Skip to main content
Dr. Michael SchymuraEconomics of AI & Innovation
Europe, MeasuredESS-06 / September 24, 2026 / revised September 27, 2026 / analysis 1.1.0

Happier, but less equally happy?

Europe got happier, however you label the scale

Between 2010/11 and 2023/24, the share of people above every cut-off on the 0-10 life-satisfaction scale rose across 17 matched countries. No relabeling of the answers can reverse that gain. Dispersion narrowed both within and between countries, although no scale-free test can certify that Europe became more equal.

Stretch the ruler See the ten cut-offsCompare two countries

By Michael Schymura / 18 min read / 58,642 respondents / 10 evidence figures

Three answers

Happier?Yes, on every ruler.+0.23points of mean life satisfaction, 95% interval +0.18 to +0.29. The share above all ten cut-offs rose; the least favourable relabeling still gives +0.03.Less equally happy?No. Less dispersed.−5.2%standard deviation, 95% interval −7.0% to −3.4%. The spread of country means fell 19%; no scale-free test can rank the spread.Everywhere?Not everywhere.12 / 17countries rose where detectable and never fell significantly. 4 show no detectable change; one, Israel, fell at the top of the scale.

A rising average can hide a falling floor. It can also hide something more basic. A 0-10 answer scale ranks feelings; it does not measure them with a ruler, and nothing guarantees that the distance from 3 to 4 equals the distance from 8 to 9. Bond and Lang (2019) showed that with such scales the ranking of two averages can be reversed by relabeling the answers, unless one distribution lies above the other at every threshold.

This study therefore asks the question twice. It first asks whether Europe became happier on the familiar equal-step scale, and then whether that answer survives every increasing relabeling of the eleven answers, which is the stronger claim a reader actually needs. For the 17 countries surveyed in both 2010/11 and 2023/24 with reviewed survey designs, the answer to both is yes. The second half of the original question, whether Europe also became less equally happy, turns out to have no scale-free answer at all. Every common ruler for spread points toward less dispersion, and the gap between countries closed fastest.

Assumption ledger

Every statistic assumes something about the scale

An average treats the distance from 3 to 4 as equal to the distance from 8 to 9. Nobody asked respondents to agree to that. This study therefore ranks its evidence by how little it assumes.

StatisticAssumes about the scaleRole here
Mean on 0-10Equal steps between all eleven answersReported, but not relied on
Standard deviationEqual steps, squared distancesOne of three dispersion rulers
Share 0-4 or 9-10That one chosen cut-off mattersTwo of the ten cut-offs tested
Threshold ladderOnly that higher answers mean more satisfactionThe decisive test
Probability of superiorityOnly the order of answersScale-free effect size

Happier? Test every cut-off, not one

A cut-off splits the scale into "this answer or lower" and "higher". The ten possible cut-offs, from 0|1 to 9|10, contain the whole distribution: the change in the mean equals the sum of the ten changes in the share above them, and the familiar tails are two of the ten. The share answering 0-4 is cut-off 4|5; the share answering 9 or 10 is cut-off 8|9.

All ten moved in the same direction. The share above the middle cut-offs, 5|6 and 6|7, rose by more than four percentage points; the share above 0|1, the step out of the very bottom answer, rose by a quarter of a point. Nine of the ten changes stay clear of zero even in a band that holds for all ten cut-offs at once. The tenth sits at the bottom of the scale, where only 1.6 percent of 2010/11 answers were, and it is small but significant in the one-sided test that dominance requires.

Fig. 01Happier at every cut-off?

The share of Europeans above every one of the ten cut-offs rose.

Change in the share answering above each cut-off on the 0-10 scale, 2010/11 to 2023/24, 17 matched countries weighted equally. Switch the lens to read the same ten estimates as a test of spread.

Outcome
Read the ladder as

All ten cut-offs moved up. 9 of ten exclude zero inside the simultaneous band (critical value 2.67); the intersection-union test establishes dominance (largest one-sided p = 0.026).

Life satisfaction: change in the share above each cut-off, 2010/11 to 2023/24Ten cut-offs from 0|1 to 9|10. All ten cut-offs moved up. 9 of ten exclude zero inside the simultaneous band (critical value 2.67); the intersection-union test establishes dominance (largest one-sided p = 0.026).+0.3+0.7+1.2+2.2+2.5+4.2+4.7+3.7+2.4+1.2
Life satisfaction: change in the share above each cut-offAll ten cut-offs moved up. 9 of ten exclude zero inside the simultaneous band (critical value 2.67); the intersection-union test establishes dominance (largest one-sided p = 0.026).+0.3+0.7+1.2+2.2+2.5+4.2+4.7+3.7+2.4+1.2

Equal-step mean change +0.23 points (95% interval +0.18 to +0.29) is the sum of these ten changes.

  • Estimate, simultaneously significant
  • Estimate, not significant after simultaneous correction
  • Pointwise 95% interval
  • Simultaneous 95% band (all ten cut-offs)

Pointwise 95% t intervals with Satterthwaite degrees of freedom; the wider band holds simultaneously for all ten cut-offs. The familiar 0-4 and 9-10 tails are cut-offs 4|5 and 8|9.

European Social Survey, ESS5 edition 3.6 and ESS11 edition 4.2; original analysis weights; Taylor-linearized design uncertainty.

Exact values and uncertainty
Life-satisfaction threshold changes, 17 matched countries
Cut-offShare above, 2010/112023/24Change95% intervalSimultaneous band
0|198.4%98.6%+0.25 pp0.00 pp to +0.50 pp−0.09 pp to +0.60 pp
1|297.2%97.9%+0.71 pp+0.36 pp to +1.05 pp+0.24 pp to +1.17 pp
2|394.9%96.2%+1.25 pp+0.79 pp to +1.71 pp+0.62 pp to +1.88 pp
3|490.9%93.0%+2.16 pp+1.52 pp to +2.81 pp+1.28 pp to +3.04 pp
4|586.7%89.2%+2.53 pp+1.74 pp to +3.32 pp+1.46 pp to +3.61 pp
5|674.6%78.9%+4.23 pp+3.21 pp to +5.25 pp+2.84 pp to +5.62 pp
6|765.9%70.5%+4.66 pp+3.53 pp to +5.80 pp+3.12 pp to +6.21 pp
7|849.2%53.0%+3.74 pp+2.61 pp to +4.86 pp+2.20 pp to +5.27 pp
8|923.7%26.0%+2.39 pp+1.53 pp to +3.25 pp+1.22 pp to +3.57 pp
9|109.6%10.9%+1.24 pp+0.64 pp to +1.84 pp+0.42 pp to +2.06 pp

Sup-t critical value 2.670 over ten cut-offs; intersection-union one-sided p-value 0.026. n = 31,322 (2010/11) and 27,320 (2023/24).

Dominance is what makes the result independent of the ruler. Any increasing relabeling that keeps the endpoints at 0 and 10 turns the mean change into a weighted average of the ten single-cut-off changes, each weighted by the width the relabeling gives its step. If all ten are positive, every weighted average is positive. The instrument below lets you test this by hand, and it lets you try the same thing on any of the seventeen countries.

Fig. 02Can a different ruler reverse it?

Stretch the ruler however you like: Europe's gain never turns into a loss.

Each column is one cut-off. Its height is the change in the share above it; its width is the step your ruler gives it. The columns' combined area is the mean change on your ruler.

Outcome
More rulers

Answer 1: 0.05 to 1.95. Neighbors stay fixed; the ruler ends remain at 0 and 10.

