1 Introduction

Adolescence is a unique and formative stage (Jaworska & MacQueen, 2015), with socioeconomically disadvantaged individuals facing an increased risk of developing mental health challenges, which may, in turn, impact their health and socioeconomic position (SEP) later in life (Reiss, 2013). Valid and comparable measures of SEP are needed to map and describe the development of health inequalities across age and time. A key challenge is that adolescents often lack the information needed to report reliably on classical SEP indicators such as parents’ income, education or occupation (Currie et al., 1997, 2008; Lin, 2011; Svedberg et al., 2016; Torsheim et al., 2016; Wardle et al., 2002). An alternative to income-based information is an asset approach (Howe et al., 2008; Sahn & Stifel, 2003). This approach assesses family SEP by asking adolescents directly about the material conditions of the family that they are likely to be familiar with (Doku et al., 2010; Wardle et al., 2002). The family affluence scale (FAS) is an example of an asset-based measure of economic resources within the family, designed for use in surveys of adolescents. It comprises items that reflect family expenditure and consumption, with the total sum providing a crude index of material wealth.

1.1 The Family Affluence Scale (FAS)

The family affluence scale was developed in the context of the international survey “Health Behaviour in School-Aged Children Study” (HBSC) (Currie et al., 1997, 2008), based on the work of Townsend (1987) and Carstairs and Morris (1990). Initially, the scale was designed for use in Scotland at the beginning of the 1990s. It included items such as telephone ownership, family cars, and whether the child had a separate bedroom (Currie et al., 1997). In the HBSC data collection of 1993/94, the telephone item was removed, and only the items of family car ownership and the child having a separate bedroom were retained (Currie et al., 2008). Over time, the scale has been tailored for use in countries across Europe and North America to reflect changes in living conditions, societal norms, and technology (Currie et al., 1997, 2008). In the data collection of 1997/98, an additional question about the number of family holidays was added, resulting in the creation of the family affluence scale I (FASI). Later, in 2001/02, an item regarding the number of computers was included, leading to the development of family affluence scale II (FASII). In the 2013/14 survey, two more items were introduced: a dishwasher at home and the number of bathrooms. Furthermore, the item related to the number of holidays was modified to specifically focus on holidays abroad, which led to the creation of the family affluence scale III (Torsheim et al., 2016).

1.2 Previous Validation Work of the Family Affluence Scales

With its high completion rates (Molcho et al., 2007), the family affluence scales have frequently been used in research papers, resulting in a large body of work describing health outcomes related to family affluence (Currie et al., 2008). Studies have documented the measurement characteristics of the scales among school-aged children and adolescents in geographically diverse samples, and they have been validated at both national and international levels (i.e., Andersen et al., 2008; Boudreau & Poulin, 2009; Currie et al., 1997; Kehoe & O’Hare, 2010; Liu et al., 2012; Molcho et al., 2007).

External validity has been assessed in a total of eight studies by comparing children’s responses to FASI (Currie et al., 1997; Wardle et al., 2002) and FASII (Boudreau & Poulin, 2009; Cho & Khang, 2010; Lin, 2011; Liu et al., 2012; Molcho et al., 2007; Svedberg et al., 2016) with their reports of other SEP indicators such as their perceived SEP, parents’ occupation, education, or income. In three studies, conformity of children’s responses to FASII (Andersen et al., 2008) and FASIII (Corell et al., 2021; Torsheim et al., 2016) with their parents’ responses to FAS or reports about their income was assessed. Low to moderate external validity was found in nine of the studies (Batista-Foguet et al., 2004; Boudreau & Poulin, 2009; Corell et al., 2021; Currie et al., 1997; Lin, 2011; Liu et al., 2012; Molcho et al., 2007; Svedberg et al., 2016; Wardle et al., 2002), whereas high external validity was found in three (Andersen et al., 2008; Cho & Khang, 2010; Torsheim et al., 2016). In addition, two studies validated FASII using macro-level indicators like Gross Domestic Product (GDP) (Boyce et al., 2006) and area deprivation index (Kehoe & O’Hare, 2010). Boyce et al. (2006) affirmed the validity of the scale. However, Kehoe and O’Hare (2010) reported variable construct validity and limitations regarding measurement reliability.

