Introduction

Dynamic, relational developmental systems (RDS)-based models are at the forefront of contemporary theory in developmental science (e.g., Mascolo & Bidell, 2020; Molenaar et al., 2014; Nucci, 2019; Overton & Molenaar, 2015). Although several different models of development have been derived from ideas in RDS metatheory (Lerner, 2018a), all models emphasize that the fundamental process of development involves systematic, holistic, integrated, embodied, and mutually-influential relations (termed “dynamic relations) across the life span (Overton, 2015). These relations are most often represented in the literature as individual \(\Leftrightarrow\) context relations (e.g., Cantor et al., 2021; Nucci, in press). When individual \(\Leftrightarrow\) context relations are mutually beneficial both the individual and the context are maintained and perpetuated across time (e.g., Gottlieb, 1998; Johanson & Edey, 1981; Tobach & Schneirla, 1968).

In the life course of humans, these mutually beneficial relations prototypically involve the provisions of developmentally-nurturant coactions with others in the individual’s social world, that is, individual \(\Leftrightarrow\) individual relationships (e.g., Bornstein, 2015; Cantor & Osher, 2021; Cantor et al., 2021). Developmentally-nurturant coactions involve relationships that provide safety for a develo** individual and, as well, the resources for healthy and positive development in youth (e.g., warm, responsive, affirming, and authoritative parenting; e.g., Bornstein et al., 2022), in ways that promote feelings of well-being, of mattering to care providers, even being loved by them, and that their individuality is known and valued (Cantor et al., 2019). These coactive relationships are foundations in RDS-based models of the life course of health and positive human development (Lerner, 2018b). Examples are the theories of Bronfenbrenner and Morris (2006), regarding individuals’ development within the nested bioecological systems comprising human development; of Fischer and colleagues (e.g., Mascolo & Bidell, 2020; Mascolo & Fischer, 2015) regarding dynamic skills development involved in cognition and holistic learning; and of Raeff (2016, 2020), in the development of complex human cognitive, affective, and cultural systems.

Moreover, the central role of developmentally-nurturant relationships in promoting health and positive development is not only a foundation of developmental theories associated with human ontogeny but, as well, of all biological life. Examples are models of the processes involved in biological evolution, particularly of hominids and, most importantly, humans (e.g., Jablonka & Lamb, 2005; Johanson & Edey, 1981; Witherington & Lickliter, 2016); comparative behavioral development involving the importance in vertebrates of conspecific social relationships (e.g., Tobach & Schneirla, 1968); and the epigenetic changes involved in social genomics, again particularly among humans (Cole, 2014; Slavich & Cole, 2013). In short, within dynamic, RDS-based theories, developmentally-nurturant relationships—relationships that provide safety and afford a sense of being known as a specific individual that matters and belongs (Lerner et al., 2015, 2018a, b)—have fundamental and, it appears, universal significance for biological well-being (health) and, more generally, for thriving (Lerner et al., 2018a, b; Masten, 2014).

There is a link, then, between developmentally-nurturant relations—relations that enable an individual young person to feel that they are known, that they matter, and in fact that they may even be loved—and thriving, or positive youth development (PYD), within RDS-based models (Lerner et al., 2015, 2018a, b). Advancing the further study of the relation between developmentally-nurturant relationships and PYD through contributing to the measurement of this key relation is the main theoretical impetus of the present study.

Past literature associated with RDS-based approaches to understanding the dynamic coactions between an individual and the context that are constitutive of developmental change have often focused on the second decade of life, on adolescence (Bowers et al., 2015; Lerner et al., 2015). This focus emerged because of the panoply of biological, psychological, behavioral, and sociocultural changes marking life across this decade and, as well, because of the enhanced role that agency and choice play in an individual youth in setting the direction of development across subsequent decades (e.g., Bowers et al., 2015; Lerner et al., 2015). The adolescent period has become, then, a sample case assessing the usefulness of RDS-based models to describe, explain, and optimize individuals’ health and positive development, both in the U.S. and internationally (e.g., Damon, 2004; Larson, 2000; Leman et al., 2017; Petersen et al., 2017).

Perhaps not surprisingly, therefore, several RDS-based models of PYD have been tested (e.g., Barbarin et al., 2020; Damon, 2008; Murry et al., 2016; Smith et al., 2013; Spencer, 2006). However, as documented in discussions by Petersen et al. (2017) and by data presented by Qi et al. (2021), the theoretical conception of PYD formulated by Lerner and Lerner (e.g., Lerner, 2021; Lerner et al., 2015, 2018a, b) has been the predominant one used in the U.S. and around the world by researchers and youth program practitioners.

The Lerner and Lerner model (e.g., Lerner et al., 2015) specifies that young people will thrive when mutually beneficial youth \(\Leftrightarrow\) adult relationships exist. These relationships are instances of developmentally-nurturant ones (Lerner, 2004), and are one of three components of out-of-school-time youth programs that have been demonstrated to promote PYD (Lerner, 2018b). Programs successful in promoting thriving integrate developmentally-nurturant relationships between an adult and a youth with life-skill-building activities within the relationship and also offer opportunities for the youth to enact these skills in valued family, school, and community activities. When these “Big Three” (Lerner, 2004; see also Tirrell et al., 2020) attributes are present in a youth-development program, young people will show PYD, marked by “Five Cs,” that is, competence, confidence, connection, character, and caring. When these Cs are developed, a youth is likely to develop a “Sixth C,” that is, contribution, to self, family, school, and community. Among youth with high levels of PYD, contribution will transform during the middle-to-late portions of the adolescent period into positive civic engagement (Zaff et al., 2011); the young person will endeavor to contribute to the institutions of civil society and democracy and work to promote social justice and equity (Lerner, 2004).

