Who Stays Single? A Longitudinal and Global Investigation Using WVS Data

Abstract

Historically, singlehood is a growing demographic trend shaped by economic, social, and personal factors. This study examines the key influences associated with this phenomenon across diverse global contexts based on empirical evidence provided by WVS (World Values Survey), which covers over 100 countries and spans four decades. A multi-technique analytical approach is applied to identify the most robust predictors of singlehood. This approach involves feature selection, cross-validation, robustness checks, and statistical modeling (parsimonious models with near-excellent or excellent classification accuracy as AUCROC > 0.9). The results indicate that age and parental status are negatively associated with singlehood, while precarious employment status is positively linked. Co-residence with parents also appears closely related to singlehood. Other factors, including education level, social class, and settlement size, also correlate with singlehood patterns, as resulting from supplemental analyses. Moreover, gender and regional analyses reveal some variations in these associations, highlighting the interplay between personal, cultural, and economic contexts. These findings also align with social and economic theories of marriage, emphasizing the impact of life course factors, financial stability, and cultural norms. They contribute to a deeper understanding of demographic shifts. They also provide meaningful and well-founded insights as well as strategic guidance for policy in areas such as youth employment, social welfare, urban planning, and demographic adaptation.

1. Introduction

Singlehood is a growing demographic trend that reflects broader societal changes in marriage patterns, family structures, and economic conditions (Mortelmans et al. 2023). We can address it by referring specifically to individuals who have never been married (often seen as single by life stage or choice) and excluding divorced, separated, or widowed (who have past marriage experience). The latter is a widely recognized operational definition in demographic and social research (Timonen and Doyle 2013; Shahrak et al. 2021; Bear and Offer 2024), particularly when distinguishing between these two categories above.

While marriage remains a key institution in many cultures, an increasing proportion of the global population delays/foregoes it altogether. This shift raises critical questions about the underlying factors influencing singlehood (Spreitzer and Riley 1974; Yu and Hara 2022; Kislev 2023) and its implications for individuals and societies. Recent studies have highlighted the global rise in singlehood, with factors such as economic instability and changing cultural norms playing significant roles (Jackson 2018; Johansen et al. 2019; Rauer and Jager 2024). Understanding these predictors is essential for addressing demographic transitions (Lesthaeghe 2014), labor market dynamics (Carr et al. 2024), and evolving cultural norms (Ochnik and Slonim 2020). Moreover, by uncovering the key influences associated with singlehood, this study offers valuable insights for data-driven strategic planning in public management. The findings can support the development and refinement of youth employment and labor market policies, the design of family support and social welfare programs, the alignment of urban planning and housing strategies, and the adaptation of public services to address emerging demographic trends. Moreover, in terms of international policy, institutions such as UNESCO and the United Nations have emphasized the importance of strengthening family structures, social cohesion, and mental well-being, particularly in response to growing trends of individualization and demographic aging. The UN 2030 Agenda for Sustainable Development, notably through the 3rd goal (Good Health and Well-being) and Goal 5 (Gender Equality), indirectly addresses the socio-demographic conditions that shape partnership and family dynamics (Perez-Alvaro and Boswell 2024; Wonde 2024). At the supranational level, the European Union has acknowledged the demographic shifts impacting family structures through initiatives like the European Pillar of Social Rights, which supports equal opportunities, work–life balance, and social protection. The EU Strategy on Demographic Change (2020) also recognizes the increase in the number of single-person households. It also promotes policies fostering social inclusion and intergenerational solidarity (De Becker 2024; Dura 2024; Meira et al. 2024). In terms of national policy responses, Nordic countries such as Sweden, Finland, and Denmark (among the global leaders in singleness rates) have adopted progressive approaches. These include universal parental leave, state-supported childcare, and strong individual social safety nets, which reduce economic dependency on traditional family units. While these policies support personal autonomy and gender equality, they may also inadvertently contribute to lower partnership formation rates, a trend increasingly noted in demographic literature (Reine et al. 2024; Engdahl et al. 2022).

The primary goal of this research is to identify the most robust predictors of singlehood across diverse global and longitudinal contexts using World Values Survey (WVS) data, while testing key hypotheses related to economic, social, and cultural factors. By employing a multi-technique analytical approach, this study aims to fill gaps in understanding demographic shifts and offer data-driven insights for policymakers in areas such as youth employment, family support, and urban planning.

Among the techniques used in this study, it is worth mentioning feature selection (Adaptive Boosting (Wibowo and Abdul-Rahman 2018), Spearman Pairwise Correlation—based selections (Homocianu and Airinei 2022), and Bayesian Model Averaging (Li et al. 2025)), cross-validation procedures (both CVLASSO (Ahrens et al. 2025) and mixed-effects modeling (Hellingman et al. 2025)), different regressions together with overfitting removal (RLASSO) (Maruejols et al. 2022), collinearity (overall based on VIF (Vatcheva et al. 2016) and pairwise (Forzani et al. 2025) based on R-squared) and mutual causation checks (Snippe et al. 2014; Homocianu 2024a).

2. Literature Review

2.1. Economic Factors and Singlehood

Prior studies consistently demonstrate that economic conditions affect marital decisions. Employment status has a consistent influence on personal capacity to form and sustain such a relationship. Research has shown that job instability and financial insecurity can delay or discourage marriage (Killewald 2016; Smock and Schwartz 2020; Harsono et al. 2024). The latter aligns with the current study’s finding that employment is positively associated with singlehood, indicating that certain job conditions may inhibit partnership formation (Modena et al. 2013; Bastianelli and Vignoli 2021; Van Wijk et al. 2021).

Moreover, co-residence with parents, often driven by economic necessity, is also positively linked to singlehood (Britton 2013; Wang 2023; Yoon and Lian 2024). Living with parents may reflect delayed economic independence and thus correlate with the postponement of marriage.

Consequently, the first two main hypotheses emerged:

H1. 

Employment status is associated with singlehood, suggesting that certain job conditions or career-related factors contribute to remaining unmarried.

H2. 

Co-residence with parents is positively associated with singlehood, indicating that individuals who remain single are more likely to live with their parents.

2.2. Cultural and Social Factors

Other research has emphasized sociodemographic influences such as age, gender, race, class, and living arrangements in shaping singlehood (Roschelle 2013; Dommaraju and Tan 2023; Zhong and Wilkinson 2025; Staff and Vuolo 2024). Age is particularly pivotal, with evidence indicating that older individuals are more likely to marry. Similarly, having children (parental status) increases the likelihood of marriage, potentially due to social expectations and economic co-dependence structures.

Consequently, the other two main hypotheses resulted:

H3. 

There is a negative correlation between age and singlehood, with older individuals being less likely to be single than younger individuals.

H4. 

There is a negative correlation between parental status and singlehood, with individuals who have children being less likely to be single than those without children.

2.2.1. Age Range and Interpretive Considerations

This study uses the full available age range from the World Values Survey, including respondents aged 13 to 103. While acknowledging that singlehood carries different social and psychological meanings at various life stages (for example, between adolescence and late adulthood), the inclusion of the full age range (Tchoukou 2024) allows for a comprehensive, population-wide understanding of singleness patterns (Avilés et al. 2024; Adamczyk et al. 2022; Ratcliffe et al. 2024). To address age heterogeneity, age is usually treated as a continuous predictor in most statistical models (Yagi et al. 2024; Chu et al. 2024). Therefore, the resulting models reveal both general and age-specific trends. Nevertheless, caution is needed when interpreting results at the extremes of the age distribution (e.g., early teens or advanced old age), as expectations and societal norms may vary substantially in these groups.

2.2.2. Additional Social, Demographic, and Contextual Factors

In addition to these four core hypotheses above, several complementary expectations arise from the literature (Mortelmans et al. 2023; Spreitzer and Riley 1974; Yu and Hara 2022; Kislev 2023; Jackson 2018; Johansen et al. 2019; Rauer and Jager 2024; Lesthaeghe 2014; Carr et al. 2024; Ochnik and Slonim 2020; Avilés et al. 2024; Adamczyk et al. 2022; Ratcliffe et al. 2024) and from the cross-validation and subgroup analyses performed in this study. They concern education, social class, settlement size, gender differences, regional variations, and temporal dynamics, as follows:

H5. 

Education level is positively associated with singlehood, suggesting that individuals with higher education are more likely to remain single, possibly due to delayed marriage or prioritization of career development.

H6. 

Urban residence is positively associated with singlehood, indicating that people living in larger settlements are more likely to remain single than those in rural areas, possibly due to lifestyle and opportunity structures.

H7. 

Social class is associated with singlehood, with individuals from upper classes being more likely to remain single than those from lower classes, reflecting socioeconomic independence and changing family norms.

H8. 

Gender moderates the relationship between employment status and singlehood, such that employment exerts a larger influence among men than among women.

H9. 

The strength and direction of predictors of singlehood (e.g., age, parental status, and employment status) vary significantly across continents, reflecting regional cultural and economic contexts.

H10. 

The prevalence of singlehood has changed over time, reflecting societal and economic transformations influencing marriage decisions globally.

2.3. Psychological Theories: Maslow’s Hierarchy of Needs

The psychological dimension of singleness can be examined through Maslow’s Hierarchy of Needs (Poduska 1992; Finkel et al. 2014; Hoffman 2017). According to this framework, individuals prioritize basic needs (e.g., economic stability) before pursuing higher-level needs such as love and belonging, which include romantic partnerships (Dougall et al. 2022). When basic needs remain unmet—due to unstable employment or insufficient income (H1)—individuals may delay marriage, reinforcing the findings above regarding employment and co-residence with parents.

Additionally, living with parents (H2) may reflect an effort to meet safety needs when financial independence is absent, further postponing progression to love/belonging needs (Johnson and Ridgeway 2023; Wellsch et al. 2023).

This framework explains why economic instability (e.g., precarious employment) may delay marriage: unmet basic needs (safety, financial security) hinder progression to belonging needs. Similarly, co-residence with parents may signal stalled independence, psychologically reinforcing singlehood (Dougall et al. 2022; Johnson and Ridgeway 2023; Wellsch et al. 2023; Al-Habies et al. 2024; Blbas et al. 2025).

