RITI: Development and Theoretical Rationale of a Four-Axis Model of Romantic Relationship Temperament

1. Introduction

Popular relationship typologies are, almost without exception, built backwards: the sixteen boxes come first, and justification is retrofitted, if it is attempted at all (Pittenger, 1993, 2005). Meanwhile the psychology of close relationships is one of the field’s most replicated literatures — five decades of work on adult attachment, emotional expressivity, and relational motivation, spanning meta-analyses, longitudinal cohorts, and field experiments. The gap between the two is not a gap in available science; it is a gap in construction method.

RITI (Relationship Initiation Type Inventory) is a method built forwards. Every axis begins as an established construct with a published measurement tradition; the typology is derived from the axes rather than the reverse; and each design decision — why these four dimensions, why binary forced choice, why types at all when the evidence says dimensions — is argued from the literature rather than asserted. The name reflects the method’s scope: relationship initiation, the phase spanning attraction, ambiguous courtship, and early coupling, is an established field of study with its own research handbook (Sprecher, Wenzel, & Harvey, 2008), and individual-difference inventories targeting this phase have academic precedent (Hall, Carter, Cody, & Albright, 2010).

This paper makes three arguments. First, it states the theoretical case for each axis, grading the supporting evidence honestly from meta-analysis to single study (Section 2). Second, it presents the method’s central architectural decision — continuous scores inside, types outside — and grounds it in the taxometric and dichotomization literatures (Section 3). Third, it derives the measurement principles of the method from the same evidential logic (Section 4). Construct definitions throughout follow the discipline recommended by Clark and Watson (2019).

2. Theoretical Framework

RITI’s four axes were selected under three constraints: each axis had to be anchored in a distinct, well-replicated research program rather than invented for the format; each had to be behaviorally visible in the initiation phase, so that concrete scenarios could target it without asking respondents to introspect on abstract traits; and the axes had to be conceptually separable, so that a four-letter profile carries four pieces of information rather than one construct repeated. Candidate dimensions that failed these tests — conflict style, which overlaps heavily with attachment; commitment readiness, which stigmatizes; disclosure pace, which Expression absorbs — were excluded during design. Figure 1 summarizes the resulting structure, and Table 1 maps each axis to its parent construct and reference measurement tradition.

The pole letters belong to the communicative display layer, not to the construct layer. They are mnemonic names for remembering the code: Affiliation (A) and Own time (O), Security (S) and Verification (V),Demonstrative (D) and Reserved (R), and Novelty (N) and Habit (H). In particular, Affiliation and Own time do not replace the first axis’ theoretical endpoints, which remain closeness and autonomy.

2.1 Two dimensions of adult attachment: Temperature and Certainty

Factor-analytic work consolidating over three hundred adult-attachment items found that self-reported individual differences reduce to two largely independent dimensions — attachment avoidance and attachment anxiety (Brennan, Clark, & Shaver, 1998; refined by item response theory in Fraley, Waller, & Brennan, 2000). In attachment theory these dimensions reflect internal working models built from relational experience: avoidance tracks discomfort with closeness and dependence, anxiety tracks vigilance about availability and fear of abandonment. They are not fixed traits; attachment orientations change with relational experience, and the standard reference volume devotes sustained attention to that change (Mikulincer & Shaver, 2016) — which is why the RITI method describes its profiles as current patterns rather than essences.

The two dimensions carry distinct behavioral signatures: avoidance manifests as deactivating strategies — distance-keeping and compulsive self-reliance — while anxiety manifests as hyperactivating strategies — proximity seeking and reassurance seeking (Mikulincer & Shaver, 2016). Both are visible outside the laboratory: avoidant adults show measurably fewer contact-seeking behaviors during real airport separations (Fraley & Shaver, 1998), and excessive reassurance seeking behaves as the relational expression of attachment anxiety (Shaver, Schachner, & Mikulincer, 2005). Meta-analytically, the two dimensions relate to relationship quality through different pathways — avoidance chiefly through lowered satisfaction, connectedness, and support; anxiety chiefly through negative expectations and conflict escalation (Li & Chan, 2012), with costs observed for both members of a couple (Candel & Turliuc, 2019) — which is the methodological reason to measure them as separate axes. The two-factor structure has also been reproduced in Korean-language adaptations of the Experiences in Close Relationships–Revised scale (Kim, 2004), supporting the model’s application in Korean.

