The aim of Study 2 was to further evaluate the hypotheses outlined within Study 1 via analysis of data from the British Children’s Play Survey.
Methods
Participants
Participants were 1919 respondents who took part in the British Children’s Play Survey, administered by YouGov, a UK public opinion research company. The sample were recruited to be approximately nationally representative and then weighted back to the national profile of all adults aged 18 + including those without internet access. All participants (54% female) were parents or caregivers of children (49% female) aged 5 to 11 years. Full details of the sample are provided in Dodd et al. [1].
Measures
The measures were identical to those detailed in Study 1, including the CPS, SDQ, PANAS and K-6.
Procedure
Participants were invited to take part via email and directed to an online survey which they completed once. Recruitment started on 4th April 2020 and data were collected 4th-15th April 2020. The UK-wide lockdown due to Covid-19 began on 26th March 2020 so all data were collected within 3 weeks of the start of lockdown. As in Study 1, respondents were given very clear instructions to answer the questions thinking about life before it changed as a result of the pandemic, with the exception of the PANAS, where they were asked to respond with the previous week in mind.
Missing Data, Distributions and Outlier Checks
There were some missing data, in particular for a single item on the PANAS. Missing data was imputed using the mice package in R [30] and pooled results are reported. Where variables were skewed they were transformed and outliers were Winsorized. See Supplementary Materials Section 3 for details.
Data Analysis Plan
Following Study 1, pooled Pearson correlations were used to evaluate associations between the play variables and four child mental health variables. A Bonferroni corrected alpha of 0.0125 (0.05/4) was used. Additionally, in keeping with Study 1, we also examined whether the results held after controlling for parent mental health and demographic variables. Linear regressions were conducted to examine whether associations remained significant after including child age, child sex, child disability, parent disability, employment status, birth order, parent age, parent education and parent mental health as predictors.
Results
The correlations are shown in Table 2. These results show that effect sizes of all associations were small. Nevertheless, hours spent playing adventurously was significantly associated with both children’s internalising scores on the SDQ and positive affect as measured using the PANAS, with children who spent more time playing adventurously having fewer internalising symptoms and more positive affect. In contrast, hours spent playing unadventurously was not related to positive affect nor to SDQ scores. Children who spent more time playing unadventurously had less negative affect. For total time spent playing outdoors, significant correlations were found with both SDQ internalising scores and PANAS positive affect scores. No significant correlations were found between any of the time spent playing variables and externalising scores. All of the significant associations were robust to controlling for demographic variables in a linear regression (see results file available: https://beta.ukdataservice.ac.uk/datacatalogue/doi/?id=8793#!#0).
Table 2 Mean and standard deviation for each child mental health measure and Pooled Pearson correlation coefficients (r) and 95% confidence intervals for relationships between play variables and children's mental health Although it was not our primary focus, in addition to the play variables, the following demographic factors were significant predictors of child mental health. For the regressions predicting SDQ internalizing scores, child disability, parent disability, birth order, parent age and parent mental health were all significant predictors (p < 0.05); children had fewer internalizing symptoms if they did not have a learning disability, physical disability or diagnosed mental health problem, were not first born, had parents who were older and parents who were experiencing less distress as measured using the K6. For regressions predicting positive affect, child age, parent education and parent mental health were significant predictors (p < 0.05), with children having more positive affect when they were younger, female, and when their reporting parent had lower distress as reported on the K6. For regressions predicting negative affect, child sex, child disability, parent disability, parent employment status, parent age, parent education and parent mental health were all significant predictors (p < 0.05). Children had less negative affect when they were male, did not have a disability, did not have a parent with a disability, when their parents were not working, older, highly educated and experiencing less distress as measured on the K6.