2.4. Integration and Gaps

Despite the breadth of existing literature (Ahmad et al. 2024; Costello et al. 2023; Hadi et al. 2024; Sihag 2024), the simultaneous consideration of these factors (economic, cultural, and psychological) and their varying impact across different cultural and economic contexts remains underexplored. This study aims to bridge this gap by applying a causal modeling approach to identify direct and indirect influences on loneliness. It also provides a broader framework for understanding the societal and personal factors influencing singlehood while standing on a statistically robust framework. Moreover, this study addresses challenges arising from demographic shifts, youth employment difficulties, and changing family dynamics, offering valuable insights for policymakers, researchers, and social planners. As societies continue to evolve, understanding the predictors of singlehood will be crucial for developing policies that support diverse life choices and respond to emerging social trends.

3. Materials and Methods

The data used belongs to the World Values Survey (WVS). The latter constitutes one of the most extensive empirical surveys of human beliefs and values, conducted across multiple countries over time and not for commercial purposes. Its most recent version (5.0-WVS_Time_Series_1981-2022_stata_v5_0.dta) has 1063 variables/fields, 443,488 raw observations, responses from more than one hundred countries, and seven waves (once every five years between 1981 and 2022). The latter is accessible online at https://www.worldvaluessurvey.org [accessed on 3 January 2025].

First, this dataset was loaded and cleaned in Stata using the REMDKNA command (Homocianu 2024b). Next, a basic binary encoding (X007bin) of the target variable (X007, Marital Status) occurred considering the coding of the original variable (top of

Table A2. (1). The most important items (version 5.0 of WVS) after ROUND 1. (2). General statistics for the most important items (version 5.0 of WVS). (3). Important WVS (v5.0) socio-demographic variables included in the cross-validation procedures of this study. (4). General statistics of important WVS (v5.0) socio-demographic items employed for cross-validation in this study.

(1)
Item/ Variable	Description				Coding
X007	Marital status (target variable—scale/original form)				1—Married; 2—Living as being married; 3—Divorced; 4—Separated; 5—Widow; 6—Single/Never married
X007bin	Marital status (target variable—binary/derived form)				1 if X007 = 6(Single/Never married); 0 if X007 is between 1 and 5
X002	Year of birth				Years with values between 1890 and 2007
X003	Age				Number of years (between 13 and 103)
X003R2	Age recoded (3 intervals)				1—15–29 years; 2—30–49 years; 3—50 and more years
X011	How many children do you have				0—No child; 1—1 child; 2—2 children; 3—3 children; 4—4 children; 5—5 children or more
X026	Do you live with your parents				0—No; 1—Yes
X028	Employment status				1—Full time; 2—Part time; 3—Self-employed; 4—Retired; 5—Housewife; 6—Students; 7—Unemployed; 8—Other
(2)
Variable		N	Mean	St.Dev.	Min.	0.25	Median	0.75	Max.
X007 (Marital status—scale form)		438,073	2.67	2.18	1	1	1	5	6
X007bin (Marital status—binary form)		438,073	0.25	0.43	0	0	0	0	1
X002 (Year of birth)		432,652	1965.05	18.16	1890	1953	1967	1979	2007
X003 (Age)		438,749	41.27	16.25	13	28	39	53	103
X003R2 (Age recoded—3 intervals)		438,749	2.01	0.77	1	1	2	3	3
X011 (How many children)		423,902	1.79	1.57	0	0	2	3	5
X026 (You live with parents)		416,283	0.3	0.46	0	0	0	1	1
X028 (Employment status)		430,456	3.31	2.16	1	1	3	5	8
(3)
Variable	Short Description				Coding Details
Continent	Africa, Asia, Europe, North America, South America or Oceania				1—Africa; 2—Asia; 3—Europe; 4—North America; 5—South America; 6—Oceania
S002VS	Chronology of WVS waves				1—1981–1984; 2—1989–1993; 3—1994–1998; 4—1999–2004; 5—2005–2009; 6—2010–2014; 7—2017–2022
S020	Year of the Survey				Years between 1981 and 2023
S024	Country & Wave (concatenation)				Values between 83 for Albania (8)+Wave 3 and 9097 for Northern Ireland (909)+Wave 7
S025	Country & Year (concatenation)				Values between 81998 for Albania (8)+1998 and 9092022 for Northern Ireland (909)+2022
X001	Sex				1—Male; 2—Female
X025R	Education level				1—Lower; 2—Middle; 3—Upper
X045	Social class (subjective)				1—Upper class; 2—Upper middle class; 3—Lower middle class; 4—Working class; 5—Lower class
X049	Settlement size				1—under 2000; 2—2000–5000; 3—5000–10,000; 4—10,000–20,000; 5—20,000–50,000; 6—50,000–100,000; 7—100,000–500,000; 8—500,000 and more
(4)
Variable		N	Mean	St.Dev.	Min.	0.25	Median	0.75	Max.
continent		388,736	2.72	1.32	1	2	2	3	6
S002VS		443,488	4.89	1.67	1	3	5	6	7
S020		443,488	2006.33	9.71	1981	1998	2006	2013	2023
S024		443,488	4586.97	2574.38	83	2333	4403	7044	9097
S025		443,488	4.60 × 106	2.60 × 106	81,998	2.30 × 106	4.40 × 106	7.00 × 106	9.10 × 106
X001		438,669	1.52	0.5	1	1	2	2	2
X025R		414,349	2.01	0.75	1	1	2	3	3
X045		380,524	3.31	0.99	1	3	3	4	5
X049		325,750	5	2.51	1	3	5	7	8

Source: (1) Own processing using the label list and tabulate commands in Stata 17. Note: The intersecting results (the six most robust variables) stand out using the Bold aspect. (2) Prepared by the author using a specific command (Univar—Stata 17, namely univar X007 X007bin X002 X003 X003R2 X011 X026 X028 (full Stata script at https://tinyurl.com/2m4h273y—accessed on 28 November 2025 and full results at https://tinyurl.com/bdw39tf9—accessed on 28 November 2025). Note: The intersecting results (the six most robust variables) stand out using the Bold aspect. (3) Prepared by the author using the ‘label list’ and ‘tabulate’ commands in Stata 17. (4) Prepared by the author using a specific command (Univar-Stata 17), namely univar continent S002VS S020 S024 S025 X001 X025R X045 X049 (full Stata script at https://tinyurl.com/ydmvrdxe—accessed on 28 November 2025 and full results at https://tinyurl.com/4ehr7wbh—accessed on 28 November 2025).

(1), Appendix A). Immediately after these two steps, a .csv export file resulted. The purpose was to use the same cleaned dataset, including the binary derivation of the target for feature selection tasks in R.

The multi-technique approach has a strong basis, namely a selection of steps grouped in rounds (

Table A1. Overview of Multi-Technique Rounds for Identifying Singlehood Predictors Using WVS Data.

Round	Purpose	Key Methods and Tools	Output
Start	Load and prepare the WVS dataset for analysis	REMDKNA (remove DK/NA values); Binary derivation of marital status (X007bin)	Cleaned .csv to use with Rattle(R) and SQL Server Analysis Services
1: Initial Selection	Identify initial robust predictors via the intersection of multiple techniques	SCDM or Spearman pairwise correlation-based selection (Stata); Adaptive Boosting (Rattle in R); Decision Tree and Naïve Bayes (SQL Server Analysis Services)	Intersected variables (common across all four methods)
2: Other Robust Selections	Further refine robust predictors using advanced modeling	Bayesian Model Averaging (BMA); Rigorous LASSO (RLASSO); Cross-Validated LASSO (CVLASSO), all in Stata	Variables passing all three techniques
3: Collinearity Checks	Detect and remove non-collinear influences	VCPR (pairwise OLS collinearity); VIF (Variance Inflation Factor); NOMOLOG (preliminary nomograms), all in Stata	Non-collinear influencers (refined variable set)
4: Mutual Causation Checks	Verify directionality and rule out reverse influences	OLOGIT/OPROBIT regressions in Stata; Compare R2, AIC, BIC in dual-role regressions	Predictors more likely to influence singlehood (confirmed causality direction)
5: Non-Random Cross-Validation	Validate models with random effects for stability	MeLOGIT models (in Stata) with random effects (Education, Class, Settlement, Year, Wave, Country-Year/Wave)	Confirmed predictors and final risk-prediction nomograms
6: Control Variable Testing	Assess effects of socio-demographic controls on core predictors	Add each socio-demographic variable individually to the top 3 models (in Stata)	Effects of controls on core predictors (e.g., robustness checks)
7: Summary Representations	Visualize and summarize findings	Two-way tables + graphs (in Stata); Additional visual insights (binary target)	Additional insights for interpretation (e.g., tables, graphs)
Results	Interpret results and ensure reproducibility	Synthesis of all rounds	Discussion of findings; Google Drive repository with code, exports, and results

and Figure A1, Appendix A). All corresponding technical terms are available and fully explained in the List of Technical Abbreviations at the end of the manuscript. Therefore, four different techniques were simultaneously served in the first selection round (ROUND 1). Moreover, only the intersecting results remained. The first technique used in ROUND 1 was a command focused on selections based on Spearman pairwise correlations (SCDM) (Homocianu 2025) in Stata 17 × 64 for multiprocessing. The second relied on using the Adaptive Boosting technique based on decision tree classifiers in Rattle (version 5.5.1) and the same binary form of the target variable. The third and the fourth involved using two classification algorithms based on decision trees (Wu et al. 2025) and Naïve Bayes (Antas et al. 2022). Both methods are implemented in the Microsoft Data Mining add-in for Excel (Microsoft Office 2013), which operates alongside SSAS 2016 on a Windows 10 Professional ×64 VM.

Only the intersecting results in ROUND 1 entered the second one (ROUND 2), which consisted of consecutively using the BMA command (reporting PIP (Bather et al. 2024)) and two additional ones from the Least Absolute Shrinkage and Selection Operator (LASSO) (Hastie et al. 2015) pack, namely RLASSO and CVLASSO.

The collinearity checks in the next round (ROUND 3) applied only to the variables remaining from the previous one (ROUND 2). Thus, VCPR (Tirnauca and Homocianu 2024) served to identify collinear variable pairs, VIF or the Variance Inflation Factor for checking the multi-collinearity in multi-input models, and NOMOLOG (Zlotnik and Abraira 2015) for creating nomograms with redundancy removal purposes.

