Abstract
Background: Person–organization (P–O) fit is among the most robust predictors of job satisfaction, commitment, and retention, yet prevailing assessment practice relies on low-dimensional models — the Big Five, HEXACO, and four-quadrant culture typologies — that necessarily aggregate finer-grained facet and contextual variance known to carry incremental predictive value.
Purpose: This conceptual article proposes a dual-assessment architecture in which organizations and applicants are profiled on a substantially expanded, commensurable set of culture, personality, and contextual dimensions. Unlike an empirical validation study, this article does not report new data; it specifies an architecture, identifies the conditions under which higher dimensionality would plausibly improve on current practice, and — critically — examines the conditions under which it would not.
Approach: We review the theoretical basis for expanded-dimensionality assessment in facet-level personality and organizational culture research, then subject the proposal to the objections it must survive to be scientifically credible: statistical overfitting from a high ratio of dimensions to validation sample size, construct redundancy (the “jangle fallacy”), the likelihood that additional self-report items increase rather than reduce opportunities for impression management in high-stakes settings, and the practical burden a large assessment battery places on applicants and on organizations that must maintain and revalidate it as culture drifts.
Key considerations: We conclude that higher-dimensional assessment is theoretically motivated but not automatically superior to current practice; its value is an empirical question that has not yet been tested at the scale proposed, and its deployment carries legal exposure illustrated by ongoing litigation over algorithmic hiring tools. We therefore present the framework as a research agenda and a set of falsifiable design commitments — explainability, adverse-impact monitoring, staged validation, and bounded dimensionality — rather than as a validated replacement for existing instruments.
Conclusion: A scientifically responsible path toward richer P–O fit assessment exists, but it runs through incremental, validated expansion of current models rather than a one-step move to a large, unvalidated dimensional space.
Keywords: person–organization fit; organizational culture; personality assessment; algorithmic hiring; adverse impact; overfitting; construct validity
1. Introduction
The alignment between individual attributes and organizational environments — person–organization (P–O) fit — is a foundational construct in organizational psychology, and meta-analytic evidence links it to job satisfaction, commitment, citizenship behavior, and turnover intention (Kristof-Brown et al., 2005; Mardiyah & Sartika, 2025). Yet the instruments that dominate research and practice remain low-dimensional: the Big Five, HEXACO, and the Competing Values Framework (CVF) reduce personality and culture to five, six, and four dimensions respectively. These models are parsimonious and well validated (Barrick & Mount, 1991; Ashton & Lee, 2007; Quinn & Rohrbaugh, 1983), but parsimony discards variance. Facet-level personality measures add incremental validity over broad traits for specific performance criteria (Dudley et al., 2006), and organizational culture researchers have long catalogued well over one hundred distinct culture dimensions in the literature, far more than any single typology represents (Ott, 1989; Van der Post et al., 1997).
This article asks whether computational advances — natural language processing, machine learning, high-dimensional matching — make it scientifically responsible to close this gap by assessing organizations and applicants on a much larger, commensurable set of dimensions. Our answer is conditional. The information-loss argument for expansion is real and well supported at the facet level. But expansion is not a free upgrade: it introduces statistical, psychometric, and practical risks that a credible proposal must confront directly rather than defer to future validation work. This article is accordingly framed as a conceptual and architectural contribution and a research agenda, not as an empirically validated model — no new instrument is administered, and no new predictive data are reported here. We view this as a precondition for scientific credibility rather than a limitation to be minimized: a proposal that outruns its evidence is more useful to the field when labeled as such.
Section 2 reviews the theoretical foundations of P–O fit and current measurement practice. Section 3 develops — and then interrogates — the case for higher-dimensional assessment. Section 4 specifies a candidate dual-assessment architecture. Section 5 addresses feasibility and burden, a consideration largely absent from prior proposals in this space. Section 6 addresses validity, fairness, and the legal landscape, including recent litigation over algorithmic hiring tools. Section 7 discusses implications, and Section 8 sets out a staged research agenda rather than a deployment roadmap.
