Conditional Parallel Trend Assumption with Repeated Cross-Sectional Difference in Difference Data

[Tara Gupta]

Hi everyone,

I am a student and i just joined this forum so I apologise in advance if this question is a little dim.
I am trying to estimate corruption in the healthcare sector during COVID. My data (as given below) has the following variables :
Deceased : If an individual died because of COVID
i_COVID19 : Whether an individual contracted COVID
respiratory : If an individual was hospitalized for a respiratory disease after 2016
i_Month and i_Year give the month and year of hospitalization for i_COVID19 and respiratory
i_costhospitalbed gives the cost of a hospital bed for i_COVID19 and respiratory
ICUBed is a dummy for whether an individual required an ICU bed or just an isolation bed.
i_HospitalSector is 1 if treatment was recieved in the Public sector and 2 if treatment was recieved in the Private sector.
Bed_Bribery_overpay & Bed_Clout_Connections are dummies for how the individual had a hospital bed allocated to them - Through personal connections or through bribery.

I tried to do a difference in difference analysis to compare the cost between other respiratory diseases and COVID. Mostly Corruption took place during the 2nd Wave of COVID, so I generated a variable period which is 1 for Pre-COVID, 2 for the 1st wave in 2020 and 3 for the 2nd wave in 2021. I also generated covorresp which is a dummy for whether an individual was hospitalized for COVID or a respiratory disease.

The goal is to estimate the impact of corruption duting the 2nd Wave. For that,
Firstly, I would like to compare the costs during respiratory diseases and COVID as it will help me remove any pre-existing corruption.
Secondly, comparing with the 2nd wave with the 1st wave will help remove any changes in the price due to changes in the demand due to the spike in COVID cases. This makes for a very complicated model and i am not sure how to progress.
Lastly, I also think the parallel trend assumption and SUTVA would be violated as due to COVID, the cost of all medical services increased. I am trying to use a conditional parallel trend assumption that controls for the trend in 2020 and 2021 due to COVID.

I have been stuck on this for a while now. Kindly help me with this. Or if you have an other recommendations or suggestions for me, I am all ears.

Here is my data:

input    byte(Deceased i_COVID19 respiratory    i_Month)    int    i_Year    long    i_costhospitalbed    byte(ICUBed    i_HospitalSector    Bed_Bribery_overpay    Bed_Clout_Connections)    float(period    DID    covorresp)
0 0 .    1 2016  10000 . 2 . . 1 0 .
0 0 .    1 2017   4500 . 2 . . 1 0 .
0 0 .    1 2018   6000 . 2 . . 1 0 .
0 0 .    1 2019  10000 . 2 . . 1 0 .
1 0 1    2 2018   5000 . 2 . . 1 0 0
0 0 .    2 2018   5000 . 2 . . 1 0 .
0 0 .    2 2018   7000 . 2 . . 1 0 .
1 0 1    3 2019   5000 . 2 . . 1 0 0
1 0 .    3 2016   6500 0 2 . . 1 0 .
0 0 .    3 2017  30000 . 2 . . 1 0 .
0 0 .    3 2018  15000 . 2 . . 1 0 .
1 0 1    4 2018  10000 0 2 . . 1 0 0
0 0 .    4 2018   3000 . 2 . . 1 0 .
1 0 .    4 2019   4100 . 2 . . 1 0 .
1 0 .    5 2017   6000 0 2 . . 1 0 .
1 0 .    5 2017   5000 0 2 . . 1 0 .
1 0 .    5 2018   2500 . 1 . . 1 0 .
1 0 1    6 2019   8000 0 2 . . 1 0 0
1 0 1    6 2019   6000 . 2 . . 1 0 0
1 0 1    6 2019  80000 . 2 . . 1 0 0
0 0 1    6 2018   2000 . 1 . . 1 0 0
1 0 1    6 2018   3000 . 2 . . 1 0 0
0 0 1    6 2017   6000 . 2 . . 1 0 0
1 0 .    6 2019   1500 . 2 . . 1 0 .
0 0 .    6 2018   5000 . 2 . . 1 0 .
1 0 .    6 2018   4000 . 2 . . 1 0 .
1 0 .    6 2019   6000 . 2 . . 1 0 .
0 0 .    6 2019  12000 . 2 . . 1 0 .
0 0 .    6 2017   3000 . 2 . . 1 0 .
1 0 1    7 2018   7500 0 2 . . 1 0 0
1 0 1    7 2017   2500 . 2 . . 1 0 0
1 0 1    7 2018   6000 . 2 . . 1 0 0
1 0 1    7 2016   7000 0 2 . . 1 0 0
0 0 .    7 2019 100000 . 2 . . 1 0 .
1 0 .    7 2018   6500 0 2 . . 1 0 .
1 0 .    7 2017   4000 . 2 . . 1 0 .
0 0 .    7 2018   4000 . 2 . . 1 0 .
1 0 1    8 2018 120000 . 2 . . 1 0 0
1 0 1    8 2019 120000 . 2 . . 1 0 0
1 0 1    8 2017   8500 0 2 . . 1 0 0
1 0 .    8 2019   5000 . 2 . . 1 0 .
0 0 .    8 2019   5000 . 2 . . 1 0 .
1 0 .    8 2019  10000 . 2 . . 1 0 .
1 0 .    8 2018   5000 0 2 . . 1 0 .
1 0 1    9 2016   1500 . 2 . . 1 0 0
1 0 1    9 2019 100000 . 2 . . 1 0 0
0 0 .    9 2019 200000 . 2 . . 1 0 .
0 0 .    9 2017  15000 . 2 . . 1 0 .
0 0 .    10 2017  10000 . 2 . . 1 0 .
0 0 .    10 2016   9000 . 2 . . 1 0 .
1 0 .    10 2018   5500 0 2 . . 1 0 .
1 0 1    11 2016   3000 . 2 . . 1 0 0
1 0 1    11 2016   8000 0 2 . . 1 0 0
1 1 .    11 2019      . 0 . 0 0 2 0 1
1 1 .    11 2019      . 0 . 0 0 2 0 1
1 0 .    11 2017   6000 0 2 . . 1 0 .
0 0 .    11 2019   3000 . 2 . . 1 0 .
0 0 .    11 2019  12000 . 2 . . 1 0 .
0 0 .    11 2019   2000 . 2 . . 1 0 .
0 0 .    11 2017   1500 . 2 . . 1 0 .
1 0 1    12 2018  20000 . 2 . . 1 0 0
1 0 1    12 2018   7500 0 2 . . 1 0 0
1 1 .    12 2019      . 0 2 0 0 2 0 1
1 0 .    12 2019   2500 0 2 . . 1 0 .
1 0 .    12 2016   5500 0 2 . . 1 0 .
1 0 .    12 2018   5000 . 2 . . 1 0 .
1 0 .    12 2016   3000 . 2 . . 1 0 .
1 0 .    12 2017   5500 . 2 . . 1 0 .
1 1 .    1 2020  35000 1 2 0 0 2 0 1
1 1 .    1 2020      . 0 2 0 0 2 0 1
1 1 .    1 2020      . 0 1 0 0 2 0 1
1 1 .    1 2020      . 0 2 0 0 2 0 1
0 0 .    1 2020  50000 . 2 . . 2 0 .
1 1 .    2 2020      . 1 . 0 0 2 0 1
1 1 .    2 2020      . 0 2 0 0 2 0 1
1 1 .    2 2020      . 0 2 0 1 2 0 1
1 1 .    2 2020  20000 0 2 0 1 2 0 1
0 1 .    2 2020   9000 0 2 0 0 2 0 1
1 1 .    2 2020      . 0 2 1 0 2 0 1
1 1 .    2 2020      . 1 . 0 0 2 0 1
1 0 1    3 2020   4000 . 2 . . 2 0 0
1 1 .    3 2020      . 1 2 0 0 2 0 1
1 1 .    3 2020   5000 1 2 0 0 2 0 1
1 1 .    3 2020      . 1 2 1 0 2 0 1
1 1 .    3 2020      . 0 2 0 0 2 0 1
1 1 .    3 2020      . 0 2 0 0 2 0 1
0 1 .    3 2020      . 0 0 0 0 2 0 1
1 1 .    3 2020      . 0 2 1 0 2 0 1
1 1 .    3 2020      . 0 2 0 0 2 0 1
0 1 .    3 2020      . 0 0 0 0 2 0 1
1 1 .    3 2020      . 0 2 1 0 2 0 1
1 0 .    3 2020  20000 . 2 . . 2 0 .
1 0 1    4 2020   5000 . 2 . . 2 0 0
1 0 1    4 2020 140000 . 2 . . 2 0 0
0 1 .    4 2020  11000 1 2 0 0 2 0 1
0 1 .    4 2020      . 0 0 0 0 2 0 1
1 1 .    4 2020      . 0 . 0 0 2 0 1
1 1 .    4 2020      . 0 2 0 0 2 0 1
1 1 .    4 2020      . 1 2 0 0 2 0 1
1 1 .    4 2020      . 1 2 0 1 2 0 1

I tried using the following code so far:

// treatment - COVID, control - other respiratory diseases
gen covorresp = .
replace covorresp = 1 if i_COVID19 == 1
replace covorresp = 0 if respiratory == 1

//across 3 time periods - pre covid, wave 1 and wave 2
gen period = .
replace period = 1 if i_Year == 2016
replace period = 1 if i_Year == 2017
replace period = 1 if i_Year == 2018
replace period = 1 if i_Year == 2019 & i_COVID19 == 0
replace period = 2 if i_Year == 2019 & i_COVID19 == 1
replace period = 2 if i_Year == 2020
replace period = 3 if i_Year == 2021
replace period = 3 if i_Year == 2022

gen DID = 0
replace DID = 1 if period == 3 & covorresp == 1

reg i_costhospitalbed covorresp i.period DID, robust

Thankyou so much for your time.

I really appreciate it.