligence hypothesis suggests that more intelligent individuals have more differentiated personalities (Brand et al. 1994), another possible consequence of this could be reflected in the personality structure, that is, in the inter-scale correlations of the personality scales. More specifically, it is hypothesized that as IQ increases, the number of personality dimensions also increases (Austin et al. 1997). Therefore, it is expected that the covariances between different personality traits might be lower when intelligence is higher. That can also be related to the idea that more intelligent people should have better quality of the data on personality questionnaires and lower response style variance. Austin et al. (2002), for example, found that the correlation between the neuroticism and psychoticism scales in Eysenck's EPQ decreased with increasing intelligence level. A study of Estonian adolescents also showed a tendency for decreasing correlations between the NEO-FFI personality dimensions with increasing cognitive ability, but also with increasing age (Allik et al. 2004). In addition, the results of the recent study by Schermer et al. (2024) showed that the inter-scale correlations for the BFI personality scales were lower in individuals with higher intelligence. On the other hand, although the results of the study by Mottus et al. (2007) showed a tendency towards lower inter-correlations in the higher ability group, these correlations mostly did not differ significantly between the different ability groups. Similar conclusions were also found in the study by Andersson et al. (2022), in which only 4 of the 10 covariance differences between different ability groups were statistically significant, and the largest difference was in the opposite direction than expected.
1.1 The Present Study
In the present study, we aimed to investigate the relationship between verbal intelligence and self-report-based personality trait levels. Our first goal was to test the differentiation of personality by intelligence hypothesis by comparing group variances. Although there are some inconsistent findings in the literature, we expect that our results would be in line with that hypothesis, that is, after forming two extreme groups based on the intelligence score distribution, we believe that the high-ability group (G3) will have significantly higher scale variances compared with the low-ability group (G1). Our second goal was to compare the reliability coefficients of personality assessment between two groups of participants with different intelligence scores. We hypothesized that higher reliability coefficients will be found in the G3 group compared to the G1 group. Finally, we also aimed to assess inter-scale correlation coefficients between five NEO/six HEXACO scale scores in the two aforementioned groups. If the higher-ability groups provide better data in general and have more differentiated personality, we hypothesized that lower inter-scale correlations (i.e., low response style variance) would be found in the G3 group compared to the G1 group.
Since past studies used various personality measures, differences in support of the differentiation hypothesis could (at least in part) be influenced by personality operationalization. It is plausible that there are some personality measures in which differences between high- and low-ability groups would be less emphasized. Also, to our knowledge, there is no study in this field that used a measure of a newer personality model—HEXACO (Ashton and Lee 2001), which operationalizes personality factors somewhat differently than the most popular five-factor personality models. Therefore, in the present study, we aimed to conduct analyses on two independent samples, who completed the same intelligence test but different personality questionnaires, NEO-FFI, and HEXACO inventory. Most previous studies also focused on general mental ability when testing the differentiation of personality by intelligence hypothesis. To our knowledge, this is the first study to use verbal intelligence, a specific facet of general mental ability, as an indicator of intelligence.
2 Materials and Methods
2.1 Participants and Procedure
This study included two independent datasets. Both samples were part of two larger independent research projects focusing on twins. Sample 1 consists of 655 individuals, who were a part of a twin study of young adults (57% female; age range: 15–22 years). This sample is roughly representative of the twin population of Zagreb (capital of Croatia) and Zagreb County. Sample 2 consists of 836 individual twins from six birth cohorts (63% female; age range: 19–28 years). This sample is roughly representative of the Croatian twin population who finished high school. All participants filled in applied measures in paper-pencil format.
2.2 Measures
2.2.1 Personality—Sample 1
In sample 1, participants' self-report-based personality trait levels were assessed with the Croatian version of the NEO Five-Factor Inventory (NEO-FFI; Bratko et al. 2002; Costa and McCrae 1992). NEO-FFI is a 60-item version of NEO inventories which measures five broad personality traits: neuroticism, extraversion, openness to experience, agreeableness, and conscientiousness. The inventory has a five-point Likert scale for responding (1 = strongly disagree; 5 = strongly agree). Cronbach's alpha coefficients in this sample were 0.81, 0.73, 0.57, 0.66, and 0.81 for neuroticism, extraversion, openness, agreeableness, and conscientiousness scales, respectively. Corresponding coefficients in the validation study were 0.71, 0.62, 0.57, 0.60, and 0.79 for males and 0.78, 0.74, 0.56, 0.59, and 0.78 for female participants (Bratko et al. 2002).
