Development and validation of statewide survey-based measures of livability in Connecticut

Abstract

Livability, or how a place and its systems (e.g., housing, transportation) supports the ability to lead a livable life, is a determinant of health. There is a lack of standard, validated measures to assess livability in the US. This study employed factor analytic methods to create measures of livability in Connecticut using data from the DataHaven Community Wellbeing Survey (DCWS) (n = 32,262). Results identified a 3-factor model (safety, opportunity, and infrastructure) as the best fit, explaining 69% of the variance in survey items. Newly created livability measures had high internal consistency, in addition to high convergent validity with other area-level measures.

1. Introduction

Geographies of inequality, or the concept that different people experience a higher quality of life based on where they live, are pervasive in the US (Bailey et al., 2017). Geographies of inequality affect how residents live and ultimately their health behaviors and outcomes (Hudson et al., 2020; Banzhaf et al., 2019; Bailey et al., 2017). Research on the neighborhood effects on health has well-documented the negative outcomes of geographies of inequality, and scholars are continually improving the measurement of constructs relevant in this field, such as gentrification and racial residential segregation (Schnake-Mahl et al., 2020), to identify intervention points in the causal pathway between place and health for the purposes of health disparity elimination (Lesorogol et al., 2020). Livability, defined as the ability of a place and its systems to reliably meet basic human needs, provide opportunities for cultural or artistic expression, and foster a general sense of community (Badland et al., 2014), is an emerging complex construct in neighborhood effects research, and is increasingly measured worldwide (Dsouza et al., 2023b; Higgs et al., 2019; Giles-Corti et al., 2022). Tools and indices of livability are also widely prevalent in urban planning, development, and public sector settings (e.g. tools such as the livability indicators in the Australian Urban Observatory, UK City LIFE, the Arts & Livability Indicators developed National Endowment for the Arts, etc.). Livability, which has significant overlap with social determinants of health frameworks, is best classified as a critical intermediary in the causal pathway of residential environment and health outcomes (Badland et al., 2014). Livable communities are hypothesized to be a fundamental cause of good health and wellbeing. However, livability in most residential contexts in the US is inequitable (Hudson et al., 2020). In the US, communities with people of lower incomes and other traditionally marginalized demographics have poorer physical and mental health due in part from living in neighborhoods with fewer amenities, lower access to public services, and higher exposure to social and environmental health hazards (Williams et al., 2016; Osypuk et al., 2009; Banzhaf et al., 2019; Grinstein-Weiss et al., 2011). These environmental injustices, resulting from systemic discriminatory policies and practices, decrease the livability of neighborhoods (Schnake-Mahl et al., 2020), despite the persistence and resilience of residents living in those places (Dsouza et al., 2023a). Literature suggests that individual-level interventions will not be successful in “unlivable areas” where environmental injustices persist, yet neighborhood-level measures of critical intermediaries, such as livability, driving these inequities are understudied and merit further exploration (Rosenberg et al., 2020; Kumanyika, 2011).

Livability as a construct is multidimensional, containing objective and subjective components. A scoping review of 24 peer-reviewed studies worldwide identified livability measurement as common, but inconsistently measured. Definitions and operationalization of livability measurement varied, but there was significant overlap in measurement domains, indicators, and methods (e.g., spatial, survey or self-report items) (Dsouza et al., 2023b). Objective components include access to built environment infrastructure and services or connectivity to transit, while subjective components include perceptions of environmental features, attitudes towards place governance and civic engagement (Badland et al., 2014). Common domains of livability include natural environment, crime and safety, education, employment and income, health and social services, housing, leisure and culture, local food and other goods, public open space, transport, and social cohesion and local democracy (Lowe et al., 2013). Operationalization of livability in the fields of economic development, real estate, and tourism results in more objective assessments than subjective assessments for the purpose of comparison and competition between residential areas. This skewed measurement of livability is leveraged mainly to incentivize tourism or, attract new residents or commercial investment, and results in an overemphasis on the aesthetic appeal or visual order and a problematic sense of public interest (Stevens, 2009). Capturing the lived experiences of residents in livability measurement possesses great utility for grassroots advocacy for strengthening neighborhood amenity infrastructure. To fully leverage livability promotion as a mechanism to eliminate health disparities, improved livability measures are required to facilitate the identification, improvement, and evaluation of mutable factors to improve health (Badland et al., 2014; Cox et al., 2010). The development and promotion of livable residential environments is a popular public policy goal across sectors, and as a result, scholars have called for increased livability measurement to promote interdisciplinary collaboration (Ruth and Franklin, 2014; Vick, 2019; Badland and Pearce, 2019). Indeed, development of livability measures, especially those containing subjective or survey-based items, can give decision-makers a window into the lived experiences of the residents they serve, while also allowing for the benchmark progress or compare geographical areas (Higgs et al., 2019; Badland and Pearce, 2019; Cantafio and Ryan, 2020).

Empirical research articles measuring livability have been published in many diverse fields, most notably sustainability, real estate economics, and social sciences (Ahmed et al., 2019). Prior studies have also established the reliability and validity of population-based survey items in assessing neighborhood social and institutional processes (e.g., social cohesion, community safety), which are theoretically related to but distinct from livability (O’Brien et al., 2015; Mooney et al., 2014). Livability measurement is growing in popularity in public health sciences as well, with scholars leveraging the concept to promote upstream thinking and policymaking for healthy environmental design (Higgs et al., 2019; Badland et al., 2014). However, there is limited measurement research assessing the psychometric properties of livability (i.e. ability of survey items to accurately assess livability perceptions of residents) (Brown et al., 2020), especially in a US context (Paul and Sen, 2020), and no research assessing the ecometric properties of livability (i. e., ability of survey items to accurately measure livability as an ecologic property of neighborhoods or towns) (Mujahid et al., 2007).

