RESEARCH DOSSIER
Cross-Cultural Comparative Design and Population Sampling
Abstract. Cross-cultural comparative design and population sampling determine which psychological differences and mechanisms may be inferred from research across culturally situated settings. This fundamental critical integrative monograph examines the historical concentration of psychological research, the definition of target populations and sampling frames, cross-national site selection, within-country heterogeneity, multilingual participation, probability and purposive designs, task adaptation, measurement comparability, clustered inference, and causal identification. It reviews foundational methodology, international survey practice, original behavioral field studies, and large multisite replication investigations. The synthesis distinguishes participant-level, site-level, national, and theory-level estimands and identifies circumstances in which convenience sampling, selection bias, translation effects, reference groups, and unmeasured ecological conditions constrain comparison. It develops a practical program of preregistered site selection, transparent sampling and response documentation, mixed-method triangulation, local scientific partnership, sensitivity analysis, and calibrated population claims. Human–artificial research is considered as an emerging application of the same principles. This critical narrative evidence synthesis reports no newly collected data and makes no claim to be a systematic review or meta-analysis.
Keywords: cross-cultural psychology; comparative research design; population sampling; WEIRD samples; sampling frame; multistage sampling; survey coverage; site selection; multilingual recruitment; measurement equivalence; cultural heterogeneity; multisite replication; multilevel inference; research generalizability; human–artificial relations
Scientific Scope and the Logic of Comparative Inference
Cross-cultural comparative design is the deliberate construction of observations and contrasts across culturally situated populations in order to test psychological propositions. Population sampling determines which persons, households, settings, communities, and societies enter those contrasts and with what probabilities or selection mechanisms. A comparison supports inference only in relation to an explicitly defined target: the recruited participants, the reachable population, the sampled sites, or a broader population for which a justified bridge exists. The present monograph examines those design decisions jointly because recruitment, measurement, and contextual variation determine the claim that can be warranted by an observed difference. The methodological lineage is developed by van de Vijver and Leung (1997) and the generalization problem is sharpened by Yarkoni (2022).
A cultural comparison can test a difference in group means, variability, developmental trajectories, individual-level associations, mechanisms, or intervention effects. These are distinct estimands. A reported mean contrast between two university samples is an estimate for those sampled settings under relevant assumptions; it is not automatically a contrast between entire nations. Likewise, an association between a local practice and performance may reflect measured exposure, selection into that practice, language, institutional arrangements, or correlated ecological conditions. Researchers should specify the level of analysis and the counterfactual or descriptive question before selecting the countries or constructing a questionnaire. The broad comparative research program described by Henrich et al. (2010) makes this discipline of inference indispensable.
This work uses a critical integrative review of foundational methods, documented empirical comparisons, multisite replication efforts, and multinational survey guidance. It is a narrative scientific synthesis; it contains no newly collected participant data, formal systematic-review search log, pooled effect estimate, or experimental trial. Its scientific task is to connect comparative questions to site-selection rules, population frames, task adaptation, data collection, and interpretable conclusions. Formal psychometric invariance, general statistical transportability, and global causal inference remain their own research domains; here they are evaluated at the operational interfaces where cultural study design can succeed or fail. Evidence about representation is considered alongside Rad et al. (2018) and Gurven (2018).
Historical Formation of Comparative Psychological Research
The problem of comparative evidence predates contemporary multinational survey infrastructure. Psychological researchers long used differences among societies, educational environments, and social institutions to ask whether a task, explanation, or developmental sequence depended on locally shared conditions. The twentieth-century expansion of experimental psychology nevertheless concentrated laboratory access and recruitment among university populations. In a documented historical critique, Sears (1986) traced the reliance of social psychology on college students and academic laboratory tasks and showed why this recurrent participant profile could condition substantive portraits of social behavior. The methodological lesson is historical as well as statistical: the place where psychology is institutionally organized shapes which forms of human life become observable.
Cross-cultural research gradually developed tools for explicit comparability rather than treating a national category as self-explanatory. van de Vijver and Leung (1997) systematized quasi-experimental comparisons, data equivalence, and interpretive design problems in a field organized around culturally diverse populations. Theories of cultural patterning also developed through contrasting cognitive tasks; for example, Nisbett et al. (2001) proposed an influential analytic–holistic account based on evidence from selected East Asian and Western populations. Such models generated hypotheses and further replication programs, while the limits of their initial populations required careful interpretation. A theory may have historical influence because it asks productive questions even when its most general formulation exceeds its evidentiary range.
A new phase of critical assessment focused on the demographic and institutional narrowness of psychology itself. Arnett (2008) examined the American concentration of authorship and samples in leading psychology journals. Henrich et al. (2010) collected comparative evidence challenging routine extrapolation from Western, educated, industrialized, rich, and democratic settings to the species. Rad et al. (2018) revisited publication practices and the representational demands of broader human science, while Ghai (2021) criticized treating WEIRD and non-WEIRD as exhaustive categories. These interventions shift the research question from whether cultural difference exists to how an evidence-generating system selects populations, defines diversity, and documents the reach of its conclusions.
Culture, Population, Site, and Unit of Analysis
A population is a defined set of eligible units at a particular place and time. A culture is a historically formed arrangement of learned practices, norms, institutions, meanings, and relational expectations, expressed and negotiated among persons. A country is an administrative and political unit; a language is a communication system and a possible recruitment variable; an ancestry category is a different kind of descriptor. These units sometimes correlate, but cannot substitute for one another without a theoretical account. National averages can be useful summaries, yet the distribution of relevant practices inside a country may be wider than the selected between-country difference. Muthukrishna et al. (2020) illustrates why cultural distance is multidimensional and depends on the selected dimensions and populations.
Researchers must declare whether the primary comparison concerns individuals nested in sites, cultural practices nested in institutional contexts, communities with specific historical experience, or countries as macro-level units. The same design may include households inside villages, villages inside districts, districts inside national sampling strata, and persons belonging to overlapping language or migration networks. Cross-border communities and diasporic groups complicate exclusive national assignment. At each level, record the corresponding sampling mechanism and key contextual attributes rather than converting a hierarchical structure into independent individual observations. Gelfand et al. (2011) offers a major example of multilevel cultural analysis while also showing the interpretive stakes of national grouping.
An operational cultural descriptor is a measurable characteristic chosen for an explanatory purpose. It may include schooling, livelihood, residential mobility, household structure, shared activity, exposure to institutions, or culturally relevant participation. None is a universal proxy for an entire cultural system. Constructing a comparison requires distinguishing measured contextual variables from the convenient site label used to recruit participants. Hruschka et al. (2018) emphasizes that apparently universal laboratory tasks may reflect skills and expectations characteristic of their settings. Thus the analytic unit and the unit from which participants are sampled should be written into the protocol before fieldwork and maintained in the public data dictionary.