2. Theoretical Foundations
2.1 Person–Organization Fit: Conceptualization and Evidence
P–O fit is most commonly conceptualized as the congruence between individual and organizational values (Kristof, 1996), formalized in the attraction–selection–attrition framework: individuals are drawn to, selected by, and retained within organizations whose values resemble their own. Kristof-Brown et al. (2005) reported meta-analytic corrected correlations between P–O fit and job satisfaction (ρ = .44), organizational commitment (ρ = .51), and turnover intention (ρ = −.35); more recent meta-analytic work confirms that P–O and person–job fit jointly predict turnover intention across sectors (Mardiyah & Sartika, 2025). Fit is also multidimensional — individuals experience distinguishable fit with their job, organization, group, and supervisor (Kristof-Brown et al., 2005) — which motivates assessing not only global cultural alignment but fit with the specific team and role an applicant will actually enter.
2.2 Personality Assessment in Hiring: Models and Validity
The Big Five (Openness, Conscientiousness, Extraversion, Agreeableness, Emotional Stability) has decades of meta-analytic support, with Conscientiousness the most consistent cross-criterion predictor (Barrick & Mount, 1991; Sackett et al., 2022). HEXACO adds Honesty–Humility, which shows incremental validity for counterproductive work behavior, and recent reviews find HEXACO explains more performance variance than the Big Five or Dark Triad (Ashton & Lee, 2007; Sackett et al., 2022). Critically, broad traits are not the optimal grain size for prediction: facet-level Conscientiousness measures — orderliness, self-discipline, deliberation — add incremental validity over the broad factor because the broad factor aggregates facets with heterogeneous, sometimes opposing, relationships to specific criteria (Dudley et al., 2006).
A separate and important complication is that self-report validity is not fixed: it appears to be attenuated in high-stakes relative to low-stakes settings, with the degree of attenuation varying by trait (Wiernik et al., 2025) — a finding we return to in Section 3.3, since it bears directly on whether adding more self-report dimensions to a hiring assessment is likely to help or hurt. Alternative methods (conditional reasoning tests, construct-driven situational judgment tests, AI-based scoring of open-ended language) are proposed as lower-fakability substitutes, but their criterion validity for job performance remains uneven and requires further field validation before being treated as solved (Sackett et al., 2025).
2.3 Organizational Culture: Frameworks and Measurement
Culture has been assessed through several traditions with distinct dimensional architectures: Chatman's (1991) Organizational Culture Profile (OCP), using value dimensions such as innovation and outcome orientation in a profile-comparison design that parallels P–O fit logic directly; and the Competing Values Framework, which crosses flexibility–stability and internal–external axes into four culture types — Clan, Adhocracy, Market, Hierarchy (Quinn & Rohrbaugh, 1983). The CVF is broadly applied and shows adequate psychometrics in cross-cultural adaptation, though its axis orthogonality and overlap with other instruments remain open questions (Mistry, 2025). More recent industry frameworks report expanded architectures with additional subdimensions and derived culture types (SHRM, 2026); we treat figures from single industry reports, however specific, as illustrative rather than as established facts pending independent replication, consistent with ordinary standards of evidence for a claim not yet appearing in peer-reviewed form. Yoo et al. (2025) validated a Korean-language OCP and found its dimensions differentially predict creative behavior, work–life balance, and thriving at work — direct evidence that collapsing culture into a handful of types discards outcome-relevant variance, the organization-side analogue of the facet argument in Section 2.2. Ott (1989) catalogued well over one hundred dimensions associated with organizational culture (Van der Post et al., 1997); this figure, not an a priori target, is the empirical anchor for the “up to roughly 100 dimensions” framing used later in this article — it describes the size of the space the literature has identified, not a claim that all one hundred should be operationalized in any given deployment.
2.4 Trait Activation and Situational Moderators
Tett and Burnett's (2003) trait activation model holds that personality is expressed in behavior only to the degree that situational cues activate the relevant trait: an agreeable person may thrive on a collaborative team and struggle in a zero-sum, individually incentivized one. This has a direct architectural implication, developed in Section 4: matching on person-side traits alone is insufficient. The situational, role-level cues that activate or suppress those traits must be measured on the organization side at comparable specificity, or the interaction the theory predicts cannot be modeled at all.
