Removing class means left a 0.50 day-to-day correlation in contact time between classes at one primary school

We tested whether daily contact time with other classes remained correlated after class differences were removed. The source was a two-day SocioPatterns record from one primary school in Lyon, France, in October 2009. The common cohort had 222 pupils in ten classes. The unadjusted Pearson correlation was 0.471. After each class mean was removed, it was 0.499. All ten class-specific correlations were positive. None of 19,999 identity shuffles within class reached the main observation. The one-sided Monte Carlo p value was 0.00005, its minimum at this draw count. Class means contributed 0.99% of summed cross-day products for time between classes, compared with 44.14% for time within class. In a secondary check, 45 pupils selected by first-day time outside class accounted for 37.52% of later endpoint exposure. Random sets with the same class counts had an exact mean of 21.27%. The result concerns one school and one day transition. It does not separate personal ties from attendance or other stable pupil properties. The package preserves code, figures, source provenance, summary results, all null draws, and replay instructions. Restricted source records remain external under their original terms.

contact networks · school contacts · class adjustment · identity persistence · conditional randomization · reproducible analysis