Why Narrow Samples Distort Psychological Theories
A research population may be systematically selected because laboratory participation requires literacy, network access, travel, time, institutional affiliation, or comfort with researchers. Such requirements shape participation before random assignment begins. Randomizing an experimental condition within a highly selected sample can identify a conditional causal contrast for enrolled participants while leaving the population-selection problem intact. Henrich et al. (2010) identifies examples of variation in perception, fairness, spatial reasoning, moral judgment, and other domains; the literature supports examining specific mechanisms across populations rather than assuming that every result is equally variable. Representation is a relationship between a sampling process and an intended target population, not a label attached to participants.
The bias is persistent across developmental research as well. Nielsen et al. (2017) reviewed the geographic concentration of studies in prominent developmental outlets and argued that reliance on accessible convenience cohorts constrains theories of childhood. Arnett (2008) showed that a broader problem extends to authorship and editorial concentration. A sample drawn from a large city in one country cannot stand in for every child in that country, and adding an urban university from a second country does not convert the design into a national comparison. Distinct age groups, socioeconomic contexts, schooling histories, migration statuses, and disability access need deliberate treatment where they can modify measured phenomena.
Sampling diversity is also multidimensional. Ghai (2021) warns that a binary partition between privileged and other populations conceals consequential variation and can reproduce simplistic representations. A design may diversify geography while preserving uniform class, internet access, or educational attainment. Alternatively, a concentrated sample may be scientifically appropriate for a sharply defined local mechanism. The key is to align the scope of theory with the coverage of people and circumstances studied. Gurven (2018) frames sample diversity and socioecological theory as complementary: well-chosen populations test the boundaries of explanations, while explicit ecological variables help explain why behavior differs.
Defining the Target Population and Sampling Frame
An operational sampling plan begins with the target population, eligibility rules, reference period, and geographic boundaries. It next distinguishes the target from the frame population actually enumerated or reachable, then specifies the sampling units and the final analytic respondents. A register of households may omit transient residents, informal settlements, institutionalized persons, and those without stable addresses. School-based sampling excludes nonschool attendees; telephone and internet recruitment omit those without reliable access. For international comparisons, such omissions differ across sites. Harkness et al. (2010) treats these practical survey procedures as components of comparability rather than incidental implementation details.
A sampling frame can be an address list, school roster, health facility register, household enumeration, geographic area frame, or a carefully delimited network recruitment structure. Its coverage should be measured wherever possible and described when measurement is infeasible. The same age rule or formal residence requirement can acquire different substantive meanings in regions with different household structures and registration practices. Teams must specify whether newcomers, seasonal workers, displaced persons, bilingual residents, and people in collective accommodation are eligible. The framing decisions define the estimand before any probability weights are calculated. Survey guidance summarized in Harkness et al. (2010) demonstrates the importance of documenting these choices systematically.
Every cross-cultural manuscript should state the source population for each site, how the roster or area frame was formed, who could be approached, how many units were selected, and how response was tracked. Probability sampling permits design-based inference to the covered frame under its assumptions, whereas convenience or purposive sampling calls for narrower population claims or explicitly modeled selection. Large sample size does not repair unknown inclusion probabilities. The general sampling concern intersects with Rad et al. (2018), which insists that theories of humanity should be constructed from evidence reflecting the variation in human populations. A transparent frame description is the first reproducible record of that relationship.
Research-Site Selection and Contextual Diversity
Choosing research sites is an act of scientific design. A site is a location and institutional ecology in which particular people can be observed through particular procedures. Its relevance follows from a proposition: schooling might condition classification, market exposure might condition bargaining, or linguistic ecology might condition a response strategy. Investigators should select sites to vary those proposed explanatory conditions and document why each site supplies informative variation. Selecting one convenient city from each of several countries creates geographic dispersion but provides little evidence about the ecological contrasts that actually matter. The comparison of psychologically relevant cultural distances in Muthukrishna et al. (2020) provides a model for thinking beyond the country-count metric.
A principled site selection matrix records language environment, recruitment institutions, age distribution, schooling and livelihood, urbanicity, migration flows, technological access, and other factors warranted by the hypothesis. It distinguishes deliberately varied dimensions from nuisance dimensions requiring measurement. A factorial site design may seek both high and low schooling exposure within regions or different ecological arrangements inside one national boundary. At the same time, there are cases where a uniquely situated community is indispensable for testing a theory. Scientific value comes from the proposed contrast and the quality of observation, not an indiscriminate demand for maximal country coverage. Gurven (2018) argues for socioecological theory capable of explaining behavioral diversity.
Researchers also select institutions, not only populations. The willingness of schools, hospitals, community organizations, and local scholars to collaborate determines the accessible frame, the framing of consent, and the credibility of data collection. Site selection should therefore be recorded before outcomes are inspected, with reasons for substitution and withdrawal. A preregistered site roster can include expected variation, exclusion criteria, contingency replacements, and an explicit account of locations that were contacted but never enrolled. Such records distinguish cultural heterogeneity from variation in research access. The multinational survey guidance in Harkness et al. (2010) supports treating field organization as an integral component of comparative methodology.
Stratification and Multistage Population Samples
Probability-based designs create known, nonzero selection probabilities for the members of a covered sampling frame. Cross-cultural surveys often use stratified, clustered, and multistage selection because complete individual registers are unavailable. A study may first sample districts, then enumeration areas, then households, then eligible members. Explicit strata can improve coverage of rural locations, minority languages, or geographic regions, provided inclusion probabilities and later weights are documented. If regions are sampled unequally, the resulting respondents should not be treated as an unweighted simple random sample of a country without examining the estimand. The applied international survey framework in Harkness et al. (2010) shows why a common protocol requires country-specific frame documentation.
Sampling stages impose dependencies. Two respondents from one household may share resources and reporting habits; several classrooms within one school share instruction, timing, and administrator selection. The effective information in such data depends on intracluster similarity, the number of independently sampled clusters, and cluster-size imbalance. Large nominal respondent totals can coexist with weak precision for a site-level comparison when the relevant contextual unit has few observations. Researchers should include the hierarchy in estimation, uncertainty calculations, and design-based variance procedures. The multilevel cultural comparison in Gelfand et al. (2011) illustrates the need to distinguish individual responses from higher-level social context.