Mean change on your ruler+0.23points on a 0-10 range / the 17-country pool / life satisfaction
Published estimate, equal steps+0.2395% interval +0.18 to +0.29

No increasing relabeling can turn this gain into a loss. The smallest possible change is +0.03 points, the limit when the whole 0-10 range is spent on the step 0|1. The presets keep every other step 0.1 wide, so they stop just short of this limit.

Change in the share above each cut-off (height, pp) × your step (width) = contribution to the mean change+0.3+0.7+1.2+2.2+2.5+4.2+4.7+3.7+2.4+1.2012345678910Height: share above each cut-off, pp. Width: your step.+0.3+0.7+1.2+2.2+2.5+4.2+4.7+3.7+2.4+1.2012345678910

Use the answer selector and position control for precise touch adjustment. With a mouse, drag an answer on the ruler; with a keyboard, focus it and use the arrow keys (Page Up/Down for whole steps). Faint outlines keep the published equal-step columns in view.

Every possible ruler for the 17-country pool: the mean change must lie inside this band

0worst +0.03best +0.47
Exact values and uncertainty
17 matched countries (pooled) / Life satisfaction / 2010/11 to 2023/24
All of the 0-10 range on stepMean changeSimultaneous bound
0|1+0.025−0.009 to +0.060
1|2+0.071+0.024 to +0.117
2|3+0.125+0.062 to +0.188
3|4+0.216+0.128 to +0.304
4|5+0.253+0.146 to +0.361
5|6+0.423+0.284 to +0.562
6|7+0.466+0.312 to +0.621
7|8+0.374+0.220 to +0.527
8|9+0.239+0.122 to +0.357
9|10+0.124+0.042 to +0.206

Every increasing relabeling that keeps 0 and 10 fixed gives a weighted average of these ten values, so its mean change lies between +0.025 and +0.466 points.

For the 17-country pool, the relabeling range uses point estimates. With the simultaneous band, its least favourable ruler is bounded below by −0.009 points; only the band at cut-off 0|1, the step out of the very bottom answer that 1.6% gave in 2010/11, reaches zero.

European Social Survey, ESS5 edition 3.6 and ESS11 edition 4.2; original analysis weights; Taylor-linearized design uncertainty.

The least favourable ruler spends the entire range on the step from 0 to 1, where the rise was smallest. Even then the change is +0.025 points. Its opposite concentrates the range on the step from 6 to 7, and every relabeling in between gives a gain between those two extremes, +0.03 and +0.47 points.

How large is the gain? On an ordinal scale the natural effect size is the probability that a randomly drawn 2023/24 answer exceeds a randomly drawn 2010/11 answer, with ties counted as half. It is 52.8 percent (95% interval 52.1 to 53.5), a modest shift that no relabeling can change. The transport view expresses the same movement in score points. The least-cost way to turn the 2010/11 distribution into the 2023/24 one moves 23.2 percent of answers up by exactly one category, and none down.

Fig. 03How much did the distribution move?

The whole shift equals 23.2% of answers moving up one step. None move down.

Answers from both rounds ordered from lowest to highest and aligned by rank, 17 matched countries. Where the category at a rank differs, the band between the bars is marked.

The whole shift equals 23.2% of answers moving up one step. None move down.Two aligned 100% bars. Monotone coupling: at equal ranks, 23.2% of answers sit one category higher in 2023/24 than in 2010/11; no rank sits lower and none sits more than one category higher.2010/11: answers ordered from lowest to highest2023/24: the same ranks5→66→77→8456789105678910
The whole shift equals 23.2% of answers moving up one step. None move down.Two aligned 100% bars. Monotone coupling: at equal ranks, 23.2% of answers sit one category higher in 2023/24 than in 2010/11; no rank sits lower and none sits more than one category higher.2010/112023/246→756789105678910

Hover or focus a band: it shows which answer at a given rank changed category.

  • Same rank, one answer higher in 2023/24
  • Same rank, same answer

Wasserstein distance 0.232 score points (simultaneous 95% bounds 0.134 to 0.330); downward movement at most 0.0009 points. This is accounting between two independent samples, not people changing their answers.

European Social Survey, ESS5 edition 3.6 and ESS11 edition 4.2; original analysis weights; Taylor-linearized design uncertainty.

Exact values and uncertainty
2010/11 answer2023/24 answer at the same rankShare of answers
001.37%
010.25%
110.47%
120.71%
221.03%
231.25%
331.89%
342.16%
441.64%
452.53%
557.82%
564.23%
664.09%
674.66%
7712.91%
783.74%
8823.18%
892.39%
9912.78%
9101.24%
10109.64%

Less equally happy? Every ruler for spread says no

Spread is harder, because there is no single ruler for it. The standard deviation assumes equal steps and squares them. On a bounded scale it also falls mechanically as the mean approaches the ceiling, since the largest possible variance at mean μ is μ(10 − μ) (Bhatia and Davis 2000). A ceiling-adjusted standard deviation divides that effect out, and an ordinal index uses only the cumulative shares. All three fell: the standard deviation by 5.2 percent, the ceiling-adjusted version by 3.1 percent and ordinal variation by 6.3 percent, each with an interval that excludes zero.

Fig. 04Less equally happy?

Every ruler for spread points the same way: Europe became less dispersed, and the gap between countries closed fastest.

Relative change in five dispersion measures between 2010/11 and 2023/24, 17 matched countries, with 95% intervals from log ratios of independent rounds.

Every ruler for spread points the same way: Europe became less dispersed, and the gap between countries closed fastest.Standard deviation: life satisfaction −5.2%, happiness −4.4%. Ceiling-adjusted SD: life satisfaction −3.1%, happiness −1.8%. Ordinal variation: life satisfaction −6.3%, happiness −5.3%. Spread of country means: life satisfaction −18.6%, happiness −27.7%. Typical spread within a country: life satisfaction −3.3%, happiness −2.4%.Standard deviationCardinal: equal steps, squared distancesCeiling-adjusted SDSD / its Bhatia-Davis maximum at the meanOrdinal variationBerry-Mielke IOV = 1 − ℓ² (order only)Spread of country meansSD of the 17 national meansTypical spread within a countryRoot mean within-country variance
Every ruler for spread points the same way: Europe became less dispersed, and the gap between countries closed fastest.Standard deviation: life satisfaction −5.2%, happiness −4.4%. Ceiling-adjusted SD: life satisfaction −3.1%, happiness −1.8%. Ordinal variation: life satisfaction −6.3%, happiness −5.3%. Spread of country means: life satisfaction −18.6%, happiness −27.7%. Typical spread within a country: life satisfaction −3.3%, happiness −2.4%.Standard deviationCardinal: equal steps, squared distancesCeiling-adjusted SDSD / its Bhatia-Davis maximum at the meanOrdinal variationBerry-Mielke IOV = 1 − ℓ² (order only)Spread of country meansSD of the 17 national meansTypical spread within a countryRoot mean within-country variance

Every interval lies left of zero: less dispersion on each ruler.

  • Life satisfaction
  • Happiness

The fall in pooled variance splits into 45% between countries and 55% within them (life satisfaction). Dispersion of a bounded self-report is not interpersonal welfare inequality.

European Social Survey, ESS5 edition 3.6 and ESS11 edition 4.2; original analysis weights; Taylor-linearized design uncertainty.