2.2.2 Personality—Sample 2
For measuring self-report-based personality trait levels in sample 2, the Croatian version of the 100-item HEXACO Personality Inventory-Revised (Babarović and Šverko 2013; Lee and Ashton 2018) was used. This instrument measures six broad personality traits defined within the HEXACO personality model (Ashton and Lee 2001). Each of the six factors is represented with 16 items, accompanied by a five-point Likert-type scale (1 = strongly disagree; 5 = strongly agree). Cronbach's alpha coefficients for the present sample were 0.81, 0.83, 0.86, 0.81, 0.81, and 0.84 for the honesty-humility, emotionality, extraversion, agreeableness, conscientiousness, and openness to experience scales, respectively. Corresponding coefficients in the validation study were 0.83, 0.78, 0.82, 0.85, 0.83, and 0.83 (Babarović and Šverko 2013).
2.2.3 Intelligence—Samples 1 and 2
As a measure of intelligence, a Croatian adaptation of a verbal subtest of the General Aptitude Battery (Tarbuk, 1977, see Bratko et al. 2012) was used. This vocabulary test is highly saturated with the g factor and consists of 20 items where participants must determine antonyms or synonyms between four words. We used this same test as a marker of verbal intelligence in both of our samples. Cronbach's alpha coefficients were 0.84 and 0.81 in sample 1 and sample 2, respectively.
2.3 Statistical Analysis
Since our two samples differ in their socio-demographic characteristics, primarily in their education level, we calculated average IQ scores. Namely, sample 2 consisted of high-school graduates who applied for the national State Matura Exams, which are prerequisites for admission to university education in Croatia. Therefore, since all participants in sample 2 did graduate from high school and were applying for college, they have a higher education level than participants in sample 1. Results showed that average IQ scores for these two samples differed, with sample 1 scoring an average of 12.07 (SD = 4.39), and sample 2 scoring an average of 13.95 (SD = 3.75). This difference can be interpreted in terms of Cohen's d = 0.46, indicating a moderate effect size (Cohen 1992), or in other words—that sample 2 scores on average half a standard deviation higher compared to sample 1.
In order to compare two extreme groups of participants in both samples, we first calculated a common mean (M12) and a common standard deviation (SD12) for the combined sample (M12 = 13.12; SD12 = 4.15). This was done in order to keep the group boundaries equal across two samples (G1[0.00–11.04]; G2[11.05–15.19]; G3[15.20–20.00]), while making group sizes as large as possible (nG1-NEOFFI = 269, nG1-HEXACO = 201; nG2-NEOFFI = 221; nG2-HEXACO = 300; nG3-NEOFFI = 165, nG3-HEXACO = 335). Group 1 (G1) consists of participants who had the lowest scores (M12 − ½ SD12), while group 3 (G3) consists of participants who had the highest scores (M12 + ½ SD12). In other words, the three groups were defined by their scores into the following categories: 0–11 for G1, 12–15 for G2, and 16–20 for G3. For the purpose of the present study, we only aimed to compare extreme groups, G1 and G3, while the middle (average) group was not included in the following analyses.
Differences in variance were tested using Levene's homogeneity of variance test. For the reliability coefficients, we calculated Cronbach's alpha. Differences between alpha coefficients were tested using the chi-squared statistic in the cocron program (Diedenhofen and Musch 2016). Inter-scale correlations between the five NEO and six HEXACO personality scales for the low-ability group (G1) and the high-ability group (G3) were calculated in order to explore whether different IQ/ability groups were differing in their error variance or in their self-report response style variance, particularly social (un)desirability. By examining both the scale reliabilities and the inter-scale correlations, we wished to see whether the lower-ability groups showed lower or higher reliabilities, and also whether the lower-ability groups showed higher inter-scale correlations. If the higher-ability groups were providing better data in general, we would expect them to show high reliabilities (i.e., low error variance), but low inter-scale correlations (i.e., low response style variance).