Mutual causation checks were used in ROUND 4 and only for the variables remaining from ROUND 3. Thus, ordinal LOGIT (OLOGIT) and PROBIT (OPROBIT) regressions in Stata 17 considered the initial scale of the variable to analyze and all other input variables that remain in this setup. The roles of these variables and the original target were alternated in paired regression models. A higher R-squared value—indicating a stronger alignment between observed data and the theoretical model—or lower AIC and BIC values—reflecting improved model fit (Mousazadeh et al. 2025) and reduced information loss (Adámková et al. 2025)—in the reversed configuration suggest a noteworthy insight into the directionality of the relationship. Specifically, when an input variable yields a better model fit (i.e., higher R-squared (Tamagusko and Ferreira 2025) or lower AIC/BIC) in predicting the original target, it may be interpreted as a more plausible explanatory factor, rather than a consequence of the target variable.

Non-random cross-validations (ROUND 5) have as a basis on consecrated socio-demographic variables (education level, social class, and settlement size). They also use some other variables related to wave chronology, survey year, combination of country and wave, and country and year. These seven act as random effects in mixed-effects models (MeLOGIT regressions implemented in Stata 17, with the remaining predictors incorporated as fixed effects to control for their influence (Vermunt 2005; Dey et al. 2025)). Moreover, two values for gender and six for continents represent IF conditions and values for additional cross-validation criteria when testing based on LOGIT and generating risk-prediction nomograms (based on the same NOMOLOG command) for these specific conditions (Homocianu et al. 2020) and the overall data. Beyond the core predictors’ validation, the nomograms augment the original visual constructs. This results when using annotations as scores/values representing the effect sizes of predictors within continental or gender-specific models. Thus, they serve as a reference used for supplementary comparative computations. Moreover, they indicate how the three main predictors act according to gender or regional contexts. They also show possible deviations from the overall model.

These three core predictors were also confirmed when using Histogram Gradient Boosting in Python (Python 3.12.12 64-bit via the Spyder IDE, version 5.5.1) as a parallel validation approach.

Moreover (ROUND 6), most of the variables previously used for cross-validation (all except continent and those two combinations of country-wave and country-year with values not tied to a specific intensity scale) were used for control purposes (Biancalana and Mancosu 2025) (in new models). This procedure consisted of sequentially incorporating each variable into the existing most robust models (those three most resilient predictors) or testing them individually.

In ROUND 7, two-way tables and visual representations (Watson and Callingham 2014; Suha and Renjith 2023) of the relations between every single relevant input variable (the most robust ones, including those in the core model together with critical socio-demographic and chronological items) and the target (on average, based on its binary form) resulted in additional insights.

The model interpretation (including risk-prediction nomograms) stands on two sections meant for introducing the corresponding results and a discussion. For replicability (Drill et al. 2024), a permanent online Google Drive repository (https://tinyurl.com/ynvyms7a, accessed on 28 November 2025) keeps all derivation, export, selection, and analysis script sequences. This container also hosts all intermediary results necessary for this study. Each original URL for a relevant component was converted into a TinyURL.

4. Results

The findings presented here correspond to the sequence of procedures described in the previous section.

For reasons related to supporting reproducibility or replication of scientific results (Homocianu 2015), version 5.0 of the dataset downloaded from the WVS site and analyzed in this paper was also shared in the online container mentioned above (both the original .dta format for Stata and the .csv, after cleaning the data and performing a binary derivation for the target variable).

The outcomes of all selection rounds, as well as the corresponding checks and tests, are available in full within the same online folder, for purposes of reproducibility. Thus, the results of the four techniques applied in R, Stata, and SQL Server Analysis Services (version 2016 × 64) (ROUND 1) are available in four dedicated online folders at https://tinyurl.com/28bsefn3 (accessed on 28 November 2025).

For each of those four, configuration captures (for Rattle in R and Analysis Services), exploration scripts (.do files for Stata), and snapshots of the results (.png, .pdf, .avi, and .txt files) are available to follow the progress of the analysis in each particular approach.

The intersecting results of the first selection round (ROUND 1) are available in Figure 1. They show that only six variables persisted across all four methods and three software applications used in ROUND 1. These are: X002—Year of birth, X003—Age, X003R2—Age recoded (3 intervals), X011—How many children do you have, X026—Do you live with your parents, and X028—Employment status. The first three are interdependent, and collinearity or redundancy naturally occurs among them. Moreover, only 382,766 non-null intersection observations remain when considering these six input variables, which result after ROUND 1 and the target one.

For the variables resulting from this first selection round, the corresponding coding (Table A2(1), Appendix A—the original coding for all variables except the binary derivation of the target-X007bin) and their descriptive statistics (Table A2(2), Appendix A—also emphasizing their high support given the large number of observations, N) are available bellow (source script at https://tinyurl.com/kbz32nj5—accessed on 28 November 2025 and primary results at https://tinyurl.com/3n7sbacx—accessed on 28 November 2025).

The second selection round (ROUND 2-results available online at https://tinyurl.com/yckz62tb—accessed on 28 November 2025) involved consecutively using BMA along with RLASSO and CVLASSO in Stata until selections stabilized with no loss. The BMA did not eliminate anything. All PIP values indicated a maximum probability of inclusion for each of those six variables. RSLASSO and CVLASSO dropped the variable indicating the year of birth (coded as X002). The latter was not a real loss because X002 was suspected of being redundant (when considered together with X003 and X003R2 as age in years and on intervals-most probably serving as the primary source for these two) at the end of the previous round.

At this point, there are more non-null observations than after ROUND 1, namely 388,736, at the intersection of the remaining five input variables resulting after ROUND 2 and the target one.

The collinearity tests in the third selection phase (ROUND 3) started from the set of five remaining variables (the first test). First, VCPR identified a collinear pair, as already suspected, namely X003 and X003R2.

Moreover, a simple OLS regression keeping all five remaining variables indicated a problematic computed VIF. The latter is more than the maximum acceptable threshold (using the formula: 1/(1 − the R-squared corresponding to a model)), namely 6.2745 (much larger than 1.8902—results available online at https://tinyurl.com/48u5vrvk—accessed on 28 November 2025).

A preliminary nomogram (Figure 2) generated (using NOMOLOG) for a logit model with these five remaining variables suggested the elimination of X003R2 from this redundant pair (X003 vs. X003R2) because of a lower magnitude of effects (a smaller red segment compared to the blue one—as seen on the bottom of Figure 2).

In the second collinearity test in ROUND 3, the remaining set of four variables (X003, X011, X026, and X028) indicated no collinearity issue, both by VCPR (no collinear pairs identified) and VIF measurements (a computed VIF of 1.4527, lower than the maximum accepted threshold of 1.8742, depending on the R-squared). These results are available online at https://tinyurl.com/4yk7jxjn (accessed on 28 November 2025).

An equal amount of non-null observations (388,736—

Table A8. Descriptive statistics for the remaining set of four most relevant WVS items used in the overall model as well as in the continental ones.

Variable	N	Mean	St.Dev.	Min.	0.25	Median	0.75	Max.
X007	388,736	2.65	2.18	1	1	1	5	6
X007bin	388,736	0.25	0.43	0	0	0	0	1
X003	388,736	41.22	16.14	13	28	39	53	103
X011	388,736	1.8	1.57	0	0	2	3	5
X026	388,736	0.29	0.46	0	0	0	1	1
X028	388,736	3.32	2.16	1	1	3	5	8

Source: Prepared by the author using a dedicated command (Univar-Stata 17, namely univar X007 X007bin X003 X011 X026 X028 if X007!=. & X003!=. & X011!=. & X026!=. & X028!=.).

, Appendix B) stands at the intersection of the remaining four input variables resulting after ROUND 3 and the target one.

This amount later supported the derivation of continental models based on 104 countries (

Table A4. Additional cross-validations using the continent criterion (and its values) in LOGIT regressions.

MODEL	(1) OVERALL	(2) Continent1 (Africa)	(3) Continent2 (Asia)	(4) Continent3 (Europe)	(5) Continent4 (North America)	(6) Continent5 (South America)	(7) Continent6 (Oceania)
X003	−0.0579 ***	−0.0778 ***	−0.0608 ***	−0.0542 ***	−0.0466 ***	−0.0425 ***	−0.0397 ***
	(0.0005)	(0.0017)	(0.0011)	(0.0009)	(0.0011)	(0.0012)	(0.0022)
X011	−1.7691 ***	−1.5259 ***	−3.4237 ***	−1.7988 ***	−0.9877 ***	−1.4297 ***	−1.5130 ***
	(0.0125)	(0.0230)	(0.0605)	(0.0276)	(0.0204)	(0.0255)	(0.0593)
X028	0.1291 ***	0.1388 ***	0.0833 ***	0.1385 ***	0.1167 ***	0.0953 ***	0.1916 ***
	(0.0023)	(0.0056)	(0.0046)	(0.0046)	(0.0064)	(0.0061)	(0.0165)
_cons	2.1855 ***	2.9628 ***	2.7223 ***	1.7423 ***	1.6689 ***	1.9515 ***	1.1305 ***
	(0.0182)	(0.0562)	(0.0397)	(0.0346)	(0.0443)	(0.0466)	(0.0864)
N	388,736	64,061	138,546	95,348	35,227	44,396	11,158
chi-squared	50,264.2664	9412.8336	8494.89	13,455.4214	5597.8870	6025.2529	1339.6273
p	0.0000	0.0000	0.0000	0.0000	0.0000	0.0000	0.0000
R-squared	0.5126	0.5480	0.6758	0.4556	0.3455	0.4389	0.4171
AIC	211,648.8812	37,378.3007	47,270.15	50,245.6568	28,101.3843	29,933.3674	6114.0808
BIC	211,692.3638	37,414.5711	47,309.5	50,283.5180	28,135.2626	29,968.1711	6143.3604
AUCROC	0.9377	0.9458	0.9725	0.9222	0.8790	0.9158	0.9126
chi-squared GOF	7.1 × 105	55,188.09	5.20 × 108	3.8 × 105	14,354.31	46,639.98	20,068.99
p GOF	0.0000	0.0000	0.0000	0.0000	0.0000	0.0000	0.0000
maxProbNlog  PenultThrsh	0.9000	0.9000	0.9000	0.8000	0.8000	0.8000	0.8000
maxProbNlog  LastThrsh	0.9500	0.9500	0.9500	0.9000	0.9000	0.9000	0.9000

Source: Prepared by the author in Stata (https://tinyurl.com/574k3fpd—accessed on 28 November 2025). Notes: Robust standard errors appear in parentheses. Coefficients denoted by *** indicate significance at the 1‰ level.