Exact values and uncertainty
MeasureLife satisfaction95% intervalHappiness95% interval
Standard deviation−5.2%−7.0% to −3.4%−4.4%−6.3% to −2.4%
Ceiling-adjusted SD−3.1%−4.6% to −1.5%−1.8%−3.5% to −0.1%
Ordinal variation−6.3%−8.1% to −4.3%−5.3%−7.2% to −3.3%
Spread of country means−18.6%−24.3% to −12.4%−27.7%−34.1% to −20.7%
Typical spread within a country−3.3%−4.9% to −1.7%−2.4%−4.2% to −0.6%

Absolute changes: SD −0.118 points; ordinal variation −0.031; between-country variance −0.230 (95% interval −0.310 to −0.149); within-country variance −0.286 (−0.428 to −0.143).

The larger movement is between countries. Equal-country pooling makes the pooled variance the sum of two parts, the average variance within a country and the variance of the country means. Both fell, but the spread of national means fell far faster, by 18.6 percent (95% interval 12.4 to 24.3), and it accounts for 45 percent of the decline in pooled variance. The countries that started lowest gained most.

Fig. 05Did countries converge?

The least satisfied countries of 2010/11 gained most; the most satisfied stood still or slipped.

Mean life satisfaction in 2010/11 against its change to 2023/24, with 95% intervals on the change. Circles rose where detectable; hollow circles show no detectable change; the square fell at the top.

The least satisfied countries of 2010/11 gained most; the most satisfied stood still or slipped.Seventeen countries. The spread of national means fell from 0.83 to 0.67 points.UAHUPTHRFRSKEESIPLDEESBEILNLNOSEFI
The least satisfied countries of 2010/11 gained most; the most satisfied stood still or slipped.Seventeen countries. The spread of national means fell from 0.83 to 0.67 points.UAHUPTHRFRSKEESIPLDEESBEILNLNOSEFI

Hover or focus a country for its level, change and interval.

SD of country means −18.6% (95% interval −24.3% to −12.4%). Sampling noise can mimic convergence, but with national standard errors near 0.05 points its expected contribution to the between-country variance is about 0.004 of 0.68. Association, not mechanism.

European Social Survey, ESS5 edition 3.6 and ESS11 edition 4.2; original analysis weights; Taylor-linearized design uncertainty.

Exact values and uncertainty
Country2010/11 meanChange95% intervalDistribution verdict
Ukraine4.89+0.24−0.03 to +0.52Up, none significantly down
Hungary5.83+0.81+0.59 to +1.02Every ruler: up
Portugal6.09+0.43+0.23 to +0.64Up, none significantly down
Croatia6.30+0.87+0.66 to +1.07Every ruler: up
France6.33+0.41+0.22 to +0.60Up, none significantly down
Slovakia6.48+0.03−0.20 to +0.26Up, none significantly down
Estonia6.57+0.35+0.20 to +0.51Up, none significantly down
Slovenia6.97+0.56+0.39 to +0.72Up, none significantly down
Poland6.98+0.46+0.29 to +0.62Every ruler: up
Germany7.24+0.27+0.11 to +0.43Up, none significantly down
Spain7.32+0.15+0.03 to +0.27Up, none significantly down
Belgium7.46−0.06−0.19 to +0.08No detectable change
Israel7.50−0.38−0.59 to −0.17Down, none significantly up
Netherlands7.77+0.100.00 to +0.20Up, none significantly down
Norway7.90−0.16−0.31 to −0.01No detectable change
Sweden7.90−0.09−0.23 to +0.05No detectable change
Finland7.92−0.05−0.16 to +0.06No detectable change

None of this certifies that Europe became more equal in a scale-free sense. The ordinal test for spread, Allison and Foster's (2004) median-preserving spread, compares two distributions that share a median and asks whether one lies further from it at every cut-off. Switch the threshold ladder to its spread lens and the obstacle becomes visible. For life satisfaction the median itself rose from 7 to 8; for happiness, answers moved toward the shared median from below but away from it above. Neither distribution is a spread of the other. Spread fell on every ruler applied here, but a reader still has to choose one of them.

Everywhere? Seventeen countries, seventeen verdicts

The pooled verdict is not a national one. Twelve countries rose where the data can detect a change and never fell significantly at any cut-off. Three of them, Croatia, Hungary and Poland, pass the full dominance test. Seven, adding Slovenia, Germany, Estonia and, narrowly, France, gained on every ruler in point estimates. For the other five, Portugal, the Netherlands, Spain, Ukraine and Slovakia, the least favourable ruler reverses a small gain.

Fig. 06Which countries are happier on every ruler?

7 countries gained on every ruler. For the others, the verdict depends on how the answers are spaced.

Life satisfaction, 2010/11 to 2023/24. Each bar is the full range of mean changes that any increasing relabeling of the eleven answers can produce; the dot is the published equal-step change.

7 countries gained on every ruler. For the others, the verdict depends on how the answers are spaced.Bars span the smallest and largest mean change over all increasing relabelings; dots mark equal steps with 95% intervals. Positive on every ruler: Croatia, Hungary, Poland, Slovenia, Germany, Estonia, France.Pooled 17 countriesCroatiaHungaryPolandSloveniaGermanyEstoniaFrancePortugalNetherlandsSpainUkraineSlovakiaBelgiumFinlandSwedenNorwayIsrael
7 countries gained on every ruler. For the others, the verdict depends on how the answers are spaced.Bars span the smallest and largest mean change over all increasing relabelings; dots mark equal steps with 95% intervals. Positive on every ruler: Croatia, Hungary, Poland, Slovenia, Germany, Estonia, France.Pooled 17CroatiaHungaryPolandSloveniaGermanyEstoniaFrancePortugalNetherlandsSpainUkraineSlovakiaBelgiumFinlandSwedenNorwayIsrael

Hover or focus a row for its relabeling range.

  • Every increasing relabeling, country rose where detectable
  • Every increasing relabeling, no detectable rise
  • Equal steps, with 95% interval

Ranges use point estimates. Sorted by the distribution verdict, then by the least favourable relabeling. A range that crosses zero means some reasonable-looking ruler reverses the sign.

European Social Survey, ESS5 edition 3.6 and ESS11 edition 4.2; original analysis weights; Taylor-linearized design uncertainty.

Exact values and uncertainty
CountryLeast favourable rulerEqual stepsMost favourable rulerVerdict
Pooled 17 countries+0.025+0.23 (+0.18 to +0.29)+0.466Every ruler: up
Croatia+0.177+0.87 (+0.66 to +1.07)+1.629Every ruler: up
Hungary+0.170+0.81 (+0.59 to +1.02)+1.628Every ruler: up
Poland+0.082+0.46 (+0.29 to +0.62)+0.785Every ruler: up
Slovenia+0.059+0.56 (+0.39 to +0.72)+1.167Up, none significantly down
Germany+0.044+0.27 (+0.11 to +0.43)+0.647Up, none significantly down
Estonia+0.034+0.35 (+0.20 to +0.51)+0.813Up, none significantly down
France+0.001+0.41 (+0.22 to +0.60)+0.952Up, none significantly down
Portugal0.000+0.43 (+0.23 to +0.64)+0.916Up, none significantly down
Netherlands−0.033+0.10 (0.00 to +0.20)+0.562Up, none significantly down
Spain−0.142+0.15 (+0.03 to +0.27)+0.602Up, none significantly down
Ukraine−0.292+0.24 (−0.03 to +0.52)+0.709Up, none significantly down
Slovakia−0.324+0.03 (−0.20 to +0.26)+0.267Up, none significantly down
Belgium−0.202−0.06 (−0.19 to +0.08)+0.043No detectable change
Finland−0.204−0.05 (−0.16 to +0.06)+0.092No detectable change
Sweden−0.436−0.09 (−0.23 to +0.05)+0.146No detectable change
Norway−0.461−0.16 (−0.31 to −0.01)+0.030No detectable change
Israel−1.168−0.38 (−0.59 to −0.17)+0.008Down, none significantly up

The country distributions show why. In Hungary and Croatia the whole distribution moved. In Spain the top thickened while the lower tail grew slightly, so a ruler that stretches the bottom of the scale turns its gain into a loss. Belgium, Finland, Norway and Sweden, which started among the most satisfied, show no detectable change at any cut-off. Israel lost at the top, with far fewer answers of 8 or more.

