Because they represent the political leanings of the country as a whole. Please list the demographics, as to the racial makeup, educational levels, average income , ruralness, etc. of each of those 19 counties, and compare them to the country as a whole for me. You must have those figures at your fingertips, if you know all that.
I identified 17 counties whose results matched the eventual winner of all presidential elections between 1980 and 2016 (I'm not sure where the 19 number is coming from--I could only find 17, so those are the ones I examined). Those are Clallam County, Washington; Bremer County, Iowa; Essex County, Vermont; Hidalgo County, New Mexico; Marquette County, Wisconsin; Juneau County, Wisconsin; Ottawa County, Ohio; Richland County, Wisconsin; Sawyer County, Wisconsin; Shiawassee County, Michigan; Valencia County, New Mexico; Van Buren County, Michigan; Vigo County, Indiana; Warren County, Illinois; Washington County, Maine; Westmoreland County, Virginia; and Wood County, Ohio. Out of those, only Clallam County, Washington had a majority vote for Joe Biden in 2020.
Out of the 16 counties that matched the winner between 1980 and 2016 but did not vote majority for Joe Biden in 2020, they were on average 91.2% White versus the the national average of 76.3%. They were on average 3.5% Black/African American versus the national average of 13.4%. They were on average 2.4% Native American versus the national average of 1.3%. They were on average 0.9% Asian versus the national average of 5.9%. They were on average 0.1% Native Hawaiian/Pacific Islander versus the national average of 0.2%. They were on average 1.9% two or more races versus the national average of 2.8%. And they were on average 11.4% Hispanic or Latino versus the national average of 18.5%. Out of the 16 counties, 15 were whiter than the national average and only 1 was blacker than the national average. 2 were more Hispanic than the national average.
Out of the 16 counties, they had slightly higher rates of high school graduates in the population over 25 years of age. The average of the 16 counties was 88.7% versus the national average of 87.7%. They had a significantly lower rate of persons over 25 with a bachelor's degree or higher. The average of the 16 counties was 20.6% versus the national average of 31.5%. Out of the 16 counties, only 1 county had a rate of higher education higher than the national average.
The average household income of the 16 counties between 2014 and 2018 was $50,059 (in 2018 dollars) versus the national average of $60,293. One of the counties had a higher-than-average income level.
On average, the 16 counties had a population density of 78.9 persons per square mile, lower than the overall US population density of 92.9 persons per square mile. The county with the highest population density is Vigo County, Indiana, with an overall density of 265 persons per square mile, with its county seat, Terre Haute, having a maximum population density of 1731 persons per square mile and a total population of 60,622. For reference, Knox County has an overall density of 850 persons per square mile, with Knoxville coming in at 1,900 persons per square mile and a total population of 187,603. Davidson County has an overall density of 1,326 persons per square mile, while Nashville has a population of 670,820. The closest Tennessee equivalent to the size of Terre Haute is Bartlett.
As for the voting trends of the 16 counties, I only compiled data going back to the 2000 presidential election, but since then, they've tended to vote, on average, close to the national popular vote margin. They have historically leaned further toward the winning side than average, with the difference being larger with Republican winners. However, this changed in 2016; in both the 2016 and 2020 elections, these counties voted overwhelmingly for Donald Trump, with margins much higher than the national average. Based on this, the deviance in these bellwethers' votes actually started during the last election cycle and continued on the same trend during this election cycle.
To put some numbers on this, the 16 counties voted with a margin 5.05 percentage points higher (more Republican) than the national popular vote total in 2000 (50.34% Bush / 45.80% Gore versus the national total 47.87% Bush / 48.38% Gore); 3.86 percentage points higher (more Republican) than the national popular vote total in 2004 (52.65% Bush / 46.33% Kerry versus the national total 50.73% Bush / 48.27% Kerry); 1.54 percentage points higher (more Democratic) than the national popular vote total in 2008 (53.60% Obama / 44.79% McCain versus the national total 52.93% Obama / 45.65% McCain); and 0.86 percentage points higher (more Democratic) than the national popular vote total in 2012 (51.32% Obama / 46.59% Kerry versus the national total 51.06% Obama / 47.20% Kerry). A huge shift happens in 2016: the 16 counties voted with a margin 16.98 percentage points higher (more Republican) than the national total (53.95% Trump / 39.06% Clinton versus the national total 46.09% Trump / 48.18% Clinton). In 2020, this lead slightly increased to a margin 20.01 percentage points higher (more Republican) than the national total (57.09% Trump / 41.03% Biden versus the national total 47.17% Trump / 51.12% Biden).
So, as you can see, the "bellwether counties" aren't a very good litmus test of the US as a whole nowadays, at least when it comes to demographics. The counties that voted for Trump this cycle tend to have demographics that tend to be favorable for Trump (when it comes to nationwide voting), and that's reflected in the numbers he got in these counties in 2016. These demographics are not representative of the nation as a whole.
Another thing that I wanted to point out that might help to process this quirk in the election--did you know that in the 2016 election, there were also 16 bellwether counties that had correct predictions back to at least 1980, but ended up being incorrect by voting a majority for Hillary Clinton? Of those 16, 14 voted a majority for Joe Biden during this election, meaning that out of the 33 (that I've confirmed) bellwethers going into the 2016 election with correct choices back to at least 1980, 1 still has 100% correct predictions, 30 have only missed on one election, and only 2 have missed on 2 elections.
As a final note, all of this info was pulled from the United States Census Bureau and state and county official voting information. I went through all the data and have it compiled in a spreadsheet if you want to review it yourself; none of this is from the media or any other political site and it can all be found with a little research into unbiased public data sources.