Due to a sparse matrix in 2010/11 it was necessary to estimate the cross-classified model in R (R Development Core Team, 2011) using lme4 (Bates et al., 2011) and then transfer the results back into Stata. The sample characteristics

and the results of the cross-classified models fitted to calculate each school’s expected mean BMI-SDS are shown in Table 1. Only a small proportion of the variation in pupil BMI-SDS was attributed to either the school or the neighbourhood in the SAHA HDAC cell line null models (intraclass correlation coefficients < 0.03). There was a significant association between socioeconomic status and BMI-SDS, with the regression coefficient for the Index of Multiple Deprivation calculated to show the mean difference in BMI-SDS between the most and least deprived LSOAs in England, based upon the trend in Devon. A subsample comprising 10 schools, approximately equally distributed across the 2006/07 Observed ranking, were selected in order that the change of rankings in some individual (anonymised) schools could be observed (Table 2). The data presented in Table 2 clearly

Gefitinib in vitro demonstrate that whilst within each year the Observed and ‘Expected’ rankings of schools are similar, the ‘Value-added’ rankings are considerably different. Furthermore, across the five years there was substantial movement in school position in each of the three rankings. The levels of agreement (concordance (ρc values)) between each of the three rankings within each year are presented in Table 3. These values confirm the observations from Table 2: within each year the agreement between the Observed and ‘Expected’ rankings were high (ρc ~ 0.9), whereas the concordances with the ‘Value-added’ rankings are much lower (ρc < 0.3). The equivalent Pearson's correlation Cell press coefficients are reported in Table S1 and the caterpillar plots in Fig. S1 of the supplementary material, which further confirm the above findings. The results of the

analyses testing how stable the rankings were across the five years are presented in Table 4. These show that within each individual ranking (Observed, ‘Expected’ and ‘Value-added’) the concordance values were small (ρc < 0.25), demonstrating that across the years the rankings varied considerably; notably, the level of agreement across the ‘Value-added’ rankings was even smaller (ρc < 0.1). These results demonstrate the lack of consistency in any of the rankings across the five years. The equivalent Pearson’s correlation coefficients are reported in Table S2 and caterpillar plots in Fig. S2; further supporting the findings presented in Table 4. The kappa values, which show the extent to which schools maintained approximately the same rankings across the five years were, 0.06 (p < 0.0001), 0.06 (p < 0.0001) and 0.05 (p < 0.0001) for the Observed, ‘Expected’ and ‘Value-added’ rankings respectively. Similar to Procter et al.