3 Results
3.1 Descriptive Statistics and Group Differences
General descriptive statistics for the two whole samples are presented in the Supporting Information (Tables S1 and S2), while means, independent sample t-tests, and d-values for two extreme groups are shown in Table 1. As can be seen from Table 1, in sample 1 G3 scored significantly higher than G1 in one out of five measured traits (openness) with a moderate effect-size Cohen's d, while G1 scored significantly higher than G3 also in one out of five traits (extraversion), showing a small effect-size. Non-significant differences were found for neuroticism, agreeableness, and conscientiousness domains. In sample 2 there were no significant differences between groups for three out of six HEXACO personality factors (emotionality, agreeableness, and conscientiousness). Participants in G3 scored significantly higher than G1 in two out of six measured traits (honesty-humility and openness) with small effect-sizes Cohen (1992), while G1 scored significantly higher than G3 in one out of five traits (extraversion), showing a small effect-size. Non-significant differences were found for the emotionality, agreeableness, and conscientiousness domains.
| Variances | Means | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| k | G1 | G3 | F | p | G1 | G3 | t | p | d | |
| Sample 1 (NEO-FFI) | ||||||||||
| Neuroticism | 12 | 80.06 | 71.20 | 3.74 | 0.05 | 20.48 | 19.28 | 1.39 | 0.16 | 0.13 |
| Extraversion | 12 | 40.44 | 53.17 | 5.34 | 0.02 | 31.34 | 29.55 | 2.69 | 0.01 | 0.26 |
| Openness | 12 | 33.18 | 41.23 | 2.64 | 0.11 | 22.18 | 26.10 | −6.59 | 0.001 | 0.63 |
| Agreeableness | 12 | 41.36 | 40.47 | 0.00 | 0.99 | 30.27 | 30.36 | −0.15 | 0.88 | 0.01 |
| Conscientiousness | 12 | 47.75 | 75.07 | 8.05 | 0.01 | 32.01 | 30.98 | 1.29 | 0.20 | 0.12 |
| Sample 2 (HEXACO-PI-R) | ||||||||||
| Honesty-humility | 16 | 0.32 | 0.30 | 0.95 | 0.33 | 3.51 | 3.61 | −2.12 | 0.04 | 0.18 |
| Emotionality | 16 | 0.25 | 0.33 | 5.11 | 0.02 | 3.33 | 3.29 | 0.68 | 0.50 | 0.06 |
| Extraversion | 16 | 0.30 | 0.35 | 1.93 | 0.17 | 3.57 | 3.39 | 3.42 | 0.001 | 0.30 |
| Agreeableness | 16 | 0.25 | 0.26 | 0.23 | 0.63 | 2.96 | 2.92 | 0.90 | 0.37 | 0.07 |
| Conscientiousness | 16 | 0.25 | 0.24 | 1.20 | 0.27 | 3.55 | 3.62 | −1.63 | 0.11 | 0.14 |
| Openness | 16 | 0.45 | 0.34 | 5.60 | 0.02 | 3.24 | 3.45 | −3.89 | 0.001 | 0.34 |
- Note: k = number of scale items; F = Levene's F; t = independent sample t-test with df(G1) = 432 and df(G3) = 534.
3.2 Homogeneity of Variance Test
Variances of every measured trait in each sample, as well as the results of the Levene's F-test of homogeneity of variance, are presented in Table 1. As can be seen from the table, in sample 1, the G3 group is significantly more variable than the G1 group in two out of five measured traits (extraversion and conscientiousness), while the G1 group is more variable in one out of five traits (neuroticism). Non-significant differences were found for the openness and the agreeableness domains.
In sample 2 there are no significant differences in variability between groups for four out of six HEXACO personality factors. Participants in the G3 group are more variable only on the emotionality scale, while for the openness scale higher variance is presented in the G1 group.