3. The Case for — and the Limits of — High-Dimensional Assessment
3.1 The Information-Loss Argument
The affirmative case is straightforward. Broad personality factors aggregate facets with heterogeneous criterion relationships (Dudley et al., 2006); a four-type culture model aggregates the more than one hundred dimensions the culture literature has catalogued (Ott, 1989); and organization-level culture types obscure differential relationships between specific cultural dimensions and specific outcomes such as creativity or thriving at work (Yoo et al., 2025). A four-type culture framework and a five-factor personality model together yield at most twenty cross-dimensional comparisons for a matching algorithm to draw on; a much larger, validated dimensional space would in principle yield many more. Whether that additional resolution translates into better real-world matches, however, is a distinct empirical question from whether it is theoretically motivated — and it is the question Section 3.3 takes up.
3.2 Computational Feasibility
The practical objection that high-dimensional assessment is computationally intractable no longer holds. NLP methods can extract culture-relevant signal from employee reviews, mission statements, and internal communications with construct validity comparable to survey-based measures in some domains (Yoo et al., 2025); machine learning models can generate personality estimates from written or spoken material, though convergence with self-report and criterion validity for performance both remain modest and require further field validation rather than assumption (Sackett et al., 2025). Dimensionality-reduction techniques (PCA, factor analysis, autoencoders) and standard distance metrics (cosine, Mahalanobis) operate efficiently at the scale proposed here. The remaining barrier is not computation but validation, governance, and regulatory compliance — the subjects of Sections 5 and 6.
3.3 Why More Dimensions Is Not Automatically Better
A scientifically responsible proposal must confront three specific risks that scale with dimensionality, none of which the facet-level literature cited above resolves, because that literature supports adding a handful of validated facets to a five-factor model, not scaling to on the order of one hundred dimensions per side.
First, overfitting. A matching architecture with roughly one hundred dimensions on each side generates on the order of ten thousand cross-dimensional terms. Estimating which of these terms carry real predictive signal, rather than fitting noise in a finite validation sample, requires sample sizes that scale with the number of parameters being estimated — a requirement most single-organization or even single-industry validation efforts will struggle to meet. Without it, a high-dimensional model risks appearing more precise than a low-dimensional one while actually being less reliable out of sample.
Second, construct redundancy, sometimes called the jangle fallacy: assigning different names to dimensions that are, statistically, largely the same underlying construct. A one-hundred-dimension inventory assembled by combining multiple existing taxonomies is at meaningful risk of this problem unless the dimensions are jointly factor-analyzed and shown to be at least partially separable, a step no prior proposal in this space has reported taking. Absent that step, apparent dimensional richness may be substantially illusory.
Third, faking dynamics may worsen, not improve, with scale. Section 2.2 noted that self-report validity is already attenuated in high-stakes settings (Wiernik et al., 2025). Expanding a self-report battery from five broad traits to dozens of narrower, more transparently job-relevant dimensions plausibly makes socially desirable responding easier, not harder, because narrower items make the “correct” answer more obvious to a motivated applicant. Any expansion of self-report dimensionality in a selection context should therefore be paired with, and partly justified by, a credible reduction in fakability — for instance through the situational-judgment or AI-based methods described in Section 2.2 — rather than treated as a benefit independent of assessment method.
None of these three risks is fatal to the case for expansion; each is a design requirement. Section 4 accordingly treats bounded, staged, and jointly-validated dimensionality as a constraint on the architecture rather than an afterthought.
4. A Proposed Dual-Assessment Architecture
4.1 Overview and Scope of the Proposal
The architecture rests on one principle: both sides of the fit equation should be assessed on commensurable dimensions at the specificity needed to capture the sources of fit and misfit that matter for outcomes, without exceeding what a given deployment can validate. It comprises organizational diagnosis, applicant diagnosis, and matching, and it is instrument-agnostic — organizations may use validated surveys, text-based assessment, behavioral data, or combinations, provided the dimensions on both sides are conceptually aligned and psychometrically examined jointly, not assembled from unrelated taxonomies and assumed to compose cleanly.
