Psychology by BogdanovaThe Encyclopedia of Psychology for the Artificial Era

RESEARCH DOSSIER

Relationship Formation and Partner Selection

Abstract. Relationship formation and partner selection are sequences of exposure, attraction, reciprocal evaluation, continued contact, and mutually recognized partnership within structured social environments. This critical narrative synthesis examines historical laboratory studies, longitudinal and demographic research, speed-dating experiments, international preference surveys, meta-analyses, online dating behavioral data, and predictive modeling. It distinguishes stated preferences from observed choice, consensual desirability from distinctive dyadic attraction, and profile interaction from established relationships. Research supports important roles for opportunity, familiarity, perceived similarity, conversational responsiveness, and institutional constraints, while showing limitations in personalized romantic prediction and the generalizability of isolated preference measures. The monograph evaluates selection bias, dyadic statistical dependence, causal identification, cross-cultural comparability, digital recommendation fairness, and research priorities for prospective, ethically governed observation of relationship formation.

Keywords: relationship formation; partner selection; romantic attraction; mate choice; partner preferences; reciprocity; speed dating; assortative partnering; dating markets; social opportunity; relational mate value; similarity; ideal preference matching; interpersonal responsiveness; online dating; algorithmic matchmaking; dyadic analysis; romantic prediction; longitudinal research; Artificial Era

The Scientific Object of Relationship Formation

Relationship formation and partner selection concern the processes by which persons encounter potential intimate partners, develop or withdraw romantic interest, evaluate reciprocal possibilities, and establish an identifiable partnership. The central scientific outcome is a transition from a pool of potentially available relationships to a mutually recognized intimate bond. Attraction is a time-varying evaluation; selection is a constrained sequence of choices; formation requires reciprocal action and a negotiated or behaviorally observable transition. These processes occur within social environments that structure who can meet, communicate, and safely pursue a relationship. Relationship science therefore needs individual-level measures of preference, dyadic measures of mutual interest, and population-level measures of opportunity rather than treating each observed couple as the inevitable result of a stable personal taste Levinger (1980) Finkel et al. (2017).

The focal construct can be decomposed into eligibility, exposure, first impressions, interaction, mutual interest, continued contact, and transition to partnership. Eligibility concerns the people a participant would consider under actual circumstances. Exposure specifies who is present or visible. First impressions may concern appearance, demeanor, or social signals. Interaction changes available information and permits one person's responses to alter the other's evaluation. Continued contact produces a selection process in which both declining and escalating interest matter. A declared partnership is one possible endpoint, while voluntary nonformation also conveys scientifically meaningful information. Studies focusing only on established couples condition on survival through all preceding stages and cannot directly recover how the unavailable, unapproached, or rejected alternatives would have behaved Kalmijn (1998) Eastwick et al. (2019).

This monograph synthesizes experimental social psychology, speed-dating studies, longitudinal relationship research, demographic evidence, online-dating behavioral records, cross-cultural surveys, and contemporary prediction research. It is a critical narrative synthesis anchored in identifiable studies and reviews, not an original experiment or a protocol-defined systematic review. The analytic questions are how opportunity shapes who becomes a candidate, which reported preferences become behavioral choices, when actual interaction transforms evaluations, and how selection processes differ across contexts and populations. The scientific boundary ends at the establishment of an intimate relationship; affection trajectories, enduring commitment, and relationship stability remain adjacent research objects, considered here only when they help explain the transition into a relationship Finkel et al. (2012) Eastwick et al. (2014).

Historical Foundations of Attraction and Mate Choice

The mid-twentieth-century tradition of interpersonal attraction investigated proximity, familiarity, physical appearance, similarity, and reinforcement as conditions of liking. Research on residential contact provided an early ecological insight: whom one has an opportunity to encounter is a prerequisite for forming a relationship. Laboratory work subsequently used controlled descriptions, photographs, and encounters to isolate individual cues. Walster, Aronson, Abrahams, and Rottman (1966) examined the importance of physical attractiveness in dating decisions; Berscheid and colleagues (1971) experimentally investigated matching and choice. These studies helped construct testable propositions, while their historical populations and task constraints also show why results must be interpreted within the circumstances of evaluation Walster et al. (1966) Berscheid et al. (1971).

A second tradition emerged in demographic and sociological studies of marriage markets. It distinguished individual preferences from the distributions of possible partners, organized social boundaries, and the influence of institutions and third parties. Kalmijn (1998) synthesized evidence on homogamy and intermarriage, arguing that preferences, social group influence, and market constraints jointly organize selection. This perspective expanded the explanandum from whom an individual likes to which relationships actually materialize. In parallel, evolutionary approaches proposed that some stated mate preferences reflect recurring reproductive and investment pressures, prompting comparative research across cultures. Buss (1989) gathered evidence from 37 samples in 33 countries, but a cross-sectional rating of a preferred trait remains a different outcome from observed mutual choice Kalmijn (1998) Buss (1989).

Later research addressed the gap between choosing hypothetical profiles and interacting with living people. Speed-dating paradigms created a sequence of brief encounters followed by independently reported desire and mutual selections. Longitudinal and diary designs began recording how acquaintanceships transition into partnership. Eastwick and Finkel (2008) showed that traditional differences in what women and men said they valued did not straightforwardly translate into corresponding differences in attraction during actual brief encounters. Subsequent reviews formalized the distinction between measured ideals and revealed evaluations. The historical shift was methodological: partner selection became a sequence of observations about real dyads instead of a single ranking of abstract features Eastwick & Finkel (2008) Eastwick et al. (2014).

Research Questions and Levels of Explanation

An adequate explanation asks at least four linked questions. First, which candidate partners become accessible given location, networks, technology, age, education, and social institutions? Second, which perceived attributes affect initial desire before people interact? Third, which qualities of reciprocal interaction alter interest beyond the two participants' independently measured tendencies? Fourth, when does emerging interest lead to a renewed encounter and a partnership decision? Each question has a different denominator. A proportion of viewed profiles receiving messages, a proportion of speed dates receiving mutual interest, and a proportion of acquaintances becoming couples describe different populations and stages Hitsch et al. (2010) Rosenfeld et al. (2019).

Psychological explanation operates at actor, partner, and relationship levels. Actor effects represent a person's tendency to be interested in many others; partner effects represent a person's tendency to attract interest across evaluators; relationship effects represent the distinctive attraction of one person to another after accounting for their general tendencies. This decomposition is especially important because a compelling perceived connection can be genuinely dyad-specific. Eastwick and Hunt (2014) studied consensus and uniqueness in evaluations, while Joel, Eastwick, and Finkel (2017) tested how much self-reported attributes and preferences predicted these different components. A powerful predictor of average desirability need not predict which particular pair will form Eastwick & Hunt (2014) Joel et al. (2017).

