Aim: I am trying to see if there is an effect of a 2011 national policy on the response to the question: "Should FGM/C discontinue?" using Difference-in-Differences between the years 1998, 2008, 2014, and 2022.
Note: Treatment versus Control is unusual because it is a national policy and we are looking only at ethnicites in the nation. So technically all respondents are treated under the policy. BUT we consider ethnicities with high prevalence(above average) of FGM/C as those who receive treatment (effected from the policy) while ethnicites with low prevalence (below averge) of FGM/C are likely to be unaffected by policy because FGM/C is already low. Simply put, If ethnicity is a high intensity for prevalence of FGM/C then that individual is part of the treated group.
Variables:
Intensity=1 if Treated... "High intensity of FGM/C Prevalence."
Discontinue= 1 if responded "Discontinue FGM/C
Y98=1 if Year== 1998
Y08=1 if Year==2008
Y14=1 if Year==2014
Y22=1 if Year==2022
Age= Age
Educ_Level= level completed (primary, seconday, or tertiary)
Religoin= Religion
Region= Region
*Policy occurs in 2011
Challenge:
The outcome variable is binary and I would like to keep the analysis at the individual level, but am unsure how to do it with a binary outcome while also controlling for age, religion, region, and education level. I cannot categorize the data as panel data becasue I do not have repeating observations (different households were interviewed each year)
Option 1: I tried using the % of people who responded "Discontiue" per ethnicity and ran the parallel assumptions trend test, but I don't know how to control for age, religion, etc and observations are limited to 11 groups each year. Could this work at the group level, or would there be too many limitations? I can categorize it as panel data and it would likely be the simplest for me, but I am unsure if 11 observations a year is too little and the lack of controlling for variables would make it not substantial evidence.
Option 2: Is there a way to generate a new variable using logit coefficients/odds ratio for each year as the probability that a respondent will respond "Discontinue" based on their age, religion, education level, and region. Then run a did with that new variable as the outcome? This way I have a continuous outcome variable and am controlling for individual caharacteristics. Is this possible or do I not understand logit models and it is not possible to run an odds ratio for each observation?
Option 3: Using the binary outcome variable, Discontinue, I have run the following for years 2008 to 2014. However, there is no control variables. Are the below findings useful? If I kept the reg the way it is, how can I add control variables?
I guess I am trying to figure out if it is possible to gen a new variable that would not be binary and would control for individual caharacteristics. Let me know if I sound like I have no idea what I am talking about! I am somewhat new to econometrics and working on a final paper. Any help is much appreciated. Thank you!!
Note: Treatment versus Control is unusual because it is a national policy and we are looking only at ethnicites in the nation. So technically all respondents are treated under the policy. BUT we consider ethnicities with high prevalence(above average) of FGM/C as those who receive treatment (effected from the policy) while ethnicites with low prevalence (below averge) of FGM/C are likely to be unaffected by policy because FGM/C is already low. Simply put, If ethnicity is a high intensity for prevalence of FGM/C then that individual is part of the treated group.
Variables:
Intensity=1 if Treated... "High intensity of FGM/C Prevalence."
Discontinue= 1 if responded "Discontinue FGM/C
Y98=1 if Year== 1998
Y08=1 if Year==2008
Y14=1 if Year==2014
Y22=1 if Year==2022
Age= Age
Educ_Level= level completed (primary, seconday, or tertiary)
Religoin= Religion
Region= Region
*Policy occurs in 2011
Challenge:
The outcome variable is binary and I would like to keep the analysis at the individual level, but am unsure how to do it with a binary outcome while also controlling for age, religion, region, and education level. I cannot categorize the data as panel data becasue I do not have repeating observations (different households were interviewed each year)
Option 1: I tried using the % of people who responded "Discontiue" per ethnicity and ran the parallel assumptions trend test, but I don't know how to control for age, religion, etc and observations are limited to 11 groups each year. Could this work at the group level, or would there be too many limitations? I can categorize it as panel data and it would likely be the simplest for me, but I am unsure if 11 observations a year is too little and the lack of controlling for variables would make it not substantial evidence.
Option 2: Is there a way to generate a new variable using logit coefficients/odds ratio for each year as the probability that a respondent will respond "Discontinue" based on their age, religion, education level, and region. Then run a did with that new variable as the outcome? This way I have a continuous outcome variable and am controlling for individual caharacteristics. Is this possible or do I not understand logit models and it is not possible to run an odds ratio for each observation?
Option 3: Using the binary outcome variable, Discontinue, I have run the following for years 2008 to 2014. However, there is no control variables. Are the below findings useful? If I kept the reg the way it is, how can I add control variables?
HTML Code:
reg Discontinue didY14 Intensityy Y14 if Y14==1Y08==1, robust |
||||||||||||||||||||||||||||||||||||||||||||||||
|
