Introduction
The prevalence of psychotic disorders (psychosis) has traditionally been thought to be similar across countries and cultures (Jablensky, Reference Jablensky2000). However, a review of studies indicates clearly that prevalence rates vary across countries (McGrath, Saha, Chant, & Welham, Reference McGrath, Saha, Chant and Welham2008). The evidence from this review also suggests that both incidence and prevalence of psychotic disorders are lower in low-and-middle-income-countries (LAMIC) than in high-income countries (HIC) (McGrath et al., Reference McGrath, Saha, Chant and Welham2008), albeit more comparative research is necessary to further substantiate these findings. Paradoxically, the prevalence of psychotic experiences (PEs) at a subclinical level of psychosis, that are often considered to be an indicator of psychosis proneness (van Os, Reference van Os2016), has been shown to be higher in LAMIC than in HIC (McGrath et al., Reference McGrath, Saha, Al-Hamzawi, Alonso, Bromet, Bruffaerts and Kessler2015) and it is unclear to date why this is the case.
Some researchers have argued that the paradox occurs due to cultural differences. People in LAMIC have been assumed to have better social support [e.g. collective culture in LAMIC (Jablensky et al., Reference Jablensky, Sartorius, Ernberg, Anker, Korten, Cooper and Bertelsen1992)], to be less distressed by PEs (Wüsten et al., Reference Wüsten, Schlier, Jaya, Fonseca-Pedrero, Peters, Verdoux and Lincoln2018), and to perceive PEs as culturally acceptable (Al-Issa, Reference Al-Issa1995; Luhrmann, Padmavati, Tharoor, & Osei, Reference Luhrmann, Padmavati, Tharoor and Osei2015) rather than as an indicator of poor mental health (McGrath et al., Reference McGrath, Saha, Al-Hamzawi, Andrade, Benjet, Bromet and Kessler2016). Together, these factors are assumed to increase the likelihood of reporting PEs and to reduce symptom-related distress. Although this is plausible, it is necessary to rule out another explanation first, namely that the higher rates of PEs in LAMIC are due to measurement invariance (MI, DeVylder & Koyanagi, Reference DeVylder and Koyanagi2018). If this were the case, people in LAMIC are interpreting the questions used in typical self-report assessments of PEs differently and we would expect the prevalence difference in PEs in LAMIC v. HIC to be significantly smaller after correcting for MI. In order to rule out the MI explanation, it is necessary to compare the frequency of PEs across countries using an instrument that has shown to be MI (Putnick & Bornstein, Reference Putnick and Bornstein2016). MI can be tested statistically at four different levels: Configural MI refers to equal factor structure of the measurement instrument across groups. Metric MI refers to equal factor loadings across groups. Scalar MI refers to equal intercepts of indicators across groups. Residual MI refers to the equal residual variance of the indicators across groups. Scalar MI is the absolute minimum requirement for comparing scores between groups (van de Schoot, Lugtig, & Hox, Reference van de Schoot, Lugtig and Hox2012). Complete MI is given if MI is shown at all four levels.
