General Discussion/Conversation

[AaryaDesai1]

Write all ideas and to-dos here.

[AaryaDesai1]

Here are the columns that I think we should include of opioids:

1. BUYER_STATE
2. BUYER_ZIP
3. BUYER_COUNTY
4. DRUG_NAME (in case we want to do extra EDA and other analysis to see the differences or weights).
5. TRANSACTION_DATE
6. CALC_BASE_WT_IN_GM
7. DOSAGE_UNIT
8. MME
9. MME_Conversion_Factor

Please confirm or suggest if this seems okay or if we need to add anything.

[osama-shawir]

Looks good, we can parse the transaction_date into date type variable and continue to filter out datapoints that are outside our timespan of interest.

[jeremymtan]

That's fine with me. We just need to make sure Texas has transaction_dates by month, identify control states to use with Texas, and have those states' transaction_dates by month

[AaryaDesai1]

That's fine with me. We just need to make sure Texas has transaction_dates by month, identify control states to use with Texas, and have those states' transaction_dates by month

Good call.

For the year range, I think we should take a good range which would be a few years before the earliest policy implement and then a few years after the latest implementation (irrespective of the states). This comes up to around 2002 to 2017-18.

Also I needed to ask how you would like to pick the control states. There are a few approaches we could take:

1. Geographical location. For this, we could choose Cali and Oregon for Washington, Arizona and Colorado for Texas and Georgia and NC for Florida.
2. Demographics. For this, we would have to look into the census data for each state and then compare what states look like the best competitors for the states of interest.
3. Trends. We could look at the opioid consumption and mortality trends for up to the year of policy implementation and then choose the states that are the most similar to the states we've chosen (this is based on the graph in Nick's pdf).
4. Comparison of policy implementation. This will take some time to research, but we could look into what states did NOT implement opioid control policies and then use them as controls. This would also need the above implementation.

[AaryaDesai1]

Hey guys, Im working on the mortality dataset right now and it's a combination of alcohol/drug induced deaths. The project PDF summary doesn't say anything about removing alcohol-related deaths and only maintaining drug-induced deaths but I think we should subset it to only the relevant data (i.e., drug induced deaths). What do you guys think?

[jeremymtan]

Naive approach of finding # of year based on parquet file doesn't work as each parquet file of each state has years from 2006-2017. We need to do another approach, which I believe might work:

• figure out how we calculate opiod shipments
• calculate the population for each county for that year (need Osama's data for this so need to merge)
• then calculate opiod shipments per capita
• find states with similar slopes before the policy implementation and those can be our control states

[AaryaDesai1]

So we would have to do the slope calculation on each parquet file (and by extension, we'd have to do 50 merges)?

[jeremymtan]

Unless I can find a better way to do this, there would be 50 merges but the datasets are small enough to find them quickly.

[AaryaDesai1]

Yeah, sounds good. Im working on the Florida merge rn. So we could just iterate that code over the rest of the states. Im just smoothing it out right now cuz there are a few problems with it :))

[AaryaDesai1]

By the way, this is a study I found about policy implementation (along with a timeline of when states implemented policies), in case we just wanna choose control states based on literature and not based on metrics:
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9914561/

[osama-shawir]

Florida merge completed, we can replicate the process for other treatment states and proceed