Building upon the preceding validation efforts, it is imperative in studies of social inequality to employ measures of SEP that are not only valid but also comparable. The following section leads to the primary focus of this study, namely, the assessment of the comparability of the scale.

1.2.1 Measurement Properties of the Scales Across Population Subgroups and Time

Bias occurs when measurements vary across different subgroups of the population and across time, with the primary factor being the test items themselves (Putnick & Bornstein, 2016; van de Vijver & Tanzer, 2004). Researchers utilize Item Response Theory (IRT) and Structural Equation Modelling (SEM) to assess and address such biases. In IRT, item bias is determined by differential item functioning (DIF) (Lord, 1980), which refers to unexpected differences in performance among groups that are supposed to be comparable (Holland & Wainer, 1993). A specific class of DIF is item parameter drift (IPD), which, in the case of the FAS, would indicate that a given level of affluence would elicit different item responses across time (Goldstein, 1983). On the other hand, SEM evaluates item bias through a measurement invariance (MI) test, which assesses a construct’s psychometric equivalence across groups and time (Millsap, 2011; Putnick & Bornstein, 2016).

Six studies have examined the measurement properties of the scale across population subgroups and time: FASI (Batista-Foguet et al., 2004), FASII (Liu et al., 2012; Makransky et al., 2014; Schnohr et al., 2008, 2013) and FASIII (Torsheim et al., 2016). Two studies found that the scale changed its measurement properties over time (Makransky et al., 2014; Schnohr et al., 2013). Three studies found DIF between countries (Batista-Foguet et al., 2004; Schnohr et al., 2008; Torsheim et al., 2016); two studies found DIF between age groups (Liu et al., 2012; Schnohr et al., 2008, 2013), and two studies found DIF between sexes (Schnohr et al., 2008, 2013).

1.3 Contribution to Literature

While previous studies have assessed the measurement properties of the family affluence scales across population subgroups and time, such studies are relatively scarce. Only two prior studies have investigated the measurement properties of the scales across historical time, drawing upon data from the HBSC surveys spanning from 2001/02 to 2009/10. Expanding the timeframe for assessment could offer further insight into potential bias in the test items.

Given the steady increase in wealth observed in Norway during the HBSC survey period (UNDP, 2019), it becomes crucial to investigate the measurement characteristics of the scale within the context of Norwegian HBSC surveys. While previous studies have included Norwegian samples (Makransky et al., 2014; Schnohr et al., 2008, 2013; Torsheim et al., 2016), none have focused explicitly on Norway. The increase in wealth may impact the response to the items in the scale, causing a given level of affluence to elicit different item responses across time (Goldstein, 1983). For instance, the probability of responding to “two computers or more at home” might increase at a given level of affluence across time. An affluent person might respond to “one computer” in 2002 but at least “two computers” in 2018 due to the changing value of the item. This signifies that accurate and proper comparisons of family affluence scores across Norwegian HBSC surveys are not possible, as results would be biased (Putnick & Bornstein, 2016). Moreover, with the increasing wealth in Norway, it becomes crucial to evaluate whether the items in the scale adequately capture the entire spectrum of affluence, spanning from low to high. Additionally, research exploring potential variation in scale responses based on age and sex remains notably limited and should be assessed further.

1.4 Aim of the Study

The primary objective of this study is to validate the application of FASII for time trend analysis within the Norwegian HBSC surveys, spanning the years 2002 to 2018. Specifically, the aim is to investigate whether the properties of the FASII change over time and across different age groups and sexes. Additionally, the study seeks to evaluate how FASII correlates with an alternative SEP measure, providing further insight into the scale’s psychometric properties.