Selection has temporal and institutional levels as well. People can become more or less available because of changing schedules, migration, economic insecurity, social norms, digital filtering, or expressed commitments. The same participant may pursue a partner in one context and decline contact in another because the perceived costs, possibilities, and obligations differ. These contextual changes become part of the causal explanation rather than statistical noise. Close-relationship theory calls attention to dispositions, dyadic regulation, and external circumstances; demographic accounts make the opportunity set explicit. A strong research design measures both, allowing the analyst to distinguish a preference that was never expressed from a preference that could not be acted upon Finkel et al. (2017) Kalmijn (1998).

Encounter Opportunities, Proximity, and Repeated Exposure

Physical and institutional proximity determines the candidate pool before preferences can operate. Shared classrooms, neighborhoods, workplaces, associations, and friendship networks produce recurring encounters and mutual knowledge. Proximity may also indicate common social position; these functions must be distinguished. Repeated exposure can enhance recognition and ease of interaction, but it also creates opportunities to discover undesirable behavior. Moreland and Beach (1992) varied the frequency with which women attended a classroom and studied the development of student affinity in the absence of direct interaction. This evidence concerns familiarity and affinity, and it cannot independently establish that exposure yields romantic partnership or that its effects are uniform across contexts Moreland & Beach (1992) Kalmijn (1998).

Experimental research with actual live interaction clarifies why meeting someone repeatedly differs from reading more isolated trait statements. Reis and colleagues (2011) varied the amount of interaction between previously unacquainted participants and found greater attraction with increased interaction, with perceived responsiveness, comfort, and knowledge implicated in the process. The participants were same-sex strangers in an interpersonal attraction design rather than couples seeking sexual partners; interpretation therefore concerns mechanisms that could contribute to romantic initiation, not a direct estimate of mate-choice outcomes. Repeated contact creates informational and relational affordances that abstract profile stimuli generally do not provide Reis et al. (2011) Montoya et al. (2008).

The opportunity perspective also explains why population-level similarity is sometimes overinterpreted. Two people can share education or class because schools, workplaces, and neighborhoods sort people before they meet. A matched pair can reflect deliberate preference for similarity, selective exposure to similar candidates, differential willingness to reciprocate, or institutions that reward particular unions. Purely descriptive assortative mating cannot separate these mechanisms. Relevant tests need measures of available alternatives, encounters, invitations, and mutual decisions, preferably before a relationship exists. Kalmijn's framework remains useful precisely because it locates partner choice inside the social conditions under which choices become feasible Kalmijn (1998) Hitsch et al. (2010).

From First Impressions to Mutual Attraction

First impressions are rapid judgments drawn from limited perceptual and contextual information. They can reflect appearance, style, apparent friendliness, conversational confidence, and inferred goals. Their predictive value depends on the task. Appearance can influence whether somebody approaches a stranger, yet initial judgments may be revised once interaction supplies new information. Walster and colleagues (1966) demonstrated the importance of attractiveness in an early dating experiment; a later experimental literature examined social desirability and matching under different rejection contingencies. Such results document bounded effects within particular encounter settings and invite attention to differences between immediate choice and sustained reciprocal involvement Walster et al. (1966) Berscheid et al. (1971).

A major theoretical challenge is that mutual attraction is a joint outcome. One person's positive evaluation may produce warmth, attention, or invitation, which changes the response of the other person; feedback can amplify or dampen the original inclination. Studies that ask only one participant whether a target is attractive measure an individual evaluation. They do not establish that the target reciprocates or that either person will organize future contact. Aron's retrospective accounts of falling in love identified noticing desirable characteristics and discovering another person's liking as prominent reported experiences, but retrospective narratives are vulnerable to memory reconstruction. They illuminate perceived transitions while requiring prospective corroboration Aron et al. (1989) Joel et al. (2017).

The actor-partner-relationship perspective helps correct an otherwise seductive error: treating attractive personal characteristics as if they fully determine pair compatibility. People can agree that someone is generally appealing while differing sharply about the chemistry of a particular encounter. Eastwick and Hunt (2014) developed the distinction between consensual mate value and unique relational evaluations. Prediction models should therefore report performance separately for who tends to be desired, who tends to desire others, and which specific dyads reciprocate. Aggregating these outcomes can make an algorithm appear effective at matchmaking when it merely ranks generally popular candidates Eastwick & Hunt (2014) Joel et al. (2017).

Formation as a Temporal Transition

Relationship formation is better represented as a series of changing evaluations and commitments than as a single moment of discovery. Levinger (1980) proposed an analysis of close relationships that distinguished attraction, building, continuation, deterioration, and ending. Later theorizing described evaluations as trajectories shaped by ascent, fluctuation, and thresholds. These formulations help identify the point at which a developing acquaintance becomes a recognizable intimate partnership. They also direct attention to alternatives that terminate before either person identifies a relationship, creating a crucial missing-data problem in studies that recruit only existing couples Levinger (1980) Eastwick et al. (2019).

Formation may occur rapidly after a brief encounter or gradually through friendship, repeated contact, and changing interpretations. A person who initially judges another as a friend can later perceive romantic possibilities without having revised any explicit demographic criterion. Consequently, prospective partner studies must retain interaction histories and allow status transitions in both directions. Designs limited to the first few minutes of meeting estimate one stage of attraction, while retrospective couple interviews are filtered through successful survival and a narrative of origins. A comprehensive account requires observation beginning before acknowledged partnership and continuing through the transition into a shared relational definition Eastwick et al. (2019) Aron et al. (1989).

Temporal models also expose a conceptual distinction between interest and action. A partner may be desired but unavailable, approached but uninterested, mutually interested but unwilling to commit, or committed while still negotiating relationship terms. These are scientifically different selection states. A useful longitudinal design records exposures, opportunities to communicate, stated intentions, encounters, reciprocal judgments, continued contact, and eventual status, each with dates. Such measurement provides a basis for estimating transitions and evaluating whether different mechanisms dominate different stages. It respects the fact that relationship science studies a developing process whose direction can change as both participants respond to one another Finkel et al. (2017) Eastwick et al. (2019).

Similarity, Perceived Compatibility, and Selection

Similarity is not a single explanatory variable. Researchers distinguish objective correspondence in attitudes, values, education, demographic background, interests, or personality from perceived correspondence, which is itself a psychological judgment. Similarity may make interaction easier, increase anticipated understanding, or arise because shared institutions bring comparable people together. Montoya, Horton, and Kirchner (2008) synthesized 460 effect sizes from 313 laboratory and field investigations and reported associations between attraction and both actual and perceived similarity. The effect of actual similarity weakened in studies involving developed relationships, whereas perceived similarity continued to predict attraction across contexts. The pattern shows that measurement time and setting are central to interpretation Montoya et al. (2008) Kalmijn (1998).