Strata should be chosen before fieldwork from scientifically material dimensions and defensible population information. Disproportionate sampling can be justified for rare communities or intersectional subpopulations, while population-level estimates require appropriate weights and transparent precision statements. A design that oversamples multilingual minority residents may improve mechanism testing even if it changes the raw sample composition. Conversely, post hoc subgroup carving can generate fragile claims, particularly with small cell sizes. Compare the actual composition of the achieved sample with the frame and with independent population benchmarks whenever they exist. The representativeness questions raised by Rad et al. (2018) are operational questions about these design and reporting decisions.
Probability, Purposive, and Hybrid Sampling
Probability sampling, purposive sampling, and respondent-driven recruitment answer different research questions under different inferential assumptions. A probability design can support descriptive estimation of a specified covered population, conditional on frame quality and response adjustment. Purposive selection can maximize theoretical contrast, recruit expertise, or expose neglected mechanisms. Snowball and online volunteer recruitment can provide access to otherwise difficult-to-reach groups but usually offer unknown inclusion probabilities and network-dependent coverage. Combining these strategies may be appropriate, provided they are recorded as distinct components rather than merged into a misleading single claim of representativeness. van de Vijver and Leung (1997) situates these choices within comparative designs that must match theory, selection, and analysis.
Hybrid studies often collect a broadly sampled quantitative survey and add ethnographic interviews or purposive task samples from contrasting settings. The components may investigate distinct estimands: prevalence or population distribution on one side and local meaning, process, or mechanism on the other. The fieldwork is integrated by common constructs, coordinated timing, explicit case-selection logic, and attention to discordant evidence. Qualitative participants are not automatically a random subset of the survey population, while a survey item cannot substitute for sustained contextual interpretation. Levitt et al. (2018) describes standards for transparent qualitative and mixed-method reporting useful for disclosing how these components relate.
A sample may also be specifically chosen to test a claimed universal process rather than estimate a national rate. Investigators can preregister a small set of populations where a theory predicts distinctive performance, implement an equivalent experimental manipulation, and examine whether the proposed effect appears. Such evidence is informative about the theory's boundary conditions but may lack design-based generalizability to entire countries. The strongest report labels its inferential population correctly and distinguishes theory-driven variation from convenient recruitment. Henrich et al. (2010) and Gurven (2018) support a comparative program in which targeted cross-population contrasts expose assumptions hidden by repeatedly studying homogeneous university cohorts.
Within-Country Heterogeneity and Mobility
A national population contains overlapping distributions of languages, social networks, educational opportunities, religious and occupational participation, and material environments. Distributions change over historical time and may differ sharply across neighborhoods or generations. Aggregating people into a single country mean can hide meaningful within-country variation and create a misleading impression of homogeneous national psychology. A culturally precise comparative design measures theoretically relevant contextual exposures and, where feasible, includes sites contrasting those exposures within the same country. The scales of cultural and psychological distance developed in Muthukrishna et al. (2020) help illustrate why demographic, linguistic, and psychological variation should be analyzed at appropriate resolutions.
Migration and transnational participation further complicate binary assignments. A person may speak one language at home, another at work, consume media from several regions, and belong to multiple communities. Schooling histories, citizenship, place of birth, current residence, and perceived cultural identification convey different information. A study of acculturation should specify duration, context, and measured adaptation rather than infer a single position from parental birthplace. Similarly, cross-border kinship networks may transmit practices independently of national administrative boundaries. The critique of broad cultural dichotomies by Ghai (2021) reinforces the importance of identifying relevant experiences directly when feasible.
A good sampling matrix includes explicit decisions for mobile populations, seasonal residence, refugee or displacement status, and persons missing from conventional household registers. Community-based enumerations may identify them better than an address frame, but sampling probabilities and privacy expectations require separate treatment. Within-country stratification and a multilevel analytic model can help examine whether observed between-country effects persist after site-level exposures are included. This does not imply that all cultural processes can be reduced to individual survey responses; institutions and shared practices may have genuine contextual effects. Comparative interpretation benefits from retaining both levels, as illustrated by Gelfand et al. (2011).
Recruitment, Response, and Coverage
Recruitment is a sequence of possible selection events: eligibility identification, successful contact, consent, questionnaire completion, item response, and analytic retention. Each stage can be associated with culturally patterned factors. Some participants work hours incompatible with the field schedule, may mistrust institutions, or need accessible formats or translation. Others may be reluctant to disclose certain experiences to local intermediaries. Rates should be calculated from clearly defined denominators and disposition categories and reported by site. A cross-country response-rate comparison is meaningful only after the contact and eligibility procedures are made comparable. Harkness et al. (2010) stresses the documentation of such field conditions.
The common use of online panels produces a special coverage challenge. Digital literacy, device access, language of platform registration, economic incentives, and familiarity with behavioral experiments may differ across recruited countries. A panel's national label identifies a market for participants, not a probability sample of a national population. Quotas on age and gender may balance selected margins while leaving unmeasured dimensions of selection intact. The appropriate analysis states whether inference is to panel members, internet users, a modeled target population, or a narrower group. Work on sampling diversity and cultural diversity in Rad et al. (2018) and Ghai (2021) provides the broader reason for keeping those distinctions visible.
Retention across multiple waves adds another layer. Follow-up may be easier among stable households and harder among mobile or economically precarious participants, changing the observed composition of longitudinal cultural comparisons. Report baseline eligibility, contact attempts, mode shifts, retention, reasons for attrition where known, and how sensitivity analyses handle unobserved differences. Community partners can help design acceptable contact practices, but their recruitment networks may also produce selection. An ethical and scientifically rigorous protocol records both the benefits and the constraints. Survey coordination recommendations in Harkness et al. (2010) treat interviewer training, translation, timing, and field monitoring as conditions of defensible comparative evidence.
Language Adaptation and Multilingual Recruitment
Language is simultaneously a practical recruitment channel, a possible cultural exposure, and a component of the measurement procedure. Translation can alter task difficulty, perceived politeness, emotional connotation, temporal reference, and the expectations attached to an answer. A comparative protocol therefore needs an explicit language inventory: which languages were offered, who translated the materials, which dialect and literacy assumptions were used, and who was excluded when an accessible version was unavailable. Translation and adaptation should be budgeted at site selection rather than added after recruitment has begun. The coordinated process described in Harkness et al. (2004) identifies translation and assessment as continuing design tasks.
A defensible adaptation workflow begins with the construct and intended use of each item. Bilingual experts propose translations, local reviewers identify conceptual and pragmatic problems, cognitive interviews investigate respondents' understanding, and pilot testing evaluates response distributions and task demands. Back-translation can reveal discrepancies but cannot establish functional equivalence by itself. The International Test Commission (2017) guidelines distinguish development and adaptation from the evidence needed to justify interpretations across language versions. Investigators should document disagreements and revisions, preserve the source and localized instruments, and explain where the task itself needs contextual modification rather than merely lexical substitution.