Fig. 07What moved inside each country?

Seventeen distributions: where the lower tail thinned, where the top thickened, and where nothing moved.

Exact 0-10 life-satisfaction distributions, 2023/24 filled and 2010/11 outlined, on a common scale. The strip under each panel marks the ten cut-offs: solid where the share above rose significantly (simultaneous band), hatched where it fell significantly.

  1. Croatia+0.87Every ruler: upCroatia: life-satisfaction distribution2010/11 outline, 2023/24 filled; mean change +0.87 points.
  2. Hungary+0.81Every ruler: upHungary: life-satisfaction distribution2010/11 outline, 2023/24 filled; mean change +0.81 points.
  3. Poland+0.46Every ruler: upPoland: life-satisfaction distribution2010/11 outline, 2023/24 filled; mean change +0.46 points.
  4. Slovenia+0.56Up, none significantly downSlovenia: life-satisfaction distribution2010/11 outline, 2023/24 filled; mean change +0.56 points.
  5. Germany+0.27Up, none significantly downGermany: life-satisfaction distribution2010/11 outline, 2023/24 filled; mean change +0.27 points.
  6. Estonia+0.35Up, none significantly downEstonia: life-satisfaction distribution2010/11 outline, 2023/24 filled; mean change +0.35 points.
  7. France+0.41Up, none significantly downFrance: life-satisfaction distribution2010/11 outline, 2023/24 filled; mean change +0.41 points.
  8. Portugal+0.43Up, none significantly downPortugal: life-satisfaction distribution2010/11 outline, 2023/24 filled; mean change +0.43 points.
  9. Netherlands+0.10Up, none significantly downNetherlands: life-satisfaction distribution2010/11 outline, 2023/24 filled; mean change +0.10 points.
  10. Spain+0.15Up, none significantly downSpain: life-satisfaction distribution2010/11 outline, 2023/24 filled; mean change +0.15 points.
  11. Ukraine+0.24Up, none significantly downUkraine: life-satisfaction distribution2010/11 outline, 2023/24 filled; mean change +0.24 points.
  12. Slovakia+0.03Up, none significantly downSlovakia: life-satisfaction distribution2010/11 outline, 2023/24 filled; mean change +0.03 points.
  13. Belgium−0.06No detectable changeBelgium: life-satisfaction distribution2010/11 outline, 2023/24 filled; mean change −0.06 points.
  14. Finland−0.05No detectable changeFinland: life-satisfaction distribution2010/11 outline, 2023/24 filled; mean change −0.05 points.
  15. Sweden−0.09No detectable changeSweden: life-satisfaction distribution2010/11 outline, 2023/24 filled; mean change −0.09 points.
  16. Norway−0.16No detectable changeNorway: life-satisfaction distribution2010/11 outline, 2023/24 filled; mean change −0.16 points.
  17. Israel−0.38Down, none significantly upIsrael: life-satisfaction distribution2010/11 outline, 2023/24 filled; mean change −0.38 points.
  • 2023/24 share
  • 2010/11 share
  • Share above rose, simultaneously significant
  • Rose, not significant
  • Fell, not significant
  • Fell, significant

Unsmoothed probability masses on the eleven answer categories; no kernel density or interpolation. Shaded columns mark the 0-4 and 9-10 tails. Select a panel to open it in the comparison desk.

European Social Survey, ESS5 edition 3.6 and ESS11 edition 4.2; original analysis weights; Taylor-linearized design uncertainty.

Exact values and uncertainty
Country012345678910
Croatia 2010/112.6%1.4%4.1%5.1%4.6%17.6%11.4%17.6%20.0%7.7%7.8%
Croatia 2023/240.9%0.7%1.9%3.2%3.2%12.5%8.1%18.1%24.0%13.1%14.1%
Hungary 2010/112.8%3.1%5.3%7.5%7.6%18.6%12.1%14.4%14.2%8.2%6.3%
Hungary 2023/241.1%1.2%3.2%4.9%4.9%13.2%12.2%19.7%20.7%9.2%9.6%
Poland 2010/111.2%1.3%2.4%3.6%3.4%12.8%9.5%15.3%25.7%13.8%11.0%
Poland 2023/240.4%0.5%1.1%2.2%2.7%11.5%8.0%17.5%26.8%12.1%17.3%
Slovenia 2010/111.4%0.7%1.7%3.3%3.5%16.9%7.1%15.8%24.9%15.0%9.7%
Slovenia 2023/240.8%0.5%0.7%1.6%2.6%9.7%7.1%16.6%29.5%17.4%13.6%
Germany 2010/110.9%0.8%1.7%3.7%3.4%10.9%6.3%17.1%26.9%15.3%12.9%
Germany 2023/240.4%0.5%1.4%2.2%3.3%8.5%7.7%14.2%29.4%18.9%13.4%
Estonia 2010/112.1%1.1%2.0%5.2%4.9%16.1%10.6%17.6%22.2%10.2%8.0%
Estonia 2023/240.7%0.6%1.9%4.1%3.9%13.5%9.2%20.1%25.4%12.2%8.3%
France 2010/113.4%1.0%4.1%6.3%4.8%14.9%10.7%16.7%21.2%8.1%8.8%
France 2023/242.5%1.2%2.7%4.2%4.3%11.3%9.6%19.0%25.4%11.1%8.8%
Portugal 2010/111.7%1.7%3.8%5.9%6.8%18.0%14.7%18.0%18.7%6.5%4.3%
Portugal 2023/241.7%0.7%2.3%4.4%5.6%16.5%12.2%19.7%21.1%9.3%6.7%
Netherlands 2010/110.3%0.2%0.3%0.9%1.2%3.3%5.0%20.1%43.3%18.6%6.8%
Netherlands 2023/240.6%0.0%0.3%0.3%1.2%3.2%5.7%17.4%40.3%24.1%6.9%
Spain 2010/110.7%0.2%0.4%1.8%3.1%8.7%9.8%22.6%30.8%12.8%9.2%
Spain 2023/240.8%0.5%1.3%1.9%2.3%7.0%7.9%21.0%29.1%13.9%14.1%
Ukraine 2010/117.0%4.8%6.4%12.7%10.2%18.5%10.1%14.3%7.9%4.7%3.4%
Ukraine 2023/249.9%3.1%6.1%7.6%7.4%25.2%7.9%10.4%10.8%3.3%8.3%
Slovakia 2010/111.7%2.0%2.2%3.8%6.5%17.2%9.8%18.8%21.1%9.5%7.4%
Slovakia 2023/240.7%0.8%1.8%7.3%8.8%14.3%11.2%17.7%18.4%9.5%9.5%
Belgium 2010/110.4%0.4%0.6%2.1%2.3%6.5%7.8%21.0%34.5%17.0%7.4%
Belgium 2023/240.9%0.1%1.3%1.2%3.5%7.2%6.9%19.5%34.4%17.7%7.2%
Finland 2010/110.1%0.1%0.7%1.3%2.0%4.0%4.5%13.4%35.8%28.9%9.2%
Finland 2023/240.3%0.0%0.8%1.7%1.9%5.2%4.5%13.8%33.0%28.7%10.1%
Sweden 2010/110.1%0.3%0.8%1.2%1.9%6.0%5.2%13.6%32.2%23.2%15.4%
Sweden 2023/240.6%0.1%0.6%1.6%2.7%3.3%5.8%16.0%35.1%21.8%12.4%
Norway 2010/110.2%0.3%0.5%1.3%1.7%5.8%6.2%14.0%30.9%23.4%15.6%
Norway 2023/240.2%0.1%0.8%2.0%2.5%5.8%7.2%16.1%29.4%22.8%13.2%
Israel 2010/111.0%0.5%1.7%3.1%3.1%9.2%8.2%12.7%24.6%15.6%20.4%
Israel 2023/240.9%1.6%1.3%2.9%3.8%8.3%10.1%22.0%24.6%12.8%11.5%

The desk puts any two countries on the same scale and applies the same tests to both. It opens on Spain, whose gain depends on the ruler, and Hungary, whose gain does not.