Developmentally-nurturant relationships are the resource (or the developmental asset; Benson, 2008) enabling these developments to unfold, and Theokas and Lerner (2006) report that parents and other family members are the major source of this developmental asset. However, the second major source of this developmental asset, both in the U.S. and internationally, is youth development programs, especially those whose theory of change aligns with the above-noted Big Three features of programs, namely youth \(\Leftrightarrow\) adult relationships (e.g., Hershberg et al., 2015; Tirrell et al., 2021a, b).

The Lerner and Lerner PYD model is, at this writing, increasingly used globally in research and practice and, in particular, in low- and middle-income countries (LMICs; e.g., Geldhof et al., 2021; Lansford et al., 2021; Petersen et al., 2017; Tirrell, Dowling, et al., 2019). In these settings, youth often face challenges to their health and general well-being or, even more, experience traumatic events because of poverty, racism, and associated inequities in education, food, housing, health resources, and employment opportunities (e.g., Lauxman et al., 2021). Both researchers and practitioners delivering youth programs would find it beneficial, for theoretical and applied reasons, respectively, to possess a measure of youth perceptions of having the developmentally-nurturant relationships emphasized in the developmental science literature and, particularly, in contemporary RDS-based PYD models, such as the Lerner and Lerner model. Unfortunately, at this writing, such a measure does not exist.

This omission represents an important theoretical lacuna in the measurement model used to test the structural model proposed by Lerner et al. (e.g., 2015). Without evidence that an RDS-based model of PYD involves the presence of youth perceptions of being known and loved, the general developmental process specified in RDS-based conceptions (see Cantor et al., 2021) is weakened. Simply, a valid measure of youth perceptions of being known and loved must be part of a research process validating the existence of theoretically-specified covariation among the constructs included in the nomological net involved in the Lerner and Lerner model of PYD.

The Present Study

This article reports enactment of a measure-development study undertaken in order to fill the theoretical gap existing because a key measure—of youth feeling known and loved—has not been included in the nomological net of constructs, that is, the measurement model, used to test the theoretical conception of the thriving process specified within the Lerner and Lerner PYD model (e.g., 2019; Lerner et al., 2015) or to evaluate the usefulness of the model when it has been used to promote PYD among youth, particularly youth in LMICs who have developed within a context wherein they have encountered poverty and the above-noted associated inequities. To undertake this work, we embedded measure-development research within an ongoing multi-nation longitudinal study of PYD among youth who participated in the programs provided by Compassion International (CI) and among a matched group of youth not participating in these programs. The CI Study of PYD (e.g., Lerner et al., 2018a, b; Lerner et al., 2021; Tirrell, Dowling, et al., 2019; Tirrell et al., 2019a, b; Tirrell et al., 2020; Tirrell et al., 2021a, b) is a researcher-practitioner collaboration that was created to generate developmental data about the pathways of PYD traversed by youth living in three of the 25 LMICs (i.e., El Salvador, Rwanda, and Uganda) wherein CI practitioners deliver their programs.

CI is a youth-serving organization that seeks to enact theory-predicated and evidence-based approaches to its youth development programs (Sim & Peters, 2014; Tirrell et al., 2019a, b). CI is a faith-based child-sponsorship and youth development organization that serves more than 2.2 million children and youth living in poverty across 25 LMICs in Central and South America, the Caribbean, Africa, and Asia. CI partners with over 8,000 local churches and projects to enact its programs, which seek to promote PYD and thriving by using a holistic, strengths-based approach to its programs. Because CI is committed to the core belief that “all children should be known, loved, and protected” (Compassion International, 2022, p. 1), it believed that it was aligned well with the Lerner and Lerner model and, as such, it created a partnership that used the model to frame its youth programs.

Financial realities associated with the costs of conducting cohort-comparative counterfactual longitudinal research precluded conducting the CI Study of PYD within all of the nations served by CI. Accordingly, we present data collected in El Salvador and Rwanda, that is, the two nations served by CI within which we were able to conduct measure-development research prior to the COVID-19 pandemic; this research enabled us to assess whether an invariant index of being known and loved could be devised across different national contexts and, as well, whether, within either nation, scores from the measure being developed had patterns of covariation that were consistent with the Lerner and Lerner model.

In regard to the characteristics of these two countries, Rwanda has the highest population density in Africa (13 million people) and is one of the world’s poorest nations. The World Bank (2022a) reported that poverty has dropped in Rwanda since 2000, but 56.5% of people still lived below the extreme poverty line in 2016 (based on $1.90 per day per capita).Footnote 1 In turn, El Salvador is the third largest economy in Central America and the most densely populated. According to the World Bank (2022b), 22.3% of people in El Salvador lived below the poverty line in 2019 (down from 39% in 2007) and 1.5% lived below the extreme poverty line in 2019 (based on $1.90 per day per capita).Footnote 2

In addition to issues of poverty, within both El Salvador and Rwanda, crime and violence threaten social development and economic growth and negatively affect the lives of many youth. In 2015, El Salvador had the highest homicide rate in the world and one of the highest homicide rates among adolescents in the world (see Centeno, 2017). According to UNICEF (2017), El Salvador is one of the deadliest places in the world for boys (behind the Syrian Arab Republic, Iraq, Venezuela, and Colombia) and for girls (behind the Syrian Arab Republic, Iraq, Afghanistan, Honduras, Somalia, and South Sudan). In the nations outside of Latin America, most adolescent deaths are due to armed conflict and war (or collective violence) but, in El Salvador, most deaths are due to homicide (UNICEF, 2017). In Rwanda, almost two in ten adolescent girls experience forced sex, and nearly half of children aged five live with a mother who is a victim of intimate partner violence (UNICEF, 2017). The 1994 genocide also continues to affect the lives of many adolescents. Many of the children and adolescents who survived the 1994 genocide—now raising the next generation—are victims of rape, torture, starvation, mutilation, and emotional abuse, have lost at least one of their family members, and/or have witnessed violence, injury, or killing. The physical and emotional trauma parents suffer lead to negative indirect impacts on the lives of Rwandese youth (see Uwizeye et al., 2022) and, as well, studies demonstrate the intergenerational transmission of this trauma (see Kahn & Denov, 2022).