2 Methods

2.1 Sample

Data stemmed from the Norwegian part of the HBSC study, which collected survey data from 11-, 13-, 15-, and 16-year-olds. The University of Bergen is responsible for its implementation in Norway. The survey sheds light on health behaviours, health perceptions, well-being, and contextual experiences among adolescents. The HBSC study employs a systematic cluster sampling method to ensure a nationally representative sample of school-aged children, with schools serving as the primary sampling units. Participation is voluntary, and informed consent is obtained from parents and students before data collection. The data collection is conducted using self-administered questionnaires during school hours, under the supervision of trained researchers or school staff (Inchley et al., 2018).

First conducted in 1983, the HBSC study has been repeated every four years since 1985, with each cycle involving around 4,000 to 7,000 pupils from all regions of Norway. The present study is retrospective, drawing on five survey cycles spanning 2002 to 2018. Non-participation occurred at both school and student levels. At the student level, response rates ranged from approximately 70–80% within each age group. The dataset, comprising a total of 27,470 valid questionnaires, was distributed as follows: 7,039 participants in 2002, 6,447 participants in 2006, 5,760 participants in 2010, 4,592 participants in 2014, and 3,632 participants in 2018 (Haug et al., 2020; Samdal, 2009; Samdal et al., 2016; Torsheim et al., 2004).

2.2 Measure

The Norwegian HBSC surveys use various measures to assess adolescent SEP. Two of them are the FASII and Perceived Family Wealth.

2.2.1 Family Affluence Scale II

The FAS version III (FASIII) could not be computed within HBSC data until 2014 (Torsheim, 2019), making it unavailable for analysis starting from 2002. Therefore, this study centres its assessment on FAS version II (FASII), comprising the following four questions (University of Bergen, 2022):

  1. 1.

    “Does your family own a car, van or truck?” Response categories were: No (= 0); Yes, one (= 1); Yes, two or more (= 2).

  2. 2.

    “Do you have your own bedroom for yourself?” Response categories were: No (= 0); Yes (= 1).

  3. 3.

    “During the past 12 months, how many times did you travel away on holiday/vacation with your family?” Response categories were: Not at all (= 0); Once (= 1); Twice (= 2); More than twice (= 3).

  4. 4.

    “How many computers does your family own?” Response categories were: None (= 0); One (= 1); Two (= 2); More than two (= 3).

In 2014, question 3 was modified in the HBSC survey from “…travel away on holiday” to “travel away on holiday abroad”. This alteration coincided with the introduction of FASIII. The FASII can be scored as a composite measure ranging from 0 to 9 or as different categories ranging from low to high (Boyce et al., 2006). The present study scored FASII as a composite measure and absolute affluence was measured by calculating a sum of items ranging from 0 (lowest affluence) to 9 (highest affluence).

2.2.2 Perceived Family Wealth

Perceived family wealth represents a subjective measure of adolescents’ SEP. It consists of a single-item question: “How would you describe the economic situation in your family?” with the response alternatives: “not at all well off”, “not so well off”, “average”, “quite well off”, and “very well off”. Responses are given numerical values ranging from 1 to 5, 1 representing “not at all well off” and 5 equalling “very well off” (Haug et al., 2020; Samdal, 2009; Samdal et al., 2016; Torsheim et al., 2004).

2.2.3 Demographic Information

Data regarding participants` age and sex were collected through specific questions. Participants were asked the question, “Are you a boy or girl?” with response options being “boy” or “girl”. Age information was obtained through a two-part inquiry: first, participants were asked to indicate their birth month, with response options ranging from January to December; second, participants were queried about their birth year, with response options that varied depending on the specific survey year.

2.3 Data Analysis

The data analysis was conducted using the software package Lavaan (Rosseel, 2012) in R for Windows (R Core Team, 2022). Data analyses were carried out in four steps: first, building a latent variable model; second, conducting confirmatory factor analysis (CFA); third, performing a polychoric correlation test and assessing perceived family wealth and FASII scores across time; and fourth, employing multi-group confirmatory factor analysis (MGCFA). The following sections explain each of these steps.

