Two Americas of Well-Being: Divergent Rural-Urban Patterns of Life Satisfaction and Happiness from 2.6 B Social Media Posts

Stefano Maria Iacus; Giuseppe Porro

cs.SINov 13, 2025

Two Americas of Well-Being: Divergent Rural-Urban Patterns of Life Satisfaction and Happiness from 2.6 B Social Media Posts

Authors: Stefano Maria Iacus, Giuseppe Porro

Organizations: Institute for Quantitative Social Science, Harvard University, 1737 Cambridge Street, Cambridge, 100190, Massachusetts, USA. · Department of Law, Economics and Culture, University of Insubria, Via S. Abbondio, 12, Como, 22100, Italy.

Abstract

Using 2.6 billion geolocated social-media posts (2014-2022) and a fine-tuned generative language model, we construct county-level indicators of life satisfaction and happiness for the United States. We document an apparent rural-urban paradox: rural counties express higher life satisfaction while urban counties exhibit greater happiness. We reconcile this by treating the two as distinct layers of subjective well-being, evaluative vs. hedonic, showing that each maps differently onto place, politics, and time. Republican-leaning areas appear more satisfied in evaluative terms, but partisan gaps in happiness largely flatten outside major metros, indicating context-dependent political effects. Temporal shocks dominate the hedonic layer: happiness falls sharply during 2020-2022, whereas life satisfaction moves more modestly. These patterns are robust across logistic and OLS specifications and align with well-being theory. Interpreted as associations for the population of social-media posts, the results show that large-scale, language-based indicators can resolve conflicting findings about the rural-urban divide by distinguishing the type of well-being expressed, offering a transparent, reproducible complement to traditional surveys.

Figures & tables

Year	N	Correlation (95% CI)	p-value
2014	3143	0.805 [0.792, 0.817]	$<10^{-300}$
2016	3140	0.496 [0.469, 0.522]	$7.5\times 10^{-195}$
2018	3143	0.605 [0.582, 0.627]	$3.5\times 10^{-313}$
2020	3143	0.558 [0.533, 0.582]	$1.2\times 10^{-256}$
2022	3140	0.442 [0.413, 0.470]	$3.5\times 10^{-150}$

Table 1: Correlation between Life Satisfaction and Happiness by Year

Figure 1: Unadjusted county-level means of lifesat and happiness by rurality code (1–9), averaged across 2014–2022. Error bars denote 95% confidence intervals. Life satisfaction increases monotonically with rurality while happiness decreases, documenting the raw rural–urban divergence in expressed well-being prior to any covariate adjustment.

Figure 2: Relationship between Twitter-based life satisfaction and happiness across U.S. counties in 2020. Each point represents a county, colored by its rurality code (1–9) using a diverging RdYlGn color scale, from green (urban, code 1) to red (most remote rural, code 9). The dashed line shows the fitted linear regression with 95% confidence band. The correlation is moderate and positive (

r=0.55

p<0.001

N=3{,}143

), indicating that, while the two affective indicators are related, they capture distinct facets of expressed well-being.

Figure 3: Relationship between Twitter-based life satisfaction and happiness by rurality code (1–9) in 2020. Each panel shows counties within the same rural classification, with a fitted linear regression (dashed line) and 95% confidence band. Colors follow the same RdYlGn diverging scale as Figure 2 . The positive association between life satisfaction and happiness is consistent across all levels of rurality, though the strength of the correlation varies by group. Notably, panels for codes 8 and 9 display an almost quadratic pattern, reflecting greater heterogeneity in expressed well-being among the most remote rural counties.

Predictor	AME	SE	z	p	lower	upper
acs5 ( $\times$ $10,000)	0.0054	0.0019	2.88	0.004	0.0017	0.0091
happiness	1.7623	0.0339	52.01	$<$ 0.001	1.6959	1.8287
margin	-0.0262	0.0100	-2.61	0.009	-0.0459	-0.0065
rural	0.0140	0.0015	9.54	$<$ 0.001	0.0111	0.0168
year = 2016	0.2302	0.0881	2.61	0.009	0.0575	0.4029
year = 2018	0.0257	0.0113	2.27	0.023	0.0035	0.0479

Table 2: Average Marginal Effects for the full lifesat model in equation ( 1 ). Clustered standard errors (HC1) by county via marginaleffects ( Arel-Bundock, 2024 ) .