In sum, challenging and often traumatic experiences are parts of the lives of youth in both El Salvador and Rwanda. However, on the other hand, contrasting historical and current (at this writing) contextual conditions also frame the lives of youth across these two nations. Accordingly, we believe there is both theoretical value and potential applied significance in interrogating three questions: (1) Can youth perceptions of being known and loved be measured in both nations?; (2) Can the measure be shown to possess at least strong invariance within and across nations?; and (3) Are scores from the measure situated in a nomological net (Cronbach & Meehl, 1955) of covariations with other measures in a manner consistent with the Lerner and Lerner model of PYD? Although answers to these questions will of course require cross-validation in other CI countries and, as well, in studies conducted by other researchers and practitioners to test and use, respectively, the Lerner and Lerner model, we believe the data reported in the present article will provide a useful foundation for this subsequent work.

Method

The goal of the present article was, first, to develop a theory-based measure of youth perceptions of being known and loved—a goal pertinent to both the theoretical usefulness of the Lerner and Lerner model of PYD (Lerner & Lerner, 2019; Lerner et al., 2015) and to practitioners’ objectives for the youth development programs they are enacting—and, second, to assess whether the known and loved measure covaried within a nomological net of scores from measures of other constructs involved in the process of PYD proposed within the Lerner and Lerner (2019; Lerner et al., 2015) model. Consistent with RDS-based models of positive youth development, being known and loved represents a contextual resource marked by the “Big Three” and, in particular, the provision to youth of developmentally-nurturant relationships (Lerner, 2004, 2018a).

Participants

Two data sets from the CI Study of PYD were used: one involving one time of testing in Rwanda; and another involving one time of testing in El Salvador. Participants from the Rwanda data set consisted of 1,204 Rwandese youth (49% female) ages 9–16 years (Mage = 11.84, SDage = 1.67) sampled from rural and urban contexts (57.8% rural, 25.3% suburban, 16.9% urban), half of whom were enrolled in CI-supported programs. Participants from the El Salvador data set included 1,205 Salvadoran youth (49.8% female) ages from 9 to 18 years (Mage = 13.03, SDage = 1.87), sampled across both urban and rural areas (32.8% rural, 0.7% suburban, 66% urban), 51% of whom were in enrolled in CI-supported programs.

Measures

The present study involved the psychometric development of a measure of youth perceptions of being known and loved. In addition, in order to assess patterns of covariation within the above-noted nomological net of other constructs related to PYD, four existing measures of constructs were used: intentional self-regulation, hopeful future expectations, transcendence, and contribution (see Lerner, 2018a, b; Lerner & Lerner, 2019; Lerner et al, 2015, 2018a, b, for full details of the nomological net of constructs involved in the Lerner and Lerner measurement model).

In the Lerner and Lerner model, intentional self-regulation (ISR) is defined as the ability to select goals and work toward achieving them despite facing challenges and setbacks (Freund & Baltes, 2002). ISR is regarded as a youth strength that contributes to the development of the Five Cs. Similarly, hopeful future expectations (HFE), defined as possessing the expectation that good things will happen in the future, at least in part because of one’s agency (as indexed by ISR) is also considered a youth strength contributing to PYD (Schmid et al., 2011). In addition, spiritual transcendence, defined as an individual’s connection to a divine or supernatural other beyond the self (King et al., 2017), is also regarded as a strength of youth associated with PYD, especially among young people who have experienced disparities and challenging or traumatic experiences in their lives. The model includes the idea that, when these strengths are fostered and aligned with a young person’s perception that the context affords being known and loved by caring and supportive adults, then thriving will occur. Such thriving is then expected to covary with positive engagement with and contributions to their settings—that is, youth actively engage with and give back to the context that is supporting them (Lerner, 2004; Lerner et al., 2005, 2015; Zaff et al., 2010, 2011).

Known and Loved

Although prior research has developed measures that are consonant with the conception of either being known or being loved, no extant measure reflected the integration of these two concepts, particularly within the conception of developmentally-nurturant relationships found with RDS-based models of PYD in general or the Lerner and Lerner model more specifically. For example, research regarding feelings of “being known” includes employees feeling known by their work project teams (Purvanova, 2013), patients feeling known by their nurses and healthcare workers (Jacobsen et al., 2015; Somerville, 2009), and students feeling known by their instructors (Chhuon & Wallace, 2014; Rodriguez-Keyes et al., 2013; Wallace et al., 2012). In the study of adolescent development, research on “being known” has focused on the school context, for instance, being known by teachers and feeling a sense of connectedness and belonging to school (Nakkula, 2008; Stafford-Brizard, 2016; Wallace et al., 2012). In turn, in regard to youth being loved, Barrett et al. (2019) noted that, despite the importance and role of love in contributing to positive development across the life span, there remained the need for a measure of “feeling loved;” as such, Barrett et al. created a short self-report “feeling loved” measure that assessed both self-love and feeling loved. The Tirrell et al. (2020) measure of the “Big Three” features of effective youth programs included a scale of perceived program safety, as well as an adult mentorship component specific to adults working in youth programs.

However, what was still needed to measure the known-and-loved construct within the theoretical model framing the present work was a non-program-specific measure to assess youth perceptions of being known and loved by adults in their lives. Accordingly, 14 items were generated as part of the CI Study of PYD to assess whether youth felt known (seven items) and loved (seven items) by at least one adult in their lives. In the present study, we present below the process of refining the original known and loved (KL) item pool to a parsimonious and robust, single-factor, five-item scale.

Intentional Self-Regulation

Nine items derived from the Freund and Baltes (2002) measure were used to assess youth perceptions of their ability to choose and commit to goals (selection), create plans and make choices to achieve those goals (optimization), and compensate for setbacks in progress toward their goals (compensation). Cronbach’s alpha coefficients were 0.97 with the Rwanda sample, and 0.88 with the El Salvador sample. Omega coefficients were 0.97 with the Rwanda sample, and 0.88 with the El Salvador sample.