Fig. 08How do two countries compare?

Compare any two of the seventeen: same scale, same tests, same rulers.

Choose a focus and a comparison country. Both columns share one scale for the distribution, the cut-off strip and the verdicts; the history below uses every round with a reviewed design.

Outcome

 

Focus

Spain

Up where detectable, never significantly down
Spain: life satisfaction distribution, 2010/11 and 2023/24Weighted shares at each answer 0-10. 2010/11 n=1885; 2023/24 n=1835.

2010/11 outline, 2023/24 filled. Shaded: 0-4 and 9-10.

Mean change
+0.1595% interval +0.03 to +0.27
Later answer higher
53.5%probability of superiority, 95% interval 51.6% to 55.3%
Every ruler
−0.14 to +0.60mean change under all increasing relabelings
Movement
0.19 up / 0.04 downscore points of transport between the two samples

Spread test: unranked in the point estimates, and the simultaneous band cannot rule out a spread ordering. Significant rises at 8|9, 9|10; significant falls at no cut-off.

Spain: mean life satisfaction by roundRounds 2, 3, 4, 5, 6, 7, 8, 9, 11. Shaded columns mark the 2010/11 and 2023/24 endpoints; both desk columns share one vertical scale.

Spain: every round with a reviewed design, mean and 95% interval.

Comparison

Hungary

Happier on every ruler
Hungary: life satisfaction distribution, 2010/11 and 2023/24Weighted shares at each answer 0-10. 2010/11 n=1555; 2023/24 n=2102.

2010/11 outline, 2023/24 filled. Shaded: 0-4 and 9-10.

Mean change
+0.8195% interval +0.59 to +1.02
Later answer higher
59.5%probability of superiority, 95% interval 56.9% to 62.1%
Every ruler
+0.17 to +1.63mean change under all increasing relabelings
Movement
0.81 up / 0.00 downscore points of transport between the two samples

Spread test: the median moved from 6 to 7, so the Allison-Foster ordering does not apply. Significant rises at 0|1, 1|2, 2|3, 3|4, 4|5, 5|6, 6|7, 7|8, 9|10; significant falls at no cut-off.

Hungary: mean life satisfaction by roundRounds 1, 2, 5, 6, 7, 8, 9, 10, 11. Shaded columns mark the 2010/11 and 2023/24 endpoints; both desk columns share one vertical scale.

Hungary: every round with a reviewed design, mean and 95% interval.

The desk loads reviewed aggregate files only and verifies their analysis fingerprint. No respondent record is queried or recalculated in the browser.

European Social Survey, ESS5 edition 3.6 and ESS11 edition 4.2; original analysis weights; Taylor-linearized design uncertainty.

Does it hold?

A result like this should survive the choices that produced it. The nine countries observed in every round tell when the shift happened: nothing detectable through 2014/15, then a rise at every cut-off from 2016/17 onward, which persists into 2023/24.

Fig. 09When did it happen?

The ladder stayed flat until 2014/15, then rose at every cut-off from 2016/17 onward.

The threshold ladder of Figure 1, re-estimated for each later round against 2010/11, for the nine countries observed in every round from Round 5 to Round 11.

  1. 2012/13+0.01 / No detectable change

    2012/13 against 2010/11: change in the share above each cut-offMean change +0.01 points; 0 cut-offs significantly up and 0 down.

  2. 2014/15−0.07 / No detectable change

    2014/15 against 2010/11: change in the share above each cut-offMean change −0.07 points; 0 cut-offs significantly up and 0 down.

  3. 2016/17+0.19 / Every ruler: up

    2016/17 against 2010/11: change in the share above each cut-offMean change +0.19 points; 8 cut-offs significantly up and 0 down.

  4. 2018-20+0.24 / Every ruler: up

    2018-20 against 2010/11: change in the share above each cut-offMean change +0.24 points; 10 cut-offs significantly up and 0 down.

  5. 2020-22+0.48 / Every ruler: up

    2020-22 against 2010/11: change in the share above each cut-offMean change +0.48 points; 10 cut-offs significantly up and 0 down.

  6. 2023/24+0.27 / Every ruler: up

    2023/24 against 2010/11: change in the share above each cut-offMean change +0.27 points; 8 cut-offs significantly up and 0 down.

Each panel compares one round with 2010/11 on the same scale.

  • Estimate, simultaneously significant
  • Estimate, not significant after simultaneous correction
  • Pointwise 95% interval
  • Simultaneous 95% band (all ten cut-offs)

Round 10 was fielded in 2020-22 under pandemic conditions; its larger rise coincides with video interviewing (up to 35% of interviews in Norway) and should not be read as a pandemic effect. The endpoint verdict does not depend on it.

European Social Survey, ESS5 edition 3.6 and ESS11 edition 4.2; original analysis weights; Taylor-linearized design uncertainty.

Exact values and uncertainty
RoundMean change95% intervalVerdictCut-offs significantly up
2012/13+0.01−0.06 to +0.09No detectable change0 of 10
2014/15−0.07−0.14 to +0.01No detectable change0 of 10
2016/17+0.19+0.12 to +0.26Every ruler: up8 of 10
2018-20+0.24+0.17 to +0.31Every ruler: up10 of 10
2020-22+0.48+0.41 to +0.55Every ruler: up10 of 10
2023/24+0.27+0.19 to +0.34Every ruler: up8 of 10

The specification grid changes one choice at a time: happiness instead of life satisfaction, population instead of equal-country weights, nine or five countries instead of seventeen, 2002/03 instead of 2010/11 as the starting point, and each of the seventeen countries left out in turn. The gain ranges from 0.16 to 0.57 points, no specification shows a significant fall at any cut-off, and every dispersion index falls in every row. The standard deviation and ordinal variation fall significantly everywhere; the ceiling-adjusted standard deviation does so in most rows. The relabeling claim is slightly less universal: in two rows, the five long-run countries weighted by population, the least favourable ruler reverses the point estimate, to −0.02 points for life satisfaction and −0.21 for happiness. Population weighting gives the largest countries more influence and the bottom cut-off less precision, so its verdict weakens from full dominance to a rise with no significant fall at any cut-off.

Fig. 10Does it hold?

Twelve specifications and seventeen leave-one-out runs: the mean gain stays positive, no cut-off falls significantly, and dispersion never rises.

Each row changes one analytical choice: outcome, country sample, time span or pooling weight. Glyphs share a scale within each column; dispersion columns show relative change.