, Appendix A, and

Table A7. Details on the continents and corresponding countries.

Continent	Continent Code (In This Paper)	Country Code (COW_ALPHA in WVS)	Country	N (No. of Obs.)
Africa	1	ALG	Algeria	2257
Africa	1	BFO	Burkina Faso	1265
Africa	1	EGY	Egypt	8769
Africa	1	ETH	Ethiopia	2726
Africa	1	GHA	Ghana	3086
Africa	1	KEN	Kenya	1230
Africa	1	LIB	Libya	2989
Africa	1	MLI	Mali	1181
Africa	1	MOR	Morocco	4619
Africa	1	NIG	Nigeria	7929
Africa	1	RWA	Rwanda	2639
Africa	1	SAF	South Africa	15,831
Africa	1	TAZ	Tanzania	1100
Africa	1	TUN	Tunisia	2251
Africa	1	UGA	Uganda	996
Africa	1	ZAM	Zambia	1500
Africa	1	ZIM	Zimbabwe	3693
Asia	2	ARM	Armenia	4255
Asia	2	AZE	Azerbaijan	3003
Asia	2	BNG	Bangladesh	3948
Asia	2	CHN	China	9298
Asia	2	GRG	Georgia	2698
Asia	2	HKG	Hong Kong SAR	3198
Asia	2	IND	India	12,255
Asia	2	INS	Indonesia	6137
Asia	2	IRN	Iran	6230
Asia	2	IRQ	Iraq	7288
Asia	2	JPN	Japan	8173
Asia	2	JOR	Jordan	3618
Asia	2	KZK	Kazakhstan	2518
Asia	2	KUW	Kuwait	1172
Asia	2	KYR	Kyrgyzstan	3548
Asia	2	LEB	Lebanon	2375
Asia	2	MAU	Macau SAR	803
Asia	2	MAL	Malaysia	3808
Asia	2	MON	Mongolia	1638
Asia	2	MYA	Myanmar	1200
Asia	2	PAK	Pakistan	5872
Asia	2	PSE	Palestine	990
Asia	2	PHI	Philippines	4792
Asia	2	QAT	Qatar	1052
Asia	2	SAU	Saudi Arabia	1502
Asia	2	SIN	Singapore	5431
Asia	2	ROK	South Korea	3605
Asia	2	TAW	Taiwan ROC	4448
Asia	2	TAJ	Tajikistan	1200
Asia	2	THI	Thailand	3931
Asia	2	TUR	Turkey	11,189
Asia	2	UZB	Uzbekistan	2691
Asia	2	DRV	Vietnam	3688
Asia	2	YEM	Yemen	992
Europe	3	ALB	Albania	1849
Europe	3	AND	Andorra	1999
Europe	3	BLR	Belarus	3579
Europe	3	BOS	Bosnia and Herzegovina	2364
Europe	3	BUL	Bulgaria	2050
Europe	3	CRO	Croatia	1152
Europe	3	CYP	Cyprus	3004
Europe	3	CZR	Czechia	3243
Europe	3	EST	Estonia	2532
Europe	3	FIN	Finland	1971
Europe	3	GMY	Germany	7503
Europe	3	GRC	Greece	1180
Europe	3	HUN	Hungary	1638
Europe	3	ITA	Italy	976
Europe	3	LAT	Latvia	1165
Europe	3	LIT	Lithuania	978
Europe	3	MAC	Macao	2027
Europe	3	MLD	Moldova	3030
Europe	3	MNG	Montenegro	1273
Europe	3	NTH	Netherlands	3708
Europe	3	NOR	Norway	2145
Europe	3	POL	Poland	1949
Europe	3	ROM	Romania	5542
Europe	3	RUS	Russia	7549
Europe	3	SRB	Serbia	4603
Europe	3	SLO	Slovakia	2726
Europe	3	SLV	Slovenia	3060
Europe	3	SPN	Spain	6225
Europe	3	SWD	Sweden	4088
Europe	3	SWZ	Switzerland	3758
Europe	3	UKR	Ukraine	6482
North America	4	CAN	Canada	5940
North America	4	DOM	Dominican Republic	378
North America	4	SAL	El Salvador	1246
North America	4	GUA	Guatemala	2087
North America	4	HAI	Haiti	1927
North America	4	MEX	Mexico	10,222
North America	4	NIC	Nicaragua	1200
North America	4	PRI	Puerto Rico	2872
North America	4	TRI	Trinidad and Tobago	1996
North America	4	USA	United States of America	7359
South America	5	ARG	Argentina	5510
South America	5	BOL	Bolivia	2031
South America	5	BRA	Brazil	7592
South America	5	CHL	Chile	6576
South America	5	COL	Colombia	6021
South America	5	ECU	Ecuador	2402
South America	5	PER	Peru	6766
South America	5	URU	Uruguay	3925
South America	5	VEN	Venezuela	3573
Oceania	6	AUL	Australia	6419
Oceania	6	MAD	Maldives	1010
Oceania	6	NEW	New Zealand	3729
TOTAL				388,736

Source: Prepared by the author in Stata (https://tinyurl.com/34cyxnt8—accessed on 28 November 2025). Notes: For all continents, the non-NULL condition involved the target variable, the tri-core predictors, and the variable X026 (Do you live with your parents-discovered as a robust influence but not passing mutual causation checks or not a predictor).

and Table A8, Appendix B) from the total of 108 existing in WVS (no intersecting data for France-FRN, Israel-ISR, Northern Ireland-NIRL, and Great Britain-UKG—results available online at https://tinyurl.com/37ycfam3—accessed on 28 November 2025).

The mutual causation checks (with OLOGIT and OPROBIT on the initial format (1–6 scale) of the variable to analyze in the fourth round of selections (ROUND 4 results available online at https://tinyurl.com/yc6sn2ap—accessed on 28 November 2025) indicated X026 (Do you live with your parents) as less likely to persist in the list of remaining four variables. They show a larger R-squared and lower AIC and BIC (

Table A9. Mutual causation tests based on OLOGIT regression with comparisons for each column pair (ROUND 4).

MODEL	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
Input (Below)\ Target Var. (Right)	X007	X003	X007	X011	X007	X026	X007	X028
X003	−0.0450 ***
(Age)	(0.0002)
X011			−0.8730 ***
(How many children)			(0.0037)
X026					2.0017 ***
(You live with parents)					(0.0077)
X028							0.1401 ***
(Employment status)							(0.0015)
X007		−0.3749 ***		−0.6121 ***		0.4521 ***		0.1347 ***
(Marital status)		(0.0015)		(0.0017)		(0.0017)		(0.0014)
N	434,411	434,411	422,102	422,102	411,971	411,971	426,316	426,316
chi-squared	33,034.1432	62,330.6628	55,747.7595	122,516.8278	67,214.9644	73,173.9650	8858.8283	8921.1370
P	0.0000	0.0000	0.0000	0.0000	0.0000	0.0000	0.0000	0.0000
R-squared	0.0500	0.0222	0.1308	0.1181	0.0849	0.1629	0.0098	0.0069
AIC	986,851.8276	3,469,984.6546	872,458.2413	1,244,777.3004	900,265.1491	419,142.4188	1,010,206.5059	1,559,130.6996
BIC	986,917.7181	3,470,973.0118	872,523.9593	1,244,843.0184	900,330.7213	419,164.2762	1,010,272.2835	1,559,218.4031

Source: Prepared by the author in Stata (https://tinyurl.com/9m75jvr9—accessed on 28 November 2025). Notes: Robust standard errors appear in parentheses. Raw coefficients denoted by *** indicate significance at 1‰ (one per mile). Colors coding highlights relative model performance variable selection: green indicates better-performing models and selected variables, while red denotes poorer-performing models and variables not selected, across model pairs (1, 2) to (7, 8).

and

Table A10. Mutual causation tests based on OPROBIT regressions with comparisons for each column pair (ROUND 4).

MODEL	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
Input (Below)\ Target Var. (Right)	X007	X003	X007	X011	X007	X026	X007	X028
X003	−0.0267 ***
(Age)	(0.0001)
X011			−0.4718 ***
(How many children)			(0.0018)
X026					1.2180 ***
(You live with parents)					(0.0044)
X028							0.0842 ***
(Employment status)							(0.0009)
X007		−0.1994 ***		−0.3392 ***		0.2706 ***		0.0769 ***
(Marital status)		(0.0008)		(0.0010)		(0.0010)		(0.0008)
N	434,411	434,411	422,102	422,102	411,971	411,971	426,316	426,316
chi-squared	43,661.1632	58,213.7572	72,661.9039	111,964.8919	76,158.4643	77,234.9165	8979.3991	9449.9554
P	0.0000	0.0000	0.0000	0.0000	0.0000	0.0000	0.0000	0.0000
R-squared	0.0538	0.0201	0.1295	0.1104	0.0877	0.1617	0.0097	0.0067
AIC	982,892.1309	3,477,355.9347	873,738.1938	1,255,708.2678	897,490.5835	419,719.3540	1,010,280.7401	1,559,445.6220
BIC	982,958.0214	3,478,344.2919	873,803.9118	1,255,773.9858	897,556.1558	419,741.2114	1,010,346.5177	1,559,533.3255

Source: Prepared by the author in Stata (https://tinyurl.com/2n5acawf—accessed on 28 November 2025). Notes: Robust standard errors appear in parentheses. Raw coefficients denoted by *** indicate significance at 1‰ (one per mile). Colors coding highlights relative model performance variable selection: green indicates better-performing models and selected variables, while red denotes poorer-performing models and variables not selected, across model pairs (1, 2) to (7, 8).

, Appendix C) when setting X026 as the target and X007 (Marital status or the original target) as the only input than the opposite (X007 as the target and X0026 as the only input). Therefore, only the three variables (X003—Age, X011—Number of children, and X028-Employment status) survived ROUND 4.