Perceived similarity can partly reflect projection: after finding another person likable, participants may infer shared beliefs without verifying them. Actual similarity can be confounded by restricted variation in the available pool or by educational and geographical sorting. This makes causal claims delicate. An experimental manipulation of apparent agreement tests something different from measuring whether two people already share political opinions or personality scores. Studies should specify whether the objective is to explain initial willingness to meet, mutual first-date attraction, pursuit of repeated contact, or the transition into a partnership. Distinguishing these stages prevents a broad correlation from becoming an unsupported account of relationship causation Montoya et al. (2008) Reis et al. (2011).

Similarity can also constrain alternatives and sustain opportunity even when it is not the preferred personal trait. People who share language, location, work schedules, or social networks can communicate and coordinate more readily. Those advantages may be misreported as psychological preference if the available candidate pool is ignored. Conversely, individuals may value complementary resources or experiences, but the evidence for specific complementarity claims varies with the trait and method. A scientific account requires separate tests of available-pool similarity, preferences before interaction, perceived compatibility after interaction, and actual mutual choices. The same observed couple can arise through distinct pathways, and the observation alone does not adjudicate among them Kalmijn (1998) Eastwick et al. (2014).

Ideal Partner Preferences and Their Predictive Validity

Ideal partner preferences are explicit ratings of the attributes a person believes an attractive partner should possess. Frequently assessed dimensions include physical attractiveness, warmth, trustworthiness, ambition, resources, intelligence, and similarity. Such reports establish the contents of a person's stated standards at the time and in the assessment context. They do not by themselves reveal how much weight the person places on an attribute during a live encounter, particularly when that encounter presents other cues and uncertain reciprocity. Eastwick and Finkel's (2008) speed-dating research directly compared stated ideals with attraction to actual partners and challenged straightforward predictions from conventional sex-differentiated preference patterns Eastwick & Finkel (2008) Eastwick et al. (2014).

The review and meta-analysis by Eastwick, Luchies, Finkel, and Hunt (2014) separated the general desirability of attributes from the distinctive validity of individual ideals. Across the reviewed attraction and relationship literature, physical attractiveness and earning prospects had associations with romantic evaluations, but classic reported sex differences did not uniformly translate into corresponding differences in effects during actual evaluations. This distinction matters for matchmaking. A platform might rank people who are widely considered desirable without learning whether a particular user obtains especially strong outcomes when a potential partner matches that user's declared profile Eastwick et al. (2014) Joel et al. (2017).

A newer worldwide registered report improved the evidential base. Eastwick and colleagues (2025) studied ideal-preference matching across 43 countries and 22 languages with a large sample, explicitly responding to inconsistent analytic approaches. Its existence changes the research question from whether ideal matching ever matters to which operationalizations, partner contexts, and effect metrics produce replicable predictive value. The appropriate inference remains calibrated: predictive validity can be context-dependent, and a statistically detectable association does not warrant a deterministic personal matching score. Researchers must distinguish trait appeal shared by many evaluators, distinctive preference fit, and the opportunity for mutual interaction Eastwick et al. (2025) Eastwick et al. (2014).

Sex Differences, Cultural Patterns, and Historical Context

Mate-preference research has often compared average stated preferences across sexes and social contexts. Buss (1989) examined ratings from 37 samples across 33 countries and reported cross-cultural differences in the relative importance placed on resource-acquisition cues and signs associated with reproductive potential. Such evidence helped establish cross-national questions, but it addressed preferences as reported in historically situated surveys. Neither the averages nor the categories describe every individual, every form of partnership, or the entire process of mutual selection. The causal interpretation requires tests of ecological, institutional, and developmental mechanisms in addition to comparisons of stated preferences Buss (1989) Walter et al. (2020).

Walter and colleagues (2020) conducted a large-scale replication across 45 countries with 14,399 participants. They found robust average sex differences in some reported mate preferences, while testing predicted associations with gender equality and pathogen prevalence. Some cross-national correlates were weaker or less general than competing explanations anticipated. The study demonstrates the value of replication beyond one national sample and the value of specifying which prediction survives which analysis. However, a cross-sectional preference survey cannot identify the actual candidate pool, partners who were approached, mutual acceptance, and long-term outcomes. Claims about biological adaptation or social role origins remain explanatory proposals requiring their own causal evidence Walter et al. (2020) Buss (1989).

A careful psychological interpretation separates within-group variation from between-group averages and acknowledges that access, safety, discrimination, and social expectations influence what respondents can pursue. The organization of partner markets differs across legal regimes, age distributions, educational pathways, and cultural practices. Researchers must also ensure that language, items, and categories have comparable meanings before aggregating surveys. The most scientifically useful comparative result is a conditional pattern with a documented sampling frame and measurement strategy, not a universally prescribed partner characteristic. The link between cultural preference surveys and partner formation becomes stronger when researchers also observe encounters, reciprocity, and actual union transitions Kalmijn (1998) Walter et al. (2020).

Evolutionary, Social, and Interdependence Explanations

Evolutionary models propose that some components of human mate preference have been shaped by recurrent selection pressures. Their empirical tests involve predicted cross-cultural patterns and relations between putative cues, preferences, and behavior. Socioeconomic and sociocultural explanations emphasize differential resources, institutional roles, and opportunity structures that shape the desirability and feasibility of partners. Relationship-science models add the observation that real attraction unfolds through reciprocal interaction and unique pair evaluations. These explanatory programs often concern different levels of analysis; their claims can be compared only after the outcome is aligned. A theory predicting a difference in stated partner ideals does not automatically predict a difference in live encounter attraction Buss (1989) Walter et al. (2020) Eastwick & Finkel (2008).

Interdependence approaches emphasize the expected consequences of interaction, available alternatives, and changes in what people experience as rewarding or costly. During initiation, expectations are provisional and updated through behavior: whether someone listens, shows interest, respects boundaries, and can participate in mutually desired contact. The scientific advantage is a focus on reciprocal processes rather than exclusively on partner traits. Finkel, Simpson, and Eastwick's (2017) integrative review positions dispositional tendencies inside ongoing interpersonal dynamics and contextual constraints. The model does not yield a simple universal formula for partner selection; it identifies classes of variables that must be jointly measured Finkel et al. (2017) Eastwick et al. (2019).

Scientific progress requires comparisons that could differentiate explanations. If a claimed universal preference changes sharply with experimentally altered opportunity or institutional incentives, social context has explanatory force. If a predicted effect appears consistently across carefully measured contexts, broader generality becomes plausible. If a trait predicts interest in many candidates but fails to predict pair-specific reciprocity, the theory may explain desirability rather than matching. An informative experiment or prospective cohort selects outcomes and counterfactuals in advance, rather than treating any association as confirmation of a favored framework. This approach integrates cultural comparison, behavioral experiments, and dyadic models without erasing their different causal targets Eastwick & Hunt (2014) Eastwick et al. (2025).

