Pay to play? Economic constraints and participation in extracurricular activities

Abstract

Research across countries shows that children from lower-income families are less likely to participate in extracurricular activities than children from more affluent families. While this income gradient in participation is by now established, the mechanisms behind the gradient are more contested. I examine whether the income gradient in extracurricular activity participation is the result of household economic constraints, using panel data methods on a nationally representative sample of Swedish adolescents. Data from the Children of Immigrants Longitudinal Study in Four European countries (CILS4EU) allow me to combine taxation register data on changes in household income with survey data on changes in extracurricular activity participation. Results from first-difference regression models show that changes in household income are not generally associated with changes in participation, but a weak association is found between changes in income and ceasing participation among adolescents in low-income households. The results largely cast doubt on theoretical explanations that emphasize household economic constraints as a substantial contributor to the income gradient in participation. Instead, results are more consistent with explanations emphasizing cultural differences in parenting logics and parental preferences for participation, as well as with explanations stressing non-economic forms of resource constraints.

Introduction

Across countries, research consistently shows that children from lower-income families are less likely to participate in extracurricular activities than children from more affluent families. This income gradient in participation is cause for concern, as it suggests that children from lower-income backgrounds have less access to age-typical and socially valued activities from which many draw meaning and social belonging. Furthermore, since it has been argued that participation in extracurricular activities supports important developmental tasks (e.g. Mahoney et al., 2005) and is often seen as beneficial for educational and occupational outcomes, scholars have suggested that participation forms part of how parents transmit advantages to the next generation (e.g. Covay and Carbonaro, 2010; Putnam, 2015; Mikus, Tieben and Schober, 2021).

The negative association between lower-income family background and participation is not a surprising finding, because participation in extracurricular activities often carries significant costs. Still, the intuitive interpretation of the income gradient in participation—that economic constraints limit participation—is not necessarily correct. Household income correlates with other parental characteristics, such as parenting logics, parental preferences, and parental working conditions, and such correlations may confound the association between income and participation. Lower participation rates among lower-income children could thus be caused by factors other than economic constraints.

The aim of the current study is to examine whether household economic constraints drive the income gradient in extracurricular activity participation or whether the differences between children in lower income and more affluent households stem from other non-economic characteristics and circumstances. Prior research examining the causes of the income gradient has relied on cross-sectional data, meaning that estimates are vulnerable to confounding by unobserved variables. Arguing that confounding factors are predominantly time-constant, I aim to identify the role of economic constraints by means of a panel approach. Using data on a nationally representative sample of Swedish adolescents, I combine survey data on changes in participation with information on changes in disposable household income drawn from taxation register data. First-difference regression models suggest a more limited role for economic constraints than proposed by prior research, instead lending support to theoretical explanations emphasizing cultural differences in parenting logics, parental preferences, or non-economic forms of resource constraints.

Extracurricular activities and the income gradient in participation

Extracurricular activities are commonly defined as activities that occur outside the school curriculum, are focussed on developing skills in some activity, and are structured around regular attendance, formal rules, and adult supervision (Vandell, Larson and Mahoney, 2015). As such, extracurricular activities include sports; cultural activities such as music, art, or theatre; as well as activities organized around other interests or beliefs.

International research shows that children and adolescents in lower-income households participate less in extracurricular activities than children in more affluent families. Although most studies come from the United States (e.g. Covay and Carbonaro, 2010; Snellman et al., 2015; Weininger, Lareau and Conley, 2015; Meier, Hartmann and Larson, 2018), an income gradient in participation is also observed in other wealthy countries, such as the UK (Holloway and Pimlott-Wilson, 2014), Canada and Australia (Guèvremont, Findlay and Kohen, 2008; Baldwin and O’Flaherty, 2018); Japan (Matsuoka, 2019); France and Germany (Coulangeon, 2018; Gülzau, 2018); and the Nordic countries (e.g. Olsson, 2007; Sletten, 2010).

What explains the income gradient in participation?

While the income gradient in extracurricular participation is well established, there is less agreement on the underlying mechanisms. Research aiming to explain the gradient has largely occurred within a broader discussion of intergenerational transmission of advantage that draws on cultural and resource constraint oriented explanations. According to Weininger, Lareau and Conley (2015), the gradient largely stems from class-specific cultural logics on parenting. This assertion originates from Lareau’s (2000, 2002, 2011) ethnographic observations of everyday practices in middle-class, working-class, and poor American families. Lareau suggested that middle-class parenting was characterized by the belief that children’s nascent skills and abilities require ‘concerted cultivation’ to be fully developed. More common among the working-class and the poor was a parenting logic Lareau termed ‘achievement of natural growth’: As long as their children were lovingly cared and provided for, skills and abilities were largely assumed to develop without deliberate parent-child cultivation. Consequently, middle-class parents were more inclined to encourage, initiate, and support their children’s engagement in extracurricular activities. In comparison, working-class and poor children’s leisure tended to be less structured and their participation in extracurricular activities less frequent, less consistent, and more often self-initiated (Lareau, 2011).

Lareau (2011) conceives the class-specific cultural logics of parenting as an instance of parents’ habitus (Bourdieu, 1990). That is, the parenting logic stems from parents’ largely stable dispositions, which in turn is shaped by parents’ prior experiences (predominantly determined by their own upbringing and structural position). In a related strand of research, researchers focussing on participation in specific forms of activities—such as sports or highbrow cultural activities—have emphasized how parents transmit their lifestyle (in terms of tastes and preferences) through socialization and exposure (e.g. Nagel, 2010; Strandbu, Bakken and Stefansen, 2020).

