Population well-being and electoral shifts
Population well-being and electoral shifts
Jeph Herrin 0
Brita Roy 0
Harlan M. Krumholz 0
Joshua L Rosenbloom, Iowa State
University, UNITED STATES
0 Section of Cardiovascular Medicine, Yale School of Medicine, New Haven, Connecticut, United States of America, 2 The Gallup Organization , Washington, DC , United States of America, 3 Department of Pediatrics, University of Cincinnati College of Medicine, Cincinnati, Ohio, United States of America, 4 Division of Critical Care Medicine, Department of Pediatrics, Cincinnati Children's Hospital Medical Center , Cincinnati, Ohio , United States of America
Population wellbeing, an aggregate measure of positive mental, physical, and emotional health, has previously been used as a marker of community thriving. We examined whether several community measures of wellbeing, and their change since 2012, could be used to understand electoral changes that led to the outcome of the 2016 United States presidential election. We found that areas of the US which had the largest shifts away from the incumbent party had both lower wellbeing and greater drops in wellbeing when compared with areas that did not shift. In comparison, changes in income were not related to voting shifts. Well-being may be more useful in predicting and understanding electoral outcomes than some more conventional voting determinants.
Funding: The authors received no specific funding
for this work. The Gallup Organization provided
support in the form of salaries for authors Dan
Witters and Diana Liu, but did not have any
additional role in the study design, data collection
and analysis, decision to publish, or preparation of
the manuscript. The specific role of these authors
is articulated in the `author contributions' section.
Gallup did not fund the pursuit of this publication.
Gallup did provide permission to use
The 2016 United States Presidential election was determined, in part, by areas of the country
that shifted support to the Republican candidate. Efforts have been made to characterize the
areas with the most shift, with prior reports identifying the areas as having higher levels of
alcohol and opioid use [
], being less healthy [
], having rising levels of unemployment
, and lower rates of college education [
]. However, to date, studies have not assessed the
association of well-being, a metric of subjective experience of life that includes elements of
physical, social, mental and emotional health, nor its recent change over time, with voting
patterns. Prior research in other contexts has found a positive relationship between
wellbeing and a voting preference for incumbents, independent of economic measures [
it is plausible that poor and worsening well-being was associated with a vote to change the
party of the President. Accordingly, we tested the hypotheses that population well-being and
four year change in population well-being were associated with shifts in voting preference
regarding the incumbent party.
Methods and materials
Our primary unit of analysis was the county (or county equivalent), which was the smallest
geographic unit for which we could obtain both voting results and well-being data. We used
Sharecare Well-Being Index data as a part of this
research. The other authors received no funding for
voting results which are available from U.S. Voting Atlas for a small fee [
]; our analysis
included voting results from every county (or county equivalent) in the U.S. except for those
in the state of Alaska, which were not available for the 2016 election at the county level as of
time of analysis [
]. For each county we used the percentage of votes earned by the Republican
nominee in each year, using all votes (i.e., including third party candidates) as the
denominator. The voting shift was calculated as the change in percentage vote for the Republican
nominee from 2012 to 2016.
The well-being survey data were collected using a complex stratified survey design. This
design precludes direct aggregation of survey responses to create area measures; such
summary measures would not account for the different patterns of response across different
counties. To compensate for disproportionalities in selection probabilities and nonresponse, we
post-stratified the well-being data for each zip code grouping using an iterative proportional
fitting (i.e., raking) algorithm to account for nonrandom nonresponse by phone status (land
line or mobile), age, sex, region, education, population density, ethnicity, and race. Targets
used for the weighting leveraged the most current data available from the Current Population
Survey administered by the U.S. Census Bureau. This `reweighting' using iterative proportional
fitting is labor intensive and inappropriate for small sample sizes, including most individual
counties in the U.S. Therefore, in order to link the voting shift to the well-being survey data,
we categorized counties into a small number of groups according to their voting shift rate; this
allowed us to construct accurately weighted estimates of well-being metrics for groups of
counties with adequate numbers of survey respondents. Specifically, the voting shift was used to
categorize U.S. counties into 6 groups, according to the percentage point shift: less than -10 (that
is, greater than 10 percentage point shift toward the Democratic nominee), -10 to -5, -5 to 0
(inclusive), 0 to 5, 5 to 10, and more than 10 percentage point shift toward the Republican
To measure well-being, we used items from the Gallup-Healthways Well-Being Index, a
nationally representative, geo-coded random digit dial outbound telephone survey for which
353,561 respondents were interviewed throughout the year in 2012 and 177,192 throughout
the year in 2016. Though the full survey instrument includes items assessing many aspects of
well-being, the Well-Being Index itself was significantly modified beginning in 2014; thus, for
this analysis we report only on a subset of items which were included in both the 2012 and
2016 versions of the survey.