Previous studies investigating MI in psychotic symptom measures have produced mixed results, both for PEs in general and for specific symptoms, such as paranoia and hallucinations. For example, Fonseca-Pedrero et al. (Reference Fonseca-Pedrero, Ortuño-Sierra, Lucas-Molina, Debbané, Chan, Cicero and Voracek2018) utilized the brief version of Schizotypal Personality Questionnaire (SPQ-B) to examine MI across samples from 14 countries from different continents. Although they found evidence for a universal existence of the schizotypy construct across countries, they could not find evidence for metric MI indicating that the contribution of the individual items to the schizotypy construct varied across countries. Similarly, Cicero, Martin, and Krieg (Reference Cicero, Martin and Krieg2019b) reported non-invariance for items related to positive symptoms of the Wisconsin Schizotypy Scales in a US sample of Asian, White, Hispanic, and Multiracial undergraduate students. In contrast, a dichotomized version of the Community Assessment of Psychic Experiences (CAPE) was shown to have MI in a sample of people with psychotic disorder, siblings, and healthy controls from Brazil, France, Italy, the Netherlands, Spain and the UK (Pignon et al., Reference Pignon, Peyre, Ferchiou, van Os, Rutten, Murray and Szöke2018). Furthermore, a study on the dichotomized version of Prodromal Questionnaire–Brief in a US sample of Asian, White, Hispanic, and Multiracial undergraduate students in two public universities (Cicero, Krieg, & Martin, Reference Cicero, Krieg and Martin2019a) was MI whereas the continuum distress scale was not. In terms of specific symptom domains, a study investigating the MI of the Psychosis Screening Questionnaire (PSQ) across samples from different ethnic groups living in the UK showed that paranoid symptoms were measured less validly among Pakistani ethnic groups in comparison to White British ethnic groups and that this exaggerated the ethnic differences in the reported frequency of paranoid symptoms (Heuvelman, Nazroo, & Rai, Reference Heuvelman, Nazroo and Rai2018). In the same study, hallucination symptoms were found to be MI, which is in line with another cross-national study using the Launay-Slade Hallucination Scale-Extended (Siddi et al., Reference Siddi, Ochoa, Laroi, Cella, Raballo, Saldivia and Preti2019). In contrast, in a sample of psychology students from the Netherlands, Nigeria, and Norway the hallucination factor of the positive symptom factor of the CAPE was found to be non-invariant whereas the paranoia and bizarre experiences factors were found to have partial scalar MI (Vermeiden et al., Reference Vermeiden, Janssens, Thewissen, Akinsola, Peeters, Reijnders and Lataster2019). Despite some promising findings, the overall picture of these studies points to problems when it comes to validly interpreting score differences in psychosis measures between different countries or ethnic groups.
In the present study, we tested the MI of the CAPE (Stefanis et al., Reference Stefanis, Hanssen, Smirnis, Avramopoulos, Evdokimidis, Stefanis and van Os2002), a self-report tool that is frequently used to assess PEs in cross-country comparisons, across country income groups (LAMIC v. HIC). The sample was a large multinational community sample drawn from individuals living in 13 countries—Canada, Colombia, France, Germany, Ghana, India, Indonesia, Mexico, Spain, Sweden, the Netherlands/Belgium, the UK, and the USA. Finally, we compared the CAPE scores across country income groups taking putative MI into account to test the hypothesis that differences of PEs across country income groups are not due to MI.
Method
Participants and procedure
The analyses were based on a large data pool of community samples from 13 countries (N = 7141), which has been described in detail elsewhere (Wüsten et al., Reference Wüsten, Schlier, Jaya, Fonseca-Pedrero, Peters, Verdoux and Lincoln2018). The data were compiled from 2016 to 2017. It consisted of data from previously published studies identified in a review article of studies using the CAPE (Mark & Toulopoulou, Reference Mark and Toulopoulou2016) and from additional data sources. We obtained community sample data from Canada, France, Germany, the Netherlands/Belgium, Sweden, Spain, and the UK. Details of the data collection procedure including ethical approvals are presented in the respective studies (Fonseca-Pedrero, Paino, Lemos-Giráldez, & Muñiz, Reference Fonseca-Pedrero, Paino, Lemos-Giráldez and Muñiz2012; Korver et al., Reference Korver, Quee, Boos, Simons and de Haan2012; Peters et al., Reference Peters, Ward, Jackson, Morgan, Charalambides, McGuire and Garety2016; Schlier, Jaya, Moritz, & Lincoln, Reference Schlier, Jaya, Moritz and Lincoln2015; Verdoux, Sorbara, Gindre, Swendsen, & van Os, Reference Verdoux, Sorbara, Gindre, Swendsen and van Os2003; Ziermans, Reference Ziermans2013). Additional data were collected via online surveys for the purpose of this study. To this aim, participants were recruited using a crowdsourcing website (Crowdflower), on which users participate in surveys for financial compensation. Participants from Colombia, Ghana, India, Indonesia, Mexico, and the USA were recruited via this website only whereas participants from Canada, Germany, and the UK were recruited via this website to complement the pre-existing data. All participants in the online surveys provided written informed consent and indicated being above 18 years of age prior to data collection. The Ethics Committee at the Universität Hamburg approved the survey.