Cultural explanations have drawn critique in research focussed on resource constraints. This position argues that parental beliefs on the importance of extracurricular activities varies little across class background. Instead, the income gradient in participation stems from resource constraints—predominantly household economic constraints—limiting participation (Chin and Phillips, 2004; Bennett, Lutz and Jayaram, 2012; Holloway and Pimlott-Wilson, 2014). Indeed, ethnographic research provides examples of low-income parents who want their children to participate in extracurricular activities but find the costs to be prohibitive (e.g. Holloway and Pimlott-Wilson 2014). Participation in extracurricular activities can entail substantial fees as well as costs for equipment and transportation. Expenses may also cluster around the start of school terms, meaning that initiating or maintaining participation is contingent on the availability of resources at specific and re-occurring time points.

Non-economic resource constraints could also contribute to the income gradient in participation. Participation may require available parental time to initiate and maintain participation. Time constraints are likely more common among parents in lower-income households, as they are more often employed in occupations where working hours are less flexible and more often involve non-standard or irregular work schedules (e.g. Gülzau, 2018; Sjödin and Roman, 2018). Additionally, if lower-income families more often reside in areas with lower availability of extracurricular activities, this could contribute to the association between income and participation (e.g. Bennett et al., 2012).

Empirical research examining the income gradient has found that a significant association between income and participation in extracurricular activities remains after adjustment for other household characteristics, such as parental education. This robustness of the income–participation association has been interpreted as evidence that economic constraints limit participation (e.g. Covay and Carbonaro, 2010; Weininger et al., 2015; Gülzau, 2018; Behtoui, 2019). Crucially, however, prior studies have used cross-sectional data, and the observable household characteristics adjusted for are unlikely to fully capture the parenting logics and parental preferences emphasized by culturally oriented explanations. In other words, the association between household income and participation could well be confounded.

The Swedish context

Sweden is often characterized as a Social Democratic welfare state (Esping-Andersen, 1990). As in neighbouring Scandinavian countries, income inequality is comparatively low, and social services are comprehensive, tend to be universal, and have low user fees. Families receive substantial direct and in-kind transfers, including universal child benefits, heavily subsidized childcare, and free schooling and healthcare for children (Kautto, 1999). Scandinavian countries, including Sweden, also have a history of popular mass movements (often with egalitarian ambitions), contributing to a sizeable membership-based voluntary sector that is especially active in sports and culture (Lundström and Svedberg, 2003; Fahlén and Stenling, 2015). As a consequence, and in contrast to the organization of extracurricular activities in, for example, the United States, extracurricular activities are detached from the school system. Sports activities are largely organized by member-based voluntary sports clubs, with activities subsidized by the state through the Swedish Sports Confederation (Støckel et al., 2010; The Swedish Agency for Youth and Civil Society, 2014). Cultural activities are predominantly organized by community arts schools, which are managed, funded, and subsidized by municipalities (SOU 2016: p. 69; Jeppsson and Lindgren, 2018).

But despite the low-income inequality, in-kind transfers, and the support for extracurricular activities from state, municipal, and volunteer efforts, children and adolescents from lower-income families participate less in extracurricular activities also in Sweden, a pattern noted in academic research (e.g. Olsson, 2007; Jonsson and Mood, 2014; Behtoui, 2019) as well as in government (SOU 2008: p. 59, 2016) and government agency reports (The Swedish National Board for Youth Affairs, 2005; Statistics Sweden, 2009; The Swedish Agency for Youth and Civil Society, 2014).

As in other countries, household economic constraints are often seen as a major cause for the income gradient in participation. Indeed, participation in extracurricular activities can involve substantial economic costs. For instance, in 2009, the Swedish Sports Confederation surveyed parents of participants in sports clubs on the costs associated with their child’s participation. The median cost among parents was slightly above SEK 5,000 (≈ €500) per year and sport—stemming from fees (on average 38 per cent), equipment (22 per cent), and transportation (39 per cent). Costs were found to be higher for adolescents than for younger children and with considerable variation between sports. A third of surveyed parents thought participation was associated with high costs, rising to 50 per cent among low-income households (The Swedish Sports Confederation, 2010). Participation in community arts schools typically involves fees similar to sports club fees (The Swedish Arts Council, 2019) and also generates costs for equipment and transportation.

The current study

The current study aims to identify the role of household economic constraints in generating the income gradient in extracurricular activity participation. To do so, I utilize panel data on a nationally representative sample of the 2010/2011 cohort of Swedish eighth grade students, drawn from the Children of Immigrants Longitudinal Survey in four European Countries (CILS4EU). The current study thus examines adolescents, contributing to knowledge on social inequalities during a period when participation in extracurricular activities is purported to support important developmental tasks, such as identity development, autonomy, and intimacy with peers (Mahoney et al., 2005). Nevertheless, it should be acknowledged that participation peaks in early adolescence and takes a downward trajectory thereafter (Mahoney et al., 2005; The Swedish Agency for Youth and Civil Society, 2014).

The analysis proceeds in two steps. The first starts from the cross-sectional bivariate association between disposable household income and extracurricular activity participation, then adjusts for non-income household characteristics in order to mirror prior research using a cross-sectional approach (e.g. Covay and Carbonaro, 2010; Weininger et al., 2015). Although this analysis cannot identify the causal effect of economic constraints, it is useful in that it can show whether the association is accounted for by observable non-income variables. If so, it is unlikely that economic constraints are the crucial mechanism. If, on the other hand, a substantive association remains between income and participation in the multivariate models, we cannot rule out a causal relationship between economic constraints and lower levels of participation. However, the variables included in these models are unlikely to fully account for the parenting logics and parental preferences emphasized in culturally oriented explanations of the income gradient.

