What precisely is the Battleground Tracker, and the place do the numbers come from? Here are 5 issues to know.
1. Going state by state to grasp this election
We take a state-by-state method to describing the race and measuring public opinion, for the reason that presidency is gained within the Electoral College, not by the nationwide widespread vote. Indeed, relying an excessive amount of on nationwide polling might be deceptive, as 2016 reminded us.
The Battleground Tracker seems to be at every particular person state, specializing in essentially the most aggressive ones. And we translate every candidate’s present help to the electoral vote scoreboard. Our state-by-state method additionally offers a way of what voters in several components of the nation suppose and really feel about this 12 months’s candidates, nationwide points and native issues.
2. More than only a ballot
While surveying voters throughout the nation is an integral a part of the Battleground Tracker, that is greater than your typical ballot. It’s actually a giant information challenge. We mix polling, voter information (from L2 Data (L2 is the agency utilized by CBS News for voter information), U.S. Census information, and historic developments to get a transparent image of what is going on on in every state.
Here’s what the info tells us:
- We know which candidates various kinds of voters are supporting from our polling, which incorporates a lot bigger pattern sizes — tens of 1000’s — than a typical ballot;
- We understand how many individuals like them are in every state and county, in addition to their turnout historical past, from voter information and U.S. Census information;
- And we all know every state’s earlier election outcomes, which permits us to anchor our 2024 estimates to latest historical past.
Our mannequin combines all this information utilizing multilevel regression with post-stratification (scroll down for extra particulars on this). A characteristic of this system is that we use developments throughout the nation to tell our image of a selected state. If we discover Hispanic voters throughout the Southwest shifting help, as an example, we use the knowledge to extra exactly estimate particular states wherein Hispanic voters reside. The similar applies for a lot of different varieties of voters. The survey lets them inform us what they’re considering, and we map that to what number of of them reside in every state.
We collaborate on information assortment and modeling with international opinion analysis agency YouGov, constructing on our profitable efforts in 2018, 2020, and 2022.
3. Think snapshots, not forecasts
We inform you the place races stand right now, explaining why and what would possibly change. We absolutely anticipate motion earlier than the primary vote is solid, so we’ll replace all the pieces commonly within the months forward.
Unlike an electoral forecast, we’re estimating every candidate’s present help, incorporating all the info we have collected up thus far. For occasion, if we estimate President Biden at 49% in a state with a margin of error of three factors, we’re assured that his help there’s between 46% and 52% right now — not that the ultimate outcome shall be in that vary.
There’s nothing right here to account for forward-looking uncertainty — nothing in regards to the economic system altering or dramatic debates, for instance. We absolutely anticipate motion earlier than the primary vote is solid, so we’ll replace all the pieces commonly all through the autumn. A race that is leaning towards a celebration right now might be reclassified as a toss-up if it turns into extra aggressive.
4. Electoral situations
That brings us to situations. Down the street, the Battleground Tracker will provide believable situations for a way the election would possibly unfold. We’ll do that utilizing a mixture of statistical simulation and by tweaking a number of the assumptions underlying our mannequin, leading to a vary of potential outcomes.
Here’s an instance. One of essentially the most difficult issues to determine shall be turnout: who’s really going to trouble to vote? Modeling who’s more likely to vote is a perennial problem that’s delicate to assumptions.
In our baseline mannequin, we estimate which voters are casting ballots primarily based on each what they inform us they’re planning on doing and historic patterns of their states. In our situations, we’ll barely alter the mannequin’s parameters to discover what might occur if, for instance, massive swaths of voters keep dwelling (maybe for concern of getting sick) or if there is a surge in voting by mail (additionally potential given intense curiosity on this election). We’ll roll out our situations later within the marketing campaign, so test again for them!
5. Models have stable observe document
While we take a unique method than conventional polling, the Battleground Tracker relies on rigorous strategies from the fields of political science, survey analysis, and statistics. Moreover, now we have a robust observe document using comparable fashions at CBS News over the previous few years.
Our 2018 mannequin carried out significantly nicely, steadily monitoring Democrats’ enchancment in key races and the eventual blue wave within the U.S. House. In truth, our high-turnout situation precisely estimated the ultimate seat breakdown, when it got here to go that historic turnout powered Democratic positive aspects.
Our correct race rankings in 2020 had been primarily based on an identical mannequin. We estimated that Democrats had constructed a lead heading into Election Day, however that Republicans might meet up with a late turnout surge. Every state we categorized as leaning Democratic wound up going to Mr. Biden, and every we rated as leaning Republican went to Trump. Of our six toss ups, Trump gained 4 and Mr. Biden gained two.
And most not too long ago, this system enabled us to precisely estimate the 2022 midterm elections. Our mannequin constantly pointed to slim Republican positive aspects within the House, moderately than the purple wave that many different analysts anticipated.
More on the statistical mannequin we use
If you wish to know extra in regards to the information and statistical mannequin we use — and do not thoughts a little bit of jargon — then maintain studying…
First, we survey 1000’s of registered voters throughout the nation and ensure to attract bigger samples in battleground states, which we anticipate to be extra aggressive. The most necessary survey questions we ask for estimation functions are how doubtless they’re to vote and which candidate they’d vote for right now.
We then decide how folks’s vote intentions are associated to their traits like age, gender, race, schooling, previous vote, the place they reside, and so forth. Each voter has a sure mixture, which we’ll name a “profile” for shorthand. For instance, one potential profile is a 45-year-old, Asian feminine, who holds a school diploma, voted for Biden 4 years in the past, and lives in Allegheny County, Pennsylvania. Change any certainly one of these traits and also you get a unique profile. We run a multilevel regression — a strategy to mannequin the connection between completely different variables — on the survey information to estimate what number of voters of every particular profile intend to vote for every main candidate. The regression consists of the voter traits above, in addition to state and district results (the degrees in “multilevel”).
The subsequent step is estimating how many individuals of every voter profile reside in every state. For this we use a mixture of U.S. census information and voter information, which incorporates counts of voters at very granular ranges, corresponding to voting precincts. In every state, we multiply the overall variety of voters of a given profile by the proportion of voters with that profile selecting a candidate (the “post-stratification” step). Aggregating throughout all voter profiles in a state, we lastly get the estimate of that candidate’s statewide vote share. In Maine and Nebraska — the 2 states that award electoral votes by congressional district — we additionally estimate candidate help in every district.