Assortative Partnering, Matching, and Sorting

Assortative partnering refers to nonrandom resemblance among people who enter relationships. The resemblance may concern education, age, religion, class, geographic background, or other social characteristics. There are several nonexclusive mechanisms: people may prefer similar partners, encounter similar partners more often, be accepted more often by similar partners, or experience structural restrictions against some unions. Kalmijn (1998) provided a framework distinguishing preferences, third-party influences, and marriage-market constraints. A statistical correlation between spouses' educational attainment is therefore a description of an outcome, not a sufficient explanation of the selection mechanism that generated it Kalmijn (1998) Hitsch et al. (2010).

Matching models ask how two-sided preferences and competition jointly produce a pattern of pairings. Hitsch, Hortaçsu, and Ariely (2010) used online dating interactions to estimate preferences and examine sorting under stable matching assumptions. Their approach demonstrated that patterns of observed matching can be compared with an explicit market model. The same evidence cannot identify what every participant would have done after offline interaction, because online behavior and actual union formation are different outcomes. Modeling also depends on visibility, strategic messaging, selective participation, and the platform's mechanism for presenting options Hitsch et al. (2010) Bruch & Newman (2018).

The concept of matching must be kept distinct from a normative claim about whom people deserve. A research model can estimate common patterns of contact and mutual acceptance without assigning intrinsic worth to persons. The term 'mate value' has been used analytically to distinguish consensus desirability from dyad-specific preference; this is a statistical construct whose meaning depends on setting and respondents. Eastwick and Hunt (2014) supplied an account of how unique evaluations arise alongside shared preferences. Measurement should report whose evaluations establish any ranking, what alternatives were observable, and how differences in representation may alter the apparent marketplace Eastwick & Hunt (2014) Hitsch et al. (2010).

Social Boundaries, Inequality, and Access to Relationships

Partner selection operates through social boundaries as well as individual attraction. Race and ethnicity, language, religion, class, citizenship, disability, geographical mobility, and family expectations can affect who becomes visible and approachable. In online settings, these boundaries can be reproduced through profile filters, rank ordering, unequal response patterns, and the platform's own presentation mechanisms. Observational studies that record initiating and reciprocating behavior can distinguish stages at which sorting occurs; they cannot by themselves isolate whether a pattern is driven by preference, discrimination, differential exposure, or a combination. A complete explanation treats the candidate pool as part of the studied system Kalmijn (1998) Ranzini et al. (2022).

Experimental profile manipulations provide more leverage on particular cues. Ranzini, Rosenbaum, and Tybur (2022) used a mock dating application with young Dutch participants to examine education, appearance, and ethnic cues. The experimental setting increases control over what users see, but the observed profile choice is still one stage of a possible relationship. It says less about subsequent reciprocal attraction and lived relationship formation. Social effects can change between browsing, messaging, meeting, and agreeing to pursue a bond. Researchers should record exclusion mechanisms and selection into app use, especially when generalizing findings to populations with different safety concerns or social opportunities Ranzini et al. (2022) Finkel et al. (2012).

The ethics of partner-selection research follow directly from its structure. No person is required to reciprocate another person's interest, and research on averages cannot prescribe an individual's intimate decisions. Studies should protect participants' sensitive preference data, avoid exposing identifiable romantic interactions, and account for the potential social consequences of algorithmic categorization. When systems optimize engagement rather than sustained mutually desired relationships, performance metrics may reward exposure or messaging without measuring actual benefit. Fairness research has begun analyzing differential recommendation quality across user groups, but its objective functions remain context-specific and need independent behavioral validation Zhao et al. (2024) Finkel et al. (2012).

Reciprocity, Responsiveness, and Emerging Interest

The emergence of mutual interest is a dynamic event. A participant's decision to communicate may change the recipient's available information, anticipation of rejection, and willingness to respond. The resulting behavior can feed back into the original participant's judgment, creating a sequence of reciprocal adjustments. This process means that initial attraction cannot be explained entirely by two sets of fixed independent attributes. Studies of reported early romantic experience repeatedly identify being liked by another person as a meaningful event, although retrospective data reveal perceived significance rather than experimentally isolated causation. The scientifically informative question is how reciprocal signals alter subsequent behavioral choices across time Aron et al. (1989) Eastwick & Hunt (2014).

Responsiveness refers to the experience that another person attends to, understands, and responds appropriately to one's expressions. It can influence perceived safety and interest during the transition from strangers to familiar interaction partners. Reis and colleagues (2011) showed that greater live interaction was accompanied by increased attraction, with perceived responsiveness among the processes implicated. Their design does not establish that every instance of attention produces romance, and respectful responsiveness must not be confused with entitlement to affection. It does suggest that reciprocal interpersonal information contributes something that cannot be captured by static ideal-partner profiles Reis et al. (2011) Finkel et al. (2017).

The practical measurement challenge is to distinguish expressed interest, perceived interest, actual willingness to continue, and consensual agreement to pursue a relationship. These are correlated but separable constructs. An individual can perceive interest incorrectly; another can respond warmly without pursuing romance; a favorable date can end because of timing or constraints. A prospective dyadic study should collect each participant's judgment separately before revealing the other's response and follow both participants through subsequent contact. Such designs can locate changes in evaluation relative to events and reduce the common-method artifact that arises when one person reports both the partner's intentions and their own attraction Joel et al. (2017) Eastwick et al. (2019).

Self-Disclosure, Familiarity, and Interpersonal Closeness

Self-disclosure can change an acquaintance's informational and emotional significance. Disclosure transmits personal information, while response quality determines whether that information contributes to feeling understood. The relational consequences depend on pacing, consent, prior trust, and the meaning of what is revealed. Aron and colleagues (1997) experimentally assigned previously unacquainted participants to escalating self-disclosure tasks and comparison conversation procedures; one study found greater closeness after the structured tasks than after small talk. The controlled induction is an important demonstration that interaction procedures can affect closeness, but closeness ratings are not direct evidence that a romantic partnership will follow Aron et al. (1997) Reis et al. (2011).

The scientific relationship between familiarity and selection is nonlinear in principle. Repeated communication can improve prediction of a partner's behavior and permit correction of unfavorable first impressions. It can also reveal incompatibilities, violations of boundaries, or goals that make formation inappropriate. The experimental evidence for increased attraction with interaction concerns particular social settings and does not warrant claims that exposure is universally beneficial. Mechanisms should be decomposed into increased knowledge, reduced uncertainty, subjective comfort, and experienced responsiveness, each measured independently when possible. The same duration of conversation may carry different consequences when intentions or power relations diverge Reis et al. (2011) Aron et al. (1997).

