Saturday, February 22, 2020


Many people say that Americans increasingly live in "bubbles" in which they don't encounter people whose political views differ from their own.  To quote a NY Times article that appeared a few weeks ago, "years of social and political sorting have left Americans of both parties less likely to know people who are different from them politically."  It didn't cite a source for this statement, but I have been looking for relevant survey evidence for some time and have found only the following.

1.  Pew surveys from 2016 and 2017 asked "Thinking about your close friends, how many are...Democrats--a lot, some, just a few or none?" and a parallel question about how many were Republicans. 
2. In 2007, a CBS/NY Times survey asked Republicans "Do any of your close friends think of themselves as Democrats?" and asked Democrats if any close friends thought of themselves as Republicans. 
3.  And way back in 1952, the American National Election Study asked how people thought their five closes friends would vote.

Those questions are all on the same general topic, but they can't be directly compared, so it's not clear whether there's actually been a change.  For what it's worth, the 1952 results were 28% thought all their friends would vote Democratic, 26% that all would vote Republican, 25% that they would be split, and 20% didn't know. 

But there is also a question of who is more or less likely to know people with different political views.  A number of people, notably David Brooks and Charles Murray, say that Democrats, more specifically college educated Democrats, are more likely to live in "bubbles." I looked at the 2017 Pew survey to see if there was any evidence for this.  I limited the analysis to non-blacks, because there was a large racial difference:  32% of blacks and only 5% of non-Hispanic whites said that none of their close friends were Republicans, while 7% of blacks and 31% of non-Hispanic whites said that a lot were.  Hispanics were in between, but a lot closer to non-Hispanic whites, so I kept them in. 

The results:  As expected, people report having more friends who support their party than support the other party.  On a scale of 1 (none) to 4 (a lot), with own party as rows and friends as columns, the means are:

                                 D            R
Democrats             3.48      2.46
Republicans           2.53      3.39

Republicans report more opposite-party friends and fewer same-party friends than Democrats, but the differences are small.  Educated people reported more own-party friends and more opposite party friends.  The differences were bigger for opposite party friends.  The general differences aren't surprising, since more educated people tend to have larger circles of friends, but it is notable that the differences are larger for the opposite party--I'll return to that. 

Finally, the relationship with education differs by party.  The relationship between education (1=HS or less, 2=some college, 3=college grad) and friends of one's own party, with Democrats as blue dots and Republicans as red:

Generally increasing with education for both, but the relationship is stronger for Democrats.

Now the relationship to opposite party friends:

Increasing for Republicans, little relationship for Democrats. 

As a table:
                                      Own        Opposite   Diff
HS, Rep                           3.26        2.38        0.88
Some, Rep                       3.47        2.46        1.01
College, Rep                    3.41        2.63        0.78

HS, Dem                          3.21        2.48        0.73
Some, Dem                      3.38        2.38        1.00
College, Dem                   3.59        2.48        1.11

If you take the difference between own-party and opposite-party friends as the measure, college-educated Democrats tend to have the thickest bubbles--score one for Brooks and Murray.  But if you take the number of opposite-party friends as the measure, then Republicans who didn't attend college and Democrats with some college have the thickest bubble (fewest friends from the other party). 

Most people don't like talking about politics, especially with people who don't agree with them.  So if you have a friend who shows signs of having different political views, you'll probably stay away from politics when you talk.  And even if you get a sense that the friend differs from you on some issues, you'll probably give them the benefit of the doubt and figure they're still a supporter of your party or at worst an independent.  Education tends to increase interest in politics and willingness to talk about politics, so more educated people are more likely to be aware of having friends who disagree with them. 
So I think the contrasting figures for college-educated Democrats and Republicans reflect the reality that college-educated people are now mostly Democratic.  Although non-college whites are now mostly Republicans, they are less likely to have accurate knowledge of their friends' political views, so there's little difference in the perceptions of Democrats and Republicans.

Wednesday, February 19, 2020


I have a post coming about the popular idea that certain kinds of people live in "bubbles" in which all of their friends share their political views.  On the way there, I ran across an unrelated bit of information.  In 1998, the Gallup Poll asked "Some people say there's not much opportunity in America today--that the average person doesn't have much chance to really get ahead. Others say there's plenty of opportunity and anyone who works hard can go as far as they want. How do you feel about this?"  They repeated the question in 2011, 2013, and 2016.  I'd noticed this question before, and thought it was potentially interesting.  But then I discovered something that made it seem more interesting:  the same question (except it said "man" rather than "person") was asked in the very first American National Election Study, back in 1952. The results:

                   Not much               Plenty         DK
1952                10%                    87%            4%
1998                17%                    81%            2%
2011                41%                    57%            2%
2013                43%                    52%            6%
2016                28%                    70%            2%

