Schools across the United States and the world have been closed in an effort to mitigate the spread of COVID-19. However, the effect of school closure on COVID-19 transmission remains unclear. We estimate the causal effect of changes in the number of weekly visits to schools on COVID-19 transmission using a triple difference approach. In particular, we measure the effect of changes in county-level visits to schools on changes in COVID-19 diagnoses for households with school-age children relative to changes in COVID-19 diagnoses for households without school-age children. We use a data set from the first 46 weeks of 2020 with 130 million household-week level observations that includes COVID-19 diagnoses merged to school visit tracking data from millions of mobile phones. We find that increases in county-level in-person visits to schools lead to an increase in COVID-19 diagnoses among households with children relative to households without school-age children. However, the effects are small in magnitude. A move from the 25th to the 75th percentile of county-level school visits translates to a 0.3 per 10,000 household increase in COVID-19 diagnoses. This change translates to a 3.2 percent relative increase. We find larger differences in low-income counties, in counties with higher COVID-19 prevalence, and at later stages of the COVID-19 pandemic.