2.3.1 Building a Latent Variable Model

First, a one-factor latent variable model of family affluence was built using the CFA method. In CFA, items that make up a construct load on a latent variable that represents the construct (Putnick & Bornstein, 2016). In the case of FASII, the latent variable is family affluence, and items that make up the construct are family cars, separate bedroom, holiday, and computer. A latent variable indicates causation and describes the relationship between observable variables and a latent variable (Bollen & Bauldry, 2011; Borsboom et al., 2003). As such, an increase in family affluence would increase the likelihood of having more cars, computers, holidays, and more space at home.

2.3.2 Confirmatory Factor Analysis

Secondly, the one-factor latent variable model was fitted to data from the different survey years. The factor structure of the models was evaluated based on CFA, using the Comparative Fit Index (CFI), Root Mean Square Error of Approximation (RMSEA), and Standardized Root Mean Square Residual (SRMR), as recommended by Brown (2015) for CFA models. The model fit is considered acceptable if SRMR values are close to 0.08 or below, RMSEA values are close to 0.06 or below, and CFI is close to 0.95 or greater (Hu & Bentler, 1999). Raykov`s reliability coefficient (RRC) was calculated for each survey year.

2.3.3 Polychoric Correlation Test and an Assessment of Perceived Family Wealth and FASII Scores Across Time

The polychoric correlation coefficient was used as the third step to assess the correlation between FASII and perceived family wealth across different survey years, age groups, and sex. The polychoric correlation coefficient measures the association for ordinal variables (Ekström, 2011). In addition, an assessment of perceived family wealth and FASII scores across time was conducted.

2.3.4 Multi-Group Confirmatory Factor Analysis

In the last step, to test MI, an MGCFA was conducted using the WLSMV estimator and delta parameterization specially designed for ordinal data (Bowen & Masa, 2015). MGCFA extends the typical CFA, but instead of fitting a single model to the data, one divides the data set into different groups (Brown, 2015). The data was divided into groups by sex (boys and girls), age group (11-, 13-, 15- and 16-year-olds), and survey years (2002–2018).

Testing for MI consists of a series of model comparisons that define more and more stringent equality constraints (Brown, 2015; Putnick & Bornstein, 2016; Vandenberg & Lance, 2000). Configural invariance tests the overall factor structure of the scale across different groups. The model is fitted onto each group, allowing factor loadings and item intercepts to vary freely. Model fit is compared across groups. Metric invariance tests whether the factor loadings are equivalent across groups. Factor loadings are constrained to be equal, while item intercepts are still allowed to vary freely. Model fit is compared to the configural model. Scalar invariance examines whether the item intercepts are equivalent across groups. The item intercepts are constrained to be equal. If scalar invariance results in a poorer fit compared to metric invariance, it indicates that item intercepts differ across groups (Putnick & Bornstein, 2016).

Even though MI is most commonly tested in the steps of configural, metric, and scalar invariance testing (Putnick & Bornstein, 2016), a particular case pertains to the testing of MGCFA with ordinal indicators (Bowen & Masa, 2015; Hirschfeld & Brachel, 2014). There are different analytical approaches one can use, but for the current study, loadings and thresholds are tested as a set (Bowen & Masa, 2015; Lubke & Muthén, 2004; McLarnon & Carswell, 2013; Millsap & Yun-Tein, 2004; Sass, 2011; Webber, 2014). As such, there is no separate test for invariant factor loadings (metric invariance test), and the test proceeds from an examination of configural to scalar invariance (Bowen & Masa, 2015; Brown, 2015, p. 370). An argument for this approach is that loadings and thresholds should be constrained and freed together because they jointly define item functioning (Sass, 2011). For a different analytical approach, see Liu et al. (2017).

A comparison was made for the model fit of an MGCFA where all item factor loadings and item intercepts were free to vary across groups (configural invariance test) with the model fit of an MGCFA where all item factor loadings and item intercepts were constrained to be equal across groups (scalar invariance test).

3 Results

3.1 Confirmatory Factor Analysis

Table 1 showed that the CFA yielded an overall good model fit in the different survey years. Structural validity was established as the model fit was acceptable (SRMR 0.08 or below, RMSEA 0.06 or below, and CFI 0.95 or greater) (Hu & Bentler, 1999).