Figure 4: Marginal effect of partisan vote margin ( margin ) on

\Pr(\texttt{lifesat}>0)

by rurality code, from the full model (Model 4). Points represent average marginal effects; bars denote 95% confidence intervals with clustered standard errors (HC1) by county. The effect is consistently negative across all rurality codes, indicating that Democratic-leaning counties display lower life satisfaction regardless of urban–rural context, though the effect appears strongest in semi-rural counties (codes 6–8) and borderline non-significant at the extremes (codes 1 and 9).

Figure 5: Predicted probability of expressing positive life satisfaction (

P(\texttt{lifesat}>0)

) across the Democratic–Republican vote margin for U.S. counties in 2022, based on the full model in equation ( 1 ). Panels correspond to counties at the 20th, 50th, and 80th percentiles of median household income (ACS 5-year estimates). Lines represent rurality codes (1–9). Life satisfaction decreases as the Democratic vote share increases, with a visually suggestive steeper decline in more-rural counties (higher rural codes); note however that the margin:rural interaction term is not statistically significant under clustered standard errors (see Figure 4 ). Confidence intervals are omitted for visual clarity given the overlap across nine rurality codes; uncertainty estimates are reported in Table 2 .

Figure 6: Predicted probability of expressing positive life satisfaction (

P(\texttt{lifesat}>0)

) across median household income levels for U.S. counties in 2022, based on the full model in equation ( 1 ). Panels correspond to Republican-leaning (

\leq-10

pp), toss-up (

\pm 10

pp), and Democratic-leaning (

\geq+10

pp) counties. Within each panel the partisan margin is held at the class mean. Lines represent rurality codes (1–9). Life satisfaction increases with household income across all partisan contexts, and the rural–urban gap widens at higher incomes, suggesting that material prosperity amplifies spatial differences in evaluative well-being. Confidence intervals are omitted for visual clarity given the overlap across nine rurality codes; uncertainty estimates are reported in Table 2 .

Predictor	AME	SE	z	p	lower	upper
acs5 ( $\times$ $10,000)	-0.0001	0.0007	-0.08	0.936	-0.0014	0.0013
lifesat	0.1409	0.0330	4.27	$<$ 0.001	0.0762	0.2057
margin	-0.0020	0.0096	-0.21	0.831	-0.0208	0.0167
rural	-0.0066	0.0017	-3.95	$<$ 0.001	-0.0099	-0.0033
year = 2016	-0.0703	0.0247	-2.84	0.004	-0.1187	-0.0219
year = 2018	-0.0062	0.0022	-2.77	0.006	-0.0106	-0.0018

Table 3: Average Marginal Effects for the full happiness model. Clustered standard errors (HC1) by county via marginaleffects ( Arel-Bundock, 2024 ) .

Year	State	County	Rural×Margin	Tweets	Poverty	Unemp	Black	Native	BA+
					%	%	%	%	%
2014	Alabama	Choctaw	9.000	15168	19.1	5.2	40.0	0.2	13.0
2014	Alabama	Perry	8.000	22413	32.8	15.7	70.8	0.0	15.2
2014	Alabama	Sumter	8.000	38680	30.4	8.2	72.3	0.1	21.8
2014	Alabama	Wilcox	9.000	13189	26.8	10.7	70.2	0.1	11.6
2014	Mississippi	Bolivar	7.000	43579	31.8	7.4	63.2	0.0	27.2

Table 4: Socioeconomic context for extreme counties (ACS 2022 5-year). Rates shown as percentages; last row reports U.S. average.

Predictor	Life Satisfaction	Happiness
Rurality ( rural )	$+++$	$---$
Partisan Margin ( margin )	$-$ †	$0$
Rural $\times$ Margin	$0$	$0$
Household Income ( acs5 )	$+$ †	$0$
Year Effects (2016–2022)	$+++$	$---$

Table 5: Comparative direction and strength of effects in the full models for Life Satisfaction and Happiness.

