Fostering pupils’ critical health literacy: examining the potential of physical education in lower secondary school

Background In Norway, the introduction of an interdisciplinary subject named Public Health and Life skills has brought about renewed attention to how health is conceptualized and taught within and across school subjects. Physical education (PE) is one subject that has traditionally been linked to health outcomes. However, a narrow focus on increased physical activity as the main outcome of PE could be counterproductive in the pursuit of health. Critical health literacy (CHL) is put forward as a resource for health that can be nurtured in the PE context; this study hypothesizes that academic achievement in PE is positively associated with some aspects of CHL. Methods This cross-sectional study included 521 pupils aged 13–15 years old from five lower secondary schools in Norway. Structural equation models were used as the primary statistical analysis to test the hypothesis. The study controlled for parents’ education, leisure physical activity, and participation in sports club activities. Results The results confirm the hypothesis, showing a positive and significant association between PE and CHL. The association remains when controlling for parents’ education, leisure physical activity, and participation in sports club activities (β^PE→CHL−C1 = 0.264, p = 0.001; β^PE→CHL−C2 = 0.351, p < 0.000). Conclusion In our sample, academic achievement in PE was associated with higher levels of CHL. This study contributes to the ongoing discussion on the health benefits of PE. We argue that a resource-based health perspective can produce the appropriate aims for health in PE contexts and that the CHL concept contributes to illuminating key areas, promoting suitable teaching strategies, and bringing balance between an individual and collective focus for future health education, both within PE and across different subjects in school contexts.


Supplementary Data
In this document we provide supplementary data that increase the transparency of the data analysis process in this research. In the first section we show model syntax and commands that were used in the SEM analysis. Short explanations are given. In the second section we provide results that we could not include in the paper for readability considerations. Items from the CHLQ-A scales with descriptive statistics is displayed along with tables of residuals from all steps in the analysis. We also present two different results from the final model, on in which we specify a lower factor loading (λ = .70) for the indicator of PE than what was specified in the model presented in the paper, and one where the loading is higher (λ = 0.90).

Model syntax and commands
As reported in the paper we estimated the measurement model with covariance between latent variables. We used the following model syntax ('lavaan' automatically estimates covariance in multiple factor models if nothing else is specified). In the subsequent steps, with regressions, covariance between latent factors must be specified. mod <-" CHLC1 =~ chl36 + chl37 + chl38 CHLC2 =~ chl39 + chl40 + chl41 PE =~ 0.8*kro" Step 1 mod1 <-" # measurement model Step 2

Commands for model estimation
In the command we used to estimate the model we specified type of estimator (e.g. ULSMV) and that the data should be treated as ordinal. The std.lv command tells lavaan to restrict the variance of latent constructs to 1 instead of the factor loading of the first indicator of each factor (which is the default scaling). In this case we specified the factor loading of the indicator of the PE construct, and therefore had to standardize the left side of the equations (the variance of the latent construct).

Supplementary Tables
In this section we present results that were not included in the paper. In the first section we present the items of CHLA-Q that were used with descriptive statistics. In the next section we show the residuals from the four steps of estimation in the SEM analysis. Lastly, we display the results from the final model with different factor loadings for PE.

Items and item statistics
The items from the CHLA-Q instrument that were used in this study with mean, kurtosis, variance, and skew.

Residuals
Four tables that show the standardized residuals from each of the main steps of the analysis.

Results from final model with different factor loadings for the latent construct PE
In this section we present the main results of models with slightly lower and higher factor loadings specified for the single-indicator latent construct of PE. We see that although there are some changes the overall results point in the same direction. If we reduce the factor loading (table S6) for the indicator of PE (λ = 0.7) the associations become slightly stronger, and the model explains more of the variance in the dependent variables. When the factor loading is higher (λ = 0.9) the pattern is opposite (table S7). ---------Standardized and unstandardized regression coefficients with standard errors (SE), p-value. R-squared are given for dependent variables. ---------Standardized and unstandardized regression coefficients with standard errors (SE), p-value. R-squared are given for dependent variables.