Attempt at a statistical prediction of the 2026 midterms

I am currently trying to dip my toes in the field of statistics. To test myself, I will try to take a shot at predicting (or guessing) the result of the 2026 US midterms, which will take place on November 3, 2026. So my prediction will be validated in 608 days.

TLDR: I predict a democrat controlled house and a republican controlled senate

The reasons I think this will happen are the following:

incumbents tend to always loose seats
republicans tend to win the senate more than the democrats when they’re in power
democrats tend to win the house more when the republicans are in power
the senate elections are of “Class II”, in which the republicans perform best
the current senate is very much dominated by republicans (53 reps vs 45 dems vs 2 independents)

1. Incumbents tend to always loose seats

There’s an anti-incumbent sentiment all around the globe. It turns out that the phenomenon even has a name in the US, it’s called the Six-Year Itch. One could argue that the Six-Year Itch doesn’t apply to Donald Trump because the his mandates are not consecutive, suggesting that the general public was sympathetic enough to reelect him. Thus, his mandate doesn’t suffer from the Six-Year Itch.

Data used for visualization is from wikipedia.

midterms.csv

year,president,presidents-party,house-seats,senate-seats
1790,George Washington ,None ,3,0
1794,George Washington ,None ,-4,3
1798,John Adams ,Federalist ,3,0
1802,Thomas Jefferson ,Democratic-Republican ,35,5
1806,Thomas Jefferson ,Democratic-Republican ,2,1
1810,James Madison ,Democratic-Republican ,13,0
1814,James Madison ,Democratic-Republican ,5,-3
1818,James Monroe ,Democratic-Republican ,13,2
1822,James Monroe ,Democratic-Republican ,34,0
1826,John Quincy Adams ,Democratic-Republican ,-9,-2
1830,Andrew Jackson ,Democratic ,-10,1
1834,Andrew Jackson ,Democratic ,0,1
1838,Martin Van Buren ,Democratic ,-3,-7
1842,John Tyler ,None ,-69,-3
1846,James K. Polk ,Democratic ,-30,2
1850,Millard Fillmore ,Whig ,-22,-3
1854,Franklin Pierce ,Democratic ,-75,-3
1858,James Buchanan ,Democratic ,-35,-4
1862,Abraham Lincoln ,Republican ,-23,1
1866,Andrew Johnson ,Democratic ,9,0
1870,Ulysses S. Grant ,Republican ,-32,-5
1874,Ulysses S. Grant ,Republican ,-93,-10
1878,Rutherford B. Hayes ,Republican ,-4,-7
1882,Chester A. Arthur ,Republican ,-29,0
1886,Grover Cleveland ,Democratic ,-16,2
1890,Benjamin Harrison ,Republican ,-93,-4
1894,Grover Cleveland ,Democratic ,-127,-4
1898,William McKinley ,Republican ,-21,6
1902,Theodore Roosevelt ,Republican ,9,0
1906,Theodore Roosevelt ,Republican ,-27,2
1910,William Howard Taft ,Republican ,-56,-9
1914,Woodrow Wilson ,Democratic ,-61,3
1918,Woodrow Wilson ,Democratic ,-22,-4
1922,Warren G. Harding ,Republican ,-77,-7
1926,Calvin Coolidge ,Republican ,-9,-6
1930,Herbert Hoover ,Republican ,-52,-6
1934,Franklin D. Roosevelt ,Democratic ,9,9
1938,Franklin D. Roosevelt ,Democratic ,-72,-7
1942,Franklin D. Roosevelt ,Democratic ,-45,-8
1946,Harry S. Truman ,Democratic ,-54,-10
1950,Harry S. Truman ,Democratic ,-28,-5
1954,Dwight D. Eisenhower ,Republican ,-18,-2
1958,Dwight D. Eisenhower ,Republican ,-48,-12
1962,John F. Kennedy ,Democratic ,-4,4
1966,Lyndon B. Johnson ,Democratic ,-47,-3
1970,Richard Nixon ,Republican ,-12,2
1974,Gerald Ford ,Republican ,-48,-4
1978,Jimmy Carter ,Democratic ,-15,-2
1982,Ronald Reagan ,Republican ,-26,0
1986,Ronald Reagan ,Republican ,-5,-8
1990,George H. W. Bush ,Republican ,-8,-1
1994,Bill Clinton ,Democratic ,-54,-9
1998,Bill Clinton ,Democratic ,4,0
2002,George W. Bush ,Republican ,8,2
2006,George W. Bush ,Republican ,-32,-6
2010,Barack Obama ,Democratic ,-63,-6
2014,Barack Obama ,Democratic ,-13,-9
2018,Donald Trump ,Republican ,-41,2
2022,Joe Biden ,Democratic ,-9,1