The aim of this study is to develop and validate a measure of livability employing data collected from a Connecticut statewide survey (using data from three cross-sectional waves collected in 2015 and 2018) about neighborhood quality of life. Subjective livability data is rarely found at the town or city level for an entire state and can complement existing place-based measures of objective components. Three time points of these data offer a unique opportunity to develop and validate a new livability measure for Connecticut and to explore how it is correlated with health status. This measure and its association with health status would not only serve public health practitioners and urban planners in Connecticut, but also provide a model for other areas in the US interested in addressing racial/ethnic disparities in livability to improve health.

2. Methods

2.1. Data source

This study is a secondary data analysis of the DataHaven Community Wellbeing Survey (DCWS) survey data collected and managed by DataHaven, a Connecticut-based nonprofit (DataHaven, 2022). The survey is a repeated cross-sectional design, administered every 3 years. Residents across the state were sampled and data was weighted to be representative of Connecticut’s demographics. Adults aged 18+ were reached via random digit dialing of a phone directory of CT residents. The survey was administered in English and Spanish.

2.2. Study context

While Connecticut contains some of the wealthiest cities in the nation, it is also home to some of the poorest cities in the US (e.g., Bridgeport, New Haven) and other geographies of concentrated poverty that experience severe racial and socioeconomic health disparities. There are 169 towns in the state of Connecticut, each with a unique culture and identity honed over the decades (Bryson, 2001). Population-level changes in household income have shown increased financial instability among Black and/or Hispanic/Latino households over the past 5 years (Davila et al., 2020). Due to its unique regional geographical positioning (e.g. – proximity to New York City, resulting in many commuter areas) and historical development, Connecticut contains significant variation between and within townships. Multidimensional area-level measures depict geographies of inequality, spatially coinciding with census tracts with higher concentrations of racial and ethnic minority populations and census tracts with poorer health outcomes (Gaul, 2014; Richardson et al., 2020). In this analysis, townships were treated as the level 2 variable in multilevel modeling due to potential of township-level governance to enact change in their areas to improve livability.

2.3. Survey variables

The DCWS survey design was designed comparably to other national and international wellbeing surveys to allow for comparisons between Connecticut and other states and geographic areas. DCWS questions for the 2015 and 2018 waves encompass the following topics: physical and mental health and wellbeing, neighborhood quality of life, civic engagement, and financial security.

2.4. Livability indicators (individual-level)

We identified livability-related survey items and matched them with empirical domains of livability identified in Badland et al., (2014) (see Table 2). Survey items had either 4-point or 5-point Likert-type response options and were recoded so that higher scores correspond with more positive perceptions of neighborhood amenities and conditions.

Table 2.

Livability-related survey items in each subscale and overall scale.

	Overall: Livability	Factor 1: Safety	Factor 2: Opportunity	Factor 3: Infrastructure
Survey question
Future of youth: Graduate from high schoola,c	X	X
Future of youth: get a job with opportunities for advancementa,c	X		X
Future of youth: be in a gangc	X	X
Future of youth: abuse drugs or alcoholc	X	X
Future of youth: get arrested for a felonyc	X	X
Many stores, banks, markets, or places to go are within easy walking distance of my homea,d	X			X
There are safe sidewalks and crosswalks on most of the streets in my neighborhooda,d	X			X
There are places to bicycle in or near my neighborhood that are safe from traffic, such as on the street or on special lanes, separate paths or trailsa,d	X			X
My neighborhood has several free or low-cost recreation facilities (parks, playgrounds, public swimming pools, etc.)a,d	X			X
I do not feel safe to go on walks in my neighborhood at nighta,d	X	X
People in this neighborhood can be trusteda,d	X	X
Children and youth in my town generally have the positive role models they need around hered	X	X
People in this neighborhood are involved in trying to improve the neighborhooda,d	X	X
How responsive local government is to the needs of residentsa,e	X		X
The availability of the goods and services that meet your needsa,e	X		X
The job done by the police to keep residents safea,e	X		X
The ability of residents to obtain suitable employmenta,e	X		X
As a place to raise childrena,e	X		X
The condition of public parks and other public recreational facilitiesa,e	X		X
The availability of affordable, high-quality fruits and vegetablesa,e	X		X
How would you describe your ability to influence local-government decision making?a	X		X
During the past 12 months, how often have you utilized arts and cultural resources within the area, such as concerts, museums, or cultural events?a,b	X
If the fire station closest to your home was going to be closed down by your city or town, how likely is it that neighborhood residents would organize to try to do something to keep the fire station open?a	X	X
Psychometric analysis results
Percent of variance explained	N/A	21.12	11.34	11.00
Mean score	50.3	50.7	50.4	50.7
Standard deviation	16.3	19.1	22.9	24.2
Cronbach’s alpha	0.84	0.74	0.88	0.61
Ecometric analysis results
Grand Mean (Y00), township-level averages of livability measures	52.1	52.9	55.5	41.7
Within-person variance	2.40	2.41	4.04	4.98
Within-township variance	0.30	0.28	0.71	1.33
Township-level reliability	0.74	0.45	0.79	0.57
ICC	0.11	0.10	0.15	0.20

^arecoded for direction so that higher values equal greater perceived livability.

^bexcluded from exploratory factor analysis due to low extraction values.

^cThis survey item was part of a block of questions beginning with the following prompt: “Now thinking about the future of young people in your neighborhood, for each of the following life events, how likely do you think it is that a typical young person in your neighborhood will experience each of the following, is it almost certain, very likely, a toss-up, not very likely, or not at all likely?”