There were 10 items collected in both 2012 and 2016 related to well-being that we report
here. Two items are Cantril's Ladder [
], which asks respondents to consider their life as
placed on a `step' between 0 and 10 of a `ladder', with the bottom step signifying the worst
possible life for them and the top step signifying the best possible life for them; the first of these
items asks respondents which step they think they are currently on, while the second asks
them which step they think they will be on in 5 years. In line with the empirically determined
reporting guidelines for these items, the first was reported as the percentages of respondents
placing themselves on steps 0±4 and 7±10, while the second was reported as the percentages of
respondents placing themselves on steps 0±4 and 8±10 [
]. One item asks respondents if they
are satisified with the city or area where they live; this we report as the percent answering yes
(versus no, don't know, or refused). The remaining 7 items ask respondents if, for ªa lotº of the
day prior, they felt or expressed: happiness, stress, enjoyment, worry, smiling, sadness, and
anger. These were also reported as the percentage of respondents in each shift category who
answered ªyesº (versus no, don't know, or refused).
2 / 6
Because income, race, and education have been suggested as factors in the voting shifts for the
2016 election, we obtained data on these variables from the 2011 and 2015 U.S. Census for
each county. Measures obtained were median household income, percent white race, and
percent college graduates [
We summarized the number of counties, number of voters, median household income in
2015, change in median income from 2011, percent white, change in percent white, percent
college educated and change in percent college educated for each of the shift categories, using
non-parametric tests of trend to assess whether the last six factors were associated with voting
shift category. Next, we calculated well-being scores for shift categories: the Gallup-Healthways
Well-Being Index survey is conducted using a stratified survey design, so that design weights
must be used to construct estimates for each category. In addition, to account for the imperfect
randomness of each sample, survey responses were reweighted by voting shift group using
iterative proportional weighting to produce demographically appropriate estimates. Thus, we
calculated uniquely weighted responses for each of the six voting shift category regions for both
2012 and 2016.
Using these scores, we looked at two different relationships between well being and voting
shift. First, for each well-being metric, we used a non-parametric test of trend to assess whether
there was a trend over categories. Finally, we calculated the change in well-being for the same
regions from 2012 to 2016, and again assessed for trend over shift categories. We report the
Pvalues for the tests of trend.
The regions comprising each shift category are reported in Table 1. Though lower household
income in 2015 was associated with greater shift to the Republican nominee, neither the
absolute nor the relative change in household income since 2011 followed the same trend; notably,
the highest relative increase in household income occurred in the group of counties which
showed the strongest shift to the Republican nominee, followed by the group of counties
which shifted most away from the Republican nominee. Percentage of population that was
white and the change in this percentage over four years were also not associated with shift
3 / 6
category, though the rate of college education as well as the change in the rate of college
education over four years both decreased with increasing voting shift (P = 0.025 for both).
The 2016 well-being responses are summarized in Table 2 according to voting shift category.
The pattern of responses was consistent for each item with areas of increasing shift towards
the Republican nominee reporting: lower percentage of respondents placing themselves on the
top of the Cantril ladder, both currently and in 5 years; higher percentage placing themselves
on the bottom, both currently and in 5 years; and less satisfaction with the city or area where
they live. These trends were significant, as were negative trends in reported happiness,
enjoyment, and smiling/laughter on any given day, and increased reported sadness. There was no
significant trend in stress, worry, or anger.