3.3 Reliability
Table 2 represents reliability coefficients for every personality scale in each sample. In sample 1, all reliability coefficients are higher for the G3 than G1 group. We additionally tested these differences using the chi-squared statistic in the cocron program (Diedenhofen and Musch 2016). Results indicated that all observed differences were statistically significant, except for the neuroticism scale. In sample 2, all reliability coefficients were higher for the G3 than G1 group, except for the openness scale. However, results of the chi-squared statistic indicate that this difference was not significant. The only significant difference in sample 2 was shown for the emotionality scale.
| k | G1 | G3 | χ2(df = 1) | |
|---|---|---|---|---|
| Sample 1 (NEO-FFI) | ||||
| Neuroticism | 12 | 0.81 | 0.84 | 1.25, p = 0.26 |
| Extraversion | 12 | 0.67 | 0.79 | 8.40, p = 0.004 |
| Openness | 12 | 0.43 | 0.65 | 9.74, p = 0.002 |
| Agreeableness | 12 | 0.65 | 0.74 | 3.70, p = 0.05 |
| Conscientiousness | 12 | 0.76 | 0.89 | 23.99, p < 0.001 |
| Sample 2 (HEXACO-PI-R) | ||||
| Honesty-humility | 16 | 0.79 | 0.82 | 1.33, p = 0.25 |
| Emotionality | 16 | 0.76 | 0.85 | 12.55, p = 0.001 |
| Extraversion | 16 | 0.85 | 0.88 | 2.79, p = 0.09 |
| Agreeableness | 16 | 0.78 | 0.82 | 2.25, p = 0.13 |
| Conscientiousness | 16 | 0.80 | 0.81 | 0.15, p = 0.70 |
| Openness | 16 | 0.85 | 0.82 | 1.80, p = 0.18 |
- Note: k = number of scale items; χ2 = chi-squared statistic of the differences between Cronbach's alpha coefficients.
If higher-ability groups were providing better data in general, we would expect them to show high reliabilities (i.e., low error variance). The trend of the observed results shown in Table 2 is in line with this hypothesis, but the results of the chi-squared statistic confirm this hypothesis for 4 out of 5 NEO personality scales and only for one out of six HEXACO scales. We continue to examine both the scale reliabilities and the inter-scale correlations in order to see whether the higher-ability groups show lower inter-scale correlations.
3.4 Inter-Scale Correlations
Table 3 represents inter-scale correlations between the five NEO and six HEXACO personality scales. In sample 1, the low-ability group G1 (above diagonal) and the high-ability group G3 (below the diagonal) inter-scale correlations show a trend of low to moderate coefficients, with seven (G1) and eight (G3) statistically significant correlations (out of 10 inter-scale correlation coefficients). In sample 2, we observe a similar trend of mostly small to moderate coefficients, but here the low-ability group (G1) indicates six, and the high-ability group (G3) only three statistically significant correlations (out of 15 inter-scale correlation coefficients).
| (1) | (2) | (3) | (4) | (5) | (6) | (7) | (8) | (9) | (10) | (11) | (12) | |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Sample 1 (NEO-FFI) | ||||||||||||
|
−0.34** | 0.14* | −0.36** | −0.32** | 0.04 | |||||||
|
−0.37** | −0.06 | 0.03 | 0.24** | −0.26** | |||||||
|
−0.01 | −0.18* | −0.17** | 0.02 | 0.09 | |||||||
|
−0.26** | 0.19* | −0.20* | 0.30** | −0.09 | |||||||
|
−0.22** | 0.26** | −0.15 | 0.22** | 0.02 | |||||||
| Sample 2 (HEXACO-PI-R) | ||||||||||||
|
0.07 | 0.02 | 0.21** | 0.23** | −0.04 | 0.09 | ||||||
|
0.10 | −0.08 | −0.30** | −0.02 | −0.13 | 0.01 | ||||||
|
−0.01 | −0.07 | 0.21** | 0.31** | 0.09 | −0.06 | ||||||
|
0.30** | −0.10 | 0.04 | 0.11 | 0.04 | −0.01 | ||||||
|
0.16** | −0.02 | 0.22** | 0.07 | 0.25** | 0.00 | ||||||
|
0.07 | 0.05 | 0.08 | 0.05 | −0.02 | 0.11* | ||||||
|
−0.18** | −0.11 | −0.02 | 0.12 | 0.04 | 0.05 | 0.00 | 0.06 | 0.06 | 0.01 | 0.08 |
- Note: Correlation for the lowest third of the sample (G1) is reported above the diagonal for personality scales and in the bottom row for the vocabulary test, and correlations for the highest third of the sample (G3) are reported below the diagonal for personality scales and in the far most right column for the vocabulary test. N = Neuroticism; E = Extraversion; O = Openness; A = Agreeableness; C = Conscientiousness; HH = Honesty-Humility; Em = Emotionality; eX = Extraversion; VT = Vocabulary test.
