Comparison of matching estimates using KiGGS and GCP.

Note: The results in the first column (KiGGS) correspond to our main set of results based on the KiGGS data but not controlling for state fixed effects. GPC A to C are based on the GCP data. In GPC A we perform a pure replication of the KiGGS results where we use only the second wave of the GCP for both outcome and control variables. GCP B presents the results when we control additionally for the set of lagged outcome variables and replace all control variables by the respective control variables from wave 1. In GPC C we repeat the strategy employed under (B) but restrict the sample to children who do not participate in a sports club in wave 1. The presented effect is the average treatment effect (ATE). p-values are computed by bootstrapping p-values of the t-statistic with 4999 replications.

注：第一列（KiGGS）的结果对应我们基于德国儿童青少年健康调查（KiGGS）数据得到的核心结果集，但未控制地区固定效应。GPC A至C组的结果均基于GCP数据集生成。在GPC A组中，我们严格复刻KiGGS的分析结果，仅使用GCP的第二波数据同时作为结果变量与控制变量。GPC B组则展示了额外加入滞后结果变量集控制，并将所有控制变量替换为第一波数据对应控制变量后的分析结果。GPC C组沿用B组的分析策略，但将样本限定为第一波调查中未参与体育俱乐部的儿童。本次展示的效应为平均治疗效应（ATE）。P值通过对t统计量进行4999次重复抽样的bootstrap法计算得到。