^dThis survey item was part of a block of questions beginning with the following prompt: “The next group of questions is about your neighborhood, that is, the area around your home that you could walk to in 10 or 15 min or that area you consider to be your neighborhood. How much do you agree or disagree with each of the following statements about your neighborhood? Do you strongly agree, somewhat agree, somewhat disagree or strongly disagree?”

^eThis survey item was part of a block of questions beginning with the following prompt: “Now I’m going to ask you to think about some aspects of life in your city or area. For each of the following, I’d like you to rate that part of life in your area as excellent, good, fair, poor or let me know if you simply don’t know enough in order to say.”

2.5. Other survey measures for validation & descriptive analyses (individual-level)

Additional survey measures used for sample descriptive analyses are questions about housing insecurity, food insecurity, and utility insecurity, all asked about if the respondent experienced the issue in the last 12 months. In responding to the questionnaire, participants were asked to rate the area they lived in by considering “the area or place you live in.”

2.6. Urban typology measure (area-level)

A calculated township-level measure was urban typology, or a composite measure incorporating characteristics of townships based on socioeconomic and demographic factors. This measure was developed by the Center for Population Research at the University of Connecticut, in which the 169 Connecticut townships were classified into different urban typologies using a combination of factor analysis, cluster analysis, and discriminate analysis methods of 1990 and 2000 Census data. Analyses identified 5 different urban typologies: wealthy (townships, such as Westport, with exceptionally high income, low poverty, and moderate population density), suburban (townships, such as Cheshire, with above average income, low poverty, and moderate population density), rural (townships, such as North Stonington, with average income, below average poverty, and the lowest population density), urban periphery (townships, such as Manchester, with below average income, average poverty, and high population density), and urban core (townships such as Bridgeport, with the lowest income, highest poverty, and the highest population density).

2.7. Other measures from external administrative datasets for scale validation (area-level)

The Area Deprivation Index is a township-level measure estimating the degree to which residents of a particular place are socioeconomically deprived (Knighton et al., 2016). The measure is calculated using 2009–2013 American Community Survey (ACS) five-year estimates, and includes 17 indicators in domains of education, employment, housing, and income and poverty.

The Index of Concentration at the Extremes (ICE) is a township-level measure estimating the extent of racialized residential segregation in a geographically bounded location, and is increasingly used in public health sciences for neighborhood effects research (Krieger et al., 2016). The measure was calculated in this study using 2015–2019 American Community Survey (ACS) five-year estimates of racial breakdown of Connecticut townships. ICE ranges from −1 to 1; townships were sorted into quartiles, with townships in Q1 having low racial residential segregation to townships in Q4 having high racial residential segregation.

The GINI ratio is a township-level measure of income inequality. The measure was calculated in this study using 2015–2019 ACS five-year estimates of household incomes in Connecticut. GINI ranges from 0 to 1; townships were sorted into quartiles, with townships in Q1 having low levels of income inequality and townships in Q4 having high level of income inequality.

The CT Opportunity Index is a township-level measure of opportunity, defined as environmental conditions or resources conducive to healthy, livable communities and associated with success in life (defined in a variety of ways) (Gaul, 2014). The measure is calculated by the Open Communities Alliance (OCA) using 2008–2012 ACS five-year estimates of 11 indicators in three domains: education, economy, and neighborhood/housing quality, and was downloaded directly from OCA for the purpose of validation. The measure ranges from −3 to 1.6; townships were sorted into quartiles, with townships in Q1 having low opportunity or impediments to opportunity to townships in Q4 having high opportunity or conduits to greater opportunity.

2.8. Statistical analysis

Survey data was weighted using inverse sampling probabilities provided by DataHaven to weight observations to achieve population estimates. More details about the sampling methodology, data collection process, and statistical weighting can be found on DataHaven’s website (DataHaven, 2022). Two waves of data (2015 and 2018) were combined to increase statistical strength, resulting in a sample size of 42.597 participants from 169 townships. Z-scores were calculated from livability-related survey items and used for confirmatory factor analysis (CFA) and exploratory factor analysis (EFA) analyses. Data cleaning and descriptive analyses were conducted in SPSS 25 (IBM, Armonk, NY).

CFA of DCWS items assessing a 1-factor model of livability was conducted in mPlus version 8.6 (Muthén & Muthén, Los Angeles, CA). Model fit statistics such as the comparative fit index (CFI) and the Tucker Lewis Index (TLI) were also calculated, with higher values indicating better fit (0.9 is a statistically accepted threshold indicating good fit). The root mean square error of approximation was also calculated, with lower values indicating better fit (0.05 is a statistically accepted threshold indicating good fit) (UCLA: Statistical Consulting Group, 2021a).

EFA of DCWS items exploring domains of livability was conducted in SPSS 25 (IBM, Armonk, NY). Prior to extraction of factors, the Kaiser-Meyer-Olkin (KMO) test and Bartlett’s Test of Sphericity were conducted to assess whether or not DCWS survey items were suitable for exploratory factor analysis purposes. The number of factors was decided based on factors with eigenvalues greater than 1, the Scree plot, and the cumulative percentage of variance explained by factors. EFA was performed using an orthogonal varimax rotation under the assumption that factors are independent to maximize differences between high and low loadings on a particular factor (UCLA: Statistical Consulting Group, 2021b). Survey items with factor loading greater than 0.40 were retained, and items that loaded to multiple factors were assigned to the factor that they loaded most strongly on to. Two survey items were dropped from factor score calculations due to low factor loadings.