The second part of the analysis distinguishes such confounding from the role of economic constraints by utilizing the panel structure of the CILS4EU-data to examine within-person changes in disposable household income and extracurricular activity participation over a 1-year period; from grade eight (ages 14–15) to grade nine (ages 15–16). Analysing within-person changes (how changes in household income relate to changes in participation) makes it possible to control for characteristics and circumstances—whether observed or unobserved—that are stable over time. As noted above, the mechanisms suggested by culturally oriented explanations are likely to be largely stable and can therefore be controlled for in this way. This is also the case for residential area related constraints, such as the availability of leisure activities. To a smaller but still substantial extent, this is also likely to be the case for parental working conditions that could hamper or facilitate participation.

If household economic constraints play a substantial role in generating the association between income and participation, we would expect adolescents to, on average, be more (less) likely to participate following periods with higher (lower) disposable household income than after periods with lower (higher) disposable household income. In such a case, changes in disposable household income will be related to changes in participation. As changes in household income are more likely to produce or alleviate economic constraints to participation in low-income households, we can expect the association between changes in income and changes in participation to be stronger in the lowest income quintiles. In addition to panel analyses using the full sample, I therefore also perform panel analyses restricting the sample to adolescents in the two lowest income quintiles.

Additionally, the effect of a change in household income could vary depending on participation at the starting point. This could be the case if, for instance, parents find it more important to maintain an activity that the adolescent is already enrolled in than to allocate resources to a new activity. To examine this, I run additional analyses, conditioning on the participation or non-participation in the first wave (eighth grade).

Data, measures, and statistical analyses

Data

CILS4EU is a school-based longitudinal survey conducted in England, Germany, the Netherlands, and Sweden (Kalter et al., 2014). This paper utilizes the Swedish part of the survey, as this has been linked to information on disposable household income drawn from taxation register data. Sampling was conducted using a stratified three-stage sampling approach. First, schools were randomly selected within four strata (based on the proportion of students with an immigrant background), oversampling schools with a high proportion of students with an immigrant background. Within sampled schools, two school classes were randomly selected. Finally, all students in the selected classes were invited to participate in the survey (CILS4EU, 2016). Official survey weights allow inferences to the 2010/2011 cohort of Swedish eighth grade students.

Statistics Sweden administered questionnaires in schools. Data collection for the first wave started in December 2010, with 90 per cent completed before the end of March 2011. Data collection for the second wave started in late February 2012, with 90 per cent completed in early May. Statistics Sweden linked survey responses to annual information from administrative registers.¹ In addition, I match the CILS4EU-data to contextual information on extracurricular activity participation in the municipalities in which the participating schools are located.²

I use the same analytical sample for the cross-sectional analyses as for the panel analyses. The gross sample consists of the 5,025 students who participated in Wave 1 (88 per cent of those sampled). I exclude 114 respondents for whom I lack data on any of the household characteristics used in the analyses, including respondents with negative or zero disposable household income (universal child benefits means these incomes are likely not accurate). I further exclude 25 observations with mischievous response patterns (as judged by the Swedish CILS4EU-team). Excluding respondents who did not respond to the question on EA-participation in Wave 1 (n = 624), was lost to panel attrition (n = 861), or who did not respond to the item in Wave 2 (n = 75), reduces the sample by 1,411 respondents. This resulted in an analytical sample of 3,475 respondents (69 per cent of the gross sample; 61 per cent of those sampled). As compared to the analytical sample, respondents excluded due to item non-response or second wave panel attrition more often come from households with lower income, less education, of immigrant descent, and that do not include both parents (see Supplementary Appendix 3). The cross-sectional association between income and participation could thus be underestimated.³ In the Supplementary Appendix 3, I repeat the main analyses employing multiple imputation to address item non-response and attrition. Results do not alter conclusions drawn from the main analyses.

Measures

Extracurricular activity participation was measured with the following item: ‘In your spare time, how often do you spend time in a club (sports/music/drama/other club)?;’ possible responses being: Every day, once or several times a week, once or several times a month, more rarely, or never.⁴ I generate two measures: First, a binary measure—weekly participation—capturing whether adolescents participate in extracurricular activities on at least a weekly basis (1) or more rarely (0). This is my preferred measure, as it best reflects the regularity of participation inherent to the definition of extracurricular activities.

Nevertheless, utilizing all frequency categories in the survey item renders a more fine-grained description of differences in participation. Furthermore, it allows for more sensitive analyses of changes in participation across waves. I thus create a second measure—frequency of participation—which treats the survey item as a continuous index ranging from 0 (Never) to 4 (Every day). It is, however, important to note that coefficients can only be interpreted as more (less) participation or increases (decreases) in participation (as the underlying variable is ordinal).

Disposable household income is the income available to the household after tax and any benefits, according to information from taxation register data. For adolescents with separated parents, I take the average disposable household income of both households. The variable is expressed in 100,000 SEK (≈ €10,000, in 2010). To decrease the influence of outliers, I top-code this income measure at three standard deviations from the mean of the average income across 2010 and 2011 (in the weighted gross sample). In addition, to reduce the influence of outliers in the panel analyses, I further top- and bottom-code the measure of income change to three standard deviations from the mean.⁵ For descriptive purposes and to restrict some analyses to lower-income households, I also generate disposable household income quintiles (referring to 2010, the first wave). Cut-offs are based on the weighted gross sample, meaning that quintiles are approximately representative of quintiles in the cohort.