Of the 7141 participants, 553 were from India (7.7%), 607 from Colombia (8.5%), 185 from Ghana (2.6%), 568 from Indonesia (8.0%), 559 from Mexico (7.8%), 681 from France (9.5%), 839 from Sweden (11.7%), 516 from the Netherlands (7.2%), 658 from Spain (9.2%), 1225 from Germany (17.2%), 257 from the UK (3.6%), 216 from the USA (3.0%), and 277 participants were from Canada (3.9%). There were 3676 male participants (51.5%) and the average age was 27.39 years (s.d. = 9.97), ranging from 18 to 80 years. Demographic characteristics and average scores of the factors of the CAPE by country are shown in Table 1.
Participants' characteristics (N = 7141)
Note. The sex distribution (χ2 (12) = 1405.81, p < 0.01) and average age (F (12, 7128) = 207.84, p < 0.01) differed significantly across countries.
The country income groups were formed according to World Bank criteria (World Bank, 2016). Participants from Colombia, Ghana, India, Indonesia, and Mexico were combined to form a LAMIC group and participants from Canada, France, Germany, the Netherlands/Belgium, Spain, Sweden, the UK, and the USA were combined to form a HIC group.
Measures
Community assessment of psychic experiences
The CAPE is a self-report measure of lifetime PEs consisting of 42 items that include positive symptoms (20 items, e.g. ‘Do you ever hear voices when you are alone?’), negative symptoms (14 items, e.g. ‘Do you ever feel that you experience few or no emotions at important events?’) and depressive symptoms (8 items, e.g. ‘Do you ever feel pessimistic about everything?’, Stefanis et al., Reference Stefanis, Hanssen, Smirnis, Avramopoulos, Evdokimidis, Stefanis and van Os2002). Each item is answered in regard to the frequency on a 4-point Likert scale (‘never’, ‘sometimes’, ‘often’, ‘nearly always’) and in regard to distress (‘not distressed’, ‘a bit distressed’, ‘quite distressed’, ‘very distressed’). The original factorial structure of the CAPE consists of three factors (positive symptoms, negative symptoms, and depression; Stefanis et al., Reference Stefanis, Hanssen, Smirnis, Avramopoulos, Evdokimidis, Stefanis and van Os2002) and the most recently published factorial structure of the CAPE has 11 factors (Schlier et al., Reference Schlier, Jaya, Moritz and Lincoln2015), that includes the three main factors called positive symptoms, negative symptoms, and depression, five factors that load into positive symptoms factor called bizarre experiences, paranoia, hallucination, magical thinking, and grandiosity factors, and three factors that load into negative symptoms factor called social withdrawal, blunted affect, and avolition factor.
In this study, the frequency scale of the CAPE was administered in paper-and-pencil (n = 3792) and online formats (n = 3349). The psychometric properties of the French, English, German, Indonesian, Spanish and Swedish version of the CAPE have been independently examined (Brenner et al., Reference Brenner, Schmitz, Pawliuk, Fathalli, Joober, Ciampi and King2007; Fonseca-Pedrero et al., Reference Fonseca-Pedrero, Paino, Lemos-Giráldez and Muñiz2012; Jaya, Reference Jaya2017; Schlier et al., Reference Schlier, Jaya, Moritz and Lincoln2015; Ziermans, Reference Ziermans2013).
Analyses
As a preliminary analysis, we computed a series of confirmatory factory analyses (CFA) to confirm that the 11 factors factorial structure of the CAPE (Schlier et al., Reference Schlier, Jaya, Moritz and Lincoln2015) had a better fit to the complete sample data than the three factors (Stefanis et al., Reference Stefanis, Hanssen, Smirnis, Avramopoulos, Evdokimidis, Stefanis and van Os2002) and the proposed seven factors factorial structure (Mark & Toulopoulou, Reference Mark and Toulopoulou2016).
Then, we used a series of multiple group CFA (MGCFA) to examine the MI of the CAPE across the two country income groups with increasingly stringent criteria for equivalence to establish configural, metric, scalar, and residual MI (van de Schoot et al., Reference van de Schoot, Lugtig and Hox2012). Where the invariance constraints were found to not meet the proposed criteria, modification indices were used to identify parameters that needed to be freed (e.g. by taking into account residual correlations, or factor loadings of items) across groups in order to test for partial MI.