Hopeful Future Expectations

Six items derived from the 4-H Study of Positive Youth Development (Schmid et al., 2011) were used to assess youth perceptions of their likelihood of having positive experiences in the future (see Tirrell et al., 2019a, b regarding how the HFE measure was refined for use in the CI Study of PYD). Cronbach’s alpha coefficients were 0.97 with the Rwanda sample, and 0.85 with the El Salvador sample. Omega coefficients were 0.96 with the Rwanda sample, and 0.83 with the El Salvador sample.

Transcendence

Four items were used to assess youth perceptions of their connection to something beyond the self, which indicates an experience of the divine or supernatural other (e.g., God; King et al., 2017). Items were derived from the Transcendence factor of the King et al., (2017, 2019) Measure of Diverse Adolescent Spirituality (MDAS) (see Tirrell et al., 2019a, b regarding how the MDAS was refined for use in the CI Study of PYD). Cronbach’s alpha coefficients were 0.94 with the Rwanda sample, and 0.79 with the El Salvador sample. Omega coefficients were 0.95 with the Rwanda sample, and 0.78 with the El Salvador sample.

Contribution

Five items were used to assess youth perceptions of their capacity and desire to make positive contributions to their communities, considered an outcome of PYD and a key indicator of thriving (Lerner et al., 2005, 2015). These items were from the King et al. (2017) MDAS and the 4-H Study of PYD (Lerner et al., 2005). Cronbach’s alpha coefficients were 0.96 with the Rwanda sample, 0.87 with the El Salvador sample. Omega coefficients were 0.96 with the Rwanda sample, and 0.87 with the El Salvador sample.

Compliance with Ethical Standards

In regard to the ethics review process for the method used in the present research, the U.S.-based research team received de-identified data and was granted exempt status for secondary data analysis by the Tufts University Institutional Review Board (IRB). The research teams in Rwanda and El Salvador are composed of CI staff and follow the CI policy of adhering to a specific nation’s governmental IRB requirements, if any. CI requires that before youth are assessed, signed parental consent forms, as well as youth assent (if below the age of consent), be obtained. All youth were told that there are no penalties if they elect not to participate, that they can decide not to answer any question, and they can end their participation at any time, again without any penalties (Tirrell et al., 2019a, b for full details).Footnote 3

Procedure

To enhance validity of the measures, researchers met with local program staff and independent data collectors in each nation to introduce the purpose of the study and to help shape the proposed survey questions and answer options. Edits were made to the survey based on the feedback that was provided, including concerns that arose during an initial piloting phase of the survey (see Tirrell et al., 2019a, b; see also Lerner et al., 2018a, b). All materials were translated from English to Spanish and Kinyarwanda (for El Salvador and Rwanda, respectively) and back-translated to English to check for accuracy.

All measures relied on youth self-reported perceptions and assessments on items asking how much they related to or identified with the items. All items were responded to using a 100-point scale, with 100 reflecting the highest level of agreement with the item (e.g., “Always like me,” “Completely agree,” “100%, certain”), and 0 reflecting the lowest level (e.g., “Never like me,” “Completely disagree,” “0%, no chance”). Initial piloting of the survey showed that all youth consistently chose responses in intervals of 5 (e.g., youth would choose a response of 85 or 90, but never 82 or 91), so the survey design was altered to use a sliding scale that moved in increments of 5 to improve ease of use (effectively creating a 21-point scale). All participating youth were given the scales in the same order. Table 1S in the Electronic Supplementary Material (ESM) presents all items for the measures used in the present article.

Following ethical review and approval and obtaining informed consent, we collected data from youth participating in CI-supported programs in Rwanda and El Salvador and from a matched comparison sample of youth from the same communities within which CI youth resided within each nation. Eligibility for CI support was based on multiple criteria, including age, household monthly income relative to number of dependents, lack of other financial child-sponsorship support, proximity to a CI-supported project site, and additional indicators of poverty (e.g., housing quality, access to resources such as water source). Comparison samples of non-CI-supported youth from the same communities as the CI program sites were selected based on meeting CI’s eligibility criteria for sponsorship.

The CI national office in each nation selected CI project sites from urban and rural locations from which to collect data (18 sites in Rwanda and 20 in El Salvador). CI staff made these selections based on their judgments of sites that represented the best program outcomes and implementation (e.g., in regard to graduation rates and extra-program activities). CI national staff held three informational meetings for collaborating facilitators at the selected project sites, which included describing the study methodology and answering questions about the survey or broader research program.

Independent data collectors were recruited from a local university and/or recruited based on previous experience with children and technology (as the surveys were tablet-based). Data collectors went through three days of training prior to data collection. The first day was dedicated to educating them about CI (e.g., history, mission, impact, child protection policy, and procedure) and informing them about the purpose of the study. They were then introduced to the survey questions and response options and trained on the methodology. On the third day, data collectors piloted the survey with CI-supported project youth. Based on concerns and questions that came up at these training sessions, edits were made and documented in the records of the research team.

The local data collection team administered one-on-one interviews by reading the survey questions and entering youth responses into an online survey. The typical survey length was about 30–45 min to complete. Participating sites were given thank-you gifts as determined by the project staff, including gift baskets, shoes, and/or clothing. Data collection took place in Rwanda in May–June 2018, and in El Salvador in June–July 2018.

Prior to data collection, members of the researcher-practitioner collaboration worked together to draft new items to include in the KL item pool. Through roundtable-type discussions, CI leadership and practitioners elaborated on CI’s belief regarding all youth deserving to be known and loved and how those constructs might be manifested in the lives of youth. The results of these discussions were that the researchers worked to create a measure of youth feeling that they were known and loved that was framed by RDS-based metatheory (emphasizing person \(\Leftrightarrow\) context relations). In seeking to maximize the chances for strong psychometric properties (see Clifton, 2020), items were iteratively drafted and revised by CI Study of PYD partners. This process resulted in a pool of 14 items (seven each reflecting “known” and “loved;” see Table 1S in the ESM), generated with the intention of trimming the pool for parsimony and robustness once data were collected and analyzed. Items were then added to the survey and translated as described above.