Multilingual recruitment also affects selection. Participants with high proficiency in the survey language may have different schooling or migration histories from those requiring translation. Offering only the majority language can convert a population comparison into a comparison of people with unequal access to institutions. Respondent language preference, interviewer language, reading support, and mode of administration should be recorded where ethical. A language-specific analysis might be relevant even when national averages are the headline result. The practical multinational survey program in Harkness et al. (2010) and the assessment principles in van de Vijver and Tanzer (2004) jointly show how translation, participation, and comparability form one methodological problem.
Tasks, Contexts, and Functional Equivalence
An identical stimulus does not necessarily constitute an identical psychological task. A hypothetical market exchange presumes experience with money, individual bargaining, and experimenter instructions; a categorization test presumes exposure to the pictured objects and response conventions. Functional equivalence asks whether the procedure elicits the same construct and comparable response opportunities in the participating settings. Standardizing every surface detail can therefore preserve artifacts, while unconstrained local modification can change the estimand. Researchers should identify a theoretically indispensable core and a separately documented set of context-sensitive elements. van de Vijver and Tanzer (2004) provides a useful conceptual vocabulary for construct, method, and item sources of bias.
Pilot work combines observation of how participants interpret instructions with analysis of administration errors and response times. Local investigators can document whether task materials are familiar, whether rewards have comparable practical value, and whether completion depends on formal schooling or computer experience. When two sites need different examples, the protocol must explain why those examples support equivalent interpretations. Parallel calibration tasks or within-site contrasts can help adjudicate alternative interpretations. Hruschka et al. (2018) stresses the importance of probing diversity in the processes elicited by laboratory tasks, rather than assuming a single laboratory format maps uniformly onto human cognition.
Functional equivalence is especially important in studies of cultural cognition. The analytic–holistic framework discussed by Nisbett et al. (2001) motivated systematic comparisons of attention, attribution, and reasoning. Follow-up work such as Norenzayan et al. (2002) examined reasoning preferences across selected groups, but local education, instruction style, and exposure to formal logic remain consequential interpretive variables. A successful comparative design reproduces both the task administration and the interpretive evidence that the task solicits a similar decision in each population. Investigators should publish both common materials and justified local adaptations so that other teams can assess how observed differences depend on procedure.
Measurement Comparability and Reference Groups
Cross-cultural mean comparisons assume that the recorded score has a comparable meaning across populations. Configural correspondence concerns the pattern of measured dimensions, metric correspondence the relationships between items and latent traits, and scalar correspondence the intercepts or thresholds that support latent mean comparisons. These are inferential conditions tested through models and substantive evidence, not binary decorations applied to a questionnaire. Davidov et al. (2014) reviews measurement equivalence in cross-national studies, and Asparouhov and Muthén (2014) develops an alignment method for estimating latent parameters when strict invariance is difficult. Operational design must preserve the sample and language documentation needed to interpret either approach.
Reference-group effects are a different source of distortion. A respondent rating personal assertiveness may compare themselves with locally salient peers; the same Likert category can encode different standards of self-evaluation. Heine et al. (2002) investigated this problem for cross-cultural subjective scales. When relevant, researchers can supplement self-reports with behavioral tasks, anchoring approaches, external criteria, and carefully constructed comparison prompts. These methods have distinct assumptions and should not be represented as automatic corrections. Cultural meaning may vary even when a statistical invariance test passes, so substantive interviews and convergent validation remain valuable alongside factor models.
The interpretation of a national difference depends on a chain of evidence: representative or deliberately bounded recruitment, equivalent measurement opportunities, defensible score models, and an estimand consistent with these decisions. An observed scale difference may reflect substantive experience, translation, response style, item meaning, sample composition, or several mechanisms together. The analysis must consider which alternatives were actually tested. van de Vijver and Tanzer (2004) identifies multiple types of cross-cultural bias, while Davidov et al. (2014) demonstrates why cross-country comparability must be established before treating latent mean rankings as direct evidence of psychological differences. A culture-specific dossier should integrate these conditions without duplicating the full mathematical theory of psychometrics.
Quasi-Experiments and Causal Identification
Country comparisons are frequently observational: people are not randomly assigned to nationality, schooling history, or childhood ecology. A difference between groups may therefore reflect the effect of a named cultural exposure, correlated institutions, selective migration, or unmeasured features of local environments. The first design task is to state a causal diagram or an equally precise explanation of proposed dependencies and define the intervention or exposure whose effect is sought. An association between national context and behavior is a valuable description, but the label attached to the country does not itself identify a manipulable cause. Yarkoni (2022) reminds researchers that the scope of any inference depends on the dimensions of variation represented by the design.
Within-site randomized experiments can establish the effect of a manipulation for participants and procedures studied, conditional on adequate implementation. Replicating that design across populations tests whether the effect changes across specified contexts. Natural experiments, policy discontinuities, longitudinal changes, and matched comparative cases may provide evidence about a cultural mechanism if assumptions are justified. In observational comparisons, adjustment for measured education or income cannot remove every unmeasured confounder, and adjusting for downstream consequences of an exposure may introduce new bias. Gelfand et al. (2011) exemplifies the ambition of linking national-level cultural variation to contextual accounts, while requiring caution about the identification of specific causes.
A comparative protocol should prespecify the primary contrast, potential moderators, measurement timing, selection mechanisms, and how deviations will be interpreted. When national contexts are few, it may be more scientifically transparent to report site-specific estimates and qualitative explanations than a heavily parameterized causal model. Strong causal claims require credible assignment, an externally imposed change, or a defensible set of identifying assumptions. Descriptive distributions, predictive associations, and causal effects are separate empirical products. van de Vijver and Leung (1997) remains relevant because it places cross-cultural quasi-experiments within a framework attentive to alternative explanations and the differences between cultural groups and experimentally assigned treatments.
Multilevel Modeling and Ecological Inference
Comparative data are usually hierarchical, and different scientific questions require different levels of analysis. A respondent-level coefficient addresses individual variability under model assumptions; a community-level association concerns contextual differences; a nation-level contrast summarizes a distribution over countries or sites. Combining all respondents in one ordinary regression while ignoring clustering can overstate precision, while analyzing country means alone can erase relevant individual heterogeneity. Researchers should specify which level owns the hypothesized mechanism and provide uncertainty calculations appropriate to the number of independent units. Gelfand et al. (2011) is a prominent multinational example in which country-level cultural characteristics are central substantive variables.