First, we tested for configural MI by computing MGCFA across the two groups according to the best fitting CAPE factorial structure found in the preliminary CFA. Here we tested whether the same latent constructs (e.g. positive symptoms factor, negative symptoms factor and depression factors) and their compositions (e.g. 20 items load on the positive symptoms factor) could be identified in LAMIC and HIC. Second, we tested for metric MI by repeating the first analysis with the additional assumption that factor loadings are equal across groups. Here we tested whether the answers of participants from LAMIC and HIC reflect the same understanding of the latent constructs (e.g. is the paranoia factor an indicator of the positive symptoms factor, it the conspiracy item an indicator of paranoia factor in both LAMIC and HIC samples). Third, we tested for scalar MI by repeating the MGCFAs with the assumption that loadings AND intercepts are equal across the two groups. Thus, here we tested whether the factor loadings AND the levels of the underlying items (i.e. the intercepts) are equal across groups. In this third step, the factor means were fixed to zero in one group and estimated in the other, while the common factor variances were freely estimated. This produced factor means, which were fixed to zero in all groups in the first and second step. Fourth, we tested for residual MI by repeating the MGCFA with the assumption that loadings, intercepts, AND residuals are equal across groups. For example, the conspiracy item is residual MI if it satisfied all the conditions specified above AND had equal error variance across LAMIC and HIC.
The CFAs and MGCFAs were conducted with structural equation modeling (SEM) using the lavaan package ver. 0.5–23 (Rosseel, Reference Rosseel2012) in R version 3.4.3. The missing data were dealt with by pairwise deletion in the MI analyses and listwise deletion in the other analyses. The data were not mean-centered. All models were estimated using diagonally weighted least squares using covariance matrices, which is recommended for ordinal data. The following fit indices, along with the proposed cut-off criteria, were used to assess model fit: CFI >0.95, RMSEA <0.06, and SRMR <0.08 (Hu & Bentler, Reference Hu and Bentler1999). The χ2 is reported but not used as a fit criterion because it tends to reject models that are based on large sample size (Bentler & Bonett, Reference Bentler and Bonett1980). To compare the fit of three factorial structures, we used the chi-square difference test following the Satorra-Bentler formula to compare nested models (Satorra & Bentler, Reference Satorra and Bentler2001). To assess MI, we followed the recommended stringent criteria from a simulation study because of our large sample size (N > 300), i.e. non-invariant is indicated by Δ CFI ⩾0.10, supplemented by Δ RMSEA ⩾0.15 or Δ SRMR ⩾0.30 (Chen, Reference Chen2007). We report the analysis of MI following a recommended convention (Putnick & Bornstein, Reference Putnick and Bornstein2016). However, because unequal sample sizes between groups have been shown to reduce MI (Yoon & Lai, Reference Yoon and Lai2018), we also report an additional MI analysis on an equalized number of participants between the LAMIC and HIC group by randomly excluding the HIC participants to match the LAMIC number of participants. Finally, we computed Cohen's d based on the pooled standard deviation of the groups in order to compare the means of the CAPE factors across groups. Cohen's d can be interpreted as small (d = 0.20), medium (d = 0.50), and large (d = 0.80) differences (Cohen, Reference Cohen1988).
Results
Confirming the factorial structure of the CAPE across countries
The fit indices of the three, seven, and 11 factors factorial structures of the total sample are presented in Table 2. All tested factorial structures met two out of three cut-off criteria for acceptable fit indicating that all had an acceptable fit. However, the 11 factors factorial structure had a significantly better fit than the other tested factorial structures. Similar results were found when we compared factorial structures within each country, except for the sample from the UK and Ghana in which the 11 factors factorial structure did not have a significantly better fit than the other factorial structures (see online Supplementary Table S1). Factor loadings of the 11 factors factorial structure in the total sample are presented in online Supplementary Table S2.
Fit indices of the confirmatory factor analyses
Note. * p < 0.05. CFI = Robust Comparative Fit Index. RMSEA = Robust Root Mean Square Error of Approximation. SRMR = Standardized Root Mean Square Residual. AIC = Akaike Information Criterion. 90% CI = 90% Confidence Interval. Comparison = Nested model comparison using a Scaled Chi-Square Difference Test with Satorra-Bentler method.