Table 1 Factor structure of confirmatory factor analysis models for the family affluence scale II (2002–2018)
Full size table

Most items consistently displayed factor loadings exceeding 0.4 across all survey years. However, the item related to holidays exhibited notably weaker factor loadings (see Table 2). These weaker factor loadings indicate that this specific item may be less suitable as an indicator for measuring affluence within the Norwegian sample. Raykov`s reliability coefficient (RRC) indicated moderate to moderately high level of internal consistency across different survey years (2002: 0.48, 2006: 0.51, 2010: 0.57, 2014: 0.52, 2018: 0.58).

Table 2 Factor loading of confirmatory factor analysis models
Full size table

3.2 Polychoric Correlation Test

When conducting a polychoric correlation test (Table 3), a positive correlation between perceived family wealth and FASII was consistently observed across groups and over time. These findings affirm the link between higher FASII scores and an increased perception of family wealth, further enhancing the validity of FASII as a measure of affluence within the Norwegian context.

Table 3 Polychoric correlation test age, survey year and sex
Full size table

3.3 Assessment of Perceived Family Wealth and FASII Scores Across Time

Figure 1 illustrates the mean FASII scores corresponding to distinct levels of perceived family wealth. The data is presented for two time points: 2002 and 2018. Notably, a clear upward trend in mean FASII scores is evident across all levels of perceived family wealth between these two years. The fact that a given level of subjective wealth reflects different mean FASII scores might indicate that the subjective value of family affluence scores has changed over time. To ensure the validity and comparability of FASII, it is imperative that the subjective value of the scale remains constant across time. Therefore, Fig. 1 underscores a potential limitation in using FASII as a consistent measure over time.

Fig. 1
figure 1

Perceived family wealth and family affluence scale II scores across time. Note: Figure 1 visually represents the mean FASII score, corresponding to distinct levels of perceived family wealth, for 2002 and 2018. Notably, the figure reveals a clear upward trend in mean FASII scores across all levels of perceived family wealth between these two years

Full size image

3.4 Multi-group Confirmatory Factor Analysis

Results from the MGCFA are displayed in Table 4. When conducting the MGCFA, the first step was to evaluate the configural invariance model. The RMSEA, SRMR, and CFI indices indicate that the overall factor structure fits well across all groups, which assumes that the latent construct is measured by the same indicators in all groups, meaning that the structure of the measurement scale is the same. Secondly, scalar invariance was tested. Full scalar invariance was not observed. MI would be established if the scalar models decreased CFI by no more than 0.010 and increased RMSEA by no more than 0.015 relative to the configural model (Chen, 2007; Cheung & Rensvold, 2002). The fit indicators changed significantly when assuming scalar invariance, meaning the item intercepts were not similar for participants in the different groups. Violation of MI was especially significant in the group survey year, but also for the age group. Constraining the factor loadings and intercepts to be equal across the groups deteriorated model fit substantially (sex: ΔCFI = − 0.019, ΔRMSEA = 0.001), (age: ΔCFI = − 0.35, ΔRMSEA = 0.043), (survey/time: ΔCFI = -0.793, ΔRMSEA = 0.083). Given the significant degradation in model fit, the practice of accepting partial invariance and proceeding with a test of mean differences or relations among constructs was not considered relevant (Putnick & Bornstein, 2016).

Table 4 Multi-group confirmatory factor analysis, age, survey year and sex
Full size table

4 Discussion

This research assessed the psychometric properties of FASII in the Norwegian HBSC surveys conducted from 2002 to 2018. The results revealed an overall good model fit in CFA and a positive correlation between FASII scores and perceived family wealth, enhancing the reliability and validity of FASII as a measure of affluence within the Norwegian context. The RRC indicated a moderate to moderately high level of internal consistency for FASII over the years. These coefficients reflect the categorical nature of the indicators, capturing a broad spectrum of the family affluence construct. While they may lack the precision of continuous indicators in comprehensively measuring the construct, the results support the overall reliability of the scale.