It's too bad the ANES didn't stick with the question--the lopsided responses in 1952 probably were a factor in making them decide it wasn't worth repeating.  We'll never know if those responses just reflect the economic conditions of the time (the unemployment rate was about 3% and falling), or something more general. The figures from 2011, 2013, and 2016 suggest that answers are sensitive to current economic conditions, so it would be interesting to see what would happen if they repeated it today.   However, unemployment was only a little higher in June 2016 than in April 1998 (4.9 vs. 4.3) but the percent who said "not much" was substantially higher.  That could be that people have become substantially more discontented in the 21st century, or that opinions on the question have become more "politicized"--in 2011, Democrats were substantially more likely to say "doesn't have much chance" although there was a Democratic president.  The individual-level data aren't available for the 1998 survey, but they are for 1952, so I will take a look at that relationship sometime. 

[Data from the Roper Center for Public Opinion Research and ANES]

Wednesday, February 12, 2020

Your income, my vote

A couple of weeks ago I had a post on the association between a county's income distribution and voting.   Larger numbers of poor (under $15,000 a year), and affluent (over $100,000 and especially over $150,000) people were associated with more support for the Democrats.  The distribution of incomes within a large middle range ($15,000 to $100,000) didn't make much difference.

These differences aren't just a reflection of differences between the voting patterns of low, middle, and high income people.  In 2016, they all split their votes in pretty much the same way.  That is, it's not your own income that matters for your vote, but the incomes of people around you.

Why would these relationships occur?  I think that the apparent effect of the number of low-income people is probably a proxy for race.  The counties with large shares of people earning less than $15,000 are mostly small rural ones, and judging from their locations I'd guess that most of them have large black or American Indian populations.

The reason for a relation between the presence of high income people and Democratic support is less straightforward.  I can think of two plausible possibilities, the first involving culture and the second involving economics.  First, that liberals have an affinity for the kinds of cultural amenities that are associated with large populations of affluent people--restaurants, museums, various kinds of "high culture."  Or you could put it the other way around--that high income people are attracted to the productions of "bohemian" culture.  Second, that where there are large numbers of people with high incomes, other people are more likely to think that we can increase spending on social programs without increasing taxes on the middle class.  Or you could say that concern about "inequality" will be a stronger force in places that have more inequality.  Even though people in Mifflin County, PA (where about 2.5% have incomes over $150,000) might know in principle that there are lots of high-income people in other parts of the country, they probably won't be as conscious of their existence as people in Fairfield County, CT (25% over $150,000).*

A final issue is that the NY Times article that started me on these posts said that the county-level relationship between average income had become considerably stronger between 1992 and 2016.  I think that the second story I gave would be more likely to account for that--changes in the numbers of high income people can occur faster than changes in the degree of residential "sorting."

*In addition, given the skewed distribution of incomes, I expect that the average incomes of the "rich" are higher in places where there are a lot of them--e. g., that in Mifflin County, even people who earn over $150,000 rarely earn much more than that, but that in Fairfield County, many of them earn a lot more than $150,000. 

Wednesday, February 5, 2020

The Decline of Tribalism

In 2018, I had a post objecting to the description of contemporary political polarization as "tribalism."  Despite my disapproval, the term has remained popular--it appeared 74 times in the New York Times in 2019 and eight so far this year (although that is down from the peak of 119 in 2018).  This post will elaborate on my objection and give some data. 

I understand political tribalism as a tendency to support or oppose a political figure because he is or is not "one of us," without respect to his positions on political issues.  Tribalism has been a strong force in American history, but is considerably weaker than it used to be. 

To see the strength of tribalism in the past, look at the elections of 1924 and 1928.   The Republicans won both easily, 54%-29% in 1924 (with 17% going to the Progressive, Robert LaFollette), and 58%-41% in 1928.  Despite getting routed in 1924, the Democrats won all eleven southern states, as they had in most elections starting in 1880 (when blacks were effectively disenfranchised).  In 1928, they won only five southern states.  What happened?  The Democratic candidate was Al Smith, the first Catholic to be nominated by a major party.   For a lot of people in the South, that was reason enough to vote against him.  On the other hand, the Democrats carried Massachusetts and Rhode Island, two states with a lot of Catholics, for the first time in many years.   Focusing on a county with a lot of Catholics, New Haven (CT), in 1924 about 120,000 votes were cast, only 36,000 (30%) for the Democrat.  In 1928, the Democratic vote rose to 82,000 out of 164,000.  In 1932, Franklin Roosevelt won the election in a landslide, but didn't match Smith's performance in New Haven County (49.9% vs. 50.1%).  So it appears that a lot of Catholics changed their votes, or came out to vote, because a Catholic was on the ballot. 