SpecificationMean changeSuperiority − 50%Least favourable rulerSDCeiling-adjusted SDOrdinal variationVerdict
Life satisfaction17 countries, 2010/11-2023/24, equal weights+0.23 +0.18, +0.29+2.8 pp +2.1 pp, +3.5 pp+0.025 bound −0.009−5.2% −7.0%, −3.4%−3.1% −4.6%, −1.5%−6.3% −8.1%, −4.3%Every ruler: up
Life satisfaction17 countries, population weights+0.31 +0.23, +0.39+3.9 pp +3.0 pp, +4.9 pp+0.016 bound −0.045−4.2% −6.8%, −1.5%−1.5% −3.8%, +0.8%−5.7% −8.5%, −2.9%Up, none significantly down
Life satisfaction9 countries in every round, equal weights+0.27 +0.19, +0.34+3.2 pp +2.3 pp, +4.2 pp+0.040 bound −0.011−7.8% −10.3%, −5.1%−5.2% −7.4%, −2.9%−8.8% −11.5%, −6.1%Every ruler: up
Life satisfaction9 countries, population weights+0.34 +0.23, +0.45+4.2 pp +2.9 pp, +5.5 pp+0.048 bound −0.126−7.3% −11.3%, −3.2%−4.6% −8.1%, −0.9%−9.3% −13.2%, −5.2%Up, none significantly down
Life satisfaction5 countries, 2002/03-2023/24, equal weights+0.57 +0.48, +0.67+6.1 pp +5.0 pp, +7.3 pp+0.081 bound −0.072−15.7% −18.1%, −13.1%−11.2% −13.5%, −8.9%−17.8% −20.3%, −15.2%Up, none significantly down
Life satisfaction5 countries, population weights+0.48 +0.32, +0.64+4.9 pp +3.1 pp, +6.7 pp−0.020 bound −0.274−12.8% −17.1%, −8.3%−9.7% −13.7%, −5.6%−14.8% −19.2%, −10.2%Up, none significantly down
Happiness17 countries, 2010/11-2023/24, equal weights+0.22 +0.17, +0.26+3.2 pp +2.5 pp, +3.8 pp+0.013 bound −0.007−4.4% −6.3%, −2.4%−1.8% −3.5%, −0.1%−5.3% −7.2%, −3.3%Every ruler: up
Happiness17 countries, population weights+0.27 +0.21, +0.34+4.0 pp +3.0 pp, +4.9 pp+0.015 bound −0.013−5.3% −7.9%, −2.5%−2.1% −4.6%, +0.4%−6.2% −8.9%, −3.4%Up, none significantly down
Happiness9 countries in every round, equal weights+0.25 +0.18, +0.31+3.7 pp +2.8 pp, +4.6 pp+0.027 bound +0.003−7.0% −9.9%, −4.0%−3.9% −6.4%, −1.4%−7.7% −10.6%, −4.7%Every ruler: up
Happiness9 countries, population weights+0.27 +0.19, +0.35+4.0 pp +2.8 pp, +5.3 pp+0.021 bound −0.020−8.0% −11.6%, −4.4%−4.8% −8.1%, −1.5%−8.3% −11.8%, −4.7%Up, none significantly down
Happiness5 countries, 2002/03-2023/24, equal weights+0.38 +0.30, +0.46+4.6 pp +3.4 pp, +5.7 pp+0.042 bound −0.079−13.5% −16.4%, −10.6%−9.4% −12.0%, −6.7%−15.3% −18.1%, −12.5%Up, none significantly down
Happiness5 countries, population weights+0.16 +0.04, +0.28+1.0 pp −0.8 pp, +2.7 pp−0.205 bound −0.506−13.9% −18.3%, −9.3%−12.3% −16.2%, −8.1%−14.5% −18.8%, −9.9%Up, none significantly down
Life satisfactionLeave one country out (17 runs)+0.19 to +0.27+2.3 pp to +3.4 pp+0.016 to +0.045−6.9% to −4.3%−4.6% to −2.4%−7.6% to −5.2%No run falls at any cut-off

Intervals are 95% t intervals. The least-favourable-ruler glyph spans the point estimate and its simultaneous lower bound; it is highlighted only where that bound stays above zero, and two population-weighted long-run rows admit a ruler that reverses the point estimate. The leave-one-out row shows the range across 17 re-estimates that each drop one country.

European Social Survey, ESS5 edition 3.6 and ESS11 edition 4.2; original analysis weights; Taylor-linearized design uncertainty.

Verdict

Europe got happier between 2010/11 and 2023/24 in the strongest sense an ordinal survey allows: the share above every cut-off rose, so no relabeling of the answers reverses the gain. It did not become less equally happy. Every ruler for spread applied here fell, and the gap between countries closed by almost a fifth. What remains open is narrower and more interesting: why the most satisfied countries stood still, and whether the video interviews that spread through their recent fieldwork explain part of it.

Failure boundary

What this evidence cannot say

  1. It does not rank latent happiness. The ruler test covers every increasing relabeling of the eleven reported answers. Bond and Lang's deeper objection concerns the unobserved feelings behind the answers; without assumptions about how people map feelings onto the card, even dominance in the answers cannot rank latent means. Evidence that people use the scale roughly linearly makes a reversal implausible, not impossible.
  2. It does not follow people. Each round is a new probability sample. The transport plan reconciles two distributions; it does not say that any person became happier.
  3. It does not identify causes. Recovery from the euro crisis, ageing, migration, survey mode and response behavior can all move a national distribution.
  4. Coverage changed in places. Ukraine's 2023/24 sample excludes Luhansk, Donetsk and Crimea, where the survey could not be conducted; its post-stratification weight is 1. Without Ukraine the pooled gain is +0.23 points.
  5. Interview mode changed where Europe stood still. Rounds 10 and 11 used video interviews in several countries. In 2023/24 almost half of Norway's interviews (48%), a quarter of Finland's (27%) and one in eight of Sweden's (12%) were by video, and these are three of the four countries whose distribution did not detectably move. A mode effect cannot be separated from real change there. Round 5 fieldwork also ran into early 2012 in Ireland and late 2011 in Croatia.
  6. Spread has no scale-free verdict. The indices agree, but the Allison-Foster ordering cannot rank the endpoints, and the 2010/11 median sits on the boundary between 7 and 8 (50.8% answered 7 or less).

Audit layer

Estimators, survey design and reproducibility

Every number on this page comes from one versioned aggregate analysis. The R survey package estimates it from respondent records with the reviewed designs; a separately written NumPy implementation reproduces every point estimate, standard error and critical value before publication.

Cumulative share and threshold change

Ft(k)=∑j≤kpt,j,k=0,…,9Δ(k)=F2023(k)−F2010(k)\begin{gathered}F_t(k)=\sum_{j\le k} p_{t,j},\qquad k=0,\dots,9\\ \Delta(k)=F_{2023}(k)-F_{2010}(k)\end{gathered}

p is the survey-weighted share at each answer 0-10. A negative threshold change means fewer answers at or below k and more above it.

The mean change is the area of the threshold changes

Δμ=∑j=010j (qj−pj)=−∑k=09Δ(k)\Delta\mu=\sum_{j=0}^{10} j\,(q_j-p_j)=-\sum_{k=0}^{9}\Delta(k)

Summation by parts: each threshold contributes one score step.

Every increasing relabeling with fixed endpoints

Δμφ=−∑k=09gk Δ(k)gk=φ(k+1)−φ(k)≥0,∑kgk=10⇒−10max⁡kΔ(k) ≤ Δμφ≤ −10min⁡kΔ(k)\begin{gathered}\Delta\mu_\varphi=-\sum_{k=0}^{9} g_k\,\Delta(k)\\ g_k=\varphi(k+1)-\varphi(k)\ge 0,\qquad \sum_k g_k=10\\ \Rightarrow\quad -10\max_k\Delta(k)\ \le\ \Delta\mu_\varphi\\ \le\ -10\min_k\Delta(k)\end{gathered}

Each relabeled mean change is a weighted average of the ten single-cut-off changes. If all ten point the same way, no relabeling can reverse the sign.