Data Analyses

We first used one time of testing from Rwanda to test and refine the new KL item pool for parsimony and robustness and to assess its psychometric properties with the Rwanda sample. We then assessed the psychometric properties of the KL measure in El Salvador. Last, we tested for cross-national measurement invariance across Rwanda and El Salvador in an effort to establish cross-national validation of the measurement model.

Analyses were conducted using Mplus Version 8 (Muthén & Muthén, 1998–2017). Across all models tested, we used multiple goodness of fit indices as recommended by Brown (2006). Absolute fit was tested by checking for χ2 significance (although an oversensitivity of this test is noted due to large sample sizes; see Rutkowski & Svetina, 2017) and the standardized root mean square residual (SRMR), with values closer to 0 indicating better fit (Brown, 2006). Parsimony–corrected fit was assessed by evaluating the root mean square error of approximation (RMSEA) and its confidence interval, with values closer to 0 indicating better model fit (Brown, 2006). The suggested upper bounds, or cut-off values, of acceptable fit for the SRMR and RMSEA are 0.08, and ideally less than 0.05 (Brown & Cudeck, 1993; Hu & Bentler, 1999; Steiger, 1990). Comparative fit, the evaluation of the specified solution in comparison to a null model in which no items are correlated, was tested with the comparative fit index (CFI) and the Tucker-Lewis Index (TLI), with values closer to 1 indicating better model fit (Brown, 2006). The suggested lower bounds, or cut-off values, of acceptable fit for the CFI and TLI are 0.90, and ideally above 0.95 (Bentler, 1990).

We used the Rwanda data to test the 14 items pertinent to KL using an exploratory factor analysis (EFA). For the EFA, Geomin rotation (Mplus default) was used for a parsimonious factor pattern matrix. A combination of criteria was used to determine the adequate number of factors to retain, including a scree plot (Cattell, 1966), parallel analysis (Horn, 1965), and the above-noted multiple goodness of fit indices as recommended by Brown (2006). Out of necessity for fitting the EFA models, we had to ignore nesting.

Following the results of the EFA, we explored Pearson product-moment correlations among the 14 KL items in the Rwanda data to determine which items to trim for parsimony. Highly correlated items (r > 0.8) were considered empirically redundant and thus further explored regarding content/wording to determine which items to retain.

The psychometric properties of the resulting refined scale were then assessed in each data set (Rwanda and El Salvador). Internal consistency reliability was assessed using Cronbach’s alpha and the omega coefficient for model-based reliability. Convergent validity was tested by including measures of youth strengths (ISR, HFE, and transcendence; believed to coact with contextual supports like KL in promoting thriving) and contribution (a theoretically specified outcome of PYD; Lerner et al., 2015) and assessing their interrelations. Between-group measurement invariance was tested across age (9–12 and 13–18 years, given the age distributions in each data set), gender (boys and girls), urban–rural area, and CI-enrollment status, respectively, within each nation using a series of confirmatory factor analyses (CFAs) with added constraints, described further below (see Little, 2013).

For all models, maximum likelihood estimation with robust standard errors (ESTIMATOR = MLR in Mplus) was used, and nesting within project sites was accommodated using a sandwich estimator (TYPE = COMPLEX in Mplus). We used the criterion suggested by Cheung and Rensvold (2002) and further supported by Kim et al. (2017) for establishing invariance if changes in CFI are less than or equal to 0.01 across subsequent models (also due to an oversensitivity of the χ2 difference test to small differences in large samples; see Rutkowski & Svetina, 2017).

Between-group measurement invariance refers to the comparability of measures across specific groups (i.e., whether measures assess the same latent constructs and demonstrate strong psychometric properties). For between-group measurement invariance, we first tested for configural invariance (that is, equivalence of the measurement model structure) across groups (younger-older, boys-girls, urban–rural, CI-enrollment status, and nation). Next, we tested for metric (weak factorial) invariance (that is, equivalence of the factor loadings within the measurement structure) by equating the factor loadings across groups and allowing the latent variances in one group (older, boys, rural, non-CI-supported, and Rwanda, respectively) to be freely estimated. Then, we tested for scalar (strong factorial) invariance (that is, equivalence of the factor loadings and item intercepts within the measurement structure) by equating the factor loadings and item intercepts across groups and allowing the latent variances and latent means in one group (older, boys, rural, non-CI-supported, and Rwanda, respectively) to be freely estimated (see Little, 1997, 2013). Establishing invariance then enabled us to compare the means of the latent constructs and their interrelations across the respective groups.

In cross-national research, establishing measurement invariance across national contexts (in different countries with different cultures) enables comparisons to be made in interpreting measurement outcomes (van de Vijver, 2019). Accordingly, we last tested for between-group (cross-national) measurement invariance across Rwanda and El Salvador to provide initial evidence of cross-national validation (see Dimitrova & Wiium, 2021; Kim et al., 2017; Rutkowski & Svetina, 2017). Model constraints in the scalar (strong factorial) invariance model revealed whether mean levels of the latent constructs differed between nations; to compare the correlations among the latent constructs between Rwanda and El Salvador, the Wald test of parameter estimates was used (MODEL TEST in Mplus).

Results

Preliminary Analyses Establishing the KL Measure with the Rwanda Data

Item-level missingness with the Rwanda data was minimal (between 0.0 and 8.1% of data missing), allowing us to assume all data were missing at random (MAR). The MAR mechanism renders missingness functionally random (Little, 2013) and is thus termed ignorable (Rubin, 1976), so no missing data procedures were necessary.