In ROUND 5, the first non-random cross-validation used MeLOGIT regressions (results available online at https://tinyurl.com/bdh43w28—accessed on 28 November 2025) and seven validation criteria (random effects). It is about the education level, social class, settlement size, wave chronology, and survey year, together with two combinations between country and wave or year, which strongly validated all three most robust core predictors (

Table A11. The cross-validation results based on seven criteria in Mixed-Effects LOGIT (MeLOGIT) regressions.

MODEL	(1)	(2)	(3)	(4)	(5)	(6)	(7)
X003	−0.0582 ***	−0.0585 ***	−0.0587 ***	−0.0587 ***	−0.0573 ***	−0.0586 ***	−0.0557 ***
(Age)	(0.0044)	(0.0034)	(0.0023)	(0.0023)	(0.0046)	(0.0019)	(0.0011)
X011	−1.7676 ***	−1.7674 ***	−1.7891 ***	−1.7891 ***	−1.7354 ***	−1.7644 ***	−1.7384 ***
(How many children)	(0.0959)	(0.1025)	(0.1050)	(0.1050)	(0.1042)	(0.1465)	(0.0653)
X028	0.1302 ***	0.1289 ***	0.1284 ***	0.1284 ***	0.1351 ***	0.1202 ***	0.1346 ***
(Employment status)	(0.0055)	(0.0073)	(0.0053)	(0.0053)	(0.0210)	(0.0151)	(0.0061)
_cons	2.1709 ***	2.1379 ***	2.1682 ***	2.1682 ***	2.0903 ***	2.2473 ***	2.0196 ***
	(0.1181)	(0.1596)	(0.0777)	(0.0777)	(0.1968)	(0.0482)	(0.0387)
var(_cons[S002VS])	0.0108 **
(S002VS = Wave Chronology)	(0.0040)
var(_cons[S020])		0.4255
(S020=Survey Year)		(0.5022)
var(_cons[S024])			0.3296 ***
(S024=Country & Wave)			(0.0681)
var(_cons[S025])				0.3296 ***
(S025 = Country & Year)				(0.0681)
var(_cons[X025R])					0.0069
(X025R = Education level)					(0.0048)
var(_cons[X045])						0.0116
(X045 = Social Class)						(0.0063)
var(_cons[X049])							0.0124 *
(X049 = Settlement size)							(0.0057)
N	407,720	407,720	407,720	407,720	385,482	360,492	311,083
AIC	221,023.6526	219,893.7355	212,383.8847	212,383.8847	210,687.7992	196,050.4390	171,021.6988
BIC	221,078.2443	219,948.3272	212,438.4764	212,438.4764	210,720.3860	196,093.6199	171,074.9379

Source: Prepared by the author in Stata (https://tinyurl.com/yrt8s6np—accessed on 28 November 2025). Notes: [var_name] shows the cross-validation criterion. Robust standard errors appear in parentheses. The raw coefficients marked using *, **, and *** are significant at 5%, 1%, and 1‰.

, Appendix C), showing their robustness for different data clusters defined for each distinct value of the seven additional variables above, acting as validation criteria.

The results of the Histogram Gradient Boosting, used as a parallel validation approach in Python, indicate that those three core predictors are among the top five of the top 25 resulting features (more details at https://tinyurl.com/mvam4u6h—accessed on 28 November 2025).

In the same phase (ROUND 5), other simple cross-validations used LOGIT (results available online at https://tinyurl.com/554tw3va—accessed on 28 November 2025) together with two values for the gender criterion (

Table A3. The results of additional cross-validations using the gender criterion (and its values) in LOGIT regressions.

MODEL	(1) OVERALL	(2) Male	(3) Female
X003	−0.0583 ***	−0.0687 ***	−0.0504 ***
	(0.0005)	(0.0008)	(0.0006)
X011	−1.7698 ***	−1.9115 ***	−1.6168 ***
	(0.0122)	(0.0211)	(0.0149)
X028	0.1314 ***	0.2059 ***	0.0843 ***
	(0.0022)	(0.0035)	(0.0031)
_cons	2.1717 ***	2.5109 ***	1.8533 ***
	(0.0177)	(0.0261)	(0.0250)
N	407,720	194,075	210,196
chi-squared	52,402.4397	27,246.1443	24,746.5071
p	0.0000	0.0000	0.0000
R-squared	0.5113	0.5694	0.4543
AIC	221,256.9285	98,753.0667	117,948.1540
BIC	221,300.6018	98,793.7707	117,989.1772
AUCROC	0.9374	0.9507	0.9218
chi-squared GOF	7.3 × 105	5.8 × 105	2.9 × 105
p GOF	0.0000	0.0000	0.0000
maxProbNlog  PenultThrsh	0.9000	0.9000	0.8000
maxProbNlog  LastThrsh	0.9500	0.9500	0.9000

Source: Prepared by the author in Stata (https://tinyurl.com/hs7k4fa6—accessed on 28 November 2025). Notes: Robust standard errors appear in parentheses. Coefficients denoted by *** indicate significance at the 1‰ level.

, Appendix A) and six for the continent (Table A4, Appendix A and supporting countries in Appendix B, Table A7) set as conditions. Moreover, a risk-prediction nomogram emerged for each condition (the overall model at the top of Figure 3). These models (overall and continental or gender-focused) additionally validated those three core predictors. Moreover, the augmented nomograms further supported comparative computations of magnitude (maximum and effective individual and summed-up scores, the weight of the firsts in the lasts, as well as the correspondence of the latter with maximum and effective probability/risk values at the model level) emphasizing peculiarities corresponding to gender and regional contexts (through deviations from the overall model—Figure 4).

While fully acknowledging the importance of interaction effects—particularly between age, gender, employment, and parental status—in understanding singleness, their explicit modeling in this study was constrained by both statistical and practical considerations. Specifically, the three core predictors refer to two multi-level categorical variables (parental status with six categories and employment status with eight) and a continuous variable. The latter is the age, which ranges from 13 to 103 and has 90 unique values emphasized by “tabulate X003”). Modeling all possible pairwise and higher-order interactions would exponentially increase the number of parameters and severely inflate model complexity (Deen et al. 2025). The latter would not only hinder interpretability and parsimony, but they would also raise issues of overfitting, multicollinearity, unstable estimates (especially in subgroup combinations with limited support), and generalizability (Loureiro et al. 2025).

That said, some theoretically motivated interactions (such as age × gender or employment × parental status) are meaningful and worthy of attention. Still, to maintain clarity, generalizability, and statistical power, the focus was on the most robust main effects in this population-wide analysis. However, interaction modeling is a significant next step. The same applies to encouraging future research (especially those employing hierarchical models or interaction-aware machine learning algorithms with large, balanced samples) and systematically testing these dynamics (Höfler 2024). A similar plan involves pursuing this direction in follow-up studies, leveraging stratified analysis or regularized interaction models to capture more nuanced behavioral patterns.

As support (number of valid responses) for the variable to analyze and those three predictors from this tri-core model/triad (X007bin together with X011, X003 and X028) and their distribution on the survey year, all indicate observations between 1981 and 2023 and a total amount between 423.902 and 438.749 valid records (Table A2(2), Appendix A) and 407.720 at the intersection of those four variables above (Table A3, Appendix A, model 1-OVERALL) or 91.93% of the total amount of 443.488 observations of the WVS dataset (version 5.0). Because X026 (Do you live with your parents) also emerged as a robust influence, the actual amount reduced to 388,736 valid observations (Table A4, Appendix A, model 1-Overall, and

Table A6. Checking (LOGIT models) employing the triad of principal, evidence-backed influences together with the standard socio-demographic factors traditionally considered in this study (ROUND 6).

MODEL	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
X003	−0.0583 ***	−0.0586 ***	−0.0586 ***	−0.0574 ***	−0.0586 ***	−0.0555 ***	−0.0582 ***	−0.0403 ***
	(0.0005)	(0.0005)	(0.0005)	(0.0005)	(0.0005)	(0.0005)	(0.0005)	(0.0005)
X011	−1.7698 ***	−1.7681 ***	−1.7682 ***	−1.7349 ***	−1.7637 ***	−1.7379 ***	−1.7526 ***	−1.6615 ***
	(0.0122)	(0.0122)	(0.0122)	(0.0124)	(0.0130)	(0.0139)	(0.0123)	(0.0125)
X028	0.1314 ***	0.1316 ***	0.1315 ***	0.1353 ***	0.1232 ***	0.1346 ***	0.1413 ***	0.1066 ***
	(0.0022)	(0.0022)	(0.0022)	(0.0023)	(0.0023)	(0.0025)	(0.0023)	(0.0024)
S002VS		0.0239 ***
		(0.0031)
S020			0.0041 ***
			(0.0005)
X025R				0.0981 ***
				(0.0075)
X045					0.0623 ***
					(0.0056)
X049						0.0376 ***
						(0.0023)
X001							−0.2950 ***
							(0.0104)
X026								1.4139 ***
								(0.0111)
_cons	2.1717 ***	2.0606 ***	−6.1271 ***	1.8995 ***	2.0042 ***	1.8505 ***	2.5682 ***	0.9946 ***
	(0.0177)	(0.0228)	(1.0650)	(0.0260)	(0.0250)	(0.0236)	(0.0232)	(0.0206)
N	407,720	407,720	407,720	385,482	360,492	311,083	404,271	388,736
chi-squared	52,402.4397	52,496.3972	52,498.7734	50,097.3965	46,164.6159	40,290.7384	52,549.8543	62,136.2502
p	0.0000	0.0000	0.0000	0.0000	0.0000	0.0000	0.0000	0.0000
R-squared	0.5113	0.5114	0.5114	0.5068	0.5085	0.4989	0.5127	0.5445
AIC	221,256.9285	221,206.5122	221,205.9374	210,680.5337	196,155.8265	171,085.0211	218,301.0639	197,830.8615
BIC	221,300.6018	221,261.1038	221,260.5290	210,734.8450	196,209.8026	171,138.2602	218,355.6132	197,885.2148
AUCROC	0.9374	0.9374	0.9374	0.9361	0.9367	0.9339	0.9375	0.9431
chi-squared GOF	7.3 × 105	1.4 × 106	2.8 × 106	9.3 × 105	1.1 × 106	1.2 × 106	8.7 × 105	4.3 × 105
p GOF	0.0000	0.0000	0.0000	0.0000	0.0000	0.0000	0.0000	0.0000
maxProbNlog  PenultThrsh	0.9000	0.9000	0.9000	0.9000	0.9000	0.9000	0.9000	0.9000
maxProbNlog  LastThrsh	0.9500	0.9500	0.9500	0.9500	0.9500	0.9500	0.9500	0.9500

Source: Prepared by the author in Stata (https://tinyurl.com/2xcsesf8—accessed on 28 November 2025). Notes: Robust standard errors appear in parentheses, while coefficients accompanied by *** indicate significance at 1‰.