Measurement invariance of the CAPE across country income groups
The fit indices of the results of MI analyses of the CAPE across country groups are presented in Table 3. We found evidence for configural MI; the three fit indices met the cut-off criteria. We found evidence for metric MI as the ΔCFI, RMSEA, and SRMR met the cut-off criteria. We did not find evidence for full scalar MI because ΔCFI was above the cut-off criterium. Then, we inspected the modification indices and found that freeing seven parameters consisting of factor loadings (i.e. the two factors grandiosity and hallucination of the positive symptoms factor, the two items ‘influenced by devices’ and ‘ thought echo’ of the bizarre experiences factor, and the item ‘being special’ of the grandiosity factor) and item intercepts (i.e. lack of hygiene” and ‘ influenced by devices’) resulted in partial scalar MI. Residual MI was not found because the ΔCFI was above the cut-off criteria. Similar results were found in the sample, in which the number of participants was equalized across country groups. However, the scalar and residual MI models achieved goodness of fit for MI (online Supplementary Table S3).
Measurement invariance of the CAPE between LAMIC (n = 2472) and HIC (n = 4669, N = 7141)
Note. LAMIC = Low-and-Middle Income-Countries (n = 2472). HIC = High Income-Countries (n = 4669). CFI = Robust Comparative Fit Index. RMSEA = Robust Root Mean Square Error of Approximation. SRMR = Standardized Root Mean Square Residual. Metric invariance model was compared with configural invariance model. Scalar and partial scalar invariance model was compared with metric invariance model. Residual and partial residual invariance model was compared with scalar invariance model.
Recomputing the MI of the CAPE across the country groups with the adjusted factorial structure (i.e. excluding the non-invariant parameters) we found evidence for configural, metric, and scalar MI. This allowed us to conduct a non-biased latent score comparison, which is reported in Table 4. This comparison indicated that the participants from LAMIC had significantly higher scores on most CAPE factors (e.g. bizarre experiences, d = 0.54, p < 0.05; affective flattening, d = 0.75, p < 0.05; positive symptoms, d = 0.83, p < 0.05; negative symptoms, d = 0.38, p < 0.05) than participants from HIC, but a significantly lower score on avolition (d = −0.42, p < 0.05).
Latent mean comparison of the CAPE factors between LAMIC (n = 2472) and HIC (n = 4669, N = 7141)
Note. * p < 0.05. d uncorrected = Latent mean differences across comparison group based on non-measurement-invariant model. d corrected = Latent mean differences across comparison group based on measurement-invariant model LAMIC = Low-and-Middle Income-Countries (n = 2472). HIC = High Income-Countries (n = 4574). Positive values indicate LAMIC>HIC. Negative values indicate LAMIC<HIC. There are no scores for hallucinations and grandiosity factors in the measurement-invariant model because they are non-invariant.
Furthermore, the CAPE with the adjusted factorial structure (MI model) produced different mean scores within the country groups than the non-adjusted version of the CAPE (non-MI model). As can be seen in Table 4, the latent mean differences for paranoia, magical thinking and avolition between LAMIC and HIC changed statistical significance after MI adjustment. For example, paranoia in the non-MI model was found to be significantly higher in LAMIC than HIC (d = 38.26, p < 0.05), but this difference was no longer significant after MI adjustment (d = 0.03, p = 0.89). In contrast, avolition was found to not differ in LAMIC and HIC (d = 0.03, p = 0.64), but after MI adjustment avolition was significantly lower in LAMIC than HIC (d = −0.42, p < 0.05). Despite numerical differences in Cohen's d estimates between the non-MI and MI adjusted models, we found the effect size category to remain the same for the remaining factors of the CAPE.
Discussion
We examined whether measurement of PEs with the CAPE is invariant between LAMIC and HIC in order to understand whether measurement bias explains previously found differences in levels of PEs across country income groups (McGrath et al., Reference McGrath, Saha, Al-Hamzawi, Alonso, Bromet, Bruffaerts and Kessler2015). We found configural and metric MI which means that the structure and understanding of PEs in the CAPE are comparable between LAMIC and HIC (e.g. positive symptoms factor, negative symptoms factor and depression factor are separate constructs, hallucinations factor are a part of the positive symptoms factor, hearing voices item is a part of the hallucinations factor). However, we did not find full scalar MI of the CAPE. Here, we found that the grandiosity factor, the hallucinations factor, and the items being influenced by devices, thought echo, and the lack of hygiene was noninvariant, meaning that their differing levels across LAMIC and HIC do not indicate true latent score differences between the groups. Moreover, we did not find residual MI, which means that differences between LAMIC and HIC in observed PE scores are partly explicable by measurement bias.
