Five criteria were used to determine whether a study should be included in the current meta-analysis: (1) The study must report a relationship between PG and at least one of the five personality traits in the five-factor model. PG was defined as any gambling problem ranging from mild gambling problems to gambling disorder, either diagnosed clinically or assessed by any instrument assessing gambling problems. (2) The personality traits of the five-factor model had to be assessed with any relevant self- or observer-rating instrument or item. (3) If the data involved group comparisons the comparator comprised a non-gambling problem group. (4) The included studies must have reported the zero-order correlations between PG and personality traits, or sufficient data for such calculations, such as N, mean scores, and standard deviations for the PG and a comparison group. (5) The studies must have reported original data. We included studies written in all European languages. There were no criteria concerning population, interventions, participant characteristics, or year of publication. Studies were excluded if they were: (1) based on case-studies, (2) were qualitative studies, (3) were reviews or only reported secondary data, or (4) measuring personality traits similar to the five-factor model, but not the actual traits.

Relevant publications were collected through a search of six research databases: Medline, Web of Science, PsychInfo, Cochrane Library, OpenGrey, and Google Scholar. The following search string was designed to locate the articles of interest and their relevance to the meta-analysis: “(gambl^*) AND (personality OR “five factor” OR neuroticism OR extraversion OR extroversion OR openness OR intellect OR agreeableness OR conscientiousness).” The searches in each database ended on February 2nd, 2021. Based on this search string we identified a total of 4,059 articles of interest. The number of articles identified from each database were: Web of Science: 1,309, PsycInfo: 1,555, Cochrane: 185, and Medline: 1,010. The searches in Google Scholar and OpenGrey were conducted to identify gray literature, but no additional literature was identified through these sources. After removing duplicates, 2,517 articles were available for screening. Of these, 2,427 were removed as they were deemed irrelevant based on title and abstract. After screening the remaining 91 full-text articles, 45 articles were included for detailed screening. In all, 17 articles [e.g., (37, 38)] were excluded as they did not fulfill the inclusion criteria (see Figure 1). Thus, a total of 28 articles were included in the present meta- analysis.

To ensure validity, as well as avoid mistakes regarding adherence to the inclusion and exclusion criteria, all articles were systematically and independently reviewed by two authors. Observer agreement was expressed as percentages. Disagreements were resolved by discussion, and if needed by consulting a third team member. Data from the selected studies (zero-order correlations between PG and the personality factors and information regarding possible moderators) were coded separately by two authors. Disagreements were resolved by discussions. For studies with appropriate data from more than one personality instrument, or in studies reporting data from at least two subdimensions of each dimension of the five-factor model, the data were entered combined.

The studies included in the meta-analysis were assessed by the Newcastle-Ottawa Scale for cross-sectional studies (39) to identify risk of bias/study quality. The Newcastle-Ottawa Scale provides predefined criteria for assessing bias/quality through a checklist, consisting of three main categories. It is possible to score a maximum of 10 stars, where a higher score indicates higher quality/less bias. The first category is “selection,” and relates to the representativeness of the sample, sample size, comparability between respondents and non-respondents, and ascertainment of the exposure. This category gives maximum of five stars. The second category is “comparability,” and concerns whether confounding factors are controlled for. This category gives maximum two stars. The third category is “outcome” and represents assessment of the outcome and the statistical tests. The maximum score on this category is three stars. In the current study, studies with five stars or more were considered to have moderate to good quality. Two authors evaluated the papers for risk of bias independently. Disagreements were resolved by discussion. Degree of agreement was calculated as percentages.