, Appendix A, model 8, with the largest AUCROC of 0.9431) or 87.65% of the total amount. This amount also supports the derivation of continental models. The results of using the tabstat command (tabstat X007bin, stat(count) by(S020); tabstat X011, stat(count) by(S020); tabstat X003, stat(count) by(S020); tabstat X028, stat(count) by(S020)) in Stata (script online at https://tinyurl.com/yck4hyxe—accessed on 28 November 2025 and results at https://tinyurl.com/4w2454d4—accessed on 28 November 2025) confirmed this above and, consequently, the fact that this study relies on all 34 years of observations (1981, 1982, 1984, 1989–1991, 1995–2014, and 2016–2023), namely all those that the underlying dataset provides.

Starting from the augmented prediction nomograms in Figure 3, the results in Figure 4 synthesize the entire computation behind the gender and continent-specific patterns (H9) in terms of deviations from the Overall Model (OVM) with three core predictors.

Figure 4 provides a comprehensive view of how the importance of key predictors in the overall statistical model varies across different subgroups. This information can be valuable for understanding the nuances of the data and potentially tailoring interventions or policies to specific populations.

Thus, those two gender-specific models and six continent-particular ones indicate the following deviations grouped by the three main predictors:

➢

→For the Number of Children (NoC) Predictor (Leftmost column in the bottom section of Figure 4):

• -The Africa (−16.79%) and North America (−16.28%) models show a significant decrease in the predictor’s importance compared to the overall model (OVM).
• -The Asian model (+27.66%) has the most substantial increase, meaning the number of children plays a more consistent role in this geographical context than in the overall model.
• -Europe (+4.96%), South America (+5.88%), and Oceania (+5.26%) show moderate increases in importance.
• -Gender-based models: The male respondents (MMR) model (−4.76%) exhibits a slight decrease in predictor importance, whereas the female respondents (MFR) model (+4.65%) shows a tiny increase.

➢

→For the Age (A) Predictor (Middle column in the bottom section of Figure 4):

• -Asia (−30.54%), Oceania (−21.05%), Europe (−8.94%), and South America (−8.13%) models all indicate decreases in the predictor impact, with the most critical drop in Asia.
• -Africa (+24.86%) and North America (+18.19%) models show consistent increases, meaning age has an increased effect on singleness in these regions than in the overall model.
• -The gender-specific models show only some tiny variations: MMR (+0.84%) suggests a slight increase in importance for male respondents, while MFR (−1.50%) indicates a minor decrease for female respondents.

➢

→For the Employment Status (ES) Predictor (Rightmost column in the bottom section of Figure 4):

• -The Asian (−57.45%), South American (−2.94%), and African (−0.94%) models exhibit a reduced impact associated with the employment status, with Asia showing an extreme drop.
• -Oceania (+75.44%), North America (+32.56%), and Europe (+9.33%) models indicate a more impactful role of employment status in explaining singleness. Oceania has the highest increase.
• -The gender-based models show MMR (+34.92%), meaning employment status has a much stronger influence in the male-only model. However, MFR (−30.23%) shows a sharp decline. The latter indicates that employment status is far less relevant in the female-only model (H8).

The Discussion section provides further insights into and observations on the differences related to the three core predictors (Figure 4). These differences appear when comparing the overall model (OVM) with the gender- and continent-specific models.

Additional controls (

Table A5. Checking (LOGIT models) all three principal and empirically supported influences (the triad/tri-core), together with the recognized socio-demographic factors included in the study (ROUND 6).

MODEL	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)	(10)
X003	−0.1204 ***
	(0.0006)
X011		−2.0967 ***
		(0.0118)
X028			0.1565 ***
			(0.0018)
S002VS				0.0004
				(0.0021)
S020					0.0004
					(0.0004)
X025R						0.4948 ***
						(0.0048)
X045							−0.1053 ***
							(0.0039)
X049								0.0571 ***
								(0.0017)
X001									−0.3695 ***
									(0.0071)
X026										2.5601 ***
										(0.0084)
_cons	3.0983 ***	0.8073 ***	−1.6547 ***	−1.1077 ***	−1.9536 **	−2.1358 ***	−0.7733 ***	−1.4249 ***	−0.5586 ***	−2.1499 ***
	(0.0183)	(0.0065)	(0.0075)	(0.0109)	(0.7296)	(0.0108)	(0.0132)	(0.0095)	(0.0110)	(0.0061)
N	434,411	422,102	426,316	438,073	438,073	410,342	376,706	324,834	433,301	411,971
chi-squared	39,035.6166	31,780.7631	7884.1269	0.0355	1.3506	10,808.1756	730.3737	1179.1757	2730.1774	92,209.5520
P	0.0000	0.0000	0.0000	0.8505	0.2452	0.0000	0.0000	0.0000	0.0000	0.0000
R-squared	0.2776	0.4537	0.0193	0.0000	0.0000	0.0224	0.0018	0.0033	0.0057	0.2293
AIC	352,228.7039	256,312.3252	468,342.3469	491,394.9171	491,393.5927	449,934.0741	419,770.9719	359,328.2871	482,266.0435	358,143.3239
BIC	352,250.6674	256,334.2312	468,364.2728	491,416.8973	491,415.5730	449,955.9236	419,792.6503	359,349.6692	482,288.0018	358,165.1813
AUCROC	0.8571	0.9113	0.5873	0.5003	0.5040	0.5981	0.5296	0.5406	0.5461	0.7751
chi-squared GOF	80,188.11	4.8 × 105	77,082.51	662.19	3737.64	1966.44	259.85	78.94	0.00	0.00
p GOF	0.0000	0.0000	0.0000	0.0000	0.0000	0.0000	0.0000	0.0000	.	.
maxProbNlog  PenultThrsh	0.9000	0.5000	0.3000	.	.	0.2000	.	.	.	0.5000
maxProbNlog  LastThrsh	0.9500	0.6000	0.4000	0.0000	0.2000	0.3000	0.3000	0.2000	0.3000	0.6000

Source: Prepared by the author in Stata (https://tinyurl.com/5x5b56b6—accessed on 28 November 2025). Notes: Robust standard errors are reported in parentheses. Coefficients marked with ** and *** are significant at 1% and 1‰.

and Table A6, Appendix A—ROUND 6) revealed the secondary importance of certain well-established socio-demographic variables (most of them already used for cross-validations and another one already confirmed as an influence, namely X026) and successfully demonstrated again the robustness of the core model with only three predictors (results available online at https://tinyurl.com/2ucyucx3—accessed on 28 November 2025). The only exception is the variable corresponding to social class (X045), originally coded with lower values for upper social classes. The letter appears to behave oppositely when considered in conjunction with the tri-core model and when analyzed independently. In the first scenario (Table A6, Appendix A), the corresponding coefficient (Model 5) indicates a positive correlation. In the second (Table A5, Appendix A), the afferent coefficient (Model 7) shows a negative correlation.

Other explanations (the original coding) and descriptive statistics (number of supporting observations, mean value, standard deviation, minimum, maximum, and quartiles) regarding these additional variables are available in Table A2(3,4) (Appendix A).

Two-way graphical representations (Figure A2, Appendix A—ROUND 7) included in this study display the relationship between the target variable (average values between 0 and 1 on the Y axis) and each of the most robust variables, along with those in the socio-demographic category or those indicating the chronology (distinct values on the X axis), all based on tabulations by mean.

Thus, the results of the relationship between the target variable, as the mean proportion of respondents who are single/never married, and those four most robust influences indicate the following:

• Mean_X007_bin vs. X003 (Age)—there is a strong negative correlation. Young individuals (teenagers and those in their 20s) have the highest probability of singleness. As age increases, the likelihood of being single decreases sharply (stabilization at low levels around 30–50) (H3). In older ages (70+), there is a slight new increase, possibly due to widowhood.
• Mean_X007_bin vs. X011 (Number of Children)—a clear negative relationship. Those with no children have the highest probability of being single. The probability drops to near zero as the number of children increases (H4).
• Mean_X007_bin vs. X028 (Employment Status)—the variation is visible across different employment categories. Students (coded as 6) have the highest probability of being single. Full-time (1), part-time (2), and self-employed (3) have lower probabilities. Retired (4), housewives (5), unemployed (7), and other categories (8) are also less likely to be single (H1).
• Mean_X007_bin vs. X026 (Living with Parents)—there is a strong positive correlation. Those who live with their parents (value 1) have a much higher probability of being single when compared to those who do not (value 0) (H2).

The results of the relationship between the target variable, as the mean proportion of respondents who are single/never married, and two variables corresponding to the chronology of waves (S002VS_) and the survey year (S020_) appear below:

• For Mean_X007_bin vs. S002VS_, the proportion of single individuals fluctuates across waves. Still, it shows a slight increase as a general trend (H10). The lowest value appears in the earlier waves (1994–1998 and 1989–1993), suggesting fewer single individuals in those years. The highest proportion is observed in Wave 1 (1981–1984), followed by some increase in later waves. The latter suggests that the percentage of never-married individuals varies across survey waves but does not follow a strict linear trend.
• For Mean_X007_bin vs. S020_, the trend shows fluctuations rather than a clear increasing or decreasing pattern. However, in recent years (2010s–2020s), the proportion of single individuals appears slightly higher on average. Some years (e.g., post-2010) show spikes, possibly due to economic crises, shifting social values, or declining marriage rates.