The changes in well-being from the same areas of the U.S. from 2012 to 2016 are shown in
Table 3. The two largest shift categories (areas of the country where at least 5% of the vote
shifted to the Republican candidate) had an increase in those reporting being on the `bottom'
of the Cantril ladder (versus a decrease for all areas which shifted away from the Republican
candidate), and reported the smallest anticipated changes in their position for 5 years hence.
These changes in current and future life evaluation, at the top and the bottom of the ladder,
Satisfied with where you
Experienced a lot yesterday
PLOS ONE | https://doi.org/10.1371/journal.pone.0193401
4 / 6
were all significantly associated with voting shift; for satisfaction with one's city and all 7 affect
items, there were no significant trends.
There are several limitations to this study. First, as an observational study it is not possible
to draw causal inferencess; any observed associations between well-being and voting shifts
may be due to some unobserved factor. And there is no evidence that the people surveyed
voted, or, if they did, if they voted differently than they did four years earlier. However, most
examinations of voting determinants have these same limitations, and we believe that these
findings offer important insights into the causes and implications of the overall election
outcome. More specific to this study, we were constrained by our data to looking at very large
groups of counties in order to construct population estimates of well-being; however, this
constraint also means that our observed associations are true across large, diverse, heterogeneous
groupings of counties, and thus more robust than a study examining a smaller or more
homogeneous population. Finally, given our design, we were unable to adjust for potential effect
modifiers; however, our estimates of well-being were standardized for demographic,
socioeconomic and population factors, accounting for much of heterogeneity known to relate to
voting. While this topic surely deserves additional study, the current analysis is the largest and
most representative examination of the relationship between well-being and voting patterns to
These findings build on earlier work which found subjective well-being positively associated
with electoral support for incumbents by linking decline in subjective well-being to decline in
electoral support for the incumbent party.
These findings are limited by the observational study design, and open to ecological fallacy
because there is no evidence that the people surveyed voted, or, if they did, if they voted
differently than they did four years earlier
In conclusion, we suggest that multidimensional measures of population well-being may be
important factors in electoral shifts and outcomes in the U.S., and that changes in population
well-being may be a particular indicator of shifts in voter support. Focusing on well-being
might serve incumbents well.
Conceptualization: Jeph Herrin, Dan Witters, Brita Roy, Carley Riley, Harlan M. Krumholz.
Data curation: Dan Witters.
Formal analysis: Jeph Herrin, Diana Liu.
Methodology: Jeph Herrin, Dan Witters.
Writing ± original draft: Jeph Herrin.
Writing ± review & editing: Dan Witters, Brita Roy, Carley Riley, Harlan M. Krumholz.
5 / 6
1. Monat SM . Deaths of Despair and Support for Trump in the 2016 Presidential Election . Research Brief. http://aese.psu.edu/directory/smm67/Election16.pdf
2. Illness as an Indicator . The Economist. 19 November 2016 .
3. Wasfy JH , Stewart C III , Bhambhani V. ( 2017 ) County community health associations of net voting shift in the 2016 U.S. presidential election . PLoS ONE 12 ( 10 ): e0185051. https://doi.org/10.1371/journal. pone. 0185051 PMID: 28968415
4. Sussman AL . States with Rising Unemployment Went Overwhelmingly for Donald Trump . Wall Street Journal. 18 November 2016 .
5. Silver N. Education , Not Income, Predicted Who Would Vote For Trump. FiveThirtyEight. 22 November 2016 .
Ward G. Is Happiness a Predictor of Election Results? Centre for Economic Performance . London School of Economics and Political Science. April 2015 .
7. Liberini F , Proto E , Redoano M. Happy Voters . 2013 . IZA Discussion Paper No. 8498 .
8. Leip D. Dave Leip's Atlas of U.S . Presidential Elections. 2017 . http://uselectionatlas.org.
9. OECD Guidelines on Measuring Subjective Well-being. Organisation for Economic Co-operation and Development (OECD). 20 Mar 2013 . Paris: OECD Publishing.
10. Gallup-Healthways Well-Being Index : Methodology Report for Indexes . 2009 . http://www.wellbeingindex.com/hubfs/WBI_Methodology.pdf
11. U.S. Census Bureau. American Community Survey. 2011 .
12. U.S. Census Bureau. American Community Survey. 2015 .