4.2 Dimensional Architecture
We propose three tiers, with dimensionality treated as a variable to be earned through validation rather than fixed in advance. Tier 1, foundational dimensions, comprises the Big Five or HEXACO factors on the individual side and CVF types or OCP values on the organizational side; these anchor the model in existing validated theory and serve as a consistency check on the higher tiers. Tier 2, contextual dimensions, captures the situational moderators trait-activation theory requires: individual work-style and motivational preferences (autonomy, mastery, structure, pace) paired with organizational decision-making, communication, reward, and leadership characteristics. Tier 2 dimensions should be selected through job analysis, not assembled generically, and weighted by role-specific relevance. Tier 3, emergent dimensions, are derived post hoc through factor analysis or clustering of assessment data rather than specified a priori; they function as a diagnostic on the completeness of Tiers 1 and 2 and should not be used in live selection decisions until independently validated against outcomes. The literature-derived figure of roughly one hundred dimensions (Ott, 1989) describes an upper bound on the space across all three tiers combined, not a per-deployment target — a specific implementation should include only as many dimensions across all tiers as its validation sample can support, per the overfitting concern in Section 3.3.
4.3 Organizational Diagnosis
Organizational diagnosis should combine survey instruments (OCP, CVF), text-based NLP extraction from internal communications and public materials, and structured behavioral observation, conducted at the team and role level rather than only the organization level, since subcultural variation within an organization can exceed variation between organizations. The goal is to capture enacted, not merely espoused, culture.
4.4 Applicant Diagnosis
Applicant diagnosis should combine self-report on foundational and select contextual dimensions with lower-fakability methods — situational judgment tests, conditional reasoning tests, and, where independently validated for the specific criterion, AI-based scoring of interview or written material (Sackett et al., 2025). Convergence across methods provides stronger evidence of trait standing than any single method; divergence is itself diagnostic and should trigger review rather than automatic aggregation.
4.5 Matching Procedure
Matching proceeds through hard filters on non-negotiable requirements, followed by dimension-weighted soft matching that specifies, per dimension, whether similarity or complementarity is the target (a team needing detail orientation may be better served by a complementary rather than similar hire), and aggregation into a two-sided score — organization-fit and candidate-fit — combined so that neither side can be optimized at the other's expense.
4.6 Explainability and Candidate Experience
Every match score should be accompanied by the specific dimensions driving it, the dimensions of mismatch, and a stated confidence level per dimension, since dimensions built on smaller item sets or newer validation will carry more uncertainty than long-established ones. Candidates should be able to access, correct, and contest their profile, and participation should be opt-in on both sides.
5. Feasibility, Burden, and Cost
A dimensional architecture this large is not cost-free, and a credible proposal should say so rather than treat feasibility as solved once computation is available. On the applicant side, longer assessment batteries increase fatigue and drop-off, which itself introduces sampling bias toward applicants with more time, patience, or test-taking comfort — a fairness concern independent of the adverse-impact issues discussed in Section 6.2. On the organizational side, culture is not static: dimensions validated at deployment can drift as an organization grows, restructures, or changes leadership, meaning a one-hundred-dimension organizational profile requires an ongoing revalidation cadence, not a single diagnostic exercise, or its accuracy will decay silently. Smaller organizations in particular may lack the applicant volume needed to validate even a fraction of Tier 2 or Tier 3 dimensions locally, which argues for pooled, industry-level validation of contextual dimensions before individual-organization deployment. These constraints do not argue against the architecture; they argue for a staged rollout — Tier 1 first, with Tier 2 added dimension-by-dimension only as each clears a predefined validation bar — rather than a full one-hundred-dimension launch.
6. Validity, Fairness, and Legal Considerations
6.1 Validity
Construct validity requires factor-analytic support for the proposed dimensional structure — ideally a joint analysis across Tier 1 and Tier 2 items to test for the redundancy risk in Section 3.3 — plus convergent, discriminant, and measurement-invariance evidence across demographic groups. Criterion validity requires the dimensional scores to predict performance, satisfaction, fit, and retention; incremental validity requires demonstrating predictive value beyond existing low-dimensional instruments specifically, not merely beyond chance. Given the overfitting risk identified above, validation samples should be pre-registered and powered for the number of parameters in the specific deployment, not for the architecture's theoretical maximum.
6.2 Adverse Impact
Personality assessment in hiring carries documented adverse-impact risk. Mean group differences on broad traits are generally small, but facet-level differences can be larger and can translate into differential selection rates (Foldes et al., 2008); broad traits tend to produce less adverse impact than facets even as facets improve prediction, a direct trade-off the architecture must manage explicitly rather than assume away. Selection rates by protected group should be monitored at every stage, with any detected disparity traced to a specific dimension, method, or weight and remediated — which is easier in an explainable, dimension-scored system than in an opaque composite score, one genuine advantage of the proposed architecture over many existing algorithmic tools.