Establishing the Factor and Trimming the Item Pool

For the EFA conducted on the 14 KL items using the Rwanda data, scree plot and parallel analysis eigenvalues suggested a one-factor solution (see Fig. 1S in the ESM). Factor loadings were all significant and ranged from 0.763 to 0.918; however, model fit was poor: χ2(65) = 849.84, p < 0.001; RMSEA = 0.100 (90% CI [0.094, 0.106]); CFI = 0.818; TLI = 0.781; SRMR = 0.056. Despite model fit improving with additional factors, cross-loadings among several items and high correlations among latent factors (e.g., r = 0.803 in the two-factor solution) suggested that the one-factor solution might indeed provide the best (conceptual) fit to the data. Accordingly, we next assessed inter-item correlations among the 14 items (see Table 1S in the ESM) to determine which items to trim to obtain a good-fitting, parsimonious, single-factor solution.

Inter-item correlations and item wording suggested several potential items to remove from the factor solution. For instance, KL Item 7 correlated highly (≥ 0.8) with KL Items 3, 4, 5, and 6. As such, we retained KL Item 7 and trimmed the other four items from the factor solution. Similarly, retaining KL Item 9 allowed us to trim Items 10 and 11. We also inspected item wording to ensure that the scale was not only empirically parsimonious but also conceptually so. For instance, KL Items 13 and 14 correlated highly (r = 0.832) so we assessed wording to determine which item to retain (in this case, we decided that having an adult “who will love me no matter what I do” was more representative of the intended construct, than having an adult “who looks forward to spending time with me”). In addition, KL Items 1, 2, 8, and 12 did not correlate highly with the remaining items; however, we decided to trim Items 1 and 12 due to language deemed less meaningful to or reflective of the intended construct, compared to the retained items. As such, we were left with a five-item solution including KL Items 2, 7, 8, 9, and 14.

Testing the Psychometric Properties of the KL Measure with the Rwanda Data

We then used the refined KL scale to confirm its factor structure and to assess its psychometric properties. Cronbach’s alpha indicated strong internal consistency reliability (α = 0.92), as did the omega coefficient (ω = 0.92). To assess model fit along with convergent validity and measurement invariance, we tested a series of CFAs with the five-item KL factor along with factors of ISR, HFE, transcendence, and contribution (see Fig. 2S in the ESM).

In a test of the full Rwanda sample, the measurement model provided excellent fit to the data: χ2(367) = 940.482, p < 0.001; RMSEA = 0.036 (90% CI [0.033, 0.039]); CFI = 0.967; TLI = 0.963; SRMR = 0.028. The model failed absolute fit as indicated by the χ2 significance; however, this result was expected due to oversensitivity of the test, for instance, to large sample sizes (see Rutkowski & Svetina, 2017). Standardized factor loadings ranged between 0.772 and 0.947 (see Table 1 for factor structure and loadings), and correlations among the latent constructs ranged from r = 0.225 (contribution with HFE) to 0.594 (KL with transcendence; and ISR with contribution), with all ps < 0.001 (see Table 2 for correlations among the latent factors). The moderate correlations among KL and the youth strengths and outcomes pertinent to CI programs (r = 0.494 with ISR; 0.384 with HFE; 0.594 with transcendence; and 0.413 with contribution), provided initial evidence for convergent validity of the KL measure.

Table 1 Factor structure and loadings for the measurement model of the factors related to known and loved, intentional self-regulation, hopeful future expectations, transcendence, and contribution across Rwanda and El Salvador
Full size table
Table 2 Correlations among the latent factors in the Rwanda and El Salvador samples, respectively
Full size table

We then tested the full measurement model for between-group measurement invariance across age, gender, urban–rural area, and CI-enrollment status in Rwanda (see Table 3 for model fit statistics). Strong invariance was established across each set of groups tested, enabling us to make meaningful comparisons across groups. In regard to age, older youth reported significantly higher levels of ISR (0.252 standard deviation units [SDUs], p = 0.042) and of contribution (0.352 SDUs, p = 0.001) compared to younger youth. In regard to gender, boys reported significantly lower levels of KL (-0.094 SDUs, p = 0.007) compared to girls. In regard to area, rural youth reported significantly lower levels of KL (− 0.475 SDUs, p = 0.001) compared to urban youth. In regard to CI-enrollment status, non-CI-supported youth reported significantly lower levels of each of the five constructs compared to CI-supported youth: KL (− 0.326 SDUs, p < 0.001); ISR (− 0.297 SDUs, p < 0.001); HFE (− 0.352 SDUs, p < 0.001); transcendence (− 0.262 SDUs, p < 0.001); and contribution (− 0.180 SDUs, p = 0.007).

Table 3 Model fit statistics for invariance tests of the measurement model of the factors related to known and loved, intentional self-regulation, hopeful future expectations, transcendence, and contribution
Full size table

Cross-Validating the KL Measure among Salvadoran Youth

Having established the KL measure with the Rwanda sample, we next tested its utility with the El Salvador sample. Item-level missingness was minimal (between 0.5 and 1.0% of data missing), so we assumed all data were missing at random (MAR). Before seeking to establish cross-national measurement invariance across the Rwanda and El Salvador samples, we first sought to assess the psychometric properties in the El Salvador sample.

Testing the Psychometric Properties of the KL Measure with the El Salvador Data

The five-item KL scale demonstrated good internal consistency reliability (Cronbach’s α = 0.78; ω = 0.77). Tested on the full El Salvador sample, the CFA of the full measurement model (see Fig. 2S in the ESM) demonstrated excellent fit to the data: χ2(367) = 605.004, p < 0.001; RMSEA = 0.023 (90% CI [0.020, 0.027]); CFI = 0.962; TLI = 0.958; SRMR = 0.037. The model failed absolute fit as indicated by the χ2 significance; however, this result was expected due to oversensitivity of the test, for instance, to large sample sizes (see Rutkowski & Svetina, 2017). Correlations among the latent constructs ranged from r = 0.392 (KL with contribution) to 0.623 (ISR with HFE), all ps < 0.001; and all correlations with KL were moderate (r = 0.392 to 0.479), providing initial evidence for convergent validity with the KL measure with the El Salvador sample (see Table 2 for correlations among the latent factors in Rwanda and El Salvador, respectively).