code

import pandas as pd
midterms = pd.read_csv("midterms.csv")

midterms[["house-seats", "senate-seats", "year"]].plot.area(x="year", figsize=(12, 4), subplots=False, stacked=False)

midterms[["house-seats", "senate-seats"]].mean()
# house-seats -25.322034 senate-seats -2.203390 dtype: float64

2. Republicans tend to win the senate

When republicans are in power, they keep control of the house only 42% of the time, while they keep control of the senate 54% of the time.

	Republicans	Democrats
% house control	42.857143	57.142857
% senate control	54.761905	45.238095

house-senate-control.csv

year,house-control,senate-control
1858 ,rep,dem
1862 ,rep,rep
1866 ,rep,rep
1870 ,rep,rep
1874 ,dem,rep
1878 ,dem,dem
1882 ,dem,rep
1886 ,dem,rep
1890 ,dem,rep
1894 ,rep,rep
1898 ,rep,rep
1902 ,rep,rep
1906 ,rep,rep
1910 ,dem,rep
1914 ,dem,dem
1918 ,rep,rep
1922 ,rep,rep
1926 ,rep,rep
1930 ,dem,rep
1934 ,dem,dem
1938 ,dem,dem
1942 ,dem,dem
1946 ,rep,rep
1950 ,dem,dem
1954 ,dem,dem
1958 ,dem,dem
1962 ,dem,dem
1966 ,dem,dem
1970 ,dem,dem
1974 ,dem,dem
1978 ,dem,dem
1982 ,dem,rep
1986 ,dem,dem
1990 ,dem,dem
1994 ,rep,rep
1998 ,rep,rep
2002 ,rep,rep
2006 ,dem,dem
2010 ,rep,dem
2014 ,rep,rep
2018 ,dem,rep
2022 ,rep,dem

code

control = pd.read_csv("house-senate-control.csv")
midterms = pd.merge(midterms, control, how="left", on="year")

dem_controlled_house = midterms[midterms["house-control"] == "dem"]
rep_controlled_house = midterms[midterms["house-control"] == "rep"]
house_total = len(rep_controlled_house) + len(dem_controlled_house)

dem_controlled_senate = midterms[midterms["senate-control"] == "dem"]
rep_controlled_senate = midterms[midterms["senate-control"] == "rep"]
senate_total = len(rep_controlled_senate) + len(dem_controlled_senate)

pd.DataFrame({
        "Republicans": {
                "% house control": len(rep_controlled_house) / house_total * 100,
                "% senate control": len(rep_controlled_senate) / senate_total * 100,
        },
        "Democrats": {
                "% house control": len(dem_controlled_house) / house_total * 100,
                "% senate control": len(dem_controlled_senate) / senate_total * 100,
        }
})

3. Class II senate elections favors the republicans

Not all senate seats are up for election each midterm, only about 30 of them are reelected.
Class II senate elections include more red states and they don’t include california, which favors the republicans.

code

# data from gh fivethirtyeight/election-results
import pandas as pd
elections = pd.read_csv("election_results_senate.csv")

reps = elections[elections["ballot_party"].str.contains("REP", na=False)]

reps.groupby(["office_seat_name"], as_index=False)["winner"].mean()

office_seat_name	winner
Class I	0.389439
Class II	0.541667
Class III	0.509615

Conclusion: Class II and III elections favor the republicans.

This explains why Biden had such fantastical midterm results even while being unpopular with the general public, losing 9 seats in the house and gaining one in the senate. The 2022 midterms were of class I.

Final words (and extra)

I think I provided a strong enough case to justify a prediction of a republican controlled senate after the midterms. Control of the house, that, I can’t really predict.

I will also include some more graphs (one I made and one from wikipedia).