Appendix figures & tables7 assets

Supplementary material from the paper’s appendix.

Appendix

	Dependent variable:
	$\Pr(\texttt{lifesat}>0)$
	(1)	(2)	(3)	(4)
rural	0.219 ∗∗∗	0.333 ∗∗∗	0.312 ∗∗∗	0.291 ∗∗∗
	(0.021)	(0.024)	(0.030)	(0.037)
acs5 ( $\times$ $10,000)		0.328 ∗∗∗	0.123 ∗∗	0.120 ∗∗
		(0.037)	(0.042)	(0.042)

Appendix

Table 6: Logit estimates of

\Pr(\texttt{lifesat}>0)

by county (2014–2022). All models are weighted by the inverse standard deviation of life satisfaction estimates. Covariates are added sequentially: rurality, partisan margin, income ( acs5 in $10,000), year fixed effects, and their interaction.

Rural.code	AME	SE	z	p	lower	upper
1	-0.0150	0.0081	-1.84	0.066	-0.0309	0.0010
2	-0.0180	0.0074	-2.44	0.015	-0.0324	-0.0036
3	-0.0204	0.0071	-2.85	0.004	-0.0344	-0.0064
4	-0.0268	0.0092	-2.91	0.004	-0.0448	-0.0088
5	-0.0272	0.0102	-2.67	0.008	-0.0472	-0.0072
6	-0.0321	0.0133	-2.41	0.016	-0.0581	-0.0060

Appendix

Table 7: Marginal effect of partisan vote margin ( margin ) on

\Pr(\texttt{lifesat}>0)

by rurality code, from the full model (Model 4). Clustered standard errors (HC1) by county via marginaleffects ( Arel-Bundock, 2024 ) .

	Dependent variable:
	$\Pr(\texttt{happiness}>0)$
	(1)	(2)	(3)	(4)
rural	$-$ 0.420 ∗∗∗	$-$ 0.584 ∗∗∗	$-$ 0.542 ∗∗∗	$-$ 0.500 ∗∗∗
	(0.114)	(0.102)	(0.089)	(0.121)
acs5 ( $\times$ $10,000)		$-$ 0.402 ∗∗∗	$-$ 0.069	$-$ 0.005
		(0.042)	(0.058)	(0.065)

Appendix

Table 8: Weighted logistic regression of above-zero happiness (

P(\texttt{happiness}>0)

) on rurality, partisan vote margin, household income, and year fixed effects.

Figure 7: Predicted probability of expressing positive happiness (

P(\texttt{happiness}>0)

) across the Democratic–Republican vote margin for U.S. counties in 2022, based on the full model in equation ( 2 ). Panels correspond to the 20th, 50th, and 80th percentiles of median household income (ACS 5-year estimates). Lines represent rurality codes (1–9). Happiness levels are uniformly high but decline slightly with rurality, consistently with the (although non statistically significant) margin coefficient in Table 8 . Confidence intervals are omitted for visual clarity given the overlap across nine rurality codes; uncertainty estimates are reported in Table 3 .

Figure 8: Predicted probability of expressing positive happiness (

P(\texttt{happiness}>0)

) across median household income levels for U.S. counties in 2022, based on the full model in equation ( 2 ). Panels correspond to Republican-leaning (

\leq-10

pp), toss-up (

\pm 10

pp), and Democratic-leaning (

\geq+10

pp) counties. Within each panel the partisan margin is held at the class mean. Lines represent rurality codes (1–9). Happiness increases modestly with income but remains consistently higher in urban counties across all partisan contexts. Political orientation has minimal influence once income and rurality are controlled, consistent with the non-significant margin coefficient in Table 8 and underscoring the predominance of the urban affective advantage over partisan context in shaping hedonic well-being. Confidence intervals are omitted for visual clarity given the overlap across nine rurality codes; uncertainty estimates are reported in Table 3 .