While it could be argued that an equivalized disposable household income measure would better capture economic constraints, due to data limitations, I can only construct equivalence scales for the first wave. Thus, changes in equivalized income would only reflect changes in income, not household composition. I therefore prefer the more transparent disposable household income measure.⁶

Parental education is intended as a proxy for the mechanisms suggested by culturally oriented explanations. This approach follows earlier research lacking direct measures of, for instance, cultural logics of parenting (e.g. Weininger et al., 2015). Following Matsuoka (2019), I measure parental education as the number of parents with any form of tertiary education. Information is drawn from administrative registers.

Immigration background has been found to be related to extracurricular activity participation (e.g. The Swedish Agency for Youth and Civil Society, 2014). As for parental education, one possible interpretation of differences in participation (when adjusted for income, education, etc.) is cultural differences in parenting logics or parental attitudes toward participation. I categorize adolescents with a majority background (at least one Swedish born parent); adolescents with a second-generation immigration background (born in Sweden to foreign-born parents); and adolescents with a first-generation immigration background (born abroad to foreign-born parents). The variable is generated based on a combination of register data (place and date of birth) and information from the survey (parents’ country of birth).

Family structure is related to extracurricular activity participation (Fransson et al., 2018), one possible interpretation being that time constraints to participation are larger when children live with single parents. I construct a binary measure distinguishing between adolescents living with both biological parents in the same household and adolescents in all other family structures. Adolescents residing with both parents are treated as the reference category. The measure is based on information drawn from administrative registers.

The number of siblings could decrease household resources (e.g. economic, time) available to the respondent. On the other hand, some resources (such as equipment) and family experience of participation could be higher with more siblings. The measure is based on survey information reported in the first wave. If a respondent reports living in two households, the measure is the sum of siblings in both households. For each individual household, the number of siblings is top-coded at seven.

Gender separates girls and boys, as reported by respondents.

Age at interview is constructed from survey responses. It is measured in days but expressed in years.

To account for the possibility that lower-income families more often reside in areas with less access to organized leisure, I include contextual measures on the municipalities where participating schools are located (in the cross-sectional models). The type of municipality could affect the availability of extracurricular activities as well as how accessible activities are in terms of transportation. I therefore include a measure of the degree of urbanization: major city; city; and small towns or rural region.

I also generate measures intended to capture the availability of extracurricular activities in the municipality (endnote 2 describes this data). Participation in organized sport in the municipality is a ratio of the sum of individual activity instances (e.g. trainings, matches) attended by individuals between seven and 20 in 2010 over the number of individuals of that age in the municipality. As individuals can attend multiple activities, the measure is best thought of as capturing how common it is for children and adolescents in the municipality to participate in sports club activities.

Participation in community arts school in the municipality is based on the number of places in subject courses (e.g. theatre, guitar, etc.). Again, I generate a ratio. The denominator is the number of children and adolescents in the municipality between six and 19 (Sjödin and Roman, 2018). Since individuals can attend multiple courses, the measure captures the availability of this type of activity, not the share of children and adolescents participating.

Statistical analyses

To examine to what extent the cross-sectional association between disposable household income and participation is accounted for by non-income household characteristics, I perform sequential OLS-regression models, reporting the share of the association accounted for. Linear probability models (LPMs) are used when analysing participation on a weekly basis (for both the cross-sectional and within-person analyses).⁷ All reported results are weighted using the official survey weights. Standard errors are clustered on the school level.

In the second part of the analyses, I run OLS-regressions of within-person changes in participation on the within-person change in disposable household income. That is, first-difference regression models.⁸ The first-differenced equation estimated by the regression models can (using notation based on Allison [2009]) be written:

$$y_{i2}-y_{i1}=\left({\mu }_2-{\mu }_1\right)+{\beta }_1\left(in{c}_{i2}-in{c}_{i1}\right)+{\beta }_2\left(x_{i2}-x_{i1}\right)+\gamma z_i+{\alpha }_i+({\epsilon }_{i2}-{\epsilon }_{i1})$$

Or (rewritten as difference scores and reduced):

$$\Delta y_i=\Delta \mu +{\beta }_1\Delta in{c}_i+{\beta }_2\Delta x_i+\Delta {\epsilon }_i$$

Using within-person changes removes time-constant characteristics or circumstances from the equation, whether these are observed (⁠$z_i$⁠) or unobserved (⁠${\alpha }_i$⁠). Time-varying covariates (⁠$x_{it}$⁠) remain in the reduced equation, meaning that the estimate for income (⁠${\beta }_1$⁠) could be biased by changes in characteristics or circumstances not adjusted for—if these are correlated with both changes in income and changes in participation (through other paths than income). In a second model, I therefore adjust for changes in family structure (which apart from changes in income could affect participation through residential disruptions or increased logistical challenges). The intercept (⁠${\mu }_t$⁠) captures average changes in participation, and the error term captures changes in participation that are unrelated to changes in income (and family structure).

Results

Descriptives

Table 1 shows weighted descriptive statistics. As shown in the table, 62 per cent of adolescents participate at least once a week in extracurricular activities (in the first wave). The more detailed measure of frequency of participation shows that the most typical frequency category is once or several times per week (49.5 per cent), while a smaller group report participating in extracurricular activities as often as every day (12.6 per cent). The modal 14- to 15-year-old adolescent thus participates in extracurricular activities on at least a weekly basis. On the other hand, the group of adolescents who do not participate on at least a weekly basis (37.9 per cent), to a substantial extent consists of adolescents who do not participate at all: almost 22 per cent (21.8 per cent) report no participation.