Multilevel models can estimate within-site associations and between-site differences, allow site-specific intercepts or slopes, and partially pool noisy estimates. Their value depends on the number, diversity, and selection of higher-level units and on assumptions about heterogeneity and residual dependence. A country random effect does not transform an opportunistic set of countries into a random sample of the world. Likewise, a regression coefficient for a country-level indicator should not be read as an individual mechanism without additional evidence. The conceptual concern over the population of inference in Yarkoni (2022) applies directly to these decisions, and the empirical cultural distance analysis in Muthukrishna et al. (2020) shows the importance of specifying what contextual dimension is represented.
Comparative reports should separate within-site, between-site, and compositional patterns where possible. For example, a cross-national relation between average education and task scores could coexist with a different relation among individuals within each country; such discrepancies are not mathematical contradictions because they concern different estimands. Contextual confounding, measurement inequivalence, and model extrapolation remain possible. Sensitivity analyses can vary site weights, individual weights, random-effect assumptions, and the inclusion of unusual sites. The objective is to recover a coherent scientific claim about persons in contexts rather than an apparently precise universal slope produced by collapsing the design hierarchy. Harkness et al. (2010) supplies the survey-design rationale for retaining this information.
Missingness, Nonresponse, Weights, and Sensitivity
Incomplete survey data can arise because a person was never covered by a frame, declined recruitment, skipped an item, discontinued a task, or was removed by a preprocessing rule. These mechanisms differ scientifically. A respondent who declines a sensitive item because of local norms may differ systematically from a respondent whose device failed. Weighting based on observed population margins can sometimes adjust unequal inclusion and differential response but cannot establish comparability on characteristics that remain unobserved. Researchers should present coverage exclusions separately from nonresponse and item missingness and should describe the assumptions supporting each adjustment. Harkness et al. (2010) provides a methodological foundation for documenting these survey operations across settings.
Design weights arise from selection probabilities, while nonresponse and calibration adjustments rely on additional models or benchmark information. If one country contributes a probability sample and another a self-selected online panel, applying weights to both does not erase the asymmetry in selection mechanisms. Comparative estimands must also specify whether every country is weighted equally, every person in the target superpopulation is weighted equally, or each site contributes under another justified rule. The numerical contrast can change when that choice changes. Rad et al. (2018) treats population representation as a substantive property of scientific evidence; it is therefore essential to disclose the rule by which selected respondents influence a reported cultural comparison.
Sensitivity analyses should examine plausible alternative definitions of eligible populations, recruitment exclusions, response assumptions, and score interpretation. A transparent report can compare complete-case estimates with justified imputation, test the influence of high-weight observations, and state the assumptions under which missing-not-at-random mechanisms might alter conclusions. Such analyses cannot make unknown selection processes disappear, but they reveal whether findings depend on fragile assumptions. Davidov et al. (2014) adds that weighting is no remedy for instruments whose scores mean different things across groups. Robustness is accordingly multi-dimensional: recruitment coverage, construct interpretation, statistical model choice, and actual response behavior must each be evaluated.
Statistical Power, Clusters, and Small Country Samples
Power in a multinational design depends on the estimand and its independent information units. For an individual-level experimental condition randomized inside many sites, precision is influenced by participant counts, within-site outcome variation, and effect heterogeneity. For a country-level association, the number and diversity of countries may be more consequential than the total number of individuals. A sample with tens of thousands of respondents across only a few countries cannot estimate rich cross-national variance components with the same precision as a design that observes many independent country contexts. Gelfand et al. (2011) provides a recognizable scale for multinational country comparisons, while Yarkoni (2022) articulates the generality problem behind such design choices.
Power calculations should account for cluster correlations, stratification, planned contrasts, multiple outcomes, missingness, and likely heterogeneity across sites. Simulations are often appropriate when interventions, sampling designs, and models interact in complicated ways. Investigators should state the minimum substantively important effect rather than treat any result detectable under an enormous aggregate N as intrinsically meaningful. Small cultural groups may be included to probe mechanisms or document variation even when their individual estimates are imprecise. In that case, interval estimates, site-specific graphical displays, and transparent descriptions are preferable to unsupported declarations of no effect. The multisite logic illustrated by Klein et al. (2018) helps distinguish aggregate replicability from universal invariance.
Many-country analyses also face a dependence problem when countries are selected by region, research network, language, or institutional capacity. Treating their selection as an independent random draw can produce optimistic uncertainty intervals. The number of countries should never substitute for a description of how those countries were chosen. A hierarchical model can partially pool observed sites but cannot recover unobserved regions without defensible extrapolation assumptions. The methodological objective is an uncertainty statement matched to actual sample construction. Muthukrishna et al. (2020) further demonstrates that geographic dispersion and psychological distance are different quantities, so an informative selection of contexts may matter more than an undifferentiated expansion in site count.
Multisite Replication and Heterogeneous Effects
Multisite replication is an especially informative approach when an experimental effect is claimed to hold across populations. Researchers harmonize a core protocol, preregister analysis, and repeat the test in settings whose relevant characteristics are recorded. The pooled result estimates an aggregate under a chosen weighting scheme; site-specific effects reveal heterogeneity and possible moderating conditions. An aggregate replication cannot by itself establish that every cultural population expresses the same magnitude, while a discrepant site does not by itself establish cultural causation. Klein et al. (2018) offers direct evidence from the Many Labs 2 project on the relationship between effect size and collection settings.
Many Labs 2 examined 28 published effects through preregistered replications conducted in 125 samples involving 15,305 participants from 36 countries and territories; different protocol sets were distributed across those samples. Using its conventional significance criterion, 15 of 28 effects showed evidence in the original direction. This is evidence about the selected effects and recruited study settings, not a representative estimate of the replication rate for psychology as a whole. The project explicitly reports variation in effect sizes and discusses the fact that the chosen effects and laboratories were not randomly sampled from all possible phenomena or sites. Its scale and transparent design make it a model for analyzing setting variation while exposing the limits of country count as a generalizability guarantee. Klein et al. (2018)
The inferential value of replication improves when a network distinguishes predictable contextual moderators from statistical noise. Site characteristics should be measured independently from outcome data and specified before comparing the directions of effects. When heterogeneity is substantial, a useful conclusion may be that a phenomenon occurs conditionally, within described institutional or task circumstances. Meta-analytic estimates with prediction intervals and investigation of protocol deviations may assist, but inference about the universe of sites remains limited by how they were selected. Camerer et al. (2018) supplies a related large-scale social-science replication effort, while Yarkoni (2022) makes explicit why generalization beyond stimuli, participants, occasions, and laboratories needs its own evidence.