We used a random-effects model in the meta-analysis of correlations between PG and personality. A random-effects model was chosen as this approach is likely to improve the external validity and generalizability of findings and is recommended when included studies are assumed to represent different populations of studies (40). Results from individual studies and syntheses were visualized as forest plots including single and accumulated effect sizes (correlation) as well as 95% confidence intervals. In order to explore factors that might explain between-study heterogeneity, we conducted a random-effects meta-regression analysis to examine whether three a priori determined moderators could explain between-study variance. The first moderator comprised level of PG. This involved whether the sample was based on participants with a diagnosed gambling disorder (coded as 1) or participants assumingly having PG, albeit without a formal diagnosis (coded as 0). The second moderator concerned comorbidity. If more than 50% of the participants had a comorbid disorder this was coded as 1, if 50% or more did not have a comorbid disorder this was coded as 0. The last moderator comprised assessment of personality. If the scale had four items or fewer this was coded as 0, and if the scale had five items or more this was coded as 1. Heterogeneity was assessed using Cochrans' Q. The I² statistic, which reflects the proportion of variation in observed effects that is due to variation in true effects (i.e., “true” heterogeneity as opposed to chance) (40), was also calculated. An I² of 0% suggests homogeneity, 25% indicates low heterogeneity, 50% indicates moderate heterogeneity, and 75% indicates high heterogeneity (41). The trim-and-fill procedure by Duval and Tweedie (42) was used for investigation of publication bias. The meta-analysis and meta-regression analyses were conducted using the Comprehensive Meta-Analysis 3.0 software (43). The observer agreement on data-coding of effect sizes, moderators and quality assessment were calculated percentwise.

The included studies were collected from 13 countries, between the years 1987 to 2020. Sample sizes in the included studies varied from 37 to 9,111. The instruments used to measure PG were, in most of the studies, the South Oaks Gambling Screen [SOGS; (44)], with a close follow up by instruments based on the criteria found in third, fourth, and fifth editions of the Diagnostic and Statistical Manual of Mental Disorders [DSM-III, DSM-IV, DSM-5; (45–47)]. Other instruments were the Problem Gambling Severity Index/Canadian Problem Gambling Index [PGSI/CPGI; (48)], the Lie/bet Questionnaire [LBQ; (49)], the Shorter PROMIS Questionnaire [SPQ; (50)], the Structured Clinical Interview for Pathological Gambling [SCI-PG; (51)], the International Classification of Disorders-10 [ICD-10; (52)], the Berlin Inventory of Gambling [BIG; (53)], the National Opinion Research Center DSM-IV Screen for Gambling Problems [NODS; (54)], the Canadian Adolescent Gambling Inventory [CAGI; (55)], the DSM-IV-Multiple Response-Juvenile Criteria to Identify Adolescent Problem Gambling [DSM-IV-MR-J; (56)], and the Gambling Problems Index [GPI; (57)]. The most common measurement tool for personality was the Revised NEO Personality Inventory Revised [NEO-PI-R; (58)], which was used in seven studies. Other measurement tools for personality were the Eysenck Personality Inventory [EPI; (59)], the Eysenck Personality Questionnaire [EPQ; (60)], the Mini-International Personality Item Pool [MINI-IPIP; (61)], the Big Five Inventory [BFI; (62)], the 16 Personality Factor Questionnaire [16PF; (63)], the International Personality Item Pool NEO 120 [IPIP NEO 120; (64)], the Estonian Personality Item Pool-NEO [EPIP-NEO; (65)], the HEXACO Personality Inventory Revised, HEXACO-60 [HEXACO-60; (66)], the NEO-Five Factor Inventory [NEO-FFI; (67)], the HEXACO-Personality-Inventory-Revised [HEXACO PI-R; (68)], the Zuckerman-Kuhlman Personality Questionnaire [ZKPQ; (69)], and the Brief HEXACO Inventory [BHI; (70)].

The number of personality dimensions of the five-factor model that each study measured varied from one to five. A summary of the characteristics of each study, including country, publication, sample sizes of PG, and control group, instrument used for assessment of personality and gambling problems, number of personality dimensions measured, and the measure of effect size, is provided in Table 1. The agreement between the independent observers regarding coding of these characteristics were 98.1%. The agreement of coding of correlation data was calculated as percentages and amounted to 75.6%. The moderators used in the meta-regression analyses were coded independently by two authors and agreement was 62.5% for diagnosis 83.3% for comorbidity and 75.0% for number of personality items, respectively.

The Newcastle-Ottawa Score results for each study are shown in Table 2. Five of the included studies obtained less than five stars and were therefore considered to be of low quality. The remaining 23 studies obtained five or more stars and were therefore considered to be of moderate or high quality. The average quality score was 5.89 (SD = 1.69). The agreement between raters were 78.6% for “selection,” 100.0% for “comparability,” and 96.4% for “outcome,” respectively.