Ecometric properties of livability measures were calculated using three-level multilevel models in HLM (Scientific Software International, Chapel Hill, NC). The level 1 model (item response level; blue portion of Fig. 1) modeled the estimated mean livability for individual x in township y, and an error term normally distributed around sigma squared. The level 2 model (individual level; yellow portion of Fig. 1) estimated the mean livability score for individuals in townships by modeling a function of a neighborhood mean and a person-specific deviation from that mean, along with a random effect term. The level 3 model (township level; red portion of Fig. 1) modeled the township-specific means as a function of an overall mean and a township-specific deviation from the overall. The township-level intraclass correlation coefficient (ICC), or percentage of variability in livability that lies between townships was calculated to assess township-level reliability (ICC divided by number of respondents in each township).

Fig. 1.

Multilevel equations for ecometric analysis.

Linear regression models with individual-level demographic and area-level variables were calculated to validate livability measures. We hypothesized that livability measures would be positively associated with opportunity (i.e., - residents living in townships with high opportunity would report high livability). We also hypothesized that livability measures would be negatively associated with deprivation, residential segregation, inequality, poverty, and cost-burdened households (e.g., - residents living in townships with high deprivation would report low livability). The reference category was decided as the group with the highest livability scores (e.g., – respondents who identified as White, respondents who had a household income of $200,000 or higher). In addition, township-level averages of livability measures were calculated to examine ecological associations with area-level variables and to spatially visualize variations across the state of Connecticut using the ArcMap GIS software version 10.8.1 (Environmental Systems Research Institute, Redlands, CA).

3. Results

Survey participants were evenly split by age: 26% of respondents were 18–34 years old, 23% of respondents were 35–49 years old, 24% of respondents were 50–64 years old, and 22% of respondents were 65 or older (see Table 1). The majority of DCWS participants identify as White (70%), followed by participants that identify as “Other” race or ethnicity (11%), Black (10%), and of Hispanic or Latino ethnicity (7%). There was a broad distribution of income levels, with a little less than a third of participants reporting income above $100,000 per year, and 23% of the sample earning below $30,000 per year. The highest proportion of respondents were living in urban periphery settings (38%), followed by suburban (28%), urban core (17%), rural (13%), and wealthy towns (5%).

Table 1.

Demographics of combined samples of the 2015 and 2018 waves of the DCWS survey of Connecticut adults (n = 32,262).

	Number (Weighted %)
Agea
18–34 years old	8517 (26.4%)
35–49 year old	7263 (22.5%)
50–64 years old	7700 (23.9%)
65 and older	6975 (21.6%)
Raceb
White	22,616 (70.1%)
Black	3129 (9.7%)
Hispanic/Latino	2355 (7.3%)
Otherc	3388 (10.5%)
Genderd
Male	15,518 (48.1%)
Female	16,744 (51.9%)
Incomee
Greater than $200k	2250 (8.3%)
Between $100k - $200k	5783 (21.2%)
Between $75k - $100k	3995 (14.5%)
Between $50k - $75k	4683 (17.2%)
Between $30k - $50k	4130 (15.2%)
Between $15k - $30k	3604 (13.2%)
Less than $15k	2833 (10.4%)
Marital statusf
Single, never married	9.804 (30.4%)
Married/civil union	15,233 (47.2%)
Widowed	2306 (7.1%)
Divorced	2855 (8.9%)
Otherg	1725 (5.3%)
Urban typology
Wealthy	1662 (5.2%)
Suburban	8947 (27.7%)
Rural	4185 (13.0%)
Urban periphery	12,135 (37.6%)
Urban core	5333 (16.5%)

^aRefused response for age is 1807 (5.6%).

^bRefused response for race is 1225 (3.8%).

^cIncludes Asian, American Indian or Alaska Native, and Native Hawaiian or other Pacific Islander.

^dGender was recorded by observation only.

^eValid total for income is 27,238 (84.4%). Missing data for race is 5024 (15.6%).

^fRefused response for marital status is (1.0%).

^gIncludes living with partner and separated.

3.1. Factor analyses

Confirmatory factor analysis results revealed that a 1-factor model was not a good solution to fit the 23 livability-related survey items. A one-factor model explained only 37.8% of the variance of livability-related survey items and had poor model fit statistics (CFI and TLI were 0.574 and 0.533 respectively, both below 0.90, the statistically accepted thresholds for both measures). In addition, the RMSEA was 0.149, higher than the recommended 0.05 threshold.

Initial tests examining suitability of DCWS data for EFA purposes revealed a high level of clustering. The KMO value was 0.91, higher than the 0.8 statistical threshold documented in literature for sufficient sampling adequacy required for factor analytic methods(Villalba et al., 2010).

Exploratory factor analysis results revealed that a 3-factor model of 22 livability-related survey items explained 69.4% of the variance (1 survey item was excluded from the EFA due to low factor loadings). The first cluster contained 9 survey items focused on neighborhood safety and collective efficacy (see Table 2). Items in Factor 1, labeled as “safety,” explained 21.1% of the variance. The second cluster contained 9 survey items inquiring about political and economic opportunities. Items in Factor 2, labeled as “opportunity,” explained 11.3% of the variance. The third cluster contained 4 survey items and were about active living and other built environment infrastructure. Items in Factor 3, labeled as “infrastructure,” explained 11% of the variance. Survey items about reliable transportation, commute time, and perception of the arts were excluded due to low factor loadings but included in the overall livability measure due to anecdotal reports from DataHaven confirming that they are important livability-related considerations for people living in Connecticut. All four measures were calculated by averaging the standardized z-score values of the survey items and scaling the scores to range from 0 to 100. Hereafter, we will refer to the three EFA-resulted measures and the overall score as the “livability measures” or by their respective labels (“overall livability, “safety,” “opportunity,” and “infrastructure”).

