Reporting of paired vote counts is confusing

[adamboche]

Currently we include paired yea/nay votes in the counts on the vote page, while other sources don't include those in their counts, so it can be confusing. We could add a remark to the display that says "including paired votes" when there are some.

Example https://voteview.com/rollcall/RH0400119

[aaronrudkin]

I added code to make paired votes be shaded slightly differently in the vote table and a note on mouseover explaining whose specific vote is paired. However, it's not clear to me the right way to adjust the total to remove paired votes and the president pseudo-vote since the page just directly reads the yea_count and nay_count attributes from the rollcall vote object.

So I think there needs to be at least one upstream change on someone else's end to reflect that and update the fields.

[JeffreyBLewis]

That's great. Thanks. Maybe the text on rollover would be more clearly stated as something like

"Paired vote: [Support|Opposition] was recorded as part of a paired voting arrangement and as such did not count toward the outcome."

I agree that it is probably best to fix the yea and nay totals in the DB (perhaps with new fields that reflect the "counted" totals rather than rebuild the total on the client side. I think this important because folks should be able to search for votes based on the margin reported in the official sources.

All of this suggests that something to add moving forward to the DB would be all of the paired dyads so that the VV would indicate who had paired with whom.

[aaronrudkin]

Yep, totally open to any revision. Was struggling to keep the text concise but explanatory.

I noticed that on one of the Johnson impeachment votes, there were precisely N=3 paired votes, and that seemed surprising to me given the definition -- so we might find that recording the dyads is impossible absent more human intervention.

[JeffreyBLewis]

No doubt. Matching the paired-vote pairings would require human intervention for sure. We should probably have an FAQ that could among other things provide more info regarding paired votes.

[JeffreyBLewis]

A quick check suggests that there are roughly 7,800 roll calls that include more than one pair of paired votes. So matching those up would be a big job.

[JeffreyBLewis]

Also, since the 101st, our data do not include info about pairs. Congress does not include pairing info in the roll call data that they report on pages like:

https://www.senate.gov/legislative/LIS/roll_call_votes/vote1011/vote_101_1_00005.xml

These Senate data do designate some votes as "Present Creating Live Pair", but I think we just code those as "Present" as it is not clear what the member's preference on the vote was or which "Abstain" that "Present" was paired with. Only 160 rollcalls since the 101st include any "present" votes, so it wouldn't be too hard to include Senate pairs since the 101st. However, there are more like 1,500 house roll calls since the 101st that include "Present" votes and there does not appear to be any notation of "live pairs" for house "Present" votes as there is for the Senate (though I am not entirely sure about that, it could just be very rarely used).

[aaronrudkin]

I can't speak to the issues since the 101st, but maybe for the initial 7800 we could quickly subdivide them between:

1. Exactly one pair of paired votes (2 voters, both identified)
2. Some other case, requires human intervention.

Then, for #2 it'd be pretty easy to have the congressional record in one frame and a list of voting members in the other frame, and then this would be basically a human grunt task (undergrad appropriate?)

Might not be worth the effort but I think this could easily be put as a kind of task where I could do the backend stuff in a single sprint if I get a bit of time.

[JeffreyBLewis]

Sounds good. Should have been more clear above. There 7,800 that require human intervention (pre-101st). There are an additional 4,500 roll calls where there is only one pairing and so no human intervention is required.

[JeffreyBLewis]

Also related are older speaker votes where sometimes the codes for paired voting are used in ways that don't mean pairing. This explains why there are a small number of votes on which lots of 2s are found and the number 2s and 5s don't match.