We found positive symptoms to be most affected by MI adjustment. This aligns with findings from studies testing MI of other scales such as the SPQ-B (Fonseca-Pedrero et al., Reference Fonseca-Pedrero, Ortuño-Sierra, Lucas-Molina, Debbané, Chan, Cicero and Voracek2018), the Wisconsin Schizotypy Scale (Cicero et al., Reference Cicero, Martin and Krieg2019b), and the PSQ (Heuvelman et al., Reference Heuvelman, Nazroo and Rai2018) that also identified problems in measuring positive symptoms invariantly across culture groups. Specifically, we found hallucinations, grandiosity, and paranoia to be most affected by MI adjustment. The lack of MI regarding the hallucinations and grandiosity factor of the CAPE has also been shown in a student sample study (Vermeiden et al., Reference Vermeiden, Janssens, Thewissen, Akinsola, Peeters, Reijnders and Lataster2019). Both Vermeiden et al. (Reference Vermeiden, Janssens, Thewissen, Akinsola, Peeters, Reijnders and Lataster2019) and our study found the paranoia factor of the CAPE to have MI across culture groups. However, both Heuvelman et al. (Reference Heuvelman, Nazroo and Rai2018) and our study found paranoia factor differences between culture group changes after MI adjustment. They found the paranoia questions in PSQ overestimate the levels of paranoia in the Pakistani ethnic group in comparison to White British in the UK (Heuvelman et al., Reference Heuvelman, Nazroo and Rai2018), while we found the paranoia factor of the CAPE to be overestimated in LAMIC in comparison to HIC. Thus, the measurement bias we found in relation to paranoia may not be a specific problem of the CAPE. Rather, it might indicate a general difficulty of constructing an invariant measure of paranoia. This difficulty may stem from the fact that a measure of paranoia has to take objective levels of threat into account, which are likely to differ between geographical locations. For example, people in some countries may objectively face more threat (e.g. due to minority sexual status), which would be reflected in their responses to the CAPE paranoia items (e.g. ‘Do you ever feel as if you are being persecuted in some way?’, ‘Do you ever feel that people look at you oddly because of your appearance?’).
In contrast, MI adjustment only had a minimal effect on the depression and negative symptoms factor of the CAPE. This is consistent with previous studies showing complete MI of depression and negative symptoms measures across different cultural groups. For example, the Patient Health Questionnaire-9 has been found to be MI in adult Dutch and Surinam Dutch patients in the Netherlands (Baas et al., Reference Baas, Cramer, Koeter, van de Lisdonk, van Weert and Schene2011) and in community samples of English and Spanish speaking Hispanic American women in the USA (Merz, Malcarne, Roesch, Riley, & Sadler, Reference Merz, Malcarne, Roesch, Riley and Sadler2011). The Center for Epidemiological Studies Depression scale has also been found to be MI on community samples of Chinese and Dutch older adults (Zhang et al., Reference Zhang, Fokkema, Cuijpers, Li, Smits and Beekman2011). The Brief Negative Symptoms Scale has been found to have residual MI in Chinese, English, German, Italian and Spanish patients (Ahmed et al., Reference Ahmed, Kirkpatrick, Galderisi, Mucci, Rossi, Bertolino and Strauss2019). Thus, MI seems to be less of an issue in assessing depression and negative symptoms across countries and ethnic backgrounds than in assessing positive symptoms.
