估计概率不能和50%直接比较吧

请问那该如何人工判断划归哪一拨呢？根据拟合的模型，我们应当要对原始数据有所判断啊

Yeah. It is more likely to be an event than being a non-event from your results.
We send mails to customers for asking credit cards application. The response rate is < 5%.  In a logistic model, the best prediction for almost all customers is always a non-response (event =0). On the other hand, we do see some groups are more tended to apply cards than others (OR >1). That is what we had learned from wasting so many junk mails.
JingJu

非常感谢！您样本中，大多数人不愿意申请信用卡（0很多），极少数人愿意申请信用卡（1很少）。拟合logistic模型后，如果我们将该模型适用于您的这个原始样本（而不是另一个群组），会出现什么样的情况呢？应当大多数样本被判定为“愿意申请信用卡（事件1）”的概率低于50%吧？但是我的这个情况更为奇怪，原始样本的1约占70%，0约占30%，然而回归模型判识几乎98%的样本应当发生事件（估计概率大于50%），请问这种情况是否正常？