请教有关DID模型的

是 否 +2 论坛币

k人 参与回答

经管之家送您一份

应届毕业生专属福利!

求职就业群

赵安豆老师微信：zhaoandou666

经管之家联合CDA

送您一个全额奖学金名额~ !

立即领取

感谢您参与论坛问题回答

经管之家送您两个论坛币！

+2 论坛币

<p>最近想用双重差分估计方法（Difference-in-Differences模型）分析政策实施效果，但看了一些文献仍不得要义（计量基础太差），望高人指点</p><p>
１．运用一般模型分析时，如果因变量是连续性的数据可以用ＯＬＳ估计，但若因变量为定序数据如：１、２、３、４这类，是不是用ordered logistic估计？</p><p>

２．在一般模型中存在的固定效应，这个模型的估计是要用到面板数据分析方法还是只要将各变量进行一阶差分后再回归即可？</p><p>

３．对于实验的自选择问题或者内生性，只有通过ＰＳＭ方法才能解决吗？看到有文献说用这种方法要求两组数据的数量相当，但是我的数据相差比较大怎么办？</p><p>

４．在Ｓｔａｔａ中使用ＰＳＭ和ＤＩＤ方法是两个步骤？需要逐步运算还是有命令一步到位？，希望高人指教，不胜感激</p><p>ＱＱ：５４８７８４０７，Ｅ－ｍａｉｌ：ｊｓｓｚｍａｙａ＠１６３．ｃｏｍ</p><p></p>

扫码加我 拉你入群

请注明：姓名-公司-职位

以便审核进群资格，未注明则拒绝

分享0 收藏5 回帖

关键词：DID模型 DID differences difference logistic 因变量 模型

1. yes

2. you still need to control forr fixed effect after first level differences, if you assume the treatment effect varies with time or group..

3. PSM assumes selection-on-observables while DD is for unobservables. Propensity score methods should be able to control the difference b/t treatment and control groups using all the variables that you have, and then you can reduce the dimensions to do whatever you wanna do, like matching, regression,etc; DD assume the treatment and control groups are very similar though you have unobservables that might bias your estimates. By differencing all the unobervables, DD is supposed to give you unbiased estimates. Read Almond and Chay's the cost of low birth weight, you will be able to know PSM a bit. For DD, read Card and Ashenfelter's (I forgot the title, but it was one of the labor econ lit)

4. pscore is a command in Stata, but for DD, you need to construct variables by yourself.

hope that helps

I have a doubt about the first level differences...Could you kindly explain why first differences is needed? As far as I know, you can run DD estimation either by including a full set of dummies for fixed effects or de-meaning with respect to fixed effects.

first level difference is just a conceptual idea. typically we identify treatment group and treatment dummies to run the regression. however, as you might read from some papers, sometimes they difference out the unobservables (simply by substracting the first equation from the second one).  they are essentially the same thing, but it depends on your data, first level difference is good for twin-kind of data.

dummies for fixed effect are used to control for selection-on-observables. After adjust for unobservables (DD) , you still need to control for observables, that's why you put in fixed effect.