Even after MI corrections, LAMIC continued to show higher levels of some positive symptoms (bizarre experiences and magical thinking) and negative symptoms (social withdrawal and blunted affect). This speaks against the assumption that the higher levels of PEs are entirely explicable by MI. Thus, higher PEs in LAMIC than HIC could reflect cultural differences (World Health Organization, 2011), a poorer mental health status (McGrath et al., Reference McGrath, Saha, Al-Hamzawi, Andrade, Benjet, Bromet and Kessler2016), or both. The interpretation of poorer mental health in LAMIC is indirectly supported by a Lancet Commission Report (Patel et al., Reference Patel, Saxena, Lund, Thornicroft, Baingana, Bolton and Jü2018), which points to the problematic discrepancy between the high prevalence of social determinants of poor mental health (e.g. higher levels of poverty, higher number of loss, trauma, and displacement due to war and disasters) and the low availability of mental health services in LAMIC compared to HIC (e.g. median number of mental health professionals 10.1 v. 50.8 per 100 000).
Nevertheless, the paradox remains that despite the somewhat higher levels of PEs in the population, the rates of clinically relevant and diagnosable psychotic disorders are lower in LAMIC than in HIC (McGrath et al., Reference McGrath, Saha, Chant and Welham2008). A putative explanation for this paradox can be derived from the theory of extended psychosis phenotype (van Os & Reininghaus, Reference van Os and Reininghaus2016), which states that an accumulation of PEs is an essential feature of transition. As there is an indication of lower PE-related distress in LAMIC compared to HIC (Wüsten et al., Reference Wüsten, Schlier, Jaya, Fonseca-Pedrero, Peters, Verdoux and Lincoln2018) and non-distressing PEs have been found to be less tightly interrelated with other PEs in network models (Murphy, McBride, Fried, & Shevlin, Reference Murphy, McBride, Fried and Shevlin2018), lower distress may be a protective factor in this regard. In addition, the low rates of psychotic disorders in LAMIC (v. HIC) may partly be due to underdiagnosis due to restricted access to medical facilities and the presence of other ways of seeking help (e.g. traditional healers).
Strengths and limitations
A strength of this study is that the community samples were recruited from different regions and cultures. Countries from the continents of Asia, Africa, North America, South America, and Europe were represented in the study. This allows us to interpret the findings as generalizable with some confidence. Furthermore, a self-report assessment, in comparison to interviews, has the advantage of being free of interviewer bias and rating inconsistencies across countries.
A limitation of the study is the heterogeneous context of the study, the participant recruitment process, and the assessment type (online v. paper-and-pencil). For example, the sample from Ghana was recruited entirely online, whereas the sample from the Netherlands/Belgium was recruited entirely via a paper-and-pencil procedure. The assessment type differences between the country income group may have introduced selection bias, which may be a source of non-MI. Besides differences in assessment type proportion between the country income groups, the significant differences in sex and age between the country income groups may also be a source of non-MI. We cannot rule out that some of the problems in MI are attributable to these factors rather than country income group. Similarly, MI problems could also arise from differences in the language version of the measure rather than from country income group. These limitations should be addressed in future studies by (a) taking care to balance samples in terms of potentially relevant factors and (b) by conducting MI analyses on further variables, such as assessment type, countries, language, age, gender, and socioeconomic status.
Conclusion
We found full configural and metric MI of the CAPE, indicating that there is a culturally universal agreement on the structure and conceptualization of PEs across LAMIC and HIC. However, we only succeeded in obtaining partial scalar MI, indicating that we can conduct latent, but not observed sum-score mean comparisons without risk of bias. After MI adjustment, we found several PEs from the positive symptom domain to be higher in LAMIC than HIC. Thus, the paradox remains that the prevalence of psychotic disorders is lower whereas subclinical PEs are higher in LAMIC than in HIC.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S0033291720003323.
Acknowledgements
The authors are grateful to Steffi Hartanto who assisted with the data analysis. Edo S. Jaya was supported by the Universitas Indonesia's internal research grant (‘Hibah Publikasi Artikel di Jurnal Internasional Kuartil Q1 dan Q2 Tahun Anggaran 2019’, Grant No: NKB-0281/UN2.R3.1/HKP.05.00/2019). Eduardo Fonseca Pedrero was supported by the Spanish Ministry of Science and Innovation (MICINN) and by the Instituto Carlos III, Center for Biomedical Research in the Mental Health Network (CIBERSAM) and by the ‘Convocatoria 2015 de Ayudas Fundación BBVA a Investigadores y Creadores Culturales’ and the ‘Ayudas Fundación BBVA a Equipos de Investigación Científica 2017’.
Conflict of interest
None.