3.2. Psychometric analysis

As indicated by the high Cronbach’s alpha of 0.84, the 25 livability-related survey items in the overall livability measure were highly internally consistent (see Table 2). Internal consistency of the items within each factor was more variable: Opportunity had a much higher internal consistency than safety and infrastructure (Cronbach’s alpha scores were 0.74, 0.88, and 0.61, respectively).

3.3. Ecometric analysis

Ecometric properties of livability measures were identified through multilevel modeling (see Table 2). Township-level reliability scores were highest for the overall livability and opportunity measures (0.74 and 0.79, respectively), and lower for infrastructure and safety (0.57 and 0.45, respectively).

Multilevel regression models depict clustering of data through the intraclass correlation coefficient (ICC), the index of the proportion of the variance in the livability measures between townships. As indicated by the ICCs, 11% of the overall variation in overall livability was occurring between townships, 10% of all the variance in safety was occurring between townships, 15% of all variation in opportunity was occurring between townships, and 20% of the overall variance in infrastructure was occurring between townships.

3.4. Livability measure descriptive statistics

The four livability measures had very similar mean scores, but when examining the grand means, or the township-level averages of livability measures, overall livability, safety, and opportunity had higher grand mean scores than infrastructure. Spatial visualizations of livability measures depict different patterning of infrastructure versus overall livability, safety, and opportunity, with a darker color representing a higher township-level average of said measure (see Fig. 2).

Fig. 2.

Maps of livability measures in Connecticut townships.

Linear models demonstrated that livability measures differed by DCWS participant demographic characteristics (see Table 3). Overall trends indicated that younger respondents, when compared to older respondents, tended to have lower perceptions of the overall livability, safety, and opportunity of their neighborhoods, but higher perceptions of neighborhood infrastructure. The highest differences were observed among respondents who were 18–34, who had lower perceptions of safety (b = −11.87, p < 0.001) when compared to respondents who were 65 and older, followed by opportunity (b = −6.89, p < 0.001) and overall livability (b = −5.47, p < 0.001). However, respondents in the age group of 18–34 had higher perceptions of infrastructure (b = 7.33, p < 0.001) than respondents who were 65 and older.

Table 3.

Associations between respondent demographics, township-level characteristics, and individual-level livability measures.

Independent Variable	Livability (23-item)	Safety (9- item)	Opportunity (9-item)	Infrastructure (4-item)
Respondent Demographics
Age
65 and older	REF	REF	REF	REF
50-64 year-olds	−3.18 (0.26)***	−4.81 (0.29)***	−3.80 (0.36)	−0.21 (0.38)
35-49 year-olds	−4.03 (0.27)***	−7.85 (0.30)***	−5.22 (0.36)	2.60 (0.38)***
18-34 year-olds	−5.47 (0.26)***	−11.87 (0.29)***	−6.89 (0.35)	7.33 (0.37)***
Race
White	REF	REF	REF	REF
Black	−6.89 (0.32)***	−9.08 (0.36)***	−12.80 (0.43)	10.01 (0.46)***
Hispanic/Latino	−5.46 (0.36)***	−8.26 (0.41)***	−9.05 (0.49)	8.60 (0.52)***
Other	−2.83 (0.31)***	−3.76 (0.35)***	−4.89 (0.42)	4.28 (0.44)***
Income
More than $200k	REF	REF	REF	REF
$100k - $200k	0.44 (0.29)	−1.47 (0.33)***	0.99 (0.40)**	0.44 (0.43)
$75k - $100k	−2.10 (0.32)***	−4.54 (0.37)***	−3.22 (0.44)	1.83 (0.48)***
$50k - $75k	−3.05 (0.31)***	−5.20 (0.35)***	−5.23 (0.42)	2.93 (0.45)***
$30k - $50k	−5.72 (0.32)***	−7.56 (0.37)***	−9.18 (0.44)	3.54 (0.47)***
$15k - $30k	−8.26 (0.33)***	−9.10 (0.38)***	−12.20 (0.46)	4.31 (0.49)***
Less than $15k	−11.93 (0.36)***	−12.56 (0.42)***	−15.92 (0.50)	4.69 (0.54)***
Self-Rated Health
Excellent	REF	REF	REF	REF
Very Good	−4.04 (0.24)***	−2.79 (0.28)***	−5.64 (0.33)	−2.43 (0.35)
Good	−8.95 (0.26)***	−6.29 (0.30)***	−12.65 (0.35)	−4.07 (0.38)
Fair	−12.84 (0.33)***	−8.97 (0.38)***	−17.41 (0.45)	−4.48 (0.49)
Poor	−15.76 (0.53)***	−9.25 (0.62)***	−20.15 (0.72)	−6.76 (0.79)
Place Satisfaction
Satisfied	REF	REF	REF	REF
Unsatisfied	−19.82 (0.22)***	−15.00 (0.27)***	−28.27 (0.30)	−9.99 (0.35)
Food Insecurity
No	REF	REF	REF	REF
Yes	−11.61 (0.27)***	−11.09 (0.31)***	−14.60 (0.37)	1.18 (0.40)**
Housing Insecurity
No	REF	REF	REF	REF
Yes	−5.75 (0.36)****	−5.65 (0.42)***	−6.65 (0.50)	1.73 (0.53)***
Utility Insecurity
No	REF	REF	REF	REF
Yes	−8.16 (0.63)***	−7.95 (0.71)***	−10.33 (0.84)	1.91 (0.86)*
Township-Level Characteristics
CT Opportunity Index
Q1 (high opportunity)	REF	REF	REF	REF
Q2	−2.38 (0.33)***	−2.13 (0.27)***	−3.56 (0.31)	−0.18 (0.35)
Q3	−6.75 (0.30)***	−4.56 (0.34)***	−9.17 (0.33)	2.45 (0.39)***
Q4 (low opportunity)	−12.14 (0.34)***	−16.75 (0.38)***	−19.30 (0.34)	6.86 (0.35)***
Area Deprivation Index
Q1 (low deprivation)	REF	REF	REF	REF
Q2	−2.79 (0.34)***	−2.15 (0.34)***	−6.48 (0.38)	0.18 (0.37)
Q3	−4.73 (0.38)***	−3.53 (0.34)***	−11.19 (0.37)	7.53 (0.38)***
Q4 (high deprivation)	−15.16 (0.35)***	−11.36 (0.29)***	−23.54 (0.36)	9.57 (0.34)***
Index of Concentration at the Extremes (ICE)
Q1 (low segregation)	REF	REF	REF	REF
Q2	2.46 (0.24)	−1.36 (0.29)***	−3.785 (0.37)	11.41 (0.34)***
Q3	−0.65 (0.28)	−6.13 (0.30)***	−6.99 (0.38)	20.08 (0.35)***
Q4 (high segregation)	−7.86 (0.27)***	−13.91 (0.30)***	−15.69 (0.38)	19.07 (0.35)***
GINI (Income Inequality)
Q1 (low inequality)	REF	REF	REF	REF
Q2	−0.71 (0.27)*	−2.16 (0.31)***	−1.76 (0.37)	2.88 (0.38)***
Q3	−1.46 (0.27)***	−2.09 (0.31)***	−3.18 (0.37)	5.28 (0.38)***
Q4 (high inequality)	−1.18 (0.26)***	−3.93 (0.30)***	−3.23 (0.36)	8.37 (0.38)***
Urban typology
Wealthy	REF	REF	REF	REF
Suburban	−4.81 (0.43)***	−2.96 (0.49)***	−7.47 (0.57)	−1.34 (0.63)*
Rural	−11.75 (0.47)***	−6.81 (0.53)***	−16.11 (0.62)	−8.92 (0.68)
Urban periphery	−9.99 (0.42)***	−11.09 (0.48)***	−18.49 (0.56)	7.73 (0.62)***
Urban core	−19.23 (0.45)***	−19.73 (0.52)***	−32.30 (0.60)	8.90 (0.66)***