RITI’s first two axes are renamings of these dimensions with reversed, non-deficit orientation. Temperature runs from closeness-oriented (display mnemonic Affiliation, code A) to autonomy-oriented (Own time, O) and corresponds inversely to attachment avoidance. Certainty runs from security-default (S) to verification-seeking (V) and corresponds inversely to attachment anxiety, operationalized through reassurance-seeking scenarios: the unanswered message, the ambiguous closing line, the partner who is “busy this week.”

2.2 Dispositional emotional expressivity: Expression

Emotional expressivity — the disposition to display felt emotion outwardly — is a stable, measurable individual difference (Kring, Smith, & Neale, 1994) with a multi-faceted structure that replicates across self-reports, peer ratings, and observed behavior (Gross & John, 1995, 1997). It is conceptually distinct from expressive suppression, an effortful regulation strategy (Gross & John, 2003); the RITI model defines its reserved pole (R) as low dispositional expressivity — a style, not suppression, and not a deficit. This distinction is the difference between describing a person and pathologizing one: a reserved style is common and functional, whereas habitual effortful suppression carries documented costs.

The relational stakes of the axis are documented at several levels of evidence. At the interaction level, emotional self-disclosure and perceived partner responsiveness drive intimacy formation (Laurenceau, Barrett, & Pietromonaco, 1998). Experimentally, instructed suppression impairs rapport and inhibits relationship formation between strangers (Butler et al., 2003) — evidence that bears directly on the initiation phase. In couples, suppression predicts emotional and relational costs for both partners (Impett et al., 2012), and expressivity correlates with marital satisfaction (King, 1993). A meta-analysis confirms the general pattern while adding a caution the model respects: expression is not uniformly beneficial — anger expression, in particular, shows negative interpersonal associations (Chervonsky & Hunt, 2017). Expressivity correlates with, but is not reducible to, attachment dimensions (Feeney, 1999), which is why Expression (demonstrative D vs. reserved R) stands as a related-but-distinct third axis rather than a facet of the first two.

2.3 Self-expansion and novelty preference: Stimulus

The self-expansion model holds that a basic relational motive is the enlargement of the self through new experiences shared with a partner (Aron & Aron, 1986; Aron, Lewandowski, Branand, Mashek, & Aron, 2022). Its evidence base spans four decades and multiple methods. A ten-week field experiment found that couples assigned to exciting shared activities gained more marital satisfaction than couples assigned to merely pleasant ones (Reissman, Aron, & Bergen, 1993); laboratory experiments replicated the effect with novel and arousing joint tasks (Aron, Norman, Aron, McKenna, & Heyman, 2000); relational boredom — the low end of the process — is a measurable construct in its own right (Harasymchuk & Fehr, 2012) and predicts satisfaction declines nine years later (Tsapelas, Aron, & Orbuch, 2009); and self-expanding activities sustain sexual desire in established couples across dyadic, diary, and experimental designs (Muise et al., 2019).

Most directly for the model, the preference for self-expansion over self-preservation has been validated as a continuous individual-difference dimension via the Self-Expansion Preference Scale (Hughes, Slotter, & Lewandowski, 2020; measurement lineage from Lewandowski & Aron, 2002), against the broader temperament background of sensation seeking (Zuckerman, 1994). RITI’s fourth axis, Stimulus, runs from novelty-seeking (N) to habit-favoring (H). The habit pole is framed as a preference for stability and ritual — the literature documents costs of boredom as a relational state, not costs of preferring routine as a disposition, and the model does not rank the poles.

2.4 Why relationship initiation

The RITI method addresses the initiation phase specifically because that is where its target constructs are most active. The central psychological task of early interaction is uncertainty management (Berger & Calabrese, 1975), and relational uncertainty is itself a measurable, multi-source construct spanning doubts about the self, the partner, and the relationship (Knobloch & Solomon, 1999). Attachment anxiety is more activated in fledgling relationships than in established ones — before a stable bond exists to regulate it (Eastwick & Finkel, 2008) — and fear of rejection distorts early signaling: people systematically believe their overtures are more visible than they are (Vorauer, Cameron, Holmes, & Pearce, 2003), which links the Certainty and Expression axes to concrete initiation outcomes.