The results of the meta-analysis on the personality trait neuroticism are presented in Figure 2. The overall correlation across 25 relevant studies was 0.273 (95% CI = 0.182, 0.358). Cochran Q was significant (Q = 544.801, df = 24, p < 0.001). The I² was 95.78% indicating a high heterogeneity. The funnel plot (Appendix in Supplementary Materials) suggested an asymmetric distribution of correlations. The trim-fill-prodedure suggested an adjusted effect size of 0.378 (95% CI = 0.258, 0.486).

The results of the meta-analysis on the personality trait conscientiousness are shown in Figure 3. The overall correlation across all 20 relevant study articles was −0.296 (95% CI = −0.400, −0.185). Cochran Q was significant (Q = 570.959, df = 19, p < 0.001), suggesting significant heterogeneity across studies. The I² was 96.670%, indicating a high heterogeneity. The funnel plot (Appendix in Supplementary Materials) suggested an asymmetric distribution of correlations. The trim-fill-prodedure suggested an adjusted effect size −0.398 (95% CI = −0.517, −0.264).

The result of the meta-analysis on the personality trait agreeableness are presented in Figure 4. The overall correlation across all 21 relevant studies was −0.163 (95% CI = −0.223, −0.101). Cochrans' Q was significant (Q = 145.237, df = 20, p < 0.001), suggesting heterogeneity. The I² was 86.229%, indicating high heterogeneity. The funnel plot (Appendix in Supplementary Materials) suggested an asymmeteic distribution of correlations. The trim-fill-prodedure suggested an adjusted effect size −0.193 (95% CI =-0.264, −0.121).

The results of the meta-analysis on the personality trait openness are presented in Figure 5. The overall correlation across all 18 relevant studies was −0.219 (95% CI = −0.308, −0.127). Cochran Q was significant (Q = 151.651, df = 17, p < 0.001), suggesting significant heterogeneity. The I² statistic was 88.790%, indicating a high heterogeneity. The funnel plot (Appendix in Supplementary Materials) suggested an asymmetric distribution of correlations. The trim-fill-prodedure suggested an adjusted effect size of −0.256 (95% CI = −0.344, −0.163).

The results of the meta-analysis on the personality trait extroversion are presented in Figure 6. The overall correlation across all 24 relevant studies was −0.083 (95% CI = −0.120, −0.046). Cochran Q was significant (Q = 51.538, df = 23, p < 0.005), suggesting heterogeneity. The I² was 55.373%, indicating a moderate heterogeneity. The funnel plot (Appendix in Supplementary Materials) suggested a symmetric distribution of correlations. The trim-fill-prodedure did not change the outcome for extroversion.

Because of the significant heterogeneity in all meta-analyses, a meta-regression analysis based on a random-effect model was conducted for each trait. The results are presented in Table 3. For the personality dimensions including neuroticism (Q = 2.62, df = 3, p = 0.455, R² = 22%), conscientiousness (Q = 0.62, df = 3, p = 0.893, R² = 0%), agreeableness (Q = 1.06, df = 3, p = 0.788, R² = 36%), and openness (Q = 5.0, df = 3, p = 0.170, R² = 8%) the regression model was not significant. The regression model was however significant for extroversion (Q = 7.91, df = 3, p = 0.048, R² = 73%). Only the predictor, diagnosis, was significant (b = −0.0797, p = 0.029), suggesting more negative effect sizes for samples with participants with a gambling diagnosis than for samples without a gambling diagnosis. There was still significant unexplained between-study variance for extroversion (Q = 51.54, df = 23, p = 0.001).