Superscripts indicate the significance level (e.g., * for <0.05, ** for <0.01, *** for <0.001).

The livability measures also varied significantly across race and ethnicity (see Table 3). Respondents who identified as Black compared to those who identified as white tended to have poorer perceptions of opportunity (b = −12.80, p < 0.001), safety (b = −9.05, p < 0.001), and overall livability (b = −6.89, p < 0.001), but higher perceptions of infrastructure (b = 10.01, p < 0.001). These trends were observed similarly for individuals that identified as Hispanic/Latino or other ethnicities.

When examining how livability measures differed across income levels, linear models showed that as household income decreases, perceptions of overall livability, safety, and opportunity also decreased, but that perceptions of infrastructure increased. In particular, when compared to those earning more than $200,000, lower perceptions of opportunity were reported by households earning less than $15,000 (b = −15.92, p < 0.001), between $15,000 and $30,000 (b = −12.20, p < 0.001), and between $30,000 and $50,000 (b = −9.18, p < 0.001).

The livability measures varied significantly across respondents who rated their health differently (see Table 3). Linear models showed that respondents who rated their health as poor, compared to respondents who rated their health as excellent, had significantly lower perceptions of opportunity (b = −20.15, p < 0.001), overall livability (b = −15.76, p < 0.001), and safety (b = −9.25, p < 0.001), and infrastructure (b = −6.76, p < 0.001).

3.5. Validation with individual- and township-level variables

Convergent validity of livability measures was established by examining if associations with place satisfaction were in the expected direction (e.g. – residents who perceive their environment as more livable report higher levels of place satisfaction). Linear models indicated that among respondents who were unsatisfied with the place they were living in, compared to residents who reported they were satisfied, perceptions of opportunity (b = −28.27, p < 0.001), overall livability (b = −19.82, p < 0.001), safety (b = −15.00, p < 0.001), and infrastructure (b = −9.99, p < 0.001) were significantly lower (see Table 3).

Correlations between the livability measures and additional individual-level variables showed variations of perceptions by different household characteristics in the expected direction (see Table 3). Linear models also indicated that perceptions of opportunity, overall livability, and safety were significantly lower among those who were food insecure (opportunity: b = −14.60, p < 0.001; overall livability: b = −11.61, p < 0.001; safety: b = −11.09, p < 0.001), utility insecure (opportunity: b = −10.33, p < 0.001; overall livability: b = −8.16, p < 0.001; safety: b = −7.95, p < 0.001), and housing insecure (opportunity: b = −6.65, p < 0.001; overall livability: b = −5.75, p < 0.001; safety: b = −5.65, p < 0.001). However, perceptions of infrastructure were significantly higher among residents who were utility insecure (b = 1.91, p < 0.05), housing insecure (b = 1.73, p < 0.01), and food insecure (b = 1.18, p < 0.001).