In the assessment of perceived family wealth and FASII scores across time, a noticeable increase in average FASII scores was observed for each perceived family wealth level between 2002 and 2018. This finding sheds light on a potential shift in the perception of family wealth and implies a broader socioeconomic transition within this period. This trend raises questions about the stability of FASII as a consistent measure over time. To delve deeper into this aspect, as well as its stability within cohorts, an evaluation of MI was undertaken. Through this analysis, measurement non-invariance across survey years, sex and age groups was identified. Similar findings have been reported in previous studies (Batista-Foguet et al., 2004; Liu et al., 2012; Makransky et al., 2014; Schnohr et al., 2008, 2013; Torsheim et al., 2016). Measurement non-invariance implies that appropriate and proper comparisons of scores across time and between cohorts are not possible (Putnick & Bornstein, 2016). It is, therefore, crucial to consider the implications of measurement non-invariance for the validity of FASII concerning time trend analysis with the HBSC data.

To measure the development of affluence over time, FASII would be utilized by conducting direct comparisons of FAS scores at different time points within the HBSC data. Such a comparison would make it tempting to conclude that increasing FAS scores reflect increasing affluence. However, the violation of MI found in the present study suggests that direct comparisons of scores across time might be biased. When the item parameters vary across time, the FASII scale is not stable, nor are time-to-time comparisons. Makransky et al. (2014) and Schnohr et al. (2013) found that the FASII changed its measurement properties over time, which is consistent with the present study. Both studies reported changes in the measurement properties of the computer item. The computer, initially regarded as a luxury item, is now owned by the majority of households in Europe (OECD, 2023). Also, in recent decades, Norwegian schools have witnessed a significant digital transformation by implementing one-to-one classrooms, providing one digital unit per student. This transformation has occurred at various times in primary and secondary education. In secondary education, students began receiving their digital unit from 2007 (Blikstad-Balas, 2012; Fjørtoft et al., 2019). Similarly, in primary education, there has been a comparable increase in the use of digital equipment over the past decade (Munthe et al., 2022). As a result, today, most pupils in primary and secondary education have their own digital unit. Such changing value of an item could lead to the erroneous conclusion that affluence is increasing, illustrating that a changing value of an item over time could have profound consequences on conclusions made when the scale is made up of only four items (Makransky et al., 2014).

The sensitivity to measurement non-invariance is not limited to the computer item alone. For instance, given the specific modification in the holiday item, focusing on holidays abroad in the 2013/14 survey and subsequent surveys, the current study anticipated the likelihood of observing some violation of MI across time. Moreover, variation in vacation could be influenced by factors such as competition and pricing levels in the tourist industry (Schnohr et al., 2008; Torsheim et al., 2016) and changing travel habits as general wealth increases. In the past few decades, Norway has experienced a noteworthy surge in international travel (Henriksen, 2020). Additionally, environmental considerations could affect travel habits in the future. Furthermore, market conditions can affect car prices, and there may be differences in car ownership between urban and rural areas of residence. Age plays a role in family holiday attendance probability, and adolescents are more likely to have separate bedrooms as they grow older, irrespective of their family’s affluence level (Boyce et al., 2006; Schnohr et al., 2008; Torsheim et al., 2016).

The lack of scalar invariance across time is not necessarily a major research limitation. When time-to-time comparisons of FASII scores are not permissible, one cannot assess the development of affluence over a certain period. However, to the authors’ knowledge, change in affluence has not been the research question in any research paper to date using HBSC data. Studies mainly aim to understand affluence ranking in relation to health outcomes. To compute the association between affluence rank and health outcomes, direct comparisons of raw FASII-scores over time are not prerequisites. CFA gave an overall good model fit, and the overall factor structure fitted well across the groups when conducting a configural invariance test, validating that FASII indicators can be used to compare and rank participants within each group: Survey year, age, and sex. Using the indicators in this manner could give important insight into FASII scores and health outcomes at, for example, different time points. As such, indicators could offer a “snapshot” of a single moment in time. Similar results obtained at different time points, assessed separately, could indicate that such coherence is similar across time. A standard measure for such comparisons is the Slope Index of Inequality (SII) and Relative Index of Inequality (RII), used to quantify, in relative and absolute terms, respectively, the linear association between affluence rank and chosen health outcome (Moreno-Betancur et al., 2015). SII and RII are based on ranked data and do not require a direct comparison across time and could be considered particularly useful.