A methodological distinction is required between an experimentally created moment of closeness and the formation of a mutually chosen ongoing bond. Researchers can manipulate the structure of conversation but cannot legitimately require genuine romantic preference. Follow-up measures should separately record whether participants request contact, whether both agree, how they interact after leaving the experimental context, and whether a recognized relationship develops. The resulting evidence could test mediation through closeness without identifying all subsequent decisions as effects of the original intervention. An interdisciplinary account therefore treats disclosure as one potential pathway through an extended selection process Aron et al. (1997) Levinger (1980).

Speed Dating as a Controlled Naturalistic Paradigm

Speed dating occupies an intermediate position between laboratory vignettes and entirely natural partner markets. Participants encounter multiple people in a short period under relatively standardized conditions. Researchers can obtain preferences beforehand, record reactions immediately afterward, and observe mutual interest or exchange of contact information. The paradigm improves the identification of actor, partner, and dyad-specific components of evaluation because each person assesses several potential partners. The estimand nevertheless concerns attraction and choice within the event's format, recruitment pool, duration, and incentive structure, not an unconditional probability of subsequent sustained partnership Eastwick & Finkel (2008) Joel et al. (2017).

The strongest designs gather measurements before and after each encounter, avoid showing the participant the other's ratings until data collection is complete, and retain all rejected as well as mutually accepted dyads. Cross-classified statistical models are needed because observations from the same individual and the same candidate partner are dependent. Dating events with highly selected student volunteers may overrepresent participants comfortable with rapid encounters and able to attend. The balance of age groups, sexual orientation, availability, and event rules limits the generality of findings. Even when the design is well controlled, the appropriate inference is about what happens under that encounter ecology Eastwick & Hunt (2014) Joel et al. (2017).

Eastwick and Finkel (2008) compared desired traits in advance of speed dates with evaluations after actual meetings. Joel and colleagues (2017) used machine learning on extensive self-report predictors in speed-dating contexts, asking whether traits could predict general tendencies toward romantic desire and the particular appeal of a specific dyad. This division of outcomes prevents an apparent rise in average classification accuracy from being mistaken for useful individualized matchmaking. The key technical issue is whether prediction remains accurate on new events, new actors, and new pairings rather than simply fitting data collected from repeated participants in the same sample Eastwick & Finkel (2008) Joel et al. (2017).

Prediction, Idiosyncratic Attraction, and Model Failure

Predicting initial romantic desire is a harder task than identifying generally desirable characteristics. In Joel, Eastwick, and Finkel's (2017) two speed-dating studies, machine-learning analyses used extensive self-report measures of traits and ideals to examine actor, partner, and relationship effects. The study found limits to predicting distinctive desire for specific partners before interaction, illustrating the difference between ranking persons and predicting pair-specific chemistry. It is important to understand the boundaries of the result: the model tested available self-report predictors within defined samples and procedures, not every conceivable observation, future interaction sequence, or psychological mechanism Joel et al. (2017) Eastwick & Hunt (2014).

Machine-learning validity requires a design appropriate to dyadic dependence. A random split of individual ratings can put the same people into training and test partitions, allowing a model to memorize stable actor or partner tendencies. Evaluation on previously unseen individuals and wholly unseen events is a more demanding test of generalization. Researchers should report separate benchmarks for trivial popularity or selectivity predictors, trait-based models, and relational signals captured during interaction. Transparent performance measures should reflect whether the application seeks to predict initial contact, mutual second-date choice, or formation of an ongoing bond. A model optimized for one target cannot be presumed valid for another Joel et al. (2017) Hitsch et al. (2010).

Research on longer-term relationship quality provides a useful warning about outcome substitution. Joel and colleagues (2020) analyzed self-reported relationship quality across many longitudinal couple studies; these findings concern people already in relationships and predictors of their experienced quality. They cannot retroactively prove that a matching algorithm could have identified the right partners before those relationships began. A successful model of relationship satisfaction and a model of initial mate selection require different training populations, observed alternatives, and causal assumptions. Keeping these targets separate improves the scientific interpretability of predictive claims and reduces the risk of advertising a predictive system beyond its evidence Joel et al. (2020) Finkel et al. (2012).

Social Networks and Pathways Into Partnership

Personal networks function as channels through which candidates are introduced, evaluated, and socially recognized. Friends, relatives, colleagues, community organizations, and local institutions affect both encounter probabilities and information about possible partners. They may also influence whose relationship is viewed as permissible or desirable. Network opportunities interact with age, geography, socioeconomic position, and the availability of digital alternatives. This perspective makes the denominator of selection visible: a person cannot form a relationship with someone they never meet or learn about. Nevertheless, observed network mediation does not determine whether the participants preferred the recommended partner or chose them for some independent relational reason Kalmijn (1998) Rosenfeld et al. (2019).

Rosenfeld, Thomas, and Hausen (2019) analyzed evidence from a nationally representative 2017 survey of American adults and documented a shift in how heterosexual couples in the United States reported first meeting. In their analysis, online encounters became the most common route for those couples, surpassing introductions through friends around 2013. The finding is a historical pattern in a specified country and relationship population, not an estimate of all contemporary relationships worldwide or of the causal quality of digital matches. It demonstrates that technologies change who can become available for mutual selection, altering the opportunity structure within which established psychological tendencies operate Rosenfeld et al. (2019) Finkel et al. (2012).

Network and platform channels may also interact. A couple may first notice each other through an app, later discover shared acquaintances, and rely on those connections to evaluate trust. A binary distinction between online and offline can conceal such sequences. Prospective research can record multiple encounter channels and social ties, tracing how each affects continuation after first contact. At the population level, demographic surveys can track shifts in meeting pathways; at the dyadic level, time-stamped contact records can reveal where reciprocal selection occurs. Joint measurement is needed to determine whether a change of channel changes who meets, how attraction develops, or which partnerships ultimately become established Rosenfeld et al. (2019) Eastwick et al. (2019).

Digital Dating, Rankings, and Reciprocal Choice

Digital platforms transform relationship search by making candidate pools searchable and contact decisions recordable. Profiles introduce a compressed representation of a person, while platform interfaces determine which alternatives are displayed, in what order, and under which conditions of communication. This creates data opportunities for research alongside substantial selection biases. Messaging behavior reflects the combined effects of preferences, interface constraints, perceived competition, and anticipated response. Finkel and colleagues (2012) distinguished the potential advantages of access and communication from stronger claims that proprietary compatibility matching algorithms had established superior long-term relationship outcomes Finkel et al. (2012) Hitsch et al. (2010).