Correlations between the livability measures and area-level variables generally indicated high convergent validity (see Table 3). Linear models showed that individuals that lived in townships with lower opportunity tended to have lower perceptions of opportunity (b = −19.30, p < 0.001), safety (b = −16.75, p < 0.001), and overall livability (b = −12.14, p < 0.001), but higher perceptions of infrastructure (b = 6.86, p < 0.001). Individuals living in townships with higher deprivation tended to have lower perceptions of opportunity (b = −23.54, p < 0.001), safety (b = −11.36, p < 0.001), and overall livability (b = −15.16, p < 0.001), but also higher perceptions of infrastructure (b = 9.57, p < 0.001). Similarly, individuals living in townships with higher segregation tended to have lower perceptions of opportunity (b = −15.69, p < 0.001), safety (b = −13.91, p < 0.001), and overall livability (b = −7.86, p < 0.001), but higher perceptions of infrastructure (b = 19.07, p < 0.001). Lastly, individuals living in townships with higher income inequality tended to have lower perceptions of safety (b = −3.93, p < 0.001), opportunity (b = −3.23, p < 0.001), and overall livability (b = −1.18, p < 0.001), but higher perceptions of infrastructure (b = 8.37, p < 0.001).

4. Discussion

The objective of this study was to develop a valid, reliable measure of livability and observe its performance across resident demographic indicators and additional area-level measures of theoretically related constructs. Results reveal that livability is a complex, multidimensional construct measurable at the community scale. The reliability of the overall livability measure and the three factor scale measures we generated from this large, statewide sample of Connecticut residents were high. The ecometric properties of the measures also depicted high township-level reliabilities, at a level consistent with reported reliabilities of other area-level measures (Mujahid et al., 2007; Raudenbush and Sampson, 1999), indicating general agreement regarding livability components among residents within townships. In general, the four measures (livability, safety, opportunity, and infrastructure) also demonstrated high convergent validity with area-based composite Census measures. Exceptions included the unexpected direction of some of these associations, especially with the infrastructure factor. In addition, some associations in the hypothesized direction at the unadjusted level, reversed direction after adjustment for demographic variables. These findings could reflect the bidirectional relationship between livability and other constructs, and the difficulty of capturing the dynamic nature of livability with a cross-sectional design.

As areas increase in their perceived livability, they can become “unlivable” for vulnerable populations already residing in the place due to rising costs and other pressures. The urgency of understanding and intervening upon livability is paramount as recent reports identify that an estimated 50% of the population in the United States is cost-burdened by housing (Harvard report). Housing is the anchor to leading a livable life because “everything is affected when you’re not stably housed” (Ezell et al., 2021), and higher housing costs result in lower residual income to cover other basic needs (e.g., food, utilities medications). At the community level, the real estate sector is focusing on development that is commercial and capitalist in nature, creating a housing stock that is largely unaffordable and inaccessible to populations with lower income. Even the most livable places contain segments of the population, usually at the margins, that warrant institutional support for improving their lived experiences and addressing these inequities. The goal of measuring livability is not to decry places as “unlivable,” or further stigmatize the neighborhoods that many live in, but to identify policies and budgetary decisions that perpetuate inequities in certain places and to develop interventions that can address these modifiable factors and reverse those inequities (Scally et al., 2021).

One of the most notable findings of this study is the three domains of livability identified (safety, opportunity, and infrastructure) and the contrast in the findings observed for the overall livability measure and the three independent livability factor measures. There are pros and cons to using a holistic versus modular approach while measuring livability. The developed measures of safety, opportunity, and infrastructure illustrate nuances that are missed in the overall livability measure, but the single, overall measure is more efficient to describe inequities and can be leveraged to communicate and collaborate across sectors as an overarching policy goal (Badland and Pearce, 2019; Laymon et al., 2021). These three measures developed from exploratory factor analysis results have high face validity, as they are well aligned with what we anecdotally know constitutes a good neighborhood (Haan et al., 2014). It’s important to note that DCWS survey items aligned with the 11 empirically identified domains of a public health-informed livability measure (Lowe et al., 2013), yet still clustered differently into these three factors, pointing to the complexity of livability as a construct. Measures of livability identified in peer-reviewed literature, especially those that were validated or used statistical approaches to identify domains, discussed the salience of these factors, even if these were not the major domains underpinning their measures (Dsouza et al., 2023b). By far, the measure with the poorest psychometric and ecometric properties was infrastructure. This is possibly attributed to the vast differences in built environments, particularly as they relate to active living, across urban and rural contexts. Low statistical properties of this measure can possibly be attributed to the fewer number of survey items that loaded on to this factor; regardless, the complex nature of infrastructure and how it contributes to perceptions of livability merit further study. While findings for the infrastructure measure are often in the opposite direction to what was expected, this could be attributed to the vast difference in infrastructure across levels of urban typology. Compared to safety and opportunity, infrastructure is unique due to its focus on transportation and connectivity, which are more pervasive in urban environments. Deeper, qualitative investigation into the nuances of infrastructure, especially as they relate to minority populations and other disadvantaged communities living in urban environments, are needed to inform future livability measures.

The highest inequalities were observed in the political and economic opportunities factor score, indicating a call to action to focus on sociopolitical dynamics among different populations to improve perceptions of livability for all. Public health as a sector has evolved to focus on upstream determinants of health; however, one criticism of the field is that it has yet to integrate a multisectoral lens needed to address factors such as safety and opportunity, and to a lesser degree, infrastructure (DeSalvo and Wang, 2018). There is a perception that public health institutions collaborate with infrastructure-related systems better, as evidenced by the rise of active living and healthy living by design work, but the lived reality is that most built infrastructure were shaped during the era of redlining and white flight, and these historic forces coupled with contemporary structural racism determine our urban systems and result in glaring urban inequalities (Hudson et al., 2020). The role of livability measurement and the creation of livability indicators at differing levels of geography is essential to the policy-making process by encouraging decision-makers adopt a “health in all policies approach” to examine upstream causes, think critically about systems change needed to advance public health, and work across departmental siloes to promote new forms of responsibility to achieve said change (Higgs et al., 2019; Browne et al., 2016; Ollila, 2011).

