Election Calculator / Simulator 2.0 released (Updated 5/5/2023)
Virginiá:
I fixed the PVI display issue for the congressional districts. Are the numbers correct? I thought those were the numbers at the time I made that feature, but now they appear to be different on wikipedia. What are the correct numbers for 2016-2020?
I'll have to look into why it's not sorting the CDs correctly though.. I couldn't find an immediately obvious cause for that.
cherry mandarin:
There's a few different factors at play here, each of which help explain part of the discrepancy between the numbers in your files:
1) Approximately a year and a half ago, the Cook Political Report updated its formula for calculating PVI figures, at both the state and congressional-district levels.
In the past, they had previously used a simple average of the two-party partisan lean of each state or district over the prior two presidential elections. They maintained this method through the 2021 release of their initial post-2020 PVI dataset, which included the 2016/2020 PVI figures that you use in the current iteration of the EV calculator on this website.
Starting in the summer of 2022, however, they tweaked their PVI formula to heavily favour the most recent presidential election—to be precise, it receives a weight of 75% in their PVI calculations, as opposed to merely 25% for the election before that.
I suspect this is due to the increasing nationalization and polarization of American politics, even at the federal level. Home-state advantages or "favourite-son" boosts no longer count for as much as they used to, making each state's ultimate result far more predictable than in bygone days. Furthermore, trends and shifts in demographic patterns have become far likelier to carry through from one presidential election into the next, rendering it that much more crucial that Cook Political Report's PVI figures remain ahead of the curve, so to speak, rather than lagging behind tomorrow's forecasted electoral realities.
Without any further ado, here's the announcement that the publication released to explain their decision to alter their approach to calculating PVI numbers moving forward:
Quote from: Dave Wasserman
Please note that the formula has been tweaked since we last released the state PVI scores in 2021. With the 2022 PVI release, we've made a slight change to how we calculate PVI scores: instead of going with a 50/50 mix of the two most recent presidential elections to assess partisanship as we've done in the past, we're switching to a 75/25 weighting in favor of the more recent presidential cycle. For the 2022 dataset, that means that the 2020 result in each state or district is weighted three times as heavily as the 2016 result.
As far as I am aware, however, Cook are the only outfit to have adopted this new 25-75 split. All other major media outlets releasing their own PVI lists either continue to employ a simple 1:1 ratio between the prior two presidential elections, or use exclusively the single most recent recent one in making their calculations.
2) Adding to the confusion even further, Cook has always calculated its statewide and congressional PVI figures based on the two-party share of both the national popular vote (for partisan-lean purposes) and each state or district's individual results. On the other hand, most other relevant publications take the whole picture into account by incorporating all votes cast for third-party candidates into their calculations, rather than discarding them altogether. Ironically, this group of analysts includes the longtime authors of the American Political Almanac, which Charlie Cook himself used to co-write and co-edit.
This difference becomes particularly impactful in cases such as that of Utah 2016, where third-party and independent support reached abnormally high levels. In such instances, Cook's PVI formula can "artificially" pad the leading candidate's edge in a state by a handful of points, at least when compared to the all-encompassing approach used by most other sources. However, the gap usually becomes significantly more muted in the district-level PVI numbers, albeit with a few notable exceptions over the years nonetheless.
3) Re-districting: at the presidential level, this obviously only affects the 5 congressional districts in the 2 states that award electoral votes by congressional district. The old PVI numbers were calculated on the 2010s lines, whereas you probably want the PVI numbers under the new boundaries, at least for the 2024 base prediction maps.
The issue here is, there does not seem to be any reliable source for 2016 presidential-level results using the 2020s congressional district boundaries. As always, Daily Kos has helpfully compiled the results of the 2020 presidential race under both sets of boundaries, but their 2016 dataset only seems to include the old ones.
Given that 2016 results are required for calculating a state or district's 2016-2020 PVI, regardless of whether you use Cook's old or new formulas, it is actually impossible to determine the correct numbers for these 5 congressional districts (unless someone's able to find or calculate the 2016 presidential results under their new lines). This also explains why I have left the 2016-to-2020 swing and trend columns empty for those 5 districts in the spreadsheet attached below.
Since we know the 2016-to-2020 swing of these 5 districts under their old boundaries, I have assumed the same 2016-to-2020 swing for the 2020s version of each district while performing my 2016-to-2020 PVI calculations below. This is because, in spite of the decennial redistricting that took place a few years back, all 5 districts have kept the majority of their previous voters from their respective 2010s iterations—in other words, this process did not result in any of them undergoing fundamental overhauls of the territory they represent.
Quote from: appropriate on January 01, 2024, 11:07:49 AM
Are the numbers correct? I thought those were the numbers at the time I made that feature, but now they appear to be different, according to Wikipedia. What should the numbers for 2016 and 2020 be?
For clarity's sake, the EVC on this website goes by Cook's official 2021 dataset, which uses their old formula (50/50) and the two-party vote share in each state. Additionally, the PVI figures for the 5 congressional districts in question follow their old (2010s) boundaries.
Of course, this is to be expected, since you retrieved these numbers at some point in 2021, when you first released the new Atlas EVC tool. Back then, the congressional-district boundaries for the 2022 midterm elections had yet to be finalized by each state, so the dataset that Cook released in 2021 still followed the previous decade's lines.
In any case, feel free to ask me for the figures, and I'll be happy to provide them for you, depending on the method requested (old "50-50" or new "25-75" formula; using Cook's two-party vote-share model or others' all-candidate totals). As for Maine's and Nebraska's congressional districts, I assume you'd prefer their new (2020s) versions rather than the old ones, at least for the main system's 2024 maps. I can certainly also send you the corresponding figures under their 2010s lines as well, however.
BigVic:
Need one for past Senate elections from 2018
President of the great nation of 🏳️⚧️:
Honestly, I totally get why you can't save a custom series of maps, but it would be so cool if you could. (and I agree with BigVic that pre-2016 Sen/Gov maps would be nice--up to 1990 should suffice, since that's as far back as Atlas proper goes--but again, you don't need to do this if you can't/won't.)
Navigation
[0] Message Index
[*] Previous page