Table 1

Descriptive statistics (weighted) of analytical sample (n = 3,475)

		Wave 1	Wave 2
Participation, on weekly basis	(1) Yes	62.11%	58.21%
	(0) No	37.89%	41.79%
Participation, frequency	Mean (Std. deviation)	2.20 (1.40)	2.14 (1.43)
	(0) Never	21.80%	22.96%
	(1) Less often	11.47%	12.74%
	(2) Once or several/month	4.62%	6.09%
	(3) Once or several/week	49.53%	44.03%
	(4) Every day	12.58%	14.18%
Disposable household income (in 100,000 SEK)	Mean (Std. deviation)	5.20 (2.79)	5.53 (3.06)
Disposable household income quintiles	Quintile 1 (lowest)	17.35%
	Quintile 2	19.19%
	Quintile 3	21.38%
	Quintile 4	21.20%
	Quintile 5 (highest)	20.88%
Number of parents with tertiary education	No parent	49.42%
	One parent	31.78%
	Both parents	18.80%
Immigration background	Majority	85.02%
	Second generation	10.46%
	First generation	4.52%
Family structure	Living with both parents	66.94%	65.83%
	Not living with both parents	33.06%	34.17%
Municipality type	Major city	34.24%
	City	38.10%
	Smaller city/rural	27.66%
Community arts school in municipality	Mean (Std. deviation)	0.15 (0.07)
Organized sport in municipality	Mean (Std. deviation)	34.61 (7.04)
Number of siblings	Mean (Std. deviation)	1.66 (1.47)
Gender	Boy	49.83%
	Girl	50.17%
Age at interview	Mean (Std. deviation)	14.63 (0.35)

The general trend in participation across waves seems to be a weak decrease. About 4 percentage points fewer adolescents report participation on a weekly basis in Wave 2. The decrease is also evident in the slightly lower mean for the frequency of participation score: from 2.20 in the first wave to 2.14 in the second.

Cross-sectional analyses

Figure 1 displays average extracurricular activity participation by disposable household income quintiles (in the first wave). First, we observe a strong positive association between disposable household income and weekly participation. Among adolescents in the lowest two quintiles, around half of the adolescents participate on a weekly basis. The corresponding share in the middle quintile is 63 per cent, and in the highest quintile, it is 75 per cent. For frequency of participation, the pattern is similar. The difference in frequency of participation score between the lowest (1.83) and the highest (2.53) income quintile is 0.70 (on the 0–4 scale).

Figure 1

Differences in weekly participation and frequency of participation across household income quintiles in the first wave (2010).

How much of the associations can be accounted for by other, non-income characteristics or circumstances? Table 2 shows the results of regression models of participation on household disposable income and other household characteristics. The two left-most models focus on weekly participation. The first model replicates the bivariate association above but uses the continuous income measure instead of quintiles and also adjusts for age and gender.⁹ For each 100,000 SEK (≈ €10,000) higher household disposable income, adolescents are on average 3.2 percentage points more likely to participate on a weekly basis.

Table 2

Regression of participation on a weekly basis (Models 1 and 2) and frequency of participation (Models 3 and 4) on disposable household income and non-income characteristics in wave 1

	Weekly participation		Frequency of participation
	Model 1	Model 2	Model 3	Model 4
Disposable household income	0.032**	0.019**	0.093**	0.052**
	(0.003)	(0.004)	(0.010)	(0.011)
Parents with tertiary education (ref. no parent)
One parent		0.122**		0.351**
		(0.024)		(0.069)
Two parents		0.130**		0.392**
		(0.028)		(0.077)
Immigration background (ref. majority)
Second generation		−0.104**		−0.314**
		(0.036)		(0.113)
First generation		−0.141**		−0.371**
		(0.047)		(0.134)
Not living with both parents		−0.067**		−0.212**
		(0.021)		(0.062)
Number of siblings		−0.002		0.011
		(0.008)		(0.020)
Type of municipality (ref. city)
Major city		−0.000		0.024
		(0.035)		(0.101)
Smaller city/rural		0.007		0.023
		(0.034)		(0.100)
Community arts school in municipality		0.165		0.478
		(0.200)		(0.520)
Organized sport in municipality		−0.002		−0.002
		(0.002)		(0.006)
Girl	0.038	0.041	0.011	0.019
	(0.024)	(0.024)	(0.065)	(0.065)
Age at interview (Wave 1)	0.018	0.038	0.080	0.138
	(0.028)	(0.029)	(0.088)	(0.089)
Intercept	0.178	−0.044	0.536	−0.178
	(0.414)	(0.441)	(1.297)	(1.350)
R-squared	0.034	0.062	0.034	0.062
Observations	3,475	3,475	3,475	3,475

Note: ** P < 0.01, * P < 0.05; Models 1 and 2 from LPMs.

The second model adds measures of parental education, immigration background, family structure, and measures on the municipality of residence. This attenuates the income coefficient to 0.019. In other words, including measures of non-income characteristics accounts for about two fifths (40.6 per cent; 0.013/0.032) of the originally estimated coefficient. Importantly, three fifths of the association remains.

The two right-most columns show the results of OLS-regressions of the frequency of participation on household income. Again, the first model adjusts for age and gender, while the second model adds non-income household characteristics. The first model shows that for each 100,000 SEK higher household disposable income, the frequency of participation score is expected to be almost 0.1 (0.093) higher (on the 0–4 index of participation). The main takeaway from the second model is that adjusting for non-income household characteristics accounts for a substantial share (44.1 per cent; 0.041/0.093) of the association between income and frequency of participation but does not fully account for the association.