- * Correlation is significant at the 0.05 level (two-tailed).
- ** Correlation is significant at the 0.01 level (two-tailed).
Bivariate correlation coefficients of the five NEO personality scales with verbal intelligence scores, as well as correlations of six HEXACO personality scales with verbal intelligence, are available for the whole samples (see Table S2 in Supporting Information) and separately for low- and high-ability groups in both samples (see Table 3). As can be seen in Table 3, most correlations are small and not statistically significant. The only exceptions are negative correlations for sample 1 between IQ-neuroticism in the low-ability group (rG1-NEOFFI = −0.18) and IQ-extraversion in the high-ability group (rG3-NEOFFI = −0.26), and positive correlations for sample 2 between IQ-openness in the high-ability group (rG3-HEXACO = 0.11).
4 Discussion
Results of the present study provide mixed support for the personality differentiation by intelligence hypothesis. Specifically, our hypotheses were mostly confirmed in sample 1, where personality was measured with NEO-FFI. Homogeneity of variance tests showed that the group of participants who scored higher on the intelligence test is indeed significantly more variable on some personality dimensions than the lowest lower-ability group. The biggest differences were established for extraversion and conscientiousness scales, while results were (marginally significant) in the opposite direction from expected for the neuroticism scale.
However, the hypothesis was not confirmed in sample 2, where HEXACO-100 was used as a personality measure. For most HEXACO scales, differences in variance between groups were not found, while results were in the opposite direction than expected for the openness scale. Interestingly, the only significant difference in the HEXACO sample consistent with the differentiation hypothesis was found for the emotionality scale.
Similar results as for variances were obtained for reliability differentiation. In sample 1, the hypothesis was mostly confirmed and in line with most previous studies (e.g., Allik et al. 2004; Andersson et al. 2022; Austin et al. 1997; Mottus et al. 2007; Navarro-Gonzalez et al. 2018), supporting the idea that intelligence can influence the psychometric properties of personality assessment. One of the biggest differences was found for the NEO-FFI openness scale, with the reliability coefficient for the low-ability group being quite small (α = 0.43). It seems that this NEO-FFI scale is the most problematic in terms of internal consistency when applied to participants with somewhat lower cognitive abilities, although it is important to note that estimates in the G3 group also did not show high reliability (α = 0.65). Once again, results in sample 2 did not support the idea that more intelligent individuals respond more reliably to personality items. Although reliability estimates are higher in G3 than in G1 for five HEXACO scales, these differences did not reach the level of statistical significance (except for emotionality). On the other hand, the G1 group shows a trend of higher reliabilities for the openness scale (even though this difference did not reach the level of statistical significance).
Some authors (e.g., Sorjonen et al. 2020) have suggested that lower scale reliability in the lower-abilities group of individuals could be caused by the difficulty of understanding reverse items, which are often found in most personality questionnaires. Andersson et al. (2022) results support this thesis since similar levels of scale reliability between two ability groups were established when they controlled for method variance related to reversed items. However, both personality inventories in our study contain reversed items in a similar percentage, so that possible confounding factor should not explain different results between samples 1 and 2.
As far as the inter-scale correlations of personality scales in low- and high-ability groups are concerned, we limit our discussion and conclusions by focusing on the results only in terms of trends. Since our hypothesis was that the higher ability groups would be more differentiated in personality, but also generally provide better data, we expected them to show higher reliabilities (i.e., low error variance), but lower inter-scale correlations (i.e., low response style variance) than the lower ability groups. However, our results do not lead to an unequivocal conclusion. In sample 1, reliability analyses indicated a clear trend of higher reliability coefficients in the high-ability group, with statistically significant differences confirmed for four out of five NEO scales. However, inter-scale correlation coefficients did not indicate that the high-ability group has lower inter-scale coefficients. On the other hand, in sample 2, reliability analyses also indicated a clear trend of higher reliability coefficients in the high-ability group (except for openness), but statistically significant differences were confirmed only for one out of six HEXACO scales. However, in this sample, inter-scale correlation coefficients do indicate a trend of lower inter-scale coefficients in the high-ability group.