The phase is now thoroughly mediated by mobile communication (Finkel, Eastwick, Karney, Reis, & Sprecher, 2012): phenomena such as ghosting are documented research topics (LeFebvre et al., 2019), and even response latency after a first date measurably shapes perceived interest (Teichmann, Petrowsky, Boecker, Soliman, & Loschelder, 2026). The RITI method therefore stages its scenarios in this native habitat — unread messages, reply timing, ambiguous closings (Figure 2) — minimizing the inferential distance between scenario and lived situation, in line with evidence that these micro-events carry real informational weight in early relationships.

Table 2 consolidates the evidence base of the model by axis and study design. The grading is deliberate: where an axis rests on meta-analysis the table says so, and where it rests on a field experiment or a diary study, it says that instead — the weight behind each claim should be visible at a glance.

3. From Dimensions to Types: The Display-Layer Architecture

The evidence on latent structure is unambiguous: taxometric analyses find adult attachment variation to be dimensional, not categorical, both in general and in relationship-specific (including romantic) representations (Fraley & Waller, 1998; Fraley, Hudson, Heffernan, & Segal, 2015), and the same conclusion holds across normal personality broadly, where dimensional findings dominate taxometric research once methodological artifacts are controlled (Haslam, Holland, & Kuppens, 2012). Type indicators in fact measure continuous traits without bimodality (McCrae & Costa, 1989); median-split typing discards information, destabilizes classification near the cut, and produces the familiar complaint of receiving a different type on retest (Pittenger, 1993, 2005; MacCallum, Zhang, Preacher, & Rucker, 2002); and types are consistently outpredicted by dimensions in head-to-head comparisons (Asendorpf, 2003).

The RITI method accepts all of this rather than arguing with it. Each axis is scored as a graded margin rather than a bare category, and the method prescribes that this margin be reported with the label. The sixteen-type code (e.g., A·V·D·N) is a display layer: a mnemonic summary located in a continuous four-dimensional space, in the spirit of person-centered summaries of dimensional data (Gerlach, Farb, Revelle, & Amaral, 2018) — cited here as precedent for profile summaries, not as evidence that sixteen natural categories exist. The wager is that the type format’s communicative power — a pronounceable identity, a code that travels in ordinary conversation, a vocabulary for discussing differences — can be kept while the dichotomization costs are contained: the method prescribes that the label always be presented alongside its per-axis margins, be described as a current pattern rather than an essence, and not be treated as a latent category in analysis.

The mnemonic words are deliberately separated from construct names. For the Temperature axis, A (Affiliation) and O (Own time) make the consumer-facing code memorable, while closeness and autonomy remain the theoretical endpoints used for interpretation. The other six mnemonic words coincide more directly with their endpoint descriptions.

Two further prescriptions follow directly from the dichotomization critique. First, axis margins are computed from an odd number of equally weighted keyed items, so no tie-forcing occurs. Second, when an axis lands at its narrowest possible margin, measurement is conditionally extended — additional items for that axis before the display code is assigned — concentrating items where classification is least stable, an application of the general insight that measurement information should be allocated where it is needed (Fraley et al., 2000). When no axis is at that boundary, the current interface gives one two-item follow-up block to the axis with the smallest baseline margin. This minimum is a completion-flow rule, not a claim that a decisive classification requires more items.

4. Measurement Principles of the RITI Method

Four principles carry the theoretical framework into measurement, each with an evidential rationale; publishing them is itself a methodological commitment to measurement transparency (Flake & Fried, 2020).

Scenario-based items. Most items present a concrete relational scenario, several of them drawn from the mobile-mediated habitat reviewed in Section 2.4, with two behaviorally phrased responses keyed to the opposing poles of a single axis. Behavioral scenarios minimize the introspective burden of trait adjectives and anchor responses in situations the literature identifies as diagnostic — the unread message for Certainty, the after-date evening for Temperature.