Bruch and Newman (2018) analyzed message-exchange patterns in large online dating markets and characterized aspirational pursuit in those particular settings. The data reveal strategic decisions about whom people approach and how they pursue contact, but observed messages are not equivalent to in-person attraction or relationship formation. Separating invitations, replies, extended conversations, in-person meetings, and mutually acknowledged partnership is essential when moving from platform activity to psychological conclusions. Missing records of offline encounters and undisplayed profiles can bias estimates of what platform users really prefer. Studying only successful replies further conditions the analysis on a selective pathway Bruch & Newman (2018) Hitsch et al. (2010).

Experimental work offers controlled tests of app presentation. Ranzini and colleagues (2022) examined profile-level cues in an experimental dating application, while Zhao and colleagues (2024) investigated fairness metrics and recommender performance in online dating algorithms. The two approaches answer different questions: one addresses how candidate attributes alter simulated choices, and the other evaluates differential recommendation behavior in a computational system. Neither alone establishes whether algorithmic exposure improves sustained relationships. Validated future systems would require prospective opt-in study, equal attention to opportunity and autonomy, independent measurement of reciprocal formation, and monitoring for unequal distribution of visibility or adverse experiences Ranzini et al. (2022) Zhao et al. (2024).

Prospective Cohorts and Longitudinal Observation

Longitudinal observation is indispensable because partner selection is a process with a history. Investigators can recruit participants before they seek a new relationship, document their social contexts and preferred characteristics, record encounters and contact over time, and observe whether interest becomes mutual. This design permits temporally ordered tests: stated preferences precede exposure; perceived chemistry follows interaction; decisions to meet again precede established partnership. It also records unsuccessful pathways rather than treating only formed couples as the relevant population. Longitudinal research on young adults has related personality and family history to romantic competence, illustrating the importance of developmental context while requiring care in attributing causality to earlier experiences Donnellan et al. (2005) Eastwick et al. (2019).

Cohort designs face selective entry, attrition, and changes in exposure. Persons who agree to participate in dating research may differ from those who do not, and participants who discontinue reporting may be especially likely to have experienced rejection or disruption. Time-varying covariates such as work schedules, relocation, and partner availability can influence both continued contact and the probability of measurement. Analyses must retain a clear risk set for each transition and distinguish missingness from a genuine decision against partnership. Repeated measures also permit investigation of changes in the stated ideal rather than assuming an unchanging preference profile over years Kalmijn (1998) Eastwick et al. (2019).

Inferences from existing-couple samples are particularly vulnerable to survivor selection. If researchers interview only established partners, their observed similarity reflects the entire prior process of meeting, pursuing, reciprocating, and maintaining involvement. Couples with unsuccessful initiation are absent. A model based solely on successful relationships can thus explain the properties of surviving matches without explaining how any one of them was formed. The ideal empirical sequence follows individuals and potential dyads from early opportunities through mutual recognition, with explicit definitions of contact, exclusivity, and partnership. Such work should use participant reports alongside behavioral records while respecting consent and confidential relationship information Joel et al. (2020) Levinger (1980).

Dyadic Modeling, Causal Inference, and Measurement

Formation data have dependencies at several levels. Each participant may evaluate many potential partners; each candidate may be evaluated by many participants; multiple assessments of the same dyad may be recorded over time. A simple regression treating all encounters as independent will misstate uncertainty and may confuse actor, partner, and unique interaction effects. Social relations models, cross-classified multilevel analyses, and time-to-event approaches address different facets of the design. They require adequate numbers of participants, repeated encounters, and a clear account of who was at risk of forming a partnership at each stage. Interpretable statistics depend on how the opportunity set has been sampled Eastwick & Hunt (2014) Joel et al. (2017).

Experiments permit stronger causal claims when they manipulate a relevant feature while holding plausible alternatives constant. Studies of repeated exposure, structured self-disclosure, and profile cues identify effects on the outcomes they actually measured. Moreland and Beach (1992) assessed affinity following exposure, Aron and colleagues (1997) assessed experimentally generated closeness, and Ranzini and colleagues (2022) studied choices on a mock app. None of those outcomes is equivalent to a durable romantic union. The distinction between manipulation, mediator, and distal outcome is central: a study may causally change immediate attraction and still leave the transition into a partnership unresolved Moreland & Beach (1992) Aron et al. (1997) Ranzini et al. (2022).

Measurement should explicitly specify the construct at each stage. Stated preference scales measure expressed standards; behavioral selection records which opportunities were pursued; mutual interest requires both parties' reports; self-reported status indexes participants' definitions of a relationship. Physiological arousal, facial expression, message length, or conversational sentiment may be correlated cues, but none independently proves affection or consent. Reliability depends on standardized questions and repeated observations, while construct validity requires showing that a measure tracks the intended process rather than general social desirability. Mechanistic claims need appropriate mediation assumptions and evidence from interventions or temporal ordering Eastwick et al. (2014) Finkel et al. (2017).

Replicability, Prediction Benchmarks, and Evidence Quality

The literature on partner selection contains consequential changes in methods and population coverage. Early experiments concentrated on visible traits and immediate choices; meta-analyses aggregated inconsistent attraction measures; contemporary studies combine preregistered international designs and machine learning. These advances can improve precision but also expose discrepancies hidden by older research traditions. Eastwick and colleagues (2025) explicitly addressed divergent analytic choices in tests of ideal-preference matching, while Walter and colleagues (2020) assessed replication of cross-cultural preference patterns. Neither sophistication nor sample size by itself removes the requirement that researchers specify the exact outcome and sampling frame Eastwick et al. (2025) Walter et al. (2020).

Reliable synthesis requires comparing like with like. Correlations from a hypothetical photograph experiment cannot be pooled without qualification with correlations from a prospective cohort tracking actual partnerships. A result about selecting a profile cannot establish a result about mutual liking, and a result about ongoing relationship quality cannot establish a result about the first meeting. Meta-analytic means are informative within defensible inclusion boundaries and with consideration of heterogeneity, measurement differences, selective reporting, and publication bias. Montoya and colleagues (2008) found different patterns for actual versus perceived similarity across research contexts, demonstrating why conceptual distinctions must inform aggregation Montoya et al. (2008) Joel et al. (2020).

Prediction research requires further safeguards. Models should be evaluated on held-out participants, relationships, and contexts; baseline models should quantify actor and partner popularity; calibration and error rates should be reported for relevant subgroups. Accuracy at predicting a first-date rating is not evidence of decision quality for a platform that seeks stable mutually chosen relationships. Training on retrospective success can also produce a collider problem: only people who previously established a bond enter the data. External validation across populations and periods is therefore essential. The appropriate scientific conclusion remains bounded by the actual target, the comparison model, and the design's ability to observe alternative outcomes Joel et al. (2017) Hitsch et al. (2010).