	Dependent variable:
	lifesat
	(1)	(2)	(3)	(4)
rural	0.004 ∗∗∗	0.005 ∗∗∗	0.005 ∗∗∗	0.004 ∗∗∗
	(0.0003)	(0.0003)	(0.0003)	(0.0004)
acs5 ( $\times$ $10,000)		0.004 ∗∗∗	0.0003	0.0003
		(0.0003)	(0.0003)	(0.0003)

Appendix

Table 9: OLS estimates of lifesat by county (2014–2022). All models are weighted by the inverse standard deviation of life satisfaction estimates. Covariates are added sequentially: rurality, partisan margin, income ( acs5 in $10,000), year fixed effects, and their interaction. Clustered standard errors (HC1) by county in parentheses.

	Dependent variable:
	happiness
	(1)	(2)	(3)	(4)
rural	$-$ 0.002 ∗∗∗	$-$ 0.003 ∗∗∗	$-$ 0.002 ∗∗∗	$-$ 0.002 ∗∗∗
	(0.0002)	(0.0002)	(0.0002)	(0.0002)
acs5 ( $\times$ $10,000)		$-$ 0.002 ∗∗∗	0.001 ∗∗∗	0.001 ∗∗∗
		(0.0003)	(0.0002)	(0.0002)

Appendix

Table 10: Linear regression of happiness on rurality, partisan vote margin, household income, and year fixed effects. Results are in line with Table 8 .

Explore similar work

Quantifying Political Partisanship for Cross-Platform Analyses
Jul 23, 2026Fathima Ameen, Christopher G. HealeyPolitical Discourse Social Media
North Carolina State University, Raleigh, NC, USA
UrbanWell: Benchmarking Multimodal Large Language Models for Spatio-Temporal Urban Wellbeing Analytics
Jun 14, 2026Yanxin Xi, Xiang Su, Jie Feng +3Urban Environments Multimodal Large Language Models
Department of Computer Science, University of Helsinki, Helsinki, Finland · Department of Computer Science, Department of Agricultural Sciences, University of Helsinki, Helsinki, Finland · Zhongguancun Academy, Beijing, China +2
When Words Divide: Diachronic Ideological Polarization in Political Discourse on Social Media
Aug 2, 2026Roy Yitzchak, Noa Lavie, Ella RabinovichIdeology Political Discourse
The Academic College of Tel-Aviv Yaffo, Israel

Jul 23, 2026, cs.SIJ/K move · Enter open · S save

Quantifying Political Partisanship for Cross-Platform Analyses

Fathima Ameen, Christopher G. Healey

North Carolina State University, Raleigh, NC, USA

Political Discourse Social Media Indonesian E-Commerce Reviews

Open paper AskPDF arXiv

Research on political polarization on social media depends on the ability to reliably measure partisanship in user-generated content. However, existing approaches are typically tailored to platform-specific properties, such as structural affordances or linguistic conventions, which hurts generalizability across platforms. This limitation is increasingly consequential as the social media ecosystem fragments and fringe, alt-tech platforms emerge alongside mainstream ones. We propose a text-based, platform-portable methodology for measuring political partisanship in social media posts, anchored by an external news-credibility signal. Posts are embedded using a transformer-based sentence encoder and clustered into topic groups, which are labeled using the aggregated AllSides media bias scores of cited news outlets. A partisanship axis is then constructed in the embedding space as the difference between centroids of oppositely labeled clusters, and individual posts are scored by projection onto this axis. We apply the method to a corpus of approximately 1.3 million posts collected from Bluesky and Truth Social during the six months preceding the 2024 U.S. presidential election, providing the first cross-platform comparison of partisanship distributions on these two ideologically asymmetric platforms. The resulting partisanship scores correlate significantly with held-out AllSides media bias scores both in-distribution and out-of-distribution on an independent Twitter corpus, and recover within-platform partisan dynamics that platform identity alone cannot explain.

Abstract

Explore similar work

Quantifying Political Partisanship for Cross-Platform Analyses

UrbanWell: Benchmarking Multimodal Large Language Models for Spatio-Temporal Urban Wellbeing Analytics

When Words Divide: Diachronic Ideological Polarization in Political Discourse on Social Media