Empirical Evidence from Global Surveys and Cultural Distance
Large comparative surveys can document distributions of values, attitudes, and other self-reported characteristics across many populations. Their scientific value depends on survey coverage, translation, recruitment, field dates, and instrument comparability. A country ranking may shift when a questionnaire is adapted, a particular subgroup is omitted, or the target population differs. Surveys can nevertheless illuminate variation that narrow university samples obscure, especially when they publish sampling procedures and allow analysts to assess the target population. Muthukrishna et al. (2020) used cross-national evidence to develop measures of cultural and psychological distance, illustrating the possibility of moving from binary cultural labels to measured multidimensional differences.
Muthukrishna and colleagues analyzed World Values Survey data drawn from a large number of respondents and societies, generating distance-based tools for characterizing populations relative to one another. Such analyses depend on the dimensions recorded by the survey: similarity on particular values cannot stand for identity across every aspect of psychological functioning. The measures are useful for selecting informative comparisons and examining how a research sample is positioned in the observed cultural space. Their uncertainty and interpretation still depend on survey representativeness, missing data, reference periods, and cross-language comparability. A deliberate design can use this information before fieldwork to choose contrasting sites rather than select them solely by geography. Muthukrishna et al. (2020)
A second example comes from the analysis of tight and loose cultural norms across 33 countries in Gelfand et al. (2011). That study links social norms and institutional environments to cross-national patterns, creating testable explanatory propositions. Its results invite further examination of the diversity inside each country, the source of selected contextual indicators, and the alternative causal pathways relating environment to reported norms. Cross-national surveys are particularly effective for describing systematic contextual covariation and generating mechanism-focused hypotheses, while strong causal interpretations require additional designs. The broader synthesis of Gurven (2018) emphasizes studying how ecological differences, social institutions, and cultural transmission can help explain observed psychological variation.
Empirical Evidence from Experiments and Behavioral Fieldwork
Cross-cultural behavioral studies can connect contextual theories with observed actions. Experiments examining reasoning, perception, or fairness often reveal useful variation in response patterns; equally important, they reveal how task conventions themselves enter the result. Norenzayan et al. (2002) compared formal and intuitive reasoning in selected cultural samples, complementing the earlier holistic–analytic research program of Nisbett et al. (2001). Such designs illuminate particular participants under specific instructions. Extending their results requires attention to education, age, task familiarity, sampling sites, and the difference between national membership and actual exposure to a relevant cognitive practice. Interpretation strengthens when subsequent work tests alternative task formulations and mechanisms rather than treating an initial contrast as definitive.
Field studies offer a different kind of ecological validity by observing decisions in familiar environments, but they also encounter uncontrolled contextual differences. The multinational civic-honesty study by Cohn et al. (2019) investigated the return of lost wallets using standardized field procedures in many settings. Its design exemplifies a behavioral outcome with immediate real-world relevance, while country- and city-level site choice, local institutions, employee behavior, and the meaning of the task remain part of interpretation. A field outcome can be measurable across settings without becoming a single measure of national character. The proper scientific claim describes the behavior sampled and the locations and circumstances in which it was observed.
Experimental and field evidence become more compelling when the same substantive question is tested through complementary procedures. Survey responses may characterize perceived norms, laboratory tasks may test controlled mechanisms, and field observation may assess behavior under ordinary institutional incentives. Agreement among methods increases confidence when their biases are sufficiently independent; disagreement can expose different estimands rather than methodological failure. Gelfand et al. (2011) offers a programmatic example of linking norms to national contexts, while Cohn et al. (2019) illustrates the distinct demands of behavioral field protocols. The resulting inference should specify which populations, settings, and behavioral domains have been observed and which remain outside the evidence.
Ethnographic Evidence and Mixed-Method Triangulation
Ethnography contributes information that a standardized instrument often cannot supply: how a practice is learned, which distinctions matter locally, and why the same task can recruit different expectations. Comparative researchers can use observation, interviews, and joint interpretation with community members to identify plausible mechanisms before constructing a broader survey. The selection of ethnographic sites is itself a sampling problem. Researchers should explain which practices, participants, and situations were observed and how contrary cases were sought. Levitt et al. (2018) emphasizes transparent qualitative reporting, including the relationship between researchers' positions, methods, evidence, and interpretation.
Mixed-method integration can be sequential or concurrent. In a sequential design, qualitative work informs item construction and the survey tests the distribution of documented practices; follow-up interviews then investigate surprising quantitative patterns. In a concurrent design, observed interactions and experimental task results may be analyzed in parallel against common hypotheses. The scientific value lies in tracing how separate forms of evidence constrain an explanation. It is inadequate to add a small illustrative interview after a quantitative study and claim general triangulation. Researchers must specify the inferential function of each source and document where the sources agree, contradict, or address different populations. Harkness et al. (2010) and van de Vijver and Leung (1997) provide complementary comparative and survey foundations for this integration.
Interpretation also requires attention to ordinary question-answering conventions. The form, order, and phrasing of survey questions influence what respondents retrieve and how they choose an answer. Schwarz (1999) reviews evidence that self-report is a contextual cognitive activity, making survey interpretation a psychological phenomenon in its own right. Interviewing and observation can detect some ambiguities but introduce their own interpersonal and analytic dependencies. Robust comparisons therefore examine possible method-specific sources of variation and document the evidence supporting a culturally equivalent reading of the results. Method diversity is strongest when it produces explicit tests of alternative interpretations.
Collaborative Fieldwork, Ethics, and Data Governance
Ethical cross-cultural research requires collaboration that begins with question formation. Local scholars and research partners can identify invalid assumptions, culturally inappropriate procedures, inaccessible recruitment routes, and harms that outside teams are unlikely to anticipate. They should receive substantive roles in research design, interpretation, authorship, and decisions about dissemination when their contributions warrant them. Engagement is not a substitute for individual consent, nor does the agreement of a community representative necessarily establish the views of all individuals in a population. Respect for autonomy, privacy, equitable participation, and scientifically valid interpretation must be negotiated in the actual social setting. Harkness et al. (2010) establishes how multinational field protocols require local expertise and coordinated quality control.
Translated consent forms, accessible explanations, confidentiality safeguards, secure storage, and clear withdrawal procedures are part of a defensible method. In small communities, the combination of age, occupation, location, and linguistic background may identify participants even when names are removed. Researchers should distinguish open metadata from potentially disclosive individual records and agree on controlled access where necessary. Sensitive cultural information can be governed by legitimate community-specific expectations while preserving enough methodological detail to assess the claims. International tests should not be exported or adapted casually when language inequivalence could cause a person or group to be misclassified. International Test Commission (2017) formalizes principles of responsible test adaptation that bear on such risks.