In addition, Models 2 and 4 demonstrate striking associations between the household characteristics adjusted for and participation in extracurricular activities—independent of disposable household income. Adolescents having no parent with a tertiary education are substantially less likely to participate on a weekly basis (more than 12 percentage points less likely); as are adolescents of immigrant descent (14 percentage points for first generation, 10 percentage points for second generation); and those not living with both parents in the same household (almost 7 percentage points).

For both weekly participation and frequency of participation, about three fifths of the association with household income remains after adjustment for non-income characteristics and circumstances. While similar findings have been interpreted as support for the notion that economic constraints are responsible for a non-negligible part of the income gradient in participation, the non-income characteristics included in these models are unlikely to adjust for all confounding factors. To better account for stable characteristics and circumstances—such as the parenting logics and parental preferences emphasized by culturally oriented explanations—I next turn to analyses of changes in income and participation.

Panel analyses

In Table 1, we observed a small decrease in participation from the first to the second wave. When assessing changes in participation, however, it is evident that the overall pattern hides substantive changes within-persons across waves (see Supplementary Appendix 6). In regard to weekly participation, 75.5 per cent maintain the same state across waves: 47.9 per cent participate at least once a week in both the first and second wave, while 27.6 per cent do not participate in either wave. Nevertheless, 24.5 per cent experience changes in participation: 14.2 per cent stop participating on a weekly basis and 10.3 per cent start participating on a weekly basis.

Unsurprisingly, there are more within-person changes in the more fine-grained measure of frequency of participation: 55.4 per cent report the same frequency in both waves. Thus, a substantial proportion, 44.6 per cent, report a different frequency of participation in the second wave: 23.6 per cent decrease participation while 21.0 per cent increase their participation. Most respondents changing frequency of participation, however, change to an adjacent frequency category, thus producing the small decrease (−0.059) in frequency score mean evidenced in Table 1.

The mean change in annual disposable household income from 2010 to 2011 is an increase of 30,000 SEK (with a standard deviation of 146,000). Figure 2 further describes the change in disposable household income by 2010 income quintile using a violin plot.¹⁰ Most households (69.8 per cent overall) experience increases in income across waves, but all quintiles also include households that experience decreases in income. Similarly, changes in income are typically fairly small, but larger changes do occur in all quintiles. Changes are larger, more common, and more often negative in the highest income quintile. In interpreting the regression coefficients below it is thus important to note that the change in income is expressed on the scale of 100,000 SEK (≈ €10,000)—a change in income that only a smaller proportion of respondents actually experience. From the bumps at the tails of the distributions, we can assess the proportion of observations affected by the top- and bottom-coding in each income quintile. The proportion is small in most quintiles (about 1 per cent), but close to 9 per cent in the highest income quintile.

Figure 2

Violin plot describing change in disposable household income from 2010 to 2011 by income quintile (in 2010).

Table 3 shows regressions of change in weekly participation and frequency of participation on the change in disposable household income. Analyses are first done for the full sample (Panel A), then restricted to respondents who were in the two lowest income quintiles in 2010 (Panel B). In each of the samples, the first model is unadjusted for any time-varying characteristics, while the second adjusts for changes in family structure.

Table 3

First-difference regression models of change in weekly participation and frequency of participation on the change in disposable household income

	Weekly participation		Frequency of participation
	Model 1	Model 2	Model 3	Model 4
Panel A. Full analytical sample
Change in disposable household income	−0.008	−0.010	−0.022	−0.022
	(0.006)	(0.006)	(0.016)	(0.017)
Change in family structure		−0.073		0.005
		(0.071)		(0.174)
Intercept	−0.037**	−0.035**	−0.052	−0.052
	(0.012)	(0.012)	(0.030)	(0.030)
R-squared	0.001	0.001	0.001	0.001
Observations	3475	3475	3475	3475
Panel B. Sample restricted to respondents in the two lowest income quintiles
Change in disposable household income	0.007	0.004	−0.020	−0.019
	(0.016)	(0.017)	(0.041)	(0.041)
Change in family structure		−0.082		0.005
		(0.088)		(0.269)
Intercept	−0.046*	−0.045*	−0.056	−0.056
	(0.018)	(0.018)	(0.047)	(0.047)
R-squared	0.000	0.001	0.000	0.000
Observations	1472	1472	1472	1472

Note: ** P < 0.01, * P < 0.05; Models 1 and 2 from LPMs.

To reiterate, if economic constraints play a substantial role in generating the income gradient in participation, we would expect increases in income to be associated with increases in participation—indicating that a lack of economic resources suppressed participation. Analyses using the full analytical sample (Panel A) evidenced a weak association in the opposite direction. A 100,000 SEK increase in household disposable income is associated with a small (0.8 percentage point) decrease in the probability of weekly participation. Adjusting for changes in family structure slightly increases the strength of this association, but in no model are coefficients close to any substantive nor statistical significance. Results for frequency of participation (Panel A, Models 3 and 4) tell a similar story in that coefficients are negative, small, and non-significant.

Theoretically, we would expect increases in disposable household income to have a greater impact on the participation of adolescents in low-income households. Panel B repeats the analyses in Panel A, but restricts the sample to respondents in the two lowest disposable household income quintiles (in Wave 1). For participation on a weekly basis, the association is now in the expected direction (an increase in income being associated with a small increase in weekly participation). Also here, however, coefficients are small and far from statistically significant. For frequency of participation (Panel B, Models 3 and 4) the association remains negative, small, and non-significant.

Models in Table 3 assume that changes in disposable household income have similar effects for initiating extracurricular activity participation as for ceasing participation. Table 4 presents results that instead examine these outcomes separately depending on the state of participation in the first wave, then regressing the change in organized leisure participation on the change in income.