Intersection-union test of dominance

H0: Δ(k)≥0 for some kis rejected if, for every k,tk=Δ^(k)/se^k<−t0.95,νk\begin{gathered}H_0:\ \Delta(k)\ge 0\ \text{for some } k\\ \text{is rejected if, for every } k,\\ t_k=\hat\Delta(k)/\widehat{\mathrm{se}}_k<-t_{0.95,\nu_k}\end{gathered}

Berger (1982): a union null is rejected at level 5% when every component test rejects at 5%. Davidson and Duclos (2013) use the same minimum-t logic for dominance.

Simultaneous sup-t band

Δ^(k)±c^ se^kc^=q0.95(max⁡k∣Zk∣),Z∼tν(0,R^)\begin{gathered}\hat\Delta(k)\pm \hat c\,\widehat{\mathrm{se}}_k\\ \hat c=q_{0.95}\Big(\max_{k}|Z_k|\Big),\qquad Z\sim t_{\nu}\big(0,\hat R\big)\end{gathered}

R is the estimated correlation of the ten threshold changes; 100,000 seeded draws; independent NumPy replication within 0.03.

Probability of superiority

A=Pr⁡(X2023>X2010)+12Pr⁡(X2023=X2010)=∑jqj[F2010(j−1)+12pj],δ=2A−1\begin{gathered}A=\Pr(X_{2023}>X_{2010})\\ +\tfrac12\Pr(X_{2023}=X_{2010})\\ =\sum_{j} q_j\Big[F_{2010}(j-1)+\tfrac12 p_j\Big],\qquad \delta=2A-1\end{gathered}

Invariant to every increasing relabeling; delta-method standard error from both share covariances.

One-dimensional transport accounting

W1=∑k∣Δ(k)∣=U+D,Δμ=U−DU=∑kmax⁡(−Δ(k),0)D=∑kmax⁡(Δ(k),0)min⁡(U,D)=12(W1−∣Δμ∣)\begin{gathered}W_1=\sum_{k}|\Delta(k)|=U+D,\qquad \Delta\mu=U-D\\ U=\sum_k\max(-\Delta(k),0)\\ D=\sum_k\max(\Delta(k),0)\\ \min(U,D)=\tfrac12\big(W_1-|\Delta\mu|\big)\end{gathered}

The monotone coupling is the optimal transport plan for every convex cost. It reconciles two independent samples; it does not follow people.

Dispersion on a bounded ordinal scale

IOV=4K−1∑k=09F(k)(1−F(k))=1−ℓ2,K=11σ~=σμ(10−μ)≤1\begin{gathered}\mathrm{IOV}=\frac{4}{K-1}\sum_{k=0}^{9}F(k)\big(1-F(k)\big)\\ =1-\ell^2,\qquad K=11\\ \tilde\sigma=\frac{\sigma}{\sqrt{\mu(10-\mu)}}\le 1\end{gathered}

Berry-Mielke index of ordinal variation (1 − ℓ² in Blair and Lacy). The Bhatia-Davis inequality bounds the variance by μ(10 − μ), so the ceiling-adjusted SD lies between 0 and 1.

Between and within countries

σpool2=1C∑cσc2+1C∑c(μc−μˉ)2\sigma^2_{\text{pool}}=\frac1C\sum_{c}\sigma_c^2+\frac1C\sum_c\big(\mu_c-\bar\mu\big)^2

Exact for the equal-country pool; the plug-in between term carries sampling noise of about (C − 1)/C² times the summed squared standard errors.

Delta method and Satterthwaite degrees of freedom

Var^ g(p^)=∇g⊤V^ ∇gν=(v1+v2)2v12/ν1+v22/ν2\begin{gathered}\widehat{\mathrm{Var}}\,g(\hat p)=\nabla g^{\top}\hat V\,\nabla g\\ \nu=\frac{(v_1+v_2)^2}{v_1^2/\nu_1+v_2^2/\nu_2}\end{gathered}

V is the Taylor-linearized covariance of the eleven shares from country-nested strata and primary sampling units.

Data and question wording

The analysis uses the eleven pinned European Social Survey integrated releases registered in the series source contract, from ESS1 edition 6.7 through ESS11 edition 4.2. Respondent files remain private; public outputs contain survey-weighted aggregates only. The primary outcome is stflife: "All things considered, how satisfied are you with your life as a whole nowadays?" (0 extremely dissatisfied, 10 extremely satisfied). The robustness outcome is happy: "Taking all things together, how happy would you say you are?" (0 extremely unhappy, 10 extremely happy). Both wordings are identical in the Round 1, 5 and 11 source questionnaires, and variable and value labels are checked in every pinned release against the frozen documentation snapshot before any value is decoded. Codes 77, 88 and 99 are refusal, don't know and no answer.

Round 10 in this study is the face-to-face integrated file (edition 3.3). The nine countries that switched to self-completion in 2020-22 are published in a separate file and are not mixed in. Round 11 edition 4.2 adds Estonia and Ukraine; Czechia's self-completion fieldwork is not in the integrated file.

Survey design and weighting

Every estimate uses the original ESS analysis weight, anweight = pspwght × pweight, and the reviewed strata and primary sampling units; historical rounds join the official design supplements under the existing source contract. Countries without an eligible reviewed design are not approximated.

The primary aggregation gives every country equal weight by normalizing analysis weights within country. It answers what happened in the average participating country. The population aggregation pools the unnormalized analysis weights, so countries count in proportion to their population aged 15 and over. Taylor-linearized standard errors use country-nested strata and PSUs, the reviewed lonely-PSU adjustment and t intervals with design degrees of freedom. Changes between rounds combine independent variances with Satterthwaite degrees of freedom.

Samples

The primary comparison keeps the 17 countries observed in both Round 5 (2010/11) and Round 11 (2023/24): Belgium, Croatia, Estonia, Finland, France, Germany, Hungary, Israel, the Netherlands, Norway, Poland, Portugal, Slovakia, Slovenia, Spain, Sweden and Ukraine. The stable history keeps the nine observed in every round from 5 to 11. The long-run comparison keeps the five observed in Rounds 1 and 11. Maximum-coverage estimates change composition and serve as context only.

Estimators and inference

All new estimands are functions of the eleven response shares. The R survey package estimates the share vector and its full design-based covariance for every country, round, pool and leave-one-out pool (542 groups). Smooth functionals, the mean, standard deviation, cumulative shares, ordinal variation, the ceiling-adjusted standard deviation and the probability of superiority, receive delta-method standard errors from that covariance. The share-based mean and standard deviation reproduce the 1.0.0 survey estimates and their standard errors, because their influence functions are identical.

Threshold changes between independent rounds receive pointwise t intervals and a sup-t simultaneous band (Montiel Olea and Plagborg-Møller 2019): the 95% quantile of the largest absolute multivariate-t statistic over the ten cut-offs, from 100,000 seeded draws using the estimated correlation of the ten changes. Dominance is established by an intersection-union test that requires all ten one-sided tests to reject at 5%. A crossing requires significant changes of both signs inside the simultaneous band. Non-smooth transport quantities, the Wasserstein distance and the counter-movement, receive conservative simultaneous bounds by projecting that band; the relabeling range is the interval between the smallest and largest of the ten single-cut-off mean changes. Relative dispersion changes use log ratios of independent rounds. The between- and within-country decomposition uses normal-approximation intervals from independent national estimates.