The institutional geography of science also affects what becomes a published result. Authors based in wealthy research centers may control the questions, data infrastructure, English-language dissemination, and editorial framing while field partners carry the recruitment burden. Improving representation consequently requires attention to the relationships through which evidence is produced and interpreted. Arnett (2008) documented the narrow institutional distribution of psychology research, and Rad et al. (2018) linked sample representation to the scope of psychological theory. Ethical collaboration and methodological validity are connected: researchers who understand local participation conditions can identify selection and measurement errors that would otherwise persist unnoticed.
Artificial Era and Comparative Human–Artificial Research
The Artificial Era introduces new settings in which people encounter language models, automated advisors, adaptive tutors, and other artificial cognitive systems. Research on human–artificial relations must account for the cultural and material conditions of access to these systems. A population recruited through an English-language online platform can differ from another population in device access, schooling, familiarity with conversational technology, regulatory conditions, and the functions assigned to artificial agents. The same digital interaction may therefore elicit different expectations of authority, privacy, reciprocity, or usefulness. Cross-cultural sampling principles apply directly to the human participants and institutional settings in which these interactions are observed. Henrich et al. (2010) and Rad et al. (2018) provide the general evidentiary basis for this caution.
Comparisons involving artificial systems must also define the unit of analysis on the artificial side. A model version, system prompt, interface language, retrieval corpus, and deployment policy can each change the observed behavior of an artificial agent. When the research question concerns a human response, these conditions should be controlled or explicitly measured across sites. When the research question concerns artificial behavior itself, sampling prompts, languages, tasks, interaction histories, and software versions becomes part of the inferential population. A narrow benchmark is a sample of tasks, not an exhaustive portrait of a system's capacities. The generalizability analysis in Yarkoni (2022) supplies a methodological analogy for extending inference across persons, circumstances, materials, and systems.
Psychology of Artificial can use cross-cultural comparative design to investigate whether human–artificial collaboration, perceived reliability, or attributed agency varies with local practices and institutions. Such studies should specify measurable constructs, behavioral indicators, and separately evaluated theoretical interpretations. The broader Artificial Era perspective motivates these research questions but does not itself establish universal psychological effects. Comparability across human and artificial systems raises distinctive construct questions beyond the scope of this dossier, particularly when artificial systems are compared with human populations. van de Vijver and Tanzer (2004) and Davidov et al. (2014) underscore why common terms or test items do not by themselves validate cross-system score equivalence.
Critical Evidence Appraisal and Future Research Program
The evidence reviewed here establishes a general methodological conclusion: psychological generalization depends on the people, contexts, questions, and procedures that generated the observations. Descriptive multi-country surveys provide distributional information under frame and measurement assumptions; laboratory experiments can identify effects of controlled manipulations in recruited samples; multisite replications test stability across selected settings; ethnography clarifies local meaning and selection processes. Each supplies important evidence, yet none alone guarantees unrestricted inference to all human populations. Henrich et al. (2010), Gurven (2018), Rad et al. (2018), and Yarkoni (2022) converge on the need to define the relevant universe of persons and circumstances explicitly.
Several open scientific problems remain. Researchers need feasible sampling frames for highly mobile and undercounted populations; methods for evaluating measurement equivalence in small linguistic groups; designs that connect local causal mechanisms with cross-site heterogeneity; strategies for allocating limited resources between additional persons and additional independent contexts; and approaches to distinguishing cultural practices from correlated education, institutional, and ecological conditions. Important replication projects such as Klein et al. (2018) help characterize setting variation but also inherit selected effects and participating laboratories. Cross-country observational results such as Gelfand et al. (2011) raise explanatory questions that require longitudinal, comparative, or intervention evidence to clarify mechanisms.
A constructive research program can prespecify multiple inferential targets: a tightly defined recruited population, a frame-covered local population, a set of deliberately selected cultural ecologies, and a broader target whose coverage can be justified. It should publish protocols and adaptations, monitor response at every stage, use independent local expertise, distinguish samples from nations, and present heterogeneity alongside pooled results. When original data cannot support a global claim, an explicitly bounded conclusion is a substantive scientific achievement because its boundary can guide the next study. The methodological synthesis of Harkness et al. (2010), van de Vijver and Leung (1997), and Levitt et al. (2018) offers tools for producing that cumulative record.
Conclusion
Cross-cultural comparative design and population sampling jointly determine the meaning and scope of psychological evidence. The decisive operations are defining the target population, constructing an adequate frame, selecting informative sites, measuring contextual variation, supporting language and task comparability, and selecting an analytic model consistent with the resulting hierarchy. These decisions govern whether an apparent cultural difference describes enrolled participants, a frame-covered population, or a theoretically justified contrast across contexts. Arnett (2008) and Henrich et al. (2010) established the historical urgency of this question in a discipline whose core evidence has often been drawn from narrow research environments.
Empirical comparisons now provide increasingly diverse forms of evidence: multilingual surveys, field experiments, replication networks, context-sensitive measurements, and collaborative field methods. Each extends the possible scope of inference while introducing particular design requirements. Klein et al. (2018) demonstrates that extensive multisite replication can quantify the stability and heterogeneity of selected effects; Muthukrishna et al. (2020) shows how cultural distances can be characterized beyond simple country labels; Cohn et al. (2019) exemplifies behavioral field comparisons; and Davidov et al. (2014) demonstrates why score equivalence matters to cross-country interpretation. Taken together, these studies support disciplined comparison rather than a universal recipe for choosing populations.
