Hi there,
(The variables named here are merely as an example and is not the actual setting of the assignment)
I need to estimate the effect of smoking while pregnant on birthweight (ATT). I will use propensity score matching to help with this.
There is a categorical variable (1,2,3,4) indicating trimester of first prenatal visit (1: no prenatal visit, 2: first prenatal visit in 1st trimester, 3: first prenatal visit in 2nd trimester, 4: first prenatal visit in 3rd trimester).
It is important that when treated units and control units are matched, they have the same value for this trimester variable (e.g. a treated individual with trimester value 3 can only be matched with controlled individuals with trimester value 3 as well). I believe this is the equivalence of exact matching (although I'm not sure). In addition, propensity score matching needs to be applied on a few other variables.
I have to follow the method of a certain study.
First, the propensity scores are estimated over the entire sample.
Then, treated and control units are matched within each trimester value (I need to use epanechnikov kernel matching with bandwidth 0.06 and 3-nearest neighbor matching with replacement and with caliper(0.01)). The article mentions the use of -psmatch2- for this.
Lastly, the ATT is estimated over the entire sample (I believe in a regression, as they add the conditioning variables as controls as well).
Also, only the area of common support must be used. I am struggling to understand when to test for this: after the estimation of the propensity score, or after creating the matched control sample (or both?).
Could anyone help me?
(The variables named here are merely as an example and is not the actual setting of the assignment)
I need to estimate the effect of smoking while pregnant on birthweight (ATT). I will use propensity score matching to help with this.
There is a categorical variable (1,2,3,4) indicating trimester of first prenatal visit (1: no prenatal visit, 2: first prenatal visit in 1st trimester, 3: first prenatal visit in 2nd trimester, 4: first prenatal visit in 3rd trimester).
It is important that when treated units and control units are matched, they have the same value for this trimester variable (e.g. a treated individual with trimester value 3 can only be matched with controlled individuals with trimester value 3 as well). I believe this is the equivalence of exact matching (although I'm not sure). In addition, propensity score matching needs to be applied on a few other variables.
I have to follow the method of a certain study.
First, the propensity scores are estimated over the entire sample.
Then, treated and control units are matched within each trimester value (I need to use epanechnikov kernel matching with bandwidth 0.06 and 3-nearest neighbor matching with replacement and with caliper(0.01)). The article mentions the use of -psmatch2- for this.
Lastly, the ATT is estimated over the entire sample (I believe in a regression, as they add the conditioning variables as controls as well).
Also, only the area of common support must be used. I am struggling to understand when to test for this: after the estimation of the propensity score, or after creating the matched control sample (or both?).
Could anyone help me?
Code:
clear all webuse cattaneo2.dta replace prenatal=prenatal+1 probit mbsmoke mmarried mhisp mrace alcohol mage medu i.prenatal predict pscore, p