Table 4

First-difference regression models of initiating participation on a weekly basis (Models 1 and 2) and ceasing participation on a weekly basis (Models 3 and 4) on changes in disposable household income (conditioning on the state of weekly participation in wave 1)

	Start participation weekly		Stop participation weekly
	Model 1	Model 2	Model 3	Model 4
Panel A. Full analytical sample
Change in disposable household income	−0.003	−0.003	0.007	0.010
	(0.010)	(0.010)	(0.008)	(0.008)
Change in family structure		−0.008		0.218*
		(0.081)		(0.090)
Intercept	0.274**	0.274**	0.227**	0.224**
	(0.018)	(0.018)	(0.014)	(0.014)
R-squared	0.000	0.000	0.001	0.005
Observations	1,450	1,450	2,025	2,025
Panel B. Sample restricted to respondents in the two lowest income quintiles
Change in disposable household income	−0.000	0.001	−0.038*	−0.033
	(0.024)	(0.025)	(0.017)	(0.017)
Change in family structure		0.025		0.176
		(0.090)		(0.105)
Intercept	0.230**	0.229**	0.319**	0.317**
	(0.022)	(0.022)	(0.024)	(0.024)
R-squared	0.000	0.000	0.008	0.010
Observations	773	773	699	699

Note: ** P < 0.01, * P < 0.05; Models are LPMs.

For analyses using the full analytical sample (Panel A, Models 1 and 2), there is essentially no association between changes in income and the probability of starting participation among respondents who did not participate in the first wave. Models 3 and 4 look at the probability of ceasing participation among respondents who did participate in the first wave. The positive coefficient here indicates that an increase in income is associated with a slightly increased probability of ceasing participation, but coefficients are small, not statistically significant, and hence do not provide evidence that changes in income affect the probability of ceasing participation.

Panel B repeats the analyses, restricting the sample to respondents who were in the two lowest income quintiles in the first wave. Again, there is no indication that a change in income affects the likelihood of starting participation (Models 1 and 2). On the other hand, income changes are associated with the risk of ceasing participation. A 100,000 SEK increase (decrease) in disposable household income is associated with a 3.8 percentage point lower (higher) probability of ceasing participation. This association is attenuated (to 3.3 percentage points) when adjusting for changes in family structure and is no longer statistically significant at the 95 per cent level (Model 4). Still, this provides some indication that economic constraints could affect the risk of ceasing participation among adolescents in the lowest income quintiles.

While not the focus of the analyses, changes in family structure appear to be strongly associated with ceasing participation for those participating on a weekly basis in the first wave. Change in family structure is associated with a 21.8 percentage point higher likelihood of ceasing participation as compared to those not experiencing a change in family structure (Panel A, Model 4). This effect size remains substantial but decreases to 17.6 percentage points and no longer reaches statistical significance when restricting the sample to the lower two income quintiles (Panel B, Model 4). The loss of significance reflects that only about 2 per cent of respondents experience a change in family structure across waves (not shown). Larger data material is needed for a robust examination of the effect of changes in family structure on participation.

Discussion

In Sweden, as in many other countries, there is a substantial income gradient in extracurricular activity participation. This paper examined whether the intuitive interpretation of this gradient—that household economic constraints limit participation—is in fact correct. I addressed this question by combining survey data on the extracurricular activity participation of Swedish adolescents with taxation register data on disposable household income. Using cross-sectional analyses, I demonstrated a substantial income gradient in participation and showed that a substantial part of the income gradient remains after adjustment for observable household characteristics. While prior research has interpreted similar findings as evidence that economic constraints limit participation, I argued that the observable household characteristics included in the current study and prior studies are unlikely to fully account for confounding factors—not least the parenting logics and parental preferences emphasized by culturally oriented explanations of the income gradient.

If economic constraints play a substantial role in generating the income gradient, as suggested by resource constraint oriented research (e.g. Chin and Phillips, 2004; Bennett et al., 2012; Holloway and Pimlott-Wilson, 2014), changes in disposable household income should, on average, imply changes in participation. I therefore used panel data to examine whether changes in income translates to changes in participation, a design that also eliminates any confounding from time-constant characteristics. Results from the panel analyses showed that, in general, changes in disposable household income were not associated with changes in participation. That said, among adolescents from lower-income households, there was a weak tendency for changes in income to affect the probability of ceasing to participate (but not of starting to participate). This suggests that while economic constraints could play some role among lower-income adolescents, the effect in generating the income gradient appears limited.

The current study thus casts doubts on theoretical explanations in which similar preferences for participation are hindered by household economic constraints (Chin and Phillips, 2004; Bennett, Lutz and Jayaram, 2012; Holloway and Pimlott-Wilson, 2014). The results instead lend more support to culturally oriented explanations that emphasize differences in the cultural logics of parenting or parental preferences for participation (Nagel, 2010; Lareau, 2011; Weininger et al., 2015) and these findings may very well extend also to other forms of parenting practices with observed socioeconomic gradients. However, other resource constraints could hinder preferences for participation from being realized. Whether the income gradient in participation predominantly stems from mechanisms emphasized by culturally oriented explanations, from varying availability of extracurricular activities (Bennett, Lutz and Jayaram, 2012), or from non-economic constraints to participation (Sjödin and Roman, 2018) remains for future research to address.