Since we used the same verbal intelligence but different personality measures in different samples in the present study, our results could imply that the influence of verbal intelligence on the personality assessment may depend on the specific personality inventory. Namely, personality questionnaires differ in their structure, item content, and complexity, as well as (in part) the factors they measure. The last issue is primarily conspicuous if questionnaires differ in their theoretical background. In sample 1, we used measures of the Five-Factor Model, but in sample 2, we operationalized personality within the HEXACO framework—a six-factor model. Besides adding a new basic personality dimension that is not included in five-factor models (i.e., honesty-humility), HEXACO has other important conceptual differences compared to other five-factor models (see Ashton and Lee 2001). Therefore, it is possible that HEXACO inventories are more suitable for individuals with lower intelligence scores. However, this is the first study to test the differentiation hypothesis when personality is assessed with the HEXACO personality model, so future studies are needed to test this assumption.
This study has a few limitations. Firstly, it is important to emphasize that we only used one intelligence measure, that is, the verbal test. Although it is highly saturated with g (Bratko et al. 2012), it is plausible that assessing intelligence with a verbal test could have a different influence on testing the differentiation hypothesis compared to when multiple intelligence measures are used or when fluid intelligence is measured (e.g., Andersson et al. 2022). Secondly, our samples consisted of younger adults, so generalization to other age groups is questionable. Thirdly, our samples differ in average IQ scores and education levels, so we urge readers to interpret and generalize these findings accordingly. Finally, our conclusions are also limited to self-assessed personality trait levels and not true personality trait levels of the five or six personality factors, so we are not able to separate effects due to true personality trait variance from the effects due to self-report-specific method variance.
5 Conclusions
In conclusion, the results of this study provide only partial support for the personality differentiation by intelligence hypothesis. In one sample, results were mostly in line with the hypothesis, showing that individuals scoring higher on the verbal intelligence test achieve more variable and more reliable results on the NEO-FFI personality measure than individuals with somewhat lower cognitive abilities. This was not confirmed in a sample where the HEXACO inventory was used. Regarding personality structure, our findings did not support the idea that more intelligent individuals will have lower inter-correlations in the NEO-FFI sample, whereas in the HEXACO sample this tendency is somewhat more evident, but only for some personality factors. If these between-sample differences are due to differences between personality inventories, this could have practical implications for which personality measure to choose depending on the target population. For example, if future studies also confirm that scale variances and reliabilities (but also the inter-scale correlations) do not depend on intelligence level when using the HEXACO inventory, this personality measure may be a better choice when assessing groups of participants with lower abilities.
Author Contributions
Conceptualization: [D.B.]; Methodology: [D.B., M.L., and T.V.H.]; Software: [D.B., M.L., and T.V.H.]; Validation: [D.B., and M.L.]; Formal analysis: [D.B., M.L., and T.V.H.]; Investigation: [D.B., M.L., and T.V.H.]; Resources: [D.B., M.L., and T.V.H.]; Data curation: [D.B., M.L., and T.V.H.]; Writing – Original draft preparation: [M.L.]; Writing – Review and editing: [T.V.H., and D.B.]; Visualization: [D.B., M.L., and T.V.H.]; Supervision: [D.B.]; Project administration: [D.B., M.L., and T.V.H.]; Funding acquisition: [D.B.].
Acknowledgments
The authors have nothing to report. Open access publishing facilitated by $WOA_OO_ELIGIBLE_INSTITUTION, as part of the Wiley - National and University Library in Zagreb Consortium Croatian Academic and Research Libraries Consortium agreement.
Ethics Statement
Ethics protocol during data collection procedure was in line with the protocol for research projects at the Department of Psychology, Faculty of Humanities and Social Sciences, University of Zagreb. All potential participants were informed of research aims and decided to participate voluntarily, anonymously, and with no compensation.
Consent
All participants gave their informed consent for the results to be published on a group level, with no personal information.
Conflicts of Interest
The authors declare no conflicts of interest.
Open Research
Data Availability Statement
The data that support the findings of this study are available from the corresponding author upon reasonable request.