The results quite closely replicate the findings from Study 1 with the exception that the correlation between adventurous play and internalising problems was notably smaller. Given that the Study 1 sample were from Northern Ireland, where human capital per head and household income are lower than across Scotland, Wales and England [31, 32], we speculated that the difference in results might be explained by demographic differences between the samples. The BCPS dataset included a binary variable labelled ‘Social Grade’. This is a classification based on the occupation of the lead income earner in the household and is closely associated with household income [33]. In the absence of household income data we chose to use this variable to explore whether results would differ between children growing up in higher income and lower income households.
Household income and play were entered as predictors in linear regression models along with their interaction. Separate regressions were conducted for each of the mental health measures and each play variable. A Bonferroni corrected alpha of 0.0125 was used given the four mental health measures. Full models are available in the results file available here: https://bit.ly/3lqxXdx. Significant interactions were found for five models. Two of these models were for SDQ internalising score; one with a significant interaction between adventurous play and household income (b = − 0.01 [− 0.02, 0.00], p = 0.002) and the second with a significant interaction between outdoor play and household income (b = − 0.01 [− 0.02, 0.00], p = 0.004). In one model for positive affect, a significant interaction between adventurous play and household income was found (b = 0.04 [0.01, 0.06], p = 0.003). Finally, for negative affect as measured using the PANAS, two further models included a significant interaction with interactions between adventurous play and household income (b = 0.00 [− 0.01, 0.00], p = 0.012) and between unadventurous play and household income (b = 0.00 [0.00, 0.01], p = 0.012).
To explore each of these interactions we divided the sample into lower and higher household income groups using the social grade variable and calculated correlations. The results show that, for low household income, there were statistically significant negative associations between SDQ internalising score and time spent playing both outdoors (r = − 0.14* [− 0.21, − 0.07]) and adventurously (r = − 0.15* [− 0.22, − 0.08]) but neither of these correlations were significant for high household income (outdoors: r = − 0.01 [− 0.07, 0.05]; adventurously: (0.01 [− 0.05, 0.07]). The relationship between positive affect and time spent playing adventurously was stronger for low household income (r = 0.25* [0.19, 0.32]) than for high household income (r = 0.11* [0.05, 0.17]), although significant for both groups. In contrast, for high household income, negative affect was negatively associated with time spent playing unadventurously (r = − 0.14* [− 0.20, − 0.08]), and positively associated with time spent playing adventurously (0.09* [0.03, 0.16]), whereas these relationships were smaller and not statistically significant for low household income (unadventurously: r = − 0.02 [− 0.09, 0.05]; adventurously: r = − 0.03 [− 0.10, 0.04]).
Linear regressions controlling for demographic variables showed that the interactions between play and household income remained significant for the models predicting internalising problems and positive affect (see results file available here: https://beta.ukdataservice.ac.uk/datacatalogue/doi/?id=8793#!#0), but did not remain significant in either model predicting negative affect.
Brief Discussion
The results supported our primary hypotheses that children who spent more time playing adventurously would have fewer internalising problems as well as more positive affect during the Covid-19 lockdown. We also hypothesised that adventurous play would be associated with less negative affect but, consistent with Study 1, this was not supported. A significant negative association was found, however, between unadventurous play and negative affect; children who typically spent more time playing unadventurously had less negative affect during the Covid-19 lockdown. For all associations, the effect sizes were small.
Complementing Study 1, we found interactions between play and household income indicating that adventurous play may be more closely related to internalising problems and positive affect for children from low-income families. Surprisingly, we also found an interaction between adventurous play and family income for negative affect, whereby children from a higher income family who spent more time playing adventurously had more negative affect, whereas no significant association was found for children from lower income families. This latter interaction did not hold after controlling for demographic characteristics and should therefore be interpreted with caution.
Overall, the pattern of results is largely consistent with Study 1 and the findings were robust after controlling for a broad range of socio-demographic factors. The results provide some indication of specificity because no significant association was found between adventurous play and externalising problems. Notably, similar patterns for outdoor play and adventurous play were found in Study 2, making it difficult to determine whether adventurous play specifically is driving these findings.