The continuing objective is a psychology capable of describing and explaining diverse human lives with explicit, reproducible evidence. Research quality grows when the population of inference is declared, local differences are investigated rather than presumed, selection limitations remain visible, and theoretical conclusions track the evidence actually gathered. The Artificial Era adds human–artificial interaction as another domain in which careful population definition and contextual comparison will be required. A rigorous comparative science therefore treats sampling and design as constitutive parts of psychological explanation, documenting the populations, settings, instruments, and assumptions through which its claims become warranted. Rad et al. (2018) Yarkoni (2022)
References
- Arnett, J. J. (2008). The neglected 95%: Why American psychology needs to become less American. American Psychologist, 63(7), 602–614. https://doi.org/10.1037/0003-066X.63.7.602
- Asparouhov, T., & Muthén, B. (2014). Multiple-group factor analysis alignment. Structural Equation Modeling: A Multidisciplinary Journal, 21(4), 495–508. https://doi.org/10.1080/10705511.2014.919210
- Camerer, C. F., Dreber, A., Holzmeister, F., Ho, T.-H., Huber, J., Johannesson, M., Kirchler, M., Nave, G., Nosek, B. A., Pfeiffer, T., Altmejd, A., Buttrick, N., Chan, T., Chen, Y., Forsell, E., Gampa, A., Heikensten, E., Hummer, L., Imai, T., ... Wu, H. (2018). Evaluating the replicability of social science experiments in Nature and Science between 2010 and 2015. Nature Human Behaviour, 2, 637–644. https://doi.org/10.1038/s41562-018-0399-z
- Cohn, A., Maréchal, M. A., Tannenbaum, D., & Zünd, C. L. (2019). Civic honesty around the globe. Science, 365(6448), 70–73. https://doi.org/10.1126/science.aau8712
- Davidov, E., Meuleman, B., Cieciuch, J., Schmidt, P., & Billiet, J. (2014). Measurement equivalence in cross-national research. Annual Review of Sociology, 40, 55–75. https://doi.org/10.1146/annurev-soc-071913-043137
- Gelfand, M. J., Raver, J. L., Nishii, L., Leslie, L. M., Lun, J., Lim, B. C., Duan, L., Almaliach, A., Ang, S., Arnadottir, J., Aycan, Z., Boehnke, K., Boski, P., Cabecinhas, R., Chan, D., Chhokar, J., D'Amato, A., Ferrer, M., Fischlmayr, I. C., ... Yamaguchi, S. (2011). Differences between tight and loose cultures: A 33-nation study. Science, 332(6033), 1100–1104. https://doi.org/10.1126/science.1197754
- Ghai, S. (2021). It's time to reimagine sample diversity and retire the WEIRD dichotomy. Nature Human Behaviour, 5, 971–972. https://doi.org/10.1038/s41562-021-01175-9
- Gurven, M. (2018). Broadening horizons: Sample diversity and socioecological theory are essential to the future of psychological science. Proceedings of the National Academy of Sciences, 115(45), 11420–11427. https://doi.org/10.1073/pnas.1720433115
- Harkness, J., Pennell, B.-E., & Schoua-Glusberg, A. (2004). Survey questionnaire translation and assessment. In S. Presser et al. (Eds.), Methods for testing and evaluating survey questionnaires. Wiley. https://doi.org/10.1002/0471654728.ch22
- Harkness, J. A., Braun, M., Edwards, B., Johnson, T. P., Lyberg, L., Mohler, P. P., Pennell, B.-E., & Smith, T. W. (Eds.). (2010). Survey methods in multinational, multiregional, and multicultural contexts. Wiley. https://doi.org/10.1002/9780470609927
- Heine, S. J., Lehman, D. R., Peng, K., & Greenholtz, J. (2002). What's wrong with cross-cultural comparisons of subjective Likert scales? The reference-group effect. Journal of Personality and Social Psychology, 82(6), 903–918. https://doi.org/10.1037/0022-3514.82.6.903
- Henrich, J., Heine, S. J., & Norenzayan, A. (2010). The weirdest people in the world? Behavioral and Brain Sciences, 33(2–3), 61–83. https://doi.org/10.1017/S0140525X0999152X
- Hruschka, D. J., Medin, D. L., Rogoff, B., & Henrich, J. (2018). Pressing questions in the study of psychological and behavioral diversity. Proceedings of the National Academy of Sciences, 115(45), 11366–11368. https://doi.org/10.1073/pnas.1814733115
- International Test Commission. (2017). The ITC guidelines for translating and adapting tests (2nd ed.). International Test Commission. https://www.intestcom.org/files/guideline_test_adaptation_2ed.pdf
- Klein, R. A., Vianello, M., Hasselman, F., Adams, B. G., Adams, R. B., Alper, S., Aveyard, M., Axt, J. R., Babalola, M. T., Bahník, Š., et al. (2018). Many Labs 2: Investigating variation in replicability across samples and settings. Advances in Methods and Practices in Psychological Science, 1(4), 443–490. https://doi.org/10.1177/2515245918810225
- Levitt, H. M., Bamberg, M., Creswell, J. W., Frost, D. M., Josselson, R., & Suárez-Orozco, C. (2018). Journal article reporting standards for qualitative primary, qualitative meta-analytic, and mixed methods research in psychology: The APA Publications and Communications Board task force report. American Psychologist, 73(1), 26–46. https://doi.org/10.1037/amp0000151
- Muthukrishna, M., Bell, A. V., Henrich, J., Curtin, C. M., Gedranovich, A., McInerney, J., & Thue, B. (2020). Beyond Western, educated, industrial, rich, and democratic (WEIRD) psychology: Measuring and mapping scales of cultural and psychological distance. Psychological Science, 31(6), 678–701. https://doi.org/10.1177/0956797620916782
- Nielsen, M., Haun, D., Kärtner, J., & Legare, C. H. (2017). The persistent sampling bias in developmental psychology: A call to action. Journal of Experimental Child Psychology, 162, 31–38. https://doi.org/10.1016/j.jecp.2017.04.017
- Nisbett, R. E., Peng, K., Choi, I., & Norenzayan, A. (2001). Culture and systems of thought: Holistic versus analytic cognition. Psychological Review, 108(2), 291–310. https://doi.org/10.1037/0033-295X.108.2.291
- Norenzayan, A., Smith, E. E., Kim, B. J., & Nisbett, R. E. (2002). Cultural preferences for formal versus intuitive reasoning. Cognitive Science, 26(5), 653–684. https://doi.org/10.1207/s15516709cog2605_4
- Rad, M. S., Martingano, A. J., & Ginges, J. (2018). Toward a psychology of Homo sapiens: Making psychological science more representative of the human population. Proceedings of the National Academy of Sciences, 115(45), 11401–11405. https://doi.org/10.1073/pnas.1721165115
- Schwarz, N. (1999). Self-reports: How the questions shape the answers. American Psychologist, 54(2), 93–105. https://doi.org/10.1037/0003-066X.54.2.93
- Sears, D. O. (1986). College sophomores in the laboratory: Influences of a narrow data base on social psychology's view of human nature. Journal of Personality and Social Psychology, 51(3), 515–530. https://doi.org/10.1037/0022-3514.51.3.515
- van de Vijver, F. J. R., & Leung, K. (1997). Methods and data analysis for cross-cultural research. SAGE Publications. https://us2.sagepub.com/en-us/nam/methods-and-data-analysis-for-cross-cultural-research/book6080
- van de Vijver, F. J. R., & Tanzer, N. K. (2004). Bias and equivalence in cross-cultural assessment: An overview. European Review of Applied Psychology, 54(2), 119–135. https://doi.org/10.1016/j.erap.2003.12.004
- Yarkoni, T. (2022). The generalizability crisis. Behavioral and Brain Sciences, 45, e1. https://doi.org/10.1017/S0140525X20001685