The results of the relationship between the target variable, as the mean proportion of respondents who are single/never married, and those four socio-demographic variables corresponding to the education level (X025R_), social class (X045_), settlement size (X049), and gender (X001) indicate the following:

• For Mean_X007_bin vs. X025R_, the relationship suggests that as the education level increases from Lower (1) to Upper (3), the mean value of SINGLE/NEVER MARRIED (Mean_X007_bin) increases. The latter means that people with higher education levels are more likely to be single (H5).
• For Mean_X007_bin vs. X045_, there is an apparent negative association, which means that as the social category moves from the Upper (1) to the Lower (5), the single proportion decreases. The latter signifies that upper-class individuals are more likely to remain single when compared to those in lower social classes (H7). Still, the minimum value of singleness corresponds to the value of 4 (working class) and not 5 (lower class) of this variable, partially confirming its inconsistent behavior (coefficient sign) in regressions (Table A6, Appendix A, Model 5, and Table A5, Appendix A, Model 7).
• For Mean_X007_bin vs. X049_, a positive association is observable. As settlement size increases from small villages (1) to large cities (8), the likelihood of being single increases. The latter means that urban residents are more likely to be single than those from rural areas (H6).
• For Mean_X007_bin vs. X001_, the male respondents (1) have a higher singleness probability than females (2), which indicates that men are more likely to remain unmarried than women.

The findings align with global policy objectives (UNESCO, UN, Agenda 2020) that promote family stability, cohesion, and well-being in the context of increasing individualization and aging (Perez-Alvaro and Boswell 2024; Wonde 2024). Moreover, the EU responds to demographic shifts through the Pillar of Social Rights and the 2020 Demographic Change Strategy, promoting equality, inclusion, and solidarity (De Becker 2024; Dura 2024; Meira et al. 2024). In addition, Nordic countries address demographic change through generous welfare policies that foster autonomy and equality, but may also reduce partnership formation (Reine et al. 2024; Engdahl et al. 2022).

5. Discussion

5.1. Main Findings

This study employs recent data from WVS (World Values Survey) to identify the most significant predictors of singlehood. WVS also defines singlehood as the condition of being single or never married. This definition emerges from the marital status variable X007. Through a rigorous selection process involving multiple tests and analyses, the research reveals a core set of three key predictors that consistently influence singlehood: the number of children (X011), age (X003), and employment status (X028). These predictors demonstrate strong resilience, passing numerous robustness checks. However, these relationships are also objects of cautious interpretations. The latter applies because they do not always imply a unidirectional causal effect.

Parental status and age exhibit a negative relationship with singlehood. Specifically, individuals with fewer or no children are more likely to be single, while those with more children are less likely to remain unmarried. While this study identifies a significant negative association between the number of children and singlehood, researchers acknowledge that the causal direction is complex. It is reasonable to argue that being single reduces the likelihood of having children, especially in societies where childbearing strongly relates to marriage. However, the modeling approach adopted here allows exploratory insight into how parenthood correlates with marital status across populations. Although including the number of children as a predictor in nomograms based on preliminary mutual causation checks using the initial scale format of the variable (Table A9 and Table A10, Appendix C) serving as a source for the binary target, the interpretation of this relationship should remain tentative, as the evidence does not justify a clear-cut causal direction (first selections also due to correlations in ROUND 1, when using four techniques, including SCDM—Figure 1). The possibility of mutual influence between parenthood and marital status remains well-supported. Further longitudinal investigation may help clarify the nature of this relationship.

Similarly, older individuals are less likely to be single than younger ones. Although the analysis includes individuals as young as 13, the findings are interpreted primarily in terms of adult behavior.

In contrast, employment status shows a positive relationship with singlehood. Individuals in precarious employment situations (such as students or the unemployed, the latter coded with higher values) face a higher risk of being single. In contrast, those in stable employment (e.g., full-time or part-time jobs, coded with lower values) are less likely to remain unmarried.

The persistent significance of these three predictors across continental and gender-specific models underscores their robustness. This robustness is further supported by mixed-effects models and cross-validations using additional criteria, such as education level, social class, settlement size, survey wave, survey year, and combined country-wave and country-year indicators.

From a theoretical perspective, these findings align with established socioeconomic theories of marriage (e.g., Maslow’s Hierarchy of Needs (Poduska 1992; Finkel et al. 2014; Hoffman 2017)). Economic stability, employment, and living arrangements are fundamental factors shaping the ability to pursue marriage and family life (Staff and Vuolo 2024; Tchoukou 2024; Avilés et al. 2024; Chandler et al. 2004; Bartošová and Fučík 2017; Kaplan and Herbst-Debby 2017; Bergström and Brée 2023; Sharma 2024; Park et al. 2025). As such, the results fully confirm three hypotheses (H1, H3, and H4), further reinforcing their consistency and reliability. These insights are valuable for policymakers addressing demographic shifts, labor market challenges, and evolving family structures, ensuring that social policies align with modern relationship dynamics.

Another significant variable, co-residence with parents (X026), acts as a strong associated influence. It passed most robustness checks, and it contributed the utmost accuracy in classification (as measured by the highest AUCROC value of 0.9431—Table A6, Appendix A, Model 8) when added to the tri-core model. Co-residence with parents positively correlates with singlehood, meaning that individuals living with their parents are more likely to be single (Homocianu 2024a; Killewald 2016; Smock and Schwartz 2020; Park et al. 2025; Simpson 2006; Chitranshi and Dhar 2023), validating hypothesis H2.

Moreover, all these findings stand on the two-dimensional plots of the relationship between the target variable and the corresponding four most robust influences (number of children, age, employment status, and co-residence with parents). They align with common life patterns: younger people, students, and those without children are more inclined to remain single, while employment, having children, and independent living (not with parents) are significantly associated with being in a relationship or married.

Beyond socioeconomic and psychological interpretations, these results can also integrate with contemporary demographic and sociological perspectives. Theories of the second demographic transition (England and Xu 2025; Oehrlein and Wolf 2025; Kamińska and Mularczyk 2025) emphasize increasing individualization of the life course coupled with growing inequalities (Han and Brinton 2022), delayed family formation (Tran 2025), and the decoupling of marriage from cohabitation and childbearing (Raymo 2022), trends consistent with the rising prevalence of singlehood (Raz-Yurovich 2025). From a life-course perspective (Elder 1994; Ermer and Keenoy 2023), singlehood reflects a flexible sequence of transitions in adulthood rather than a fixed status (Fomby and Bosick 2013; Jager et al. 2024). These theories provide a robust interpretive lens for understanding the big why behind the rising prevalence of singlehood, particularly among younger, childless individuals in urban settings. From a Maslowian point of view (Poduska 1992; Finkel et al. 2014; Hoffman 2017), precarious employment and economic instability hinder progression beyond basic physiological and safety needs, thereby delaying or obstructing the pursuit of intimacy, belonging, and family formation. This helps explain why students, the unemployed, and those in unstable jobs exhibit significantly higher singlehood rates. Simultaneously, the second demographic transition (England and Xu 2025; Oehrlein and Wolf 2025; Kamińska and Mularczyk 2025) sheds light on how evolving social norms increasingly decouple marriage from traditional milestones such as parenthood, homeownership, and financial security. In this context, urban residence and higher education function not merely as correlates but as enablers of individualized life trajectories, where personal fulfillment, career advancement, and self-actualization take precedence over institutionalized partnership. These normative shifts are further reinforced by structural changes (e.g., rising housing costs, delayed economic independence, and the weakening of intergenerational support systems) that consolidate singlehood as a rational and adaptive life strategy rather than a residual demographic condition. Taken together, these psychological and sociological dynamics reveal why the observed patterns persist, namely that singlehood is less a failure of partnership formation (Reine et al. 2024; Engdahl et al. 2022) than a reflection of evolving priorities and constraints in late-modern societies. Moreover, changing gender norms and the reorganization of work and care roles (England 2010; Esping-Andersen and Billari 2015; Hook and Paek 2020) help explain the gender-specific patterns identified in this study. Integrating these perspectives situates the findings within broader demographic transformations observed worldwide.

In terms of strict relevance for data-driven strategic approaches to public management, understanding the key influences of singlehood as resulting from this paper can help strategic approaches to define the following:

• -Youth employment policies and labor market interventions.
• -Family support programs and social welfare initiatives.
• -Urban planning and housing policies.
• -Addressing demographic shifts and their consequences for public services.

5.2. Additional Findings

The first set of additional findings highlights socio-demographic influences of singlehood, including gender (X001), education level (X025R), social class (X045), and settlement size (X049). The chronology of responses does not indicate a relevant role (lack of significance when considered individually in regressions—Table A5, Appendix A, Models 4 and 5), as evidenced by the influences of the survey wave (S002VS) and survey year (S020). Among these influences, gender (X001) brought the highest classification accuracy (as measured by the AUCROC value of 0.9375—Table A6, Appendix A, Model 7) when added to the tri-core model.

Education level and settlement size demonstrate positive correlations with the likelihood of being single. Specifically, individuals with higher education levels (coded with larger values) and those from larger urban areas (coded with higher values) are more likely to be single. In contrast, gender shows a negative relationship with singlehood, with women (coded as 2) being less likely to be single compared to men (coded as 1). Social class behaves the opposite way when considered in conjunction with the tri-core model and individually. This statement stands on the opposite coefficient sign in the corresponding regressions. It shows a non-linear but generally negative relation with the target variable, with a minimum value of singleness for the working class or a value of 4 from the maximum of 5 on its corresponding scale (Mean_X007bin vs. X045, bottom-left of Figure A2, Appendix A, and Table A5, Appendix A, Model 7).

All the additional findings in this section also rely on two-way graphical representations of the relationship between the target variable and others corresponding to the chronology of waves, survey year and the four from the socio-demographic category. They suggest that marriage patterns have changed over time, but there is a non-linear and fluctuating relationship with singlehood. The latter has the lowest values for waves 3 (1994–1998) and 7 (2017–2022). It suggests changing cultural and economic factors affecting marriage rates. Moreover, the relationship between the target variable and the survey year reinforces this fluctuating pattern. It highlights year-to-year fluctuations. These fluctuations may result from economic cycles, global events, or some policy shifts. The highest value of the target variable at the mean (almost 0.6) corresponds to the year 2016. Other higher values (between 0.3 and 0.35) corresponded to the following years: 1981, 1984, 2000, 2002, 2003, 2021, and 2023. In addition, individuals with higher education, higher social class, urban residence, and male gender are more likely to be single. Part of these additional findings align with previous results in the literature, reinforcing the consistency of the observed patterns. For example, some studies have shown that education level and social class significantly influence singlehood, with higher education often delaying marriage (Perelli-Harris and Lyons-Amos 2015; Van Den Berg and Verbakel 2021; Chaloupkova 2023).

