Data is the new gold. In education, this debate focusses specifically on test scores; whether used for evaluating schools or teachers, assessing social mobility or investigating children’s skill development. However, the perfect data do not exist. The value of any data analysis depends crucially on how the data were collected, which individuals are covered and how the data were treated. This thesis investigates how to use the potential, while avoiding the pitfalls of rich education data. Four separate studies assess the consequences of selective data collection and selective testing, non-response bias, measurement error and the use of aggregated test scores.