4.1 Suggestions for Future Research

This study found that the holiday-related item showed weak factor loadings, indicating that this item might be less fitting for measuring family affluence in Norway. Torsheim et al. (2016) found ceiling effects for several items in the Norwegian sample, indicating that they are too prevalent to capture the distribution of family affluence. These findings indicate a need for indicators that are more fitting for a Norwegian context. To further optimize the measurement of the scale, developing country-specific items with higher local relevance should be a focus for future research. From an inequality perspective, capturing the entire continuum of affluence is essential. If items focus on the lower end of the continuum, in countries high in affluence, one would mainly be able to assess individuals at the poverty line, not the affluent (Torsheim et al., 2016). Developing country-specific indicators for Norway would require a shift in focus from not only necessities but also desirable goods, making it possible to stratify the population at the higher end of the affluence continuum (Torsheim et al., 2016).

In addition to developing country-specific indicators, the current indicators should be better tailored for the Norwegian context. Three main units are used in primary and secondary school: Laptops, iPads, and Chromebooks (Gilje et al., 2020). Asking Norwegian adolescents about computer ownership could be confusing as it is unclear if this question also includes iPads and Chromebooks. Digital units should be specified. Further, car ownership could be considered an increasingly ambiguous question as car-sharing has become common, especially in urban areas in Norway (Hjorteset & Böcker, 2020). In such cases, the family can access multiple cars without direct ownership. This question could be tailored further.

Collecting qualitative data through interviews could help provide insight into how adolescents interpret and understand the items and further what might be considered desirable goods in the Norwegian context for additional development of new indicators.

4.2 Limitations

The study is not without limitations. Given that Norway is recognized as a particularly affluent nation, it is plausible that findings may not be replicated in other countries. This raises questions regarding the generalizability of the present results beyond the Norwegian context. Replicating the methodologies employed in this study across other HBSC countries would allow for an assessment of the reproducibility of findings in diverse socio-economic settings. Nevertheless, focusing on a nation characterized by high levels of prosperity can be considered a unique strength of this study. It offers insight into how the indicators operate within such socio-economic context, thus highlighting the higher end of the affluence continuum. Such insight is valuable for informing the refinement and advancement of indicators pertaining to high levels of affluence.

A notable limitation inherent to the scale rather than the study itself is the predominance of items fitting the lower end of the affluence continuum. This could impede studies seeking to encompass the full range of affluence, especially in nations with high overall wealth.

5 Conclusion

This study aimed to validate the use of FASII for time trend analysis within the Norwegian HBSC survey conducted between 2002 and 2018. The study’s results revealed an overall good model fit in CFA and a positive correlation between FASII scores and perceived family wealth. However, the analysis uncovered measurement non-invariance across survey years, sex, and age groups. Measurement non-invariance hampers direct time-to-time comparisons of FASII scores, impeding the assessment of affluence development. Despite this limitation, FASII maintains its utility for ranking affluence and measuring health outcomes at single time points. As such, this study provides valuable insight into the suitability of FASII for time trend analysis within the Norwegian HBSC data and, more broadly, within HBSC data in general. Additionally, it contributes to broader research on social inequality.

The study recommends further development of the scale for the Norwegian context. Current indicators might need refinement, particularly by specifying digital units, and car ownership should be explicitly defined to avoid confusion with car-sharing. Given the weaker factor loadings of the holiday item compared to other indicators, reconsidering its inclusion in Norwegian surveys is advisable. Additionally, incorporating indicators representing higher levels of affluence is necessary to stratify the Norwegian population at the higher end of the affluence continuum.