Comments on draft - Githubissues

nickeubank commented 4 years ago

A very nice draft, and very clean diff-in-diff results! A couple thoughts:

As we've discussed, showing that turnout rose is important, but also not super surprising. Descriptively, the difference between caususes and primaries is so large across so many different states and elections, it's hard to imagine there isn't a causal. So it's great to see the result, but I think you should continue forward by...
Showing me changes in voter composition! I think that's gonna be the most interesting thing. For these same states, how did the composition of the primary electorate change from 2016 to 2020?! Because in some sense, it's not clear we even care about turnout in and of itself (some people might, but others might not) -- what seems to clearly be important is whether the types of people voting (and thus the people who are likely to be elected) have changed!

I know you're also working at a single-state voter registry study, but first I'd do the state-level primary-voter composition analysis!

katevcoulter commented 4 years ago

Hi @nickeubank ! We are having some issues finding data, but I wanted to follow up with what we do have access to and see your thoughts on it. We cannot find any voter composition data for the states we have in our treatment group. However, we do have for Maine county level turnout for 2016 and 2020 in terms of just numbers (not by demographic, just a lump sum). We also have county level composition not by turnout but just in general. We could infer using the latter what the former could have looked like by composition in some way, correct? Or is there a better option? Let us know!

nickeubank commented 4 years ago

hmmmm... proving harder to find than I'd thought it would be. Let me ask a colleague...

The Maine strategy is solid.

katevcoulter commented 4 years ago

@nickeubank hi Nick! We are pulling data now and noticed a hiccup. It appears that Maine had an open caucus in 2016 for both Dem and Repub, skewing our numbers a ton. I'm wondering if it was the case that because people could literally show up and decide which caucus to participate in (it appears a huge number of whom decided to participate in the republican caucus) we cannot use these numbers. We are getting our numbers from two different websites (politico and our initial data source) and it seems that it is either the case that our initial data source is wrong or politico is counting our numbers so wrong that we get a .2% turnout rate from politico while our initial gives us around 4%. We aren't sure if we should continue with this huge discrepancy or what you would advise us to do otherwise. Any advice?

nickeubank commented 4 years ago

I'm just hearing "try exit polls". Roper has 2016 data: https://ropercenter.cornell.edu/exit-polls/state-primary-exit-polls But not sure I can find 2020. May have to stick to an in-state data.

0.2% is clearly insanely low.

From the voter rolls, can't you just see what voters voted?

katevcoulter commented 4 years ago

Roper doesn't have any of our 4 states :(

nickeubank commented 4 years ago

damn.

Do voter rolls have turnout?

katevcoulter commented 4 years ago

you mean voter registration numbers? i’m looking at maine.gov http://maine.gov/ currently and it seems no

On Apr 23, 2020, at 5:23 PM, Nick Eubank notifications@github.com wrote:

damn.

Do voter rolls have turnout?

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/MIDS-at-Duke/unifying-data-science-final-project-primaries/issues/15#issuecomment-618679362, or unsubscribe https://github.com/notifications/unsubscribe-auth/AKLBR556SNRNGCBDQ35G7NDROCWT7ANCNFSM4MNNG3YQ.

nickeubank commented 4 years ago

State registrars keep track of who votes in every election and the data is public. Often times if you can find a voter roll – the individual level data on voters – they will include columns on whether people have voted in certain elections.

nickeubank commented 4 years ago

So if:

You can't find aggregate turnout by race, and
You can't find individual level turnout data,
then I'd hunt for county-level primary turnout for any of your states and use the trick you were gonna use anyway to do some race imputations by county.

nickeubank commented 4 years ago

the far right column here has data sources listed: http://www.electproject.org/2020p

aberman6 commented 4 years ago

Hi Nick - Serious thank you for looking into this for us. For some inexplicable reason, we can't seem to find caucus turnout for Maine in 2016, only delegate votes by county. I don't know where electproject.org got their data and it's not listed in the notes column at the url you linked us.

After much data searching, I think our best bet is to pivot to Colorado. I was not able to find voter roll, but I was able to find 2016 and 2020 election turnout. If you're curious, the merged dataset is in 20_intermediate_files/turnout_colorado.csv. Once we add demographic data in we can do the same type of county-level analysis we were planning for Maine.

MIDS-at-Duke / unifying-data-science-final-project-primaries

Comments on draft #15