We then tested the full measurement model for between-group measurement invariance across age, gender, urban–rural area, and CI-enrollment status in El Salvador. For each group, strong measurement invariance was established (see Table 3), enabling meaningful comparisons of the latent constructs to be made. For age, older youth reported significantly lower levels of ISR (-0.169 SDUs, p = 0.020), of HFE (− 0.198 SDUs, p = 0.003), of transcendence (− 0.184 SDUs, p = 0.016), and of contribution (− 0.383 SDUs, p < 0.001) compared to younger youth. For gender, boys reported significantly lower levels of ISR (− 0.112 SDUs, p = 0.026) compared to girls. For urban–rural area, rural youth reported significantly higher levels of contribution (0.105 SDUs, p = 0.042) compared to urban youth. For CI-enrollment status, CI-supported youth reported significantly lower levels of transcendence (− 0.112 SDUs, p = 0.013) compared to non-CI-supported youth.

Testing for Cross-National Invariance

As shown in Table 3, we established strong measurement invariance of the KL measure and the broader measurement model (including ISR, HFE, transcendence, and contribution) across the Rwanda and El Salvador samples, providing initial empirical evidence of cross-national validation of KL and the present measurement model, at least across these two diverse national contexts. Establishing cross-national invariance also enabled us to make comparisons across the samples on the latent constructs. In comparing latent means, youth in Rwanda reported significantly lower levels of each of the five constructs compared to youth in El Salvador: KL (− 0.526 SDUs, p < 0.001); ISR (− 0.323 SDUs, p < 0.001); HFE (− 0.312 SDUs, p < 0.001); transcendence (− 0.468 SDUs, p < 0.001); and contribution (− 0.161 SDUs, p < 0.001).

Table 4 presents results from comparing the correlations among the latent constructs cross-nationally, using the Wald tests of parameter estimates. Of the 10 correlations compared, six were significantly different between the Rwanda and El Salvador samples. For KL with ISR, the Rwanda sample showed a stronger correlation (r = 0.494, compared to r = 0.485 in El Salvador; Wald statistic = 5.318, p = 0.021). For KL with HFE, the El Salvador sample showed a stronger correlation (r = 0.457, compared to r = 0.385 in Rwanda; Wald statistic = 6.332, p = 0.012). For KL with transcendence, the Rwanda sample showed a stronger correlation (r = 0.595, compared to r = 0.461 in El Salvador; Wald statistic = 10.054, p = 0.002). For KL with contribution, the Rwanda sample showed a stronger correlation (r = 0.413, compared to r = 0.396 in El Salvador; Wald statistic = 7.275, p = 0.007). For ISR with transcendence, the Rwanda sample showed a stronger correlation (r = 0.557, compared to r = 0.509 in El Salvador; Wald statistic = 9.242, p = 0.002). For HFE with transcendence, the El Salvador sample showed a stronger correlation (r = 0.610, compared to r = 0.517 in Rwanda; Wald statistic = 5.586, p = 0.018).

Table 4 Comparing correlations among the latent constructs of known and loved, intentional self-regulation, hopeful future expectations, transcendence, and contribution between the Rwanda and El Salvador samples using the Wald test of parameter estimates
Full size table

Discussion

Dynamic, RDS-based models of development (e.g., Cantor et al., 2021; Fischer & Bidell, 2006; Lerner, 2018a, b; Mascolo & Bidell, 2020; Overton, 2015) emphasize that developmentally-nurturant relationships between adults and youth, ones that elicit in young people feelings of being known and loved, are foundations of PYD (e.g., Lerner et al., 2015; Lerner, 2018b). Because measures of youth perceptions of being known and loved did not exist when we undertook the present study, these theoretical models, and their applications to programs seeking to promote PYD, remained incompletely tested or evaluated, respectively. Accordingly, the goal of this study was to construct a measure of such youth perceptions and to also ascertain its covariation within the nomological net of constructs within the Lerner and Lerner model of PYD (e.g., Lerner & Lerner, 2019; Lerner et al., 2015), the most widely-used instance of RDS-based PYD models (e.g., Petersen et al., 2017; Qi et al., 2021). To pursue this goal, we capitalized on data sets in two nations (Rwanda and El Salvador) wherein an ongoing multi-national study of PYD was being used in the design of youth programs provided by Compassion International (CI).

The results of the present research established the psychometric quality of the measure of youth perceptions of being known and loved. Both within and across the data sets from Rwanda and El Salvador our findings supported the reliability and within- and across-nation invariance of the measure and, as well, demonstrated the presence of theoretically-predicated covariations between scores from the known and loved measure and scores from other constructs (i.e., intentional self-regulation, hopeful future expectations, transcendence, and contribution) within the nomological net proposed in the Lerner and Lerner PYD model (e.g., Lerner & Lerner, 2019; Lerner et al., 2015).

Beyond the psychometric and validational contributions of the known and loved measure for the PYD model we tested, the results of this research illustrated the usefulness of having a measure of youth perceptions of being known and loved in illuminating variation in the processes involved in thriving, both within and across samples of youth develo** within LMICs and, as well, among youth participating or not participating in youth development programs designed to promote PYD. For instance, results of the between-group measurement invariance tests revealed that, in the Rwanda sample, older youth reported significantly higher levels of ISR and contribution than younger youth, and that boys and rural youth reported significantly lower levels of KL than girls and urban youth, respectively. We also found that CI-supported youth in Rwanda reported higher levels of each of the five PYD constructs we measured compared to non-CI-supported youth. In turn, in the El Salvador sample, older youth reported significantly lower levels of ISR, HFE, transcendence and contribution compared to younger youth; boys reported significantly lower levels of ISR compared to girls; rural youth reported significantly higher levels of contribution compared to urban youth; and CI-supported youth reported significantly lower levels of transcendence compared to non-CI-supported youth.