Reconciliation and versions

Version 1.1.0 extends the accepted 1.0.0 analysis rather than replacing it: every 1.0.0 summary, response distribution, country change and headline estimate is reproduced unchanged, and the three 1.0.0 R output files are byte-identical. A separately written pandas and NumPy implementation recomputes every share vector directly from respondent weights (within 1e-8), re-derives every delta-method standard error from the R covariance (within a relative 1e-7), repeats each sup-t simulation with an independent generator (critical values within 0.03) and checks the estimator identities: the mean change equals minus the sum of the threshold changes and the net transport shift, the Wasserstein distance equals the transport cost, the transport plan reproduces both distributions, the tails equal cut-offs 4|5 and 8|9, and the pooled variance equals the within- plus between-country variance.

Literature

  • Allison, R. A. and Foster, J. E. (2004). Measuring health inequality using qualitative data. Journal of Health Economics 23(3), 505-524. doi:10.1016/j.jhealeco.2003.10.006
  • Berger, R. L. (1982). Multiparameter hypothesis testing and acceptance sampling. Technometrics 24(4), 295-300. doi:10.1080/00401706.1982.10487790
  • Berry, K. J. and Mielke, P. W. (1992). Assessment of variation in ordinal data. Perceptual and Motor Skills 74(1), 63-66. doi:10.2466/pms.1992.74.1.63
  • Bhatia, R. and Davis, C. (2000). A better bound on the variance. American Mathematical Monthly 107(4), 353-357. doi:10.1080/00029890.2000.12005203
  • Blair, J. and Lacy, M. G. (2000). Statistics of ordinal variation. Sociological Methods & Research 28(3), 251-280. doi:10.1177/0049124100028003001
  • Bond, T. N. and Lang, K. (2019). The sad truth about happiness scales. Journal of Political Economy 127(4), 1629-1640. doi:10.1086/701679
  • Clark, A. E., Flèche, S. and Senik, C. (2016). Economic growth evens out happiness: Evidence from six surveys. Review of Income and Wealth 62(3), 405-419. doi:10.1111/roiw.12190
  • Cliff, N. (1993). Dominance statistics: Ordinal analyses to answer ordinal questions. Psychological Bulletin 114(3), 494-509. doi:10.1037/0033-2909.114.3.494
  • Davidson, R. and Duclos, J.-Y. (2013). Testing for restricted stochastic dominance. Econometric Reviews 32(1), 84-125. doi:10.1080/07474938.2012.690332
  • Delhey, J. and Kohler, U. (2011). Is happiness inequality immune to income inequality? New evidence through instrument-effect-corrected standard deviations. Social Science Research 40(3), 742-756. doi:10.1016/j.ssresearch.2010.12.004
  • Dutta, I. and Foster, J. (2013). Inequality of happiness in the U.S.: 1972-2010. Review of Income and Wealth 59(3), 393-415. doi:10.1111/j.1475-4991.2012.00527.x
  • Kaiser, C. and Oswald, A. J. (2022). The scientific value of numerical measures of human feelings. PNAS 119(42), e2210412119. doi:10.1073/pnas.2210412119
  • Kaiser, C. and Vendrik, M. C. M. (2020). How threatening are transformations of happiness scales to subjective wellbeing research? IZA Discussion Paper 13905. docs.iza.org/dp13905.pdf
  • Kalmijn, W. and Veenhoven, R. (2005). Measuring inequality of happiness in nations: In search for proper statistics. Journal of Happiness Studies 6(4), 357-396. doi:10.1007/s10902-005-8855-7
  • Kaminska, O. (2023). Guide to using weights and sample design indicators with ESS data, version 1.2. European Social Survey ERIC. europeansocialsurvey.org
  • Montiel Olea, J. L. and Plagborg-Møller, M. (2019). Simultaneous confidence bands: Theory, implementation, and an application to SVARs. Journal of Applied Econometrics 34(1), 1-17. doi:10.1002/jae.2656
  • Vallender, S. S. (1974). Calculation of the Wasserstein distance between probability distributions on the line. Theory of Probability and Its Applications 18(4), 784-786. doi:10.1137/1118101
  • Vargha, A. and Delaney, H. D. (2000). A critique and improvement of the CL common language effect size statistics of McGraw and Wong. Journal of Educational and Behavioral Statistics 25(2), 101-132. doi:10.3102/10769986025002101

Coverage: every country and round with a reviewed design

Life-satisfaction respondents by country and round, where a reviewed survey design exists. Membership: E = 17-country endpoint sample, 9 = every round 5-11, 5 = rounds 1 and 11.
CountrySamples1234567891011
Albania-1,197
Austria-1,7922,0082,4952,346
BelgiumE 91,7021,8681,7691,7631,7641,3401,593
Bulgaria-2,2482,1732,7012,221
CroatiaE1,6371,8031,5841,559
Cyprus-1,110777679
Czechia-2,9771,9952,3781,9792,1412,2652,3862,456
Denmark-1,4941,4971,6091,5721,5001,565
EstoniaE 91,7912,3732,0452,0191,9031,5401,291
FinlandE 9 51,9992,0191,8932,1951,8762,1942,0851,9231,7531,5751,562
FranceE 9 51,5001,8061,9862,0701,7281,9671,9112,0681,9981,9711,738
GermanyE3,0272,9563,0412,8462,3522,418
Greece-2,7832,747
HungaryE 9 51,6681,4851,5551,9961,6811,6031,6461,8162,102
Iceland-748878859899840
Ireland-2,2782,6182,3812,7522,1951,7592,002
IsraelE 52,4762,2492,5402,544902
Italy-9482,5862,7122,5452,811
Kosovo-1,279
Latvia-9131,246
Lithuania-2,0852,2312,0811,8191,6191,341
Luxembourg-1,5421,633
Montenegro-1,1911,2651,599
NetherlandsE 91,8281,8451,9181,6791,6731,4531,675
North Macedonia-1,409
NorwayE 91,5481,6221,4331,5431,4041,4031,333
PolandE1,7401,8871,6071,6801,4801,429
PortugalE 92,1332,1361,2601,2681,0501,8281,372
Russia-2,4192,4962,5832,4582,413
Serbia-2,0201,544
SlovakiaE1,8391,8341,0731,3951,419
SloveniaE 9 51,4971,4391,4701,3941,2531,2191,3061,3141,2461,236
SpainE1,6531,8642,5521,8851,8841,9221,9511,6621,835
SwedenE1,4961,8441,7881,5431,5351,226
Switzerland-1,4911,5321,5241,5391,5201,377
UkraineE2,0021,9641,7781,8942,1282,630
United Kingdom-1,8882,3892,2682,2611,9572,2011,1481,679

Interview mode in the pandemic round

Share of interviews conducted by video (unweighted interview counts, ESS mode variable). Round 10 in this study is the face-to-face integrated file; the nine self-completion countries are not included.
CountryRound 10 (2020-22)Round 11 (2023/24)
Belgium1.2%0.2%
Germanynot in file0.0%
Estonia15.6%0.0%
Spainnot in file0.0%
Finland15.2%26.8%
France2.3%4.1%
Croatia6.0%0.0%
Hungary0.0%0.0%
Israelnot in file0.0%
Netherlands16.9%4.6%
Norway34.8%48.4%
Polandnot in file0.0%
Portugal0.4%0.0%
Swedennot in file12.0%
Slovenia0.0%0.0%
Slovakia0.0%0.6%
Ukrainenot in file0.0%

Downloads

Source releases: ESS1 edition 6.7 through ESS11 edition 4.2. Public files contain aggregate estimates only. Analysis 1.1.0 (supersedes 1.0.0, whose estimates it reproduces unchanged); fingerprint d58fb0809920; 100,000 seeded simulation draws per band.