Population Diversity and the Meaning of Partnership

Relationship formation varies across life stages, populations, and social institutions. Youthful dating events offer a useful laboratory for first impressions, while older adults, people seeking partners after separation, and individuals forming relationships under distinct legal or cultural conditions may experience different opportunities and expectations. Research also needs to include same-sex and gender-diverse relationships rather than translating assumptions from heterosexual samples into universal propositions. Sexual orientation, age, disability, and cultural norms can change who is visible in the candidate pool and what it means to declare a relationship. A study's sampling frame and category definitions must be part of its scientific interpretation Rosenfeld et al. (2019) Kalmijn (1998).

Cross-cultural research can identify regularities in stated preferences, but equivalence is an empirical matter. An item such as 'financial prospects' or 'marriage intention' may carry different meanings across welfare systems, family arrangements, and legal contexts. Walter and colleagues (2020) show the reach of international survey replication, and Eastwick and colleagues (2025) extend comparison of preference matching across languages and countries. The next evidentiary task is linking such ratings to observed reciprocal selection with culturally appropriate outcomes. Researchers should report measurement invariance, geographic coverage, recruitment strategy, and limitations to generalizing from educated or digitally connected populations Walter et al. (2020) Eastwick et al. (2025).

Research ethics must recognize that intimate decisions are personal and involve at least two autonomous agents. The right to refuse contact, withhold disclosure, redefine the relationship, and end interaction sets the conditions for valid participation and valid measurement. Public or commercial access to messaging logs creates significant confidentiality risks, especially when unusual relationship histories could identify participants despite removing names. Ethically designed research limits data collection to scientific purposes, protects information about third parties, provides meaningful withdrawal options, and avoids feedback interventions that pressure participants into contact. Scientific descriptions of preferences must not be converted into prescriptions about whose preferences are legitimate Finkel et al. (2012) Zhao et al. (2024).

Artificial Era, Algorithmic Mediation, and Open Problems

The Artificial Era makes the role of artificial systems in human partner selection a concrete research issue. Dating applications and recommender systems already affect the sequence of people encountered and the amount of information available before a meeting. Their algorithms can change selection opportunities without altering users' underlying stated preferences. An artificial system that predicts message replies or displays profiles is an intermediary in a human relationship market; such predictive behavior should be evaluated with clear measures of access, reciprocal consent, and relationship outcomes. Fairness analyses of dating recommendations show that the design of ranking objectives is scientifically consequential, while behavioral validation remains necessary before claims about improved partnerships Zhao et al. (2024) Bruch & Newman (2018).

More ambitious claims concern artificial agents capable of evaluating dialogue or participating in relationship mediation. The methodological issue is whether a model learns socially shared cues, latent actor popularity, or signals unique to the particular human dyad. The limitations of pre-interaction self-report prediction documented by Joel and colleagues (2017) remain a demanding benchmark. A system may estimate interest from observed conversation more accurately than from biographical profiles and still lack evidence that its recommendations causally benefit the people involved. Research must distinguish algorithmic inference, intervention effects, personal autonomy, and the status of the people whose experiences generate the training data Joel et al. (2017) Finkel et al. (2012).

Major open questions concern how attractions change with repeated encounters, which preferences survive realistic opportunity constraints, and when mutual responsiveness becomes a decisive causal ingredient. Investigators also need genuinely prospective evidence on long-term consequences of algorithmically mediated meetings, culturally representative studies of same-sex and gender-diverse formation, and external validation of individualized matching models. The appropriate intellectual program combines psychology, sociology, network science, and computational methods. It treats the identity of the emerging relationship as an empirical outcome constructed through mutual choices and interactions, with no presumption that a platform's engagement metric is a proxy for a flourishing partnership Eastwick et al. (2019) Kalmijn (1998).

Conclusion

Relationship formation and partner selection are multi-stage processes through which available persons become acquainted, evaluate each other, exchange signals of interest, decide whether to continue contact, and establish a mutually recognized intimate bond. The research evidence identifies important contributions from opportunity structures, physical and perceived characteristics, familiarity, similarity, reciprocal responsiveness, and social context. Their effects differ by stage, measurement, and population. The central conceptual result is that attraction to a generally desirable person, unique interest in a particular partner, and eventual reciprocal partnership are distinct scientific outcomes. None can be assumed from another without following the processes that connect them Finkel et al. (2017) Eastwick & Hunt (2014).

Historical experiments isolated individual cues, demographic research documented structured markets, speed dating revealed dyadic variation, and contemporary computational methods demonstrated both new possibilities and clear limitations of prediction. Research on stated ideals and their behavioral validity has become more rigorous through meta-analysis and international replication. Digital platform data offer unprecedented evidence of search and approach behavior while also recording outcomes strongly shaped by visibility algorithms and self-selection. The body of evidence supports an integrative explanation grounded in actual opportunities, changing interpersonal evaluations, and reciprocal choice, with explicit boundaries around claims drawn from hypothetical tasks, existing couples, or algorithmic forecasts Eastwick et al. (2014) Eastwick et al. (2025) Rosenfeld et al. (2019).

Future progress will come from prospective designs that observe the full sequence from exposure to established relationship, cross-classified models that separate actor, partner, and dyadic effects, experiments that identify causal pathways, and representative international research that measures available alternatives. Validated outcomes must reflect informed, mutually desired partnership rather than mere contact or response probability. Within the Artificial Era, the same standards apply to digitally mediated encounters: artificial systems may change opportunities and supply predictions, but scientific evaluation depends on what they measurably change for the participants. The resulting research object remains the formation of a particular relationship through a traceable history of reciprocal decisions Joel et al. (2017) Eastwick et al. (2019) Zhao et al. (2024).