By exploring possible demographic moderators of invariance, these analyses were undertaken in the service of ascertaining the robustness of the cross-national invariance that was found (Duncan et al., 2014). Although these robustness analyses were hypothesis searching rather than hypothesis testing (Cattell, 1966), they do provide potentially provocative instances of variation that may be explored in future research in Rwanda and El Salvador, in other nations within which CI has programs, in LMICs within which CI does not work and, perhaps as well, in nations within the minority world. For instance, it might be expected that older youth will demonstrate stronger intentional self-regulation skills and higher contribution beyond the self, as was found in the Rwanda sample. In what nations and under what individual \(\Leftrightarrow\) context coactions will comparable versus divergent findings be identified and what dynamic processes are involved in such variation? In turn, the finding that older youth were lower on nearly all constructs in El Salvador is puzzling and may suggest either developmental differences or cultural differences in response styles. Future research, both in El Salvador and in other nations would help address this question. In any case, the invariance that was identified through the present research, which underscores the robustness of data supporting cross-national consistency in the measurement model pertinent to being known and loved, enables questions such as the ones we have suggested to be appropriately interrogated.

Similarly, future research might build on the findings (1) that youth in Rwanda tended to report significantly lower levels of each of the five constructs within the Lerner and Lerner nomological net of constructs we assessed compared to youth in El Salvador and, as well, (2) that patterns in the correlations among the constructs differed between the Rwanda and El Salvador samples. Perhaps these differences between Rwandese and Salvadoran youth reflect cultural variations in the development of these indicators of positive development, and/or cultural differences in the willingness of youth to assess themselves as having high scores on these constructs. Whether these differences were due to differences in the broader ecological contexts across the two LMICs, or due to contextual or cultural differences in response style (e.g., differences in survey biases such as social desirability bias) remains unknown and requires further exploration. As the CI Study of PYD continues, qualitative research (e.g., involving assessments of the semantic space within these constructs that exist in each nation) could illuminate whether our conjectures here are reasonable.

Nonetheless, establishing measurement invariance and cross-national validation across two LMICs in Africa and Central America is timely and important, perhaps especially given the calls by international development organizations including UNICEF, USAID, and World Bank to better implement, evaluate, and document PYD programs in LMICs with robust and consistent measurement models, for instance, in assessing progress toward the 2030 Sustainable Development Goals (see United Nations, 2015; YouthPower Learning, 2017).

Limitations and Future Directions

Our interpretations of the findings need to be considered with an understanding of the limitations of this research. As we have noted, we were limited by only having one time of testing. With additional times of testing, developmental trajectories might be assessed, including models testing predictive relations among the constructs (e.g., mediation; see Maxwell & Cole, 2007), as well as more powerful time-series data analysis techniques (e.g., dynamic structural equation modeling, etc.; see McNeish & Hamaker, 2020; Yu et al., 2021). Future research might include additional times of testing to ascertain the use of such models and, if sufficient power is present, to use the Bornstein (2017, 2019) specificity principle to assess what features of what programs promote what facets of thriving among what specific youth (e.g., see Rhodes, 2020, and Cantor et al., 2021, for discussions of the importance of such specificity analyses in PYD research). Nonetheless, we were able to demonstrate a preliminary assessment of a measurement model assessing person\(\Leftrightarrow\)context relations in youth programs.

Measurement issues also limit our present findings. Our study included only five variables (KL, ISR, HFE, transcendence, and contribution) and therefore unmeasured variables might enable better conceptualization of mediation and moderation effects. Moreover, all of our measures were collected using interviewer-led youth self-reports, which are susceptible to self-report biases (e.g., response-style bias, social desirability bias, or responses biased by stereotype threat). Last, the hypothesized interrelations among the person- and context-related variables might be better assessed by triangulation with additional measures, methods of assessing constructs, and/or sources of information, for instance, parents or program staff.

Sampling and diversity were also limitations in the present study. Two data sets were used—one from Rwanda and one from El Salvador—involving youth enrolled in CI-supported programs and comparison youth who were eligible economically for CI support but were not involved in CI programs. Although comparison youth were selected to meet eligibility criteria for enrollment in CI programs—meaning all participants were living in contexts of poverty—diversity of life experiences and developmental histories were not assessed. Future studies might include youth from more diverse backgrounds including diverse socioeconomic statuses, diverse geographic locations, cultures (including nationalities, languages, ethnicities, religions, identities, etc.), and ages to more thoroughly examine demographic and individual differences and specificity (Bornstein, 2017, 2019) and to generalize results and applications of the measure across broader contexts.

Furthermore, regarding sampling, and as noted in the Procedure section, project sites were selected for study in the CI Study of PYD to represent the best program outcomes. As such, findings may not be generalizable to less-exemplary sites, or to youth in other communities and nations. This limitation further emphasizes the importance of attending to specificity in future research with different samples across varied demographic groups and contexts (Bornstein, 2017, 2019).

Conclusions

In all RDS-based conceptions of development and, more specifically, of PYD, developmentally-nurturant relationships between specific youth and specific contexts (including specific adults in the social ecology of a young person) are the foundation of both youth thriving and of programs effective in promoting such thriving. Accordingly, both researchers studying, and practitioners delivering programs to promote, PYD conceptualized within RDS-based models of youth thriving will find it beneficial to possess a measure of youth perceptions of the outcome of such relationships, that is, feeling known and loved. This present research provides evidence that an at least first step in develo** such a measure has been taken and, as well, that there is evidence of the usefulness of such a measure within a widely-used RDS-based model of PYD.

We hope that the readers of this article will find that the results of the present research are therefore both timely and important. Indeed, as developmental scientists seeking to describe, explain, and optimize human development, including promoting thriving among diverse youth, taking such intentional and rigorous approaches to develo** measures will be contributing essential scholarship that can continue to refine developmental theory and its applications in program evaluation efforts that enhance the success of youth development programs.