6.3 Legal Landscape
The legal risk of algorithmic hiring tools at this scale is not hypothetical. In Mobley v. Workday, Inc. (N.D. Cal. No. 23-cv-00770-RFL), a federal court denied Workday's motion to dismiss claims that its applicant-recommendation system disparately impacted candidates by age, race, and disability, and in May 2025 granted preliminary certification of a nationwide collective action under the Age Discrimination in Employment Act covering applicants aged 40 and over rejected since September 2020. Court filings put the scale of the underlying system at roughly 1.1 billion processed applications over the relevant period, and the court held that a vendor whose system actively scores and ranks candidates — rather than merely applying an employer's stated criteria — can be held liable as the employer's agent under federal anti-discrimination law. This is directly relevant to the architecture proposed here: a matching system that actively weights and ranks candidates on dozens of algorithmically-scored dimensions sits closer to the liability profile at issue in Mobley than a system that merely reports scores for human review, which argues for keeping a human decision-maker meaningfully in the loop rather than allowing the match score to function as a de facto hiring decision. New York City's Local Law 144, requiring annual independent bias audits of automated employment decision tools, and the traditional Uniform Guidelines on Employee Selection Procedures framework for adverse-impact analysis both apply, though neither was written with a hundred-dimension matching system in mind, and organizations should treat compliance as requiring bespoke legal review rather than a checklist exercise.
6.4 Fairness, Transparency, and Privacy
Beyond legal compliance, candidates should know which dimensions are assessed, how they are weighted, and how a match score is derived; an architecture that cannot produce this explanation for an individual candidate or an external auditor should not be used for high-stakes selection decisions, regardless of its predictive performance. Assessment methods must accommodate candidates with disabilities and neurodivergent candidates, with accommodations available on request. The scale of personal data generated by a hundred-dimension assessment also raises data-minimization questions the architecture should address directly: dimensions that are collected but not used in a given role's weighting scheme arguably should not be collected for that role at all.
7. Discussion
The theoretical contribution of this framework is to extend P–O fit theory in a specific, testable direction: modeling fit at the grain of individual dimensions and their situational moderators, consistent with trait activation theory (Tett & Burnett, 2003), rather than as a single global score. Its practical value, if the validation program in Section 8 bears out, would be more accurate matching and richer, more actionable feedback to both organizations and candidates than current typologies provide. But the discussion in Sections 3.3, 5, and 6 also establishes what the framework is not: it is not a proven improvement over existing five- and four-dimension models, and treating it as one before that evidence exists would repeat, at greater scale, the overfitting and redundancy risks the framework is meant to solve. The most defensible reading of the evidence assembled here is that expansion beyond current models is theoretically well-motivated and technically feasible, but that its net benefit over current practice, once overfitting, redundancy, faking, and burden are accounted for, is genuinely unknown and should be established empirically before broad deployment.
8. Limitations and a Research Agenda
This is a conceptual article; it proposes an architecture and does not report new validation data, and its central claims about the value of expanded dimensionality should be read as hypotheses rather than findings. We propose a staged empirical agenda rather than a deployment roadmap. Stage one would validate a modest expansion (Tier 1 plus a small, job-analysis-derived set of Tier 2 dimensions, on the order of ten to twenty total) against real hiring and retention outcomes in a single industry, with pre-registered adverse-impact monitoring, to establish whether even modest expansion improves on current low-dimensional models before any move toward the full architecture. Stage two would test for construct redundancy across combined Tier 1/Tier 2 item pools via joint factor analysis, addressing the jangle-fallacy risk directly. Stage three would compare self-report versus lower-fakability methods (situational judgment, AI-scored language) for the same dimensions in matched high-stakes and low-stakes samples, to test whether the faking-attenuation concern in Section 3.3 is confirmed or whether alternative methods resolve it. Only evidence from stages of this kind, not architectural elegance alone, should determine whether dimensionality is expanded further toward the upper bound the culture literature has identified. Cross-cultural generalizability of both personality and culture dimensions is a further open question the agenda above does not yet address and that should precede any claim of international applicability.