References

  • Aron, A., Dutton, D. G., Aron, E. N., & Iverson, A. (1989). Experiences of falling in love. Journal of Social and Personal Relationships, 6(3), 243–257. https://doi.org/10.1177/0265407589063001
  • Aron, A., Melinat, E., Aron, E. N., Vallone, R. D., & Bator, R. J. (1997). The experimental generation of interpersonal closeness: A procedure and some preliminary findings. Personality and Social Psychology Bulletin, 23(4), 363–377. https://doi.org/10.1177/0146167297234003
  • Berscheid, E., Dion, K., Walster, E., & Walster, G. W. (1971). Physical attractiveness and dating choice: A test of the matching hypothesis. Journal of Experimental Social Psychology, 7(2), 173–189. https://doi.org/10.1016/0022-1031(71)90065-5
  • Bruch, E. E., & Newman, M. E. J. (2018). Aspirational pursuit of mates in online dating markets. Science Advances, 4(8), eaap9815. https://doi.org/10.1126/sciadv.aap9815
  • Buss, D. M. (1989). Sex differences in human mate preferences: Evolutionary hypotheses tested in 37 cultures. Behavioral and Brain Sciences, 12(1), 1–14. https://doi.org/10.1017/S0140525X00023992
  • Donnellan, M. B., Larsen-Rife, D., & Conger, R. D. (2005). Personality, family history, and competence in early adult romantic relationships. Journal of Personality and Social Psychology, 88(3), 562–576. https://doi.org/10.1037/0022-3514.88.3.562
  • Eastwick, P. W., & Finkel, E. J. (2008). Sex differences in mate preferences revisited: Do people know what they initially desire in a romantic partner? Journal of Personality and Social Psychology, 94(2), 245–264. https://doi.org/10.1037/0022-3514.94.2.245
  • Eastwick, P. W., Luchies, L. B., Finkel, E. J., & Hunt, L. L. (2014). The predictive validity of ideal partner preferences: A review and meta-analysis. Psychological Bulletin, 140(3), 623–665. https://doi.org/10.1037/a0032432
  • Eastwick, P. W., & Hunt, L. L. (2014). Relational mate value: Consensus and uniqueness in romantic evaluations. Journal of Personality and Social Psychology, 106(5), 728–751. https://doi.org/10.1037/a0035884
  • Eastwick, P. W., Finkel, E. J., & Simpson, J. A. (2019). Relationship trajectories: A meta-theoretical framework and theoretical applications. Psychological Inquiry, 30(1), 1–28. https://doi.org/10.1080/1047840X.2019.1577072
  • Eastwick, P. W., Sparks, J., Finkel, E. J., Meza, E. M., Adamkovič, M., Adu, P., Ai, T., Akintola, A. A., Al-Shawaf, L., Apriliawati, D., Arriaga, P., Aubert-Teillaud, B., Baník, G., Barzykowski, K., Batres, C., Baucom, K. J., Beaulieu, E. Z., Behnke, M., Butcher, N., ... Coles, N. A. (2025). A worldwide test of the predictive validity of ideal partner preference matching. Journal of Personality and Social Psychology, 128(1), 123–146. https://doi.org/10.1037/pspp0000524
  • Finkel, E. J., Eastwick, P. W., Karney, B. R., Reis, H. T., & Sprecher, S. (2012). Online dating: A critical analysis from the perspective of psychological science. Psychological Science in the Public Interest, 13(1), 3–66. https://doi.org/10.1177/1529100612436522
  • Finkel, E. J., Simpson, J. A., & Eastwick, P. W. (2017). The psychology of close relationships: Fourteen core principles. Annual Review of Psychology, 68, 383–411. https://doi.org/10.1146/annurev-psych-010416-044038
  • Hitsch, G. J., Hortaçsu, A., & Ariely, D. (2010). Matching and sorting in online dating. American Economic Review, 100(1), 130–163. https://doi.org/10.1257/aer.100.1.130
  • Joel, S., Eastwick, P. W., & Finkel, E. J. (2017). Is romantic desire predictable? Machine learning applied to initial romantic attraction. Psychological Science, 28(10), 1478–1489. https://doi.org/10.1177/0956797617714580
  • Joel, S., Eastwick, P. W., Allison, C. J., Arriaga, X. B., Baker, Z. G., Bar-Kalifa, E., Bergeron, S., Birnbaum, G. E., Brock, R. L., Brumbaugh, C. C., Carmichael, C. L., Chen, S., Clarke, J., Cobb, R. J., Coolsen, M. K., Davis, J., de Jong, D. C., Debrot, A., DeHaas, E. C., ... Wolf, S. T. (2020). Machine learning uncovers the most robust self-report predictors of relationship quality across 43 longitudinal couples studies. Proceedings of the National Academy of Sciences, 117(32), 19061–19071. https://doi.org/10.1073/pnas.1917036117
  • Kalmijn, M. (1998). Intermarriage and homogamy: Causes, patterns, trends. Annual Review of Sociology, 24, 395–421. https://doi.org/10.1146/annurev.soc.24.1.395
  • Levinger, G. (1980). Toward the analysis of close relationships. Journal of Experimental Social Psychology, 16(6), 510–544. https://doi.org/10.1016/0022-1031(80)90056-6
  • Montoya, R. M., Horton, R. S., & Kirchner, J. (2008). Is actual similarity necessary for attraction? A meta-analysis of actual and perceived similarity. Journal of Social and Personal Relationships, 25(6), 889–922. https://doi.org/10.1177/0265407508096700
  • Moreland, R. L., & Beach, S. R. (1992). Exposure effects in the classroom: The development of affinity among students. Journal of Experimental Social Psychology, 28(3), 255–276. https://doi.org/10.1016/0022-1031(92)90055-O
  • Ranzini, G., Rosenbaum, J. E., & Tybur, J. M. (2022). Assortative (online) dating: Insights into partner choice from an experimental dating app. Computers in Human Behavior, 127, 107039. https://doi.org/10.1016/j.chb.2021.107039
  • Reis, H. T., Maniaci, M. R., Caprariello, P. A., Eastwick, P. W., & Finkel, E. J. (2011). Familiarity does indeed promote attraction in live interaction. Journal of Personality and Social Psychology, 101(3), 557–570. https://doi.org/10.1037/a0022885
  • Rosenfeld, M. J., Thomas, R. J., & Hausen, S. (2019). Disintermediating your friends: How online dating in the United States displaces other ways of meeting. Proceedings of the National Academy of Sciences, 116(36), 17753–17758. https://doi.org/10.1073/pnas.1908630116
  • Walster, E., Aronson, V., Abrahams, D., & Rottmann, L. (1966). Importance of physical attractiveness in dating behavior. Journal of Personality and Social Psychology, 4(5), 508–516. https://doi.org/10.1037/h0021188
  • Walter, K. V., Conroy-Beam, D., Buss, D. M., Asao, K., Sorokowska, A., Sorokowski, P., Aavik, T., Akello, G., Alhabahba, M. M., Alm, C., Amjad, N., Anjum, A., Atama, C. S., Atamtürk Duyar, D., Ayebare, R., Batres, C., Bendixen, M., Bensafia, A., Bizumic, B., ... Zupančič, M. (2020). Sex differences in mate preferences across 45 countries: A large-scale replication. Psychological Science, 31(4), 408–423. https://doi.org/10.1177/0956797620904154
  • Zhao, Y., Wang, Y., Zhang, Y., Wisniewski, P., Aggarwal, C., & Derr, T. (2024). Leveraging opposite gender interaction ratio as a path towards fairness in online dating recommendations based on user sexual orientation. Proceedings of the AAAI Conference on Artificial Intelligence, 38(20), 22547–22555. https://doi.org/10.1609/aaai.v38i20.30263