The Catholic/Protestant divide was still important in 1960.  In April 1960, a Gallup Poll asked "if Nixon were the Republican candidate and (John) Kennedy were the Democratic candidate, which would you like to see win (the 1960 presidential election)?" 48% said Kennedy and 45% said Nixon.  It also asked "as you may know, Kennedy is a Catholic in his religion. Suppose Kennedy were not a Catholic--which man would you like to see win: (Richard) Nixon or (John) Kennedy?"  51% said Kennedy, and 40% said Nixon.  These questions were not asked to different parts of the sample--they were both asked the the whole sample, one right after the other.   That is, about 5-6% of the sample came right out and said that they would vote differently if Kennedy were not Catholic.  In October 1960, 5% said that the fact that Kennedy was a Catholic made them more in favor of him, and 19% that it made them less in favor. 

Region, especially South/non-South, also made a difference.  In 1976, Jimmy Carter seemed to gain not just in Georgia, but across the whole South.  Bill Clinton got some benefit, but not as much, in 1992.  By 2000, it didn't seem to make any difference for Al Gore.

Under some circumstances, "tribalism" can promote political polarization.  But in the United States, it generally limited it.  The Catholic with conservative views would often be a Democrat because his family had always been Democrats, while the Protestant with liberal views would often be a Republican because his family had always been Republican (an example), blurring the differences between the parties.  In the last 40 years or so, tribalism has declined and been replaced by ideology.  Of course, there are still group differences in support for the parties, but they have a clear basis in the current positions of the parties, and don't depend on the identity of the candidates.  For example, Tim Scott, the black Republican Senator from South Carolina, gets strong support from whites and not much from blacks.

[Data from the Roper Center for Public Opinion Research] 

Saturday, February 1, 2020

More than I bargained for

A few days ago, the NY Times had an article on the increasing association between the prosperity and voting patterns of counties.  It noted that in 1992, there was a small tendency for poorer counties to give more support to the Republicans, but that this tendency has become stronger, and that in 2016 "the Republican Party won almost twice the share of votes in the nation’s most destitute counties — home to the poorest 10 percent of Americans — than it won in the richest."  There is a mistake in that sentence--the poorest 10% of counties are not home to the poorest 10% of Americans (although of course they are home to some of them).  I point that out not just to be pedantic, but because it will play a role in this post.  

The article also discussed some examples--Montgomery County in Ohio and Macomb in Michigan.  It said that both had declined in relative income, and had moved towards the Republicans.  It also mentioned Gallatin County, Montana, which had gained and moved to the Democrats.  However, these examples don't shed much light on the issue.  If there is a relationship between county affluence and party support, then it follows that places that change their relative standing will tend to change their voting.  But in order to understand why the relation arose in the first place, you should look at poor places that shifted to the Republicans and rich places that shifted to the Democrats.  

I don't have data on voting at the county level going back to 1992, but I do have data for 2016, so I decided to look more deeply at the relationship between economic conditions and party support in that election.  My general though was that it might depend on the distribution as well as the average level of income.  I tried to download statistics on average household income by county from the Census, but found that I had the percentage in sixteen different income categories.  Since I had them, I regressed the Republican lead in 2016 (Republican share minus Democratic share) on the the percentages.  Based on those results, I made five groups:  less than $10,000 a year, 10-15,000, 15-45,000, 45-100,000, 100-150,000, and 150,000 and up.  I will call those "very poor," "poor," "lower," "middle," "upper middle," and "upper." The regression results:

Constant     1.1
v. poor       -3.8
poor           -0.9
lower           0      [reference]
middle       -0.4 
u. middle   -1.1
upper         -3.2

So the predicted performance of Republicans is best when more people have moderate (or moderately low) incomes, and worse when more people are well-off or poor.  That is, the Democratic vote increases not just general affluence, but also the degree of inequality.  

In order to try to understand the relationship, I identified some counties with extreme predicted values and small residuals.  Some pro-Democratic examples:

                                  Predicted lead                                Actual lead
Westchester, NY           -.29                                       -.32  (32%-64%)
Fairfield, CT                 -.27                                       -.20  (38%-58%)
Norfolk, MA                 -.22                                       -.28  (33%-61%)

Some pro-Republican examples:

Baxter, AR                      .55                                        .54 (75%-21%)
Mifflin, PA                      .54                                        .57 (77%-20%)
Wabash, IN                     .53                                         .51 (73%-21%)

On the Republican side,  I restricted it to counties with at least 10,000 households.  As it happens, I know something about all of the Democratic examples--affluent suburbs of large metropolitan areas.  I didn't know anything about the others, but between the data and Wikipedia a few things emerge:
1.  They are not "destitute":  they're places where a lot of people have moderate incomes, but few have incomes over $100,000.  
2.  They are rural or small town; none of them are within commuting distance of major cities.  
3.   It doesn't seem that any of them are or were particularly dependent on industry--they seem to have fairly diversified economies.  

I will say more in future posts, but that gives me enough to think about for now.