The studies included in the present meta-analysis originated from a variety of countries. Still, few studies were conducted in non-Western and developing countries. Furthermore, the studies showed a broad variation in terms of study quality, as shown by the Newcastle-Ottawa assessment in Table 2. For instance, Roy et al. (86) and Barnes and Parwani (71) received rather low scores, which may reflect that those studies were relatively old and thus, not fulfill common criteria of scientific scrutiny of today. Other studies consisted of relatively small sample sizes with a specific group in focus. The studies of Chiu and Storm (31), MacLaren et al. (78), and McGrath et al. (33), for instance, consisted of students, while Müller et al. (25), Myrseth et al. (21), and Mann et al. (80) included problem gamblers seeking treatment. In contrast, there were only one cross-sectional study consisting of a large sample of more than 9,000 randomly selected participants (22). Hence, further studies should comprise large samples of randomly selected participants. Another major limitation of the identified literature is that very few studies (30, 75, 87) used longitudinal designs. We did, therefore, not conduct a meta-analysis of such data. Still, longitudinal data on the relationship between PG and personality are of vital importance in order to elucidate directionality and potential causality between PG and personality.

The present meta-analysis has several strengths. It was conducted in line with the PRISMA guidelines (36). Searches were conducted across several databases, including search in gray literature, although the gray literature yielded no unique hits. To ensure reliability regarding quality assessment and effect size data, the included studies were coded independently by two authors where disagreements were resolved by discussion or by consulting a third team member. Records showed overall good initial agreement between the raters, suggesting high interrater reliability. Authors were contacted when relevant data were missing.

The main limitation of the present meta-analysis is that definitions of PG were not the same across studies. It cannot be ruled out that the associations between PG and personality may vary as a function of the various operationalizations of both PG and personality. The large number of operationalizations did, however, preclude meaningful subgroup analysis of results from specific operationalizations. In addition, there were few studies comprising randomized samples. In addition, the target population (problems gamblers), consisted of various types of gamblers, reflecting large heterogeneity in terms of types/groups of gamblers. This implies restrictions in terms of external validity. Also, limited generalizability relates to the broad usage of personality measures and the large range of gamblers included (from students to pathological gamblers). Another limitation pertains to the screening process of relevant studies, which initially was based on title and abstract. Hence, it cannot be ruled out that a few relevant studies may have been excluded, although relevant information usually is presented in the title or abstract. A final limitation is that the funnel plots showed asymmetry in four out of five meta-analyses, suggesting publication bias. The trim and fill procedure did however aim to correct for this. The discrepancies between the original and adjusted effect sizes were small to moderate, suggesting that the data and conclusions were not influenced by publication bias.

The positive association between PG and neuroticism should alert clinicians to the vulnerability of those suffering from PG regarding experiencing dysphoric states. Therapists should therefore consider to address this vulnerability specifically. In this context clinicians should also be aware that many gamblers gamble in order to escape dysphoric emotional states (97). Neuroticism has also been associated with treatment relapse (23), hence relapse prevention should be given extra thought when treating patients with high scores on neuroticism. Conscientiousness was inversely related to PG. Hence many patients have low scores on this trait. Clinicians may therefore make an extra effort to motivate these patients in terms of compliance (e.g., with home tasks) as this has been shown to be lowered in those with low scores on conscientiousness (98) and also include exercises helping patients with long-term goal planning. Low scores on conscientiousness has also been associated with treatment relapse (23), suggesting that increased focus on relapse prevention might be called for. Agreeableness was inversely related to PG. Therapists should be aware that forming a good therapeutic alliance with those having low scores on agreeableness is more difficult than with those with higher scores (98). Those with low scores on this trait often cope with interpersonal difficulties in destructive ways (99), which may be addressed specifically in therapy. Clinicians should also be aware that those with low scores on agreeableness seem to drop out of treatment more often than those with higher scores (23). Also, openness was inversely related to PG. Those with low scores on this trait may be less receptive to the idea of psychotherapy itself. In therapy they may however be particularly responsive to concrete and practical suggestions (100). Extroversion showed the weakest association with PG. This trait has been found to be related to gambling motives. Those with high scores on extroversion seem to be incentivized to gamble due to social motives (101). In a similar vein, an inverse association between online gambling and extroversion has been found (102). Hence, clinicians may consider gambling motives in relation to extroversion in therapy. Overall, several of the traits in the five-factor model of personality are related to several therapeutic challenges which clinicians should be aware and take into account when treating patients with PG.