Some caveats deserve mentioning. First, the current study addresses participation in any form of extracurricular activity, but economic constraints could be more important for participation in specific, more expensive activities than for participation in general. Second, it is important to note that the current study assesses explanations for the income gradient in participation among the general population. The influence of economic constraints could be more pronounced among the economically most vulnerable groups. Third, the first-difference models used in the panel analyses identify the effect of economic constraints by looking at short-term changes in income in adolescence. If economic constraints predominantly exert an influence at younger ages, the current study will not pick this up. I can therefore not rule out the possibility that some part of the cross-sectional association stems from economic constraints having a larger impact in earlier ages. However, the fact that costs for participation are generally higher in adolescence than in earlier childhood (e.g. The Swedish Sports Confederation, 2010) speaks against this possibility. To examine this issue further, future research could apply the panel approach used in the current study to panel data covering younger age groups.

Caveats aside, results of this study suggest that raising the disposable income of lower-income households would not in itself translate to an equalization of children’s extracurricular activity participation. This does not suggest that state, municipal, and volunteer efforts to reduce economic costs of participation are necessarily misguided. On the contrary, one possible interpretation of this study is that the Swedish approach to organizing extracurricular activities is largely successful in removing economic barriers to participation among the general population. If so, however, the Swedish approach does not appear as successful in addressing other dimensions of social inequality, as the current study documents striking gaps in participation between adolescents with different educational and immigration backgrounds—net of household income.

To what extent are these findings specific to the Swedish context? It is of course possible that the income gradient in participation has different underlying mechanisms in different countries. Economic constraints could have a greater impact where social services are less comprehensive and income inequality more amplified. Parenting logics and parental preferences could matter less where schools organize extracurricular activities. Nevertheless, the current study paves the way to further studies by demonstrating that failing to account for relevant unobserved confounding factors likely leads to an overemphasis of the importance of economic constraints in determining participation.

Acknowledgements

This study has benefited from detailed reads and constructive comments from Lawrence Berger, Viveca Östberg, Mattias Strandh, Curt Hagquist, Martin Hällsten, Magnus Bygren, Anni Erlandsson, Sara Kjellson, Per Engzell, Peter Fallesen, Stephanie Plenty, and Carina Mood. Valuable feedback from participants at the internal seminar of the Level-of-Living unit at the Swedish Institute for Social Research and seminars at the Sociology Department, Stockholm University are gratefully acknowledged.

Funding

This work was supported by funding from the Swedish Research Council for Health, Working Life and Welfare (FORTE; grant numbers 2017-00947 and 2016-07099). The CILS4EU-project was funded by the New Opportunities for Research Funding Agency Co-operation in Europe (NORFACE).

Data Availability

The study relies on the CILS4EU Swedish national data files. As these include linked administrative register data, data cannot be shared publicly (due to privacy and ethical reasons). Reduced versions of the data are available from GESIS Data Archive, Cologne (ZA5353/ZA5656). Municipal level data are publicly available from the respective data owner (see Supplementary Appendix 2) and available on request from the corresponding author.

Simon Hjalmarsson is a postdoctoral researcher at the Swedish Institute for Social Research, Stockholm University. His research interests center on living conditions in childhood and youth, with a special interest in associations between economic resources and social participation. His work has been published in the Journal of Youth and Adolescence and Children and Youth Services Review.

Footnotes

1

Access to disposable household income from taxation register data is a clear benefit, but that it is only available annually might be problematic. If, for instance, an improvement in income from 2010 to 2011 began early in 2011, late respondents in the first wave might have already increased their participation at the time of the survey. If so, this could downwardly bias the panel results. By excluding respondents who were surveyed at a later time, we can get some indication about the potential bias. Excluding the 39 per cent of respondents in the analytical sample who responded after January 2011 (the first salary of the year arriving in late January) does not alter conclusions (see Supplementary Appendix 1).

2

Contextual data is drawn from the Swedish Sports Confederation and the Swedish Arts Council. For more information, see Supplementary Appendix 2.

3

The panel estimation is not biased (relying on changes within individuals), but generalisability to the target population is affected.

4

In Swedish: ‘På fritiden, hur ofta… deltar du i föreningsaktivitet på fritiden (t ex idrott/musik/teater/annat)? Varje dag, en eller flera gånger i veckan, en eller flera gånger i månaden, mer sällan, aldrig)’.

5

Top-coding affects few observations in the analytical sample: Top-coding the cross-sectional measures affects 12 observations in 2010 and 18 in 2011. The panel top- and bottom-coding affects 65 observations. In the cross-sectional analyses, income coefficients are smaller using the non-top-coded measure, but conclusions are not substantively altered. For the panel analyses, using the non-top-coded measure does not substantively alter results (see Supplementary Appendix 5).

6

A replication of the main analyses using an equivalized measure (and an extended discussion) is available in the Supplementary Appendix 4. While there are minor differences in results, conclusions remain unchanged.

7

In contrast to log-odds ratios or odds ratios from logistic regression models, LPMs are comparable across models and can be intuitively interpreted as percentage point differences (e.g. Mood, 2010). Average marginal effects (AME) from logistic regression yields results very similar to the LPM estimates. In Table 2, Model 1, LPM = 0.032, AME = 0.038, and in Model 2, LPM = 0.019, AME = 0.023.

8

For analyses using panels with two periods, first-difference models are equivalent to fixed-effects models with individual and time-fixed effects.

9

Since I expect a decreasing effect of income on participation (in the cross-section), I have also tried models including a quadratic income term (not shown). For both weekly participation and frequency of participation, I find support for a non-linear association in the first model but not for the second model (when adjusting for non-income characteristics). I therefore do not include a quadratic income term in any of the models.

10

Violin plots combine a box plot and a display of the probability distribution by adding a kernel density plot to each side of the box plot (Hintze and Nelson, 1998).