请问现在比较OLS和heckit的回归结果treatment effect

数据如下



对OLS回归
. reg reg re78 treat age education black hispanic married nodegree re75

处理效应是806

对heckit是

replace re78= . if re78==0
heckman re78 treat age education black hispanic married nodegree re75 , select (treat age education black hispanic married nodegree re75 ) twostep

处理效应是-33

为啥结果差这么多？OLS不是一般会低估处理效应吗，为啥还比heckit高？

如能帮助，多谢多谢鞠躬谢！