9. Conclusion
Person–organization fit assessment is genuinely constrained by low-dimensional instruments, and the facet-level and organizational-culture literatures give real reason to expect that richer assessment could improve on current practice. But dimensionality is not a benefit that accrues automatically with scale: overfitting, construct redundancy, faking dynamics, and practical burden all worsen as dimensionality grows, and none of the prior proposals in this space — including earlier drafts of this one — addressed them directly. The contribution offered here is an architecture designed around those constraints from the outset, paired with a staged, falsifiable research agenda and an explicit acknowledgment that its central premise remains to be tested. That, rather than a specific dimension count, is what would make high-dimensional P–O fit assessment a scientific advance rather than a plausible-sounding one.
References
American Bar Association. (2026). Adverse impact analysis in applicant flow: The challenge of artificial intelligence tools. American Bar Association.
Ashton, M. C., & Lee, K. (2007). Empirical, theoretical, and practical advantages of the HEXACO model of personality structure. Personality and Social Psychology Review, 11(2), 150–166.
Barrick, M. R., & Mount, M. K. (1991). The Big Five personality dimensions and job performance: A meta-analysis. Personnel Psychology, 44(1), 1–26.
Chatman, J. A. (1991). Matching people and organizations: Selection and socialization in public accounting firms. Administrative Science Quarterly, 36(3), 459–484.
Dudley, N. M., Orvis, K. A., Lebiecki, J. E., & Cortina, J. M. (2006). A meta-analytic investigation of conscientiousness in the prediction of job performance. Journal of Applied Psychology, 91(1), 40–57.
Foldes, H. J., Duehr, E. E., & Ones, D. S. (2008). Group differences in personality: Meta-analyses comparing five U.S. racial groups. Personnel Psychology, 61(3), 579–616.
Kristof, A. L. (1996). Person-organization fit: An integrative review of its conceptualizations, measurement, and implications. Personnel Psychology, 49(1), 1–49.
Kristof-Brown, A. L., Zimmerman, R. D., & Johnson, E. C. (2005). Consequences of individuals' fit at work: A meta-analysis. Personnel Psychology, 58(2), 281–342.
Mardiyah, K. F., & Sartika, D. (2025). Meta analysis study: Person-organization fit and person-job fit on turnover intention. EKOMBIS REVIEW, 13(2), 7123.
Mistry, J. (2025). Examining the correlation between the Competing Values Framework and other scales of measure. The Plymouth Student Scientist, 18(2), Article 17.
Mobley v. Workday, Inc., No. 23-cv-00770-RFL (N.D. Cal.) (order granting preliminary certification May 16, 2025).
Ott, J. S. (1989). The organizational culture perspective. Dorsey Press.
Quinn, R. E., & Rohrbaugh, J. (1983). A spatial model of effectiveness criteria: Towards a competing values approach to organizational analysis. Management Science, 29(3), 363–377.
Sackett, P. R., Zhang, C., Berry, C. M., & Lievens, F. (2022). Revisiting meta-analytic estimates of validity in personnel selection. Journal of Applied Psychology, 107(11), 2040–2068.
Sackett, P. R., et al. (2025). Novel methods of personality assessment in organizational psychology: From innovative psychological tests to AI-based approaches. In Reference Module in Neuroscience and Biobehavioral Psychology. Elsevier.
SHRM. (2026). 2026 Global Workplace Culture Report. Society for Human Resource Management. [Figures cited from this report are treated as illustrative pending independent replication; see Section 2.3.]
Tett, R. P., & Burnett, D. D. (2003). A personality trait-based interactionist model of job performance. Journal of Applied Psychology, 88(3), 500–517.
Van der Post, W. Z., de Coning, T. J., & Smit, E. v. d. M. (1997). An instrument to measure organizational culture. South African Journal of Business Management, 28(4), 147–168.
Wiernik, B. M., et al. (2025). Personality test validity differs between low-stakes and high-stakes employment settings. International Journal of Selection and Assessment, 33(1), e12456.
Workforce Bulletin. (2025). Artificial Intelligence Bias: Harper v. Sirius XM challenges algorithmic discrimination in hiring. Proskauer Rose LLP.
Yoo, S., et al. (2025). Cultivating “good work”: Validating organizational culture profile in South Korea. European Journal of Training and Development, 49(9), 929–951.