A second set of findings reveals gender- and continent-specific deviations from the overall model, as visualized through augmented prediction nomograms (Figure 4). For instance, the importance of the number of children as a predictor significantly decreases in Africa and North America but increases markedly in Asia. Europe, South America, and Oceania show only slight increases. Similarly, the importance of the age predictor declines consistently in Asia and Oceania while rising significantly in Africa and North America. Europe and South America exhibit minor decreases.

Employment status demonstrates the most notable variation across continents. Its importance increases sharply in Oceania and North America but drops significantly in Asia. Europe shows a moderate increase, while Africa and South America indicate slight decreases. Gender-specific analyses further reveal that employment status has a much more profound impact on singlehood for men than for women. In contrast, the number of children and age show only minor gender-related variations, with the former being slightly more influential for women and the latter for men.

These findings underscore the importance of considering cultural, economic, and regional contexts when analyzing predictors of singlehood.

5.3. Contextual Applicability of Findings Across Continents

While this study identifies robust predictors of singlehood that are generally consistent across global models, the observed continental deviations (Figure 4) suggest that interventions or policy responses should be context-sensitive rather than universally applied. The tri-core model (comprising children count, age, and employment status) offers a valuable baseline, but its operationalization must reflect the regional sociocultural and economic specificities.

For instance, the significantly greater importance of employment status in Oceania (+75.44%) and North America (+32.56%), compared to its markedly diminished relevance in Asia (−57.45%), implies that employment-focused policies or economic incentives to encourage marriage may be effective in the former but largely inadmissible in the latter. In Asia, where employment status is weakly associated with singlehood, interventions must account for broader structural or cultural dynamics beyond the labor market.

Similarly, the number of children carries heightened importance in Asia (+27.66%), a region where traditional family values and intergenerational expectations remain influential. The latter makes childcare policies, family tax incentives, or other pronatalist interventions potentially more effective. However, in Africa (−16.79%) and North America (−16.28%), where the explanatory power of this predictor declines, such strategies may not yield the intended outcomes—reflecting the inadmissibility of direct policy transfers.

Moreover, the predictive importance of age varies across continents. Africa (+24.86%) and North America (+18.19%) show a strengthened relationship between age and singlehood, while Asia (−30.54%) and Oceania (−21.05%) exhibit significantly weaker associations. The latter suggests that age-targeted interventions (such as programs supporting young adults entering the workforce or initiatives for delaying or encouraging earlier marriage) may be more effective in Africa and North America. In contrast, such strategies may prove less relevant or even counterproductive in Asia or Oceania, where familial pressure, socioeconomic constraints, or entrenched cultural norms may overshadow age as a key influence (e.g., in some Asian contexts, extended family expectations and housing affordability may influence singlehood more than chronological age).

The results indicate a solid policy transfer point: the regions must handle it with attention to detail. The EU experience, often rooted in social welfare frameworks, gender equality initiatives, and labor market interventions, may not align with other economic structures and cultural traditions (e.g., those of Asia or Africa). Therefore, while global models can guide high-level understanding, effective interventions require local calibration.

In sum, the tri-core framework proves to be regionally adaptive rather than prescriptive. Its utility depends on local validation, paying attention to regional variances in predictor strength, cultural norms, labor market dynamics, and demographic patterns. It is only at that point that strategies grounded in data and aligned with cultural contexts emerge.

5.4. Further Research Directions

Future research should investigate additional cultural and psychological predictors of this phenomenon of singlehood. Recent studies suggest that cultural attitudes play a crucial role in marriage-related decisions. Moreover, psychological traits, such as attachment styles, play a significant role in these decisions (Sassler and Lichter 2020; Shahi et al. 2023; Apostolou et al. 2023), and incorporating them alongside the quantitative approach used in this study would be beneficial. Furthermore, adopting longitudinal frameworks (Tessler 2023) could enhance our understanding of the dynamic interplay between individual and societal changes in shaping the phenomenon of singlehood. Specifically, inquiry into the influence of cultural attitudes, personal values, and psychological traits, such as attachment styles and personality characteristics, on singlehood may provide a more nuanced perspective. Qualitative methods, including interviews and focus groups, could augment the present quantitative findings by elucidating personal motivations, social pressures, and emotional factors relevant to singlehood. Longitudinal studies would also offer valuable insights into the temporal evolution of these predictors across diverse life stages, thereby facilitating a more robust differentiation between correlation and causation.

Moreover, subsequent research endeavors should examine the broader economic landscape (Marco-Gracia 2018; English 2019) influencing singlehood. Factors such as labor market regulations (Wojnicka et al. 2024), housing affordability (Balatonyi 2025), and parental leave policies may significantly influence the status of being single. The role of technology (particularly the impact of social media platforms, online dating applications, and digital communication modalities) warrants exploration. This exploration is necessary to elucidate the evolving dynamics of contemporary relationships (Diesen et al. 2025). Moreover, further investigation of gender and regional disparities is justified. The focus should be on understanding the differential impact of employment status on male singlehood and the influence of distinct cultural settings on singlehood trends. Examining life course transitions, including divorce, separation, widowhood, career shifts, health alterations, and migration events, would also provide valuable insights into how significant life milestones modulate the likelihood of remaining single.

Future research could also explore the integration aspect of national-level statistics (such as marriage and divorce rates from official statistical agencies). While this study focused exclusively on individual-level data from WVS, using country-level data in a multilevel modeling framework may offer important contextual insights. Such approaches would help bridge micro-level attitudes and macro-level trends, provided that appropriate variance and modeling structures are in place.

To address the underlying reasons why individuals remain single (beyond the statistical associations already identified) future research should incorporate a mixed-methods design that complements quantitative patterns with interpretive depth. Qualitative approaches such as in-depth interviews, focus groups, or narrative inquiries with never-married adults across diverse life stages and cultural contexts could reveal the subjective meanings and social dynamics that shape singlehood. These methods would help uncover personal motivations (e.g., intentional autonomy versus constrained opportunity), perceived social pressures (e.g., familial expectations, gender norms, stigma), and psychological factors (e.g., attachment styles, fear of relational failure, or prioritization of self-development) (Shahrak et al. 2021; Band-Winterstein and Manchik-Rimon 2014; Koren and Atamneh 2025; Van Gasse et al. 2025; Xu et al. 2025; Heta and Phillips 2025; Dennett et al. 2025; Rowlingson and McKay 2005). Such insights are essential for understanding why precarious employment may disproportionately affect men’s partnership formation (e.g., Oceania); why the number of children correlates more strongly with singlehood (e.g., Asia); or why urban, highly educated individuals delay or opt out of marriage despite economic independence. Longitudinal qualitative panels or diary studies could further illuminate how these motivations shift over time, helping to disentangle correlation from causation and identify key inflection points in the life course. When integrated with multilevel modeling of macro-level indicators (such as housing affordability, gender equality, or digital dating cultures), this approach is expected to support a more comprehensive theoretical framework for explaining the global rise in singlehood and guiding culturally responsive policy interventions.

Addressing these avenues of inquiry will enable future studies to develop a more comprehensive framework for understanding singlehood across diverse populations and social contexts.

6. Conclusions

This study presents a comprehensive analysis of the influences closely related to singlehood, drawing on recent data from WVS (World Values Survey) and employing a multi-method approach to ensure robust findings. Through rigorous feature selection, cross-validation, robustness checks, and statistical modeling (compact models with near-excellent or excellent classification accuracy, with AUCROC > 0.9), the analysis identifies robust and consistent statistical relationships between singlehood and several socio-demographic variables, notably age, parental status, and employment conditions. While these variables have strong correlations with singlehood in the models applied and remain stable across demographic, regional, and gender-based analyses, confirming their broad relevance, the results do not establish causal relationships. The analyses rely on observational data. They allow the detection of significant associations and temporal sequences, but not definitive causal inference. These results are best interpreted as correlational, emphasizing observed relationships without implying deterministic causality. Co-residence with parents also emerged following a simple logic: the individuals who remain single are more likely to live with their parents.

Other relevant correlations with education level, social class, settlement size, and gender emerged from additional analyses and contribute to a deeper understanding of the singlehood patterns. They also highlight the effects of economic and social security on marriage decisions. Therefore, they align with established theoretical frameworks such as Maslow’s Hierarchy of Needs.

The study reveals notable variations in the strength of these influences across gender and geographical contexts. Employment status has a more pronounced correlation with singlehood for men, and also in Oceania and North America. The number of children is more significant in these terms in Asia. These variations underscore the importance of considering personal characteristics, cultural and economic contexts to gain a deep understanding of singlehood.

The findings directly address the hypotheses of this study, confirming economic, social, and cultural influences on singlehood with supporting evidence. Employment status (H1) and co-residence with parents (H2) have positive associations. In these terms, some precarious jobs (e.g., students and unemployed) correlated with higher singlehood and two-way tables indicated a notable difference in co-residence between non-singles and those more likely to be singles. Age (H3) and parental status (H4) exhibit negative correlations, with older age and having children reducing the likelihood of being single. By contrast, higher education (H5) and urban residence (H6) positively correlate with singlehood. The social class (H7) correlates in the sense that the minimum likelihood of singleness corresponds to the working class (a higher value in the related original scale). In addition, the gender moderates effects (H8), with employment being more influential for men. Predictors vary by continent (H9). In fact, the employment counts more in Oceania and North America (increases of 75.44% and 32.56%, respectively), while the parental status in Asia (increase of 27.66% from the overall model). Moreover, the prevalence of singlehood has increased over time (H10), reflecting societal shifts.

By uncovering universal and context-specific influences, this study advances the understanding of the role of demographic shifts, evolving family structures, and social and economic conditions in shaping singleness. It also equips researchers, policymakers, and stakeholders with clear and valuable insights to address the multifaceted challenges of singlehood in contemporary societies.

Beyond advancing academic understanding, the findings from this study provide practical guidance for strategic public planning, ensuring better support for employment, welfare, housing, and service policies in response to changing family dynamics and demographic trends.