Single-axis binary forced choice. The method requires that both options be written to be socially acceptable, so that neither pole is the obviously creditable answer; whether a given item meets that requirement is an empirical question about its response distribution, not something the format guarantees. RITI adopts the unidimensional pair not for faking resistance but for interpretability: because each item opposes poles of a single dimension rather than statements from different dimensions, every response remains attributable to one construct and within-axis tallies avoid the classical interpretability problems of ipsative scoring (Hicks, 1970). Its low-stakes, self-selected context makes deliberate faking a secondary concern. Model-based approaches to forced-choice data (Brown & Maydeu-Olivares, 2011) define the formal standard for such responses; RITI’s current scoring is a simple within-axis tally, deliberately transparent but not a latent-trait estimate, and the model-based route is the natural upgrade path.

Graded margins beneath the code. Each axis is scored by simple majority among an odd number of equally weighted keyed items — an odd count precludes ties — and the margin is retained rather than discarded at the cut. The margin is a graded score whose resolution is set by the item count, not a continuous latent estimate. The display code concatenates the four majority poles, but the margins travel with it: the method prescribes reporting both, so the graded information is surfaced rather than hidden behind the label (MacCallum et al., 2002).

Instrument parameters. The current implementation administers five keyed items per axis (twenty scored items in total), each a binary forced choice presented with the two options in randomized order. An axis whose split is three-to-two triggers the conditional extension described below, adding two further items for that axis. If no axis has a three-to-two split, the axis with the smallest normalized baseline margin receives one two-item follow-up block. The resulting extension is therefore two, four, six, or eight items. Scoring is the unweighted within-axis tally described above. The instrument is written in Korean and takes roughly three minutes. Item wording is held out of this report; what is specified here is the construction rationale and the scoring rule.

Conditional extension at the boundary. An axis at its narrowest margin is exactly where a dichotomous label is least trustworthy (Pittenger, 2005). The method therefore gives priority to contested axes before assigning the code. When no axis is contested, it gives the least pronounced baseline axis one two-item block so that every completed session has a short follow-up phase. The two cases should not be conflated: the first allocates information at a classification boundary; the second is a user-experience floor and is not presented as psychometrically necessary. Because two added items cannot overturn a decisive four-to-one or five-to-zero margin, the procedure economizes respondent time rather than manufacturing precision; boundary stability must come from item quality rather than from the extension itself. The allocation principle nonetheless echoes information-based item selection in modern psychometrics (Fraley et al., 2000).

5. Scope and Boundary Conditions

This is a conceptual and methodological paper: it documents the construction rationale of the RITI method and reports no empirical data.

The RITI method describes the temperament an individual brings to relationship initiation. Three boundaries follow from the evidence it is built on. First, it profiles individuals, not dyads: the strongest available review of algorithmic matching concludes that predicting couple compatibility from individual profiles lacks evidential support (Finkel et al., 2012), and the method accordingly makes no compatibility or relationship-outcome claims. Second, its sixteen types are a communicative vocabulary over continuous dimensions, not natural categories (Section 3), and profiles are read as current patterns, consistent with the evidence that attachment orientations change with experience (Mikulincer & Shaver, 2016). Third, the framework is descriptive, not evaluative or clinical: the model ranks no pole. That is a descriptive and ethical commitment about how profiles are reported, not a claim that the poles are equivalent in outcome — Section 2.1 reviews evidence that the attachment dimensions underlying two of the axes do predict differences in relationship quality.

6. Concluding Remarks

RITI illustrates that a sixteen-type format need not be built backwards. Each of its four axes stands on an established construct with its own measurement tradition — adult attachment’s two dimensions, the expressivity tradition, the self-expansion model — chosen for behavioral visibility in the initiation phase that the field of relationship initiation research has mapped. Its typology is honest about what types are: a language layered over graded measurement, with measurement effort concentrated exactly where labels are weakest. The result is a construction method whose axis choices, exclusions, and scoring rules are argued from the published experimental record rather than asserted — the standard this document exists to make auditable.

Author Note

The author is an independent researcher and develops a consumer application that implements the method described here. This paper reports the method’s theoretical construction only; it makes no claims about the reliability or validity of any implementation. No external funding was received.

Generative AI tools were used to assist with drafting, literature search, and reference verification. The author reviewed and verified all content, including the existence and content of every cited source; each of the works listed below was checked against its publisher record or DOI.