Election 2020: can the Democrats win the Senate? | The Economist


who's going to win america's election

today was the day of a thousand polls

it's a question

everyone is trying to answer joe biden

up over president trump donald trump is

within the margin of

error most of the focus is on the


but there's another crucial race with

democrats and republicans battling for

control of the senate in 2020

the democrats hope this is their chance

to win back the senate

after five years the democrats are in

with a shot of taking back america's


but with 35 individual races to follow

the result is harder to predict than the


so how do you work out who will win the


might have the answer

control of congress can make or break a


especially one intent on delivering


if you just look at the presidency of

barack obama

he passed a bunch of major legislation

during the first two years of his term

the health care reform also the

financial reform and then once the


won the house in 2010 that was that no


legislative achievements for six years

mitch mcconnell the senate majority

leader at the end of barack obama's time

in office

saw blocking democratic motions as part

and parcel

of his job i was in charge of what we

did the last two years of the obama

administration i give you

full credit for that there is no doubt

in my mind that if joe biden wins the


mitch mcconnell will respond to his

nominees and legislative initiatives

exactly the way

he did to barack obama which is to block

everything that he possibly can

there are 35 senate seats up for grabs

in november

the republicans are defending 23 seats

the democrats 12. to take control

the democrats will need to gain at least

four three if biden wins the presidency

as the vice president decides on tied


but trying to predict the result is not

an easy task

there's so many different races with so

many different candidates and so many

different polls i mean there's just no

way that

one person could keep track of it all in

their head and

fortunately we have computers to do that

for us

earlier this year dan began work on

creating a model to predict the result

of the senate vote

it was probably a good three months

of non-stop work including nights and


this july august september it's about as

big as

a project as anything i've ever done

all of that work for one crucial number

the percentage chance of a victory

as of october 6th our model gave the

democrats a 71

chance of winning the senate if you

click on the link above

it'll take you to the most up-to-date


but why trust this number more than the


the economist model takes in two sources

of information to produce its


the first is those polls

you can get quite accurate predictions


how legislative elections are going to

go using polls

when you have a lot of quality polls

taken really close to an election

the problem is the quality and quantity

of polls for senate races

vary widely state by state and every


counts that leaves you

very very in the dark in races that

don't have much or any polling

so the model also looks at a range of

historical factors

things like a state's voting history

whether a candidate is the incumbent

their political experience or how much

money they've raised

political science nerds call these

factors fundamentals

you can produce predictions of

results in these races based on all of


sort of benchmark quote-unquote

fundamental factors

without seeing a single poll but first

the model has to estimate the impact

each of those factors

is likely to have on the result to work

that out

it tests it on an election for which we

already know the result

you say okay let's give our model

the data from every election year except


say 1976. and based on everything we

knew from

every year but 1976 what would we have

predicted for 1976

it does terribly you say oops

that was over fit what were the least

useful variables in here the ones we

could most afford to do without and

let's get rid of them

and then even the really strong

variables let's maybe trim their effect

size by like 10

or something because you just never know

and you basically have the computer

just try a zillion of these permutations

until it finds the amount of shrinkage

that yields the best predictions on the

unseen dab

why go to all this trouble because

projections that combine polling data

with fundamental factors

tend to be much more stable than those

looking at polling alone

take the 1992 election where george h.w

bush lost the presidency to bill clinton


line marks his eventual share of the


but in the months before the election

you'd have been hard-pressed to foresee


the polls were all over the place the

economist model would have been within

the margin of error throughout that

election year

so much for elections that have long

since been decided

what about the 2020 senate race

here's an example of how we rate the

chances that a republican senator

like susan collins from maine will

retain her seat

collins won the 2014 election for her

senate seat by a landslide

suggesting it's more likely she would be

voted in again

she also has the advantage of incumbency

something particularly important in


but the democratic candidate sarah

gideon has raised more money in


based on these and other fundamentals

the model had collins

eight points ahead but that's not the

end of the story

and then each time a poll comes in you


your prediction it will move

incrementally in the direction of

whatever the new pole you're adding says

there have been a number of high quality

poles in maine

the kind of pole our model weighs


with those factored in the forecast

swings in favor of her democratic


so which other races should you keep an

eye on

according to the economist model alabama

is very likely to switch from democrat

to republican

whilst arizona colorado and north


like maine have a good chance of

flipping the other way

if biden wins the presidency this map

would give the democrats their senate


of course there's another challenge in

predicting the result of these

or any other race in the 2020 election

however good the model while in a

pandemic votes cast by mail are expected

to set a new record this november while

the experts say voting by mail is safe

it's yet another example of how it's


and it could be even more challenging

come november my

model does not make any

explicit adjustments for the fact that

this election

is occurring during the cobit pandemic

i think covet is a huge deal it's

obviously killed a huge amount of people

and of course it will cause changes in

the way we vote it could indeed have

significant unexpected you know

unmodelable effects


i'll need to see the evidence

without that evidence predicting how the

pandemic's going to affect this election

would call for forces far beyond science

i'm dan rosenhack the economist data

editor you can keep up to date on all of

our election coverage by clicking on the

link opposite

thank you for watching