This research builds upon previous research establishing validated measures of livability across differing types of geography. No studies we are aware of have examined livability in a US residential context, nor captured variations of livability across urban typology in non-urban contexts. One of the major findings from this analysis is the importance of urban typology, and the rationale for examining livability differently in urban, suburban, and rural contexts. There is a growing acknowledgement that perceptions of neighborhood life are very different in an urban versus rural context. Scholars are tasked with producing research sensitive to the growing political divide observed between urban and rural settings, along with the different yet aligned challenges of creating economic opportunities in both contexts. Despite this, though, one of the largest differences across urban typology is in infrastructure, as indicated by calculated ICCs during the ecometric multilevel analysis process. While building infrastructure for active living is more easily facilitated in urban contexts and needed to accommodate higher housing density, the need for active living infrastructure in rural settings cannot be ignored (Barnidge et al., 2013; Vick, 2020). From a research perspective, there is much to explore with the potentially intersectional role of place and identity, and many future directions that livability and health research can take. Survey items, which are unlicensed and adapted from common place-based measures such as the Gallup Poll, are easily replicable and can be administered with minimal effort and cost.

These livability measures paint a picture of the varying neighborhood quality of life in the state of Connecticut. Connecticut is lauded as a relatively healthy and wealthy state; however, the state’s overall performance on national indicators masks stark inequities in livability. Geographical considerations of livability measurement in Connecticut are paramount. The governance of Connecticut is at the township level; however, certain townships, especially those more urban in nature, are much more heterogenous than rural townships. While urban typology was adjusted for in this analysis, it’s important to consider that domains of livability differ significantly among different communities, especially active living infrastructure (which tends to be built solely in more densely populated areas). Spatial visualization of the infrastructure measure shows a distinct and compelling pattern of townships with higher infrastructure scores, aligned with the routes of major highways and rail lines bisecting the state. In addition, the spatial visualization of the opportunity measure depicts some clustering around the New York City metropolitan region. As the state of Connecticut continues to rapidly urbanize due to its proximity to the megacity of NYC, regional planning experts are engaging with how to address racial residential segregation and the lack of socially connected communities by providing greater local opportunity (Kata and Kaplan-Macey, 2021). Measurement of livability can provide a benchmark for their efforts to understand and address inequities, in addition to providing an opportunity to engage residents in the planning process. These findings are transferrable to other contexts and other states in the US, despite the uniqueness of the state of Connecticut reflected in the measures (e.g., while other states may not have the same highways or rail lines, there is still patterning of infrastructure around urban development). Consistent with other states, livability promotion efforts in Connecticut originated through examination of aging, and considerations of older populations (The Commission on WomenConnecticut). Examining livability across the life course can identify opportunities to improve places to serve a wide spectrum of age ranges, not just older adults.

Results from this study indicate that livability measure creation from survey questions is a sound way to measure neighborhood-level quality of life, especially if the process of testing the newly created measures is robust. Livability measures performed in acceptable ways, in that livability was correlated with area-level measures in expected direction. However, measure distributions also depict township-level variations in livability, safety, opportunity, and infrastructure which are not captured by other area-level scales, indicating that these measures of perception contain valuable information about these unique constructs that are not captured in other area-level measures. Validated measures of livability are needed to not only further investigate causal pathways between residential environments and health outcomes, but also to identify meso-level opportunities for intervention across various levels of the sociological model. In addition, there is much to explore with the intersectionality of livability, and how place identity can act as a moderator in associations between structural determinants and health outcomes. Fields such as community science (rebranded from the previously named citizen science but rebranded to be more inclusive of undocumented and other immigrant populations without the privilege of citizenship), and methods such as community-based participatory research, can capture community voices essential to creating livability measures with higher validity for minoritized populations and/or geographies of inequality (Badland and Pearce, 2019; Zwald et al., 2016, Minkler, 2010).

5. Limitations and strengths

This study has several limitations. The cross-sectional design does not allow for causal inference, and especially hampers our ability to identify directionality of associations. The measured constructs are also dynamic: demographic changes in a geographic area can affect the perceived livability by residents and of that area, and changes to the livability of a certain area can potentially affect who remains versus leaves a certain neighborhood, or who is attracted to the neighborhood as a potential area of residence. Test-retest reliability is ideal for proper validation of this measure to build on this secondary data analysis. Survey items in the factor analysis were treated similarly, regardless of if they were a block of prompts under an overall question or if they were written as independent questions. Also, there is some potential misclassification bias based on the different geographical areas considered for different survey items: some survey questions asked about “city or area,” and others asked about “neighborhood.” This analysis does not take into account changes to livability that have occurred due to COVID-19. Future research can examine changes in livability, safety, opportunity, and infrastructure after the onset of the COVID-19 pandemic with the most recently available 2021 wave of DCWS data.

Despite these limitations, this study has many strengths. The DCWS contains data from a statewide representative sample, capturing variations in perceived livability across different urban typologies. Comparisons of livability across urban and rural contexts are rare, contributing greatly to our understanding of the interdependence between those types of areas. The sample size of the DCWS is high, providing us with sufficient statistical power to create and test livability measures. The unique context provided by the state of Connecticut, with great variance in socioeconomic status especially, is ideal to test the performance of livability measures.

6. Conclusion

Livability measurement is growing in public health sciences and has significant utility in advancing cross-sector collaboration needed to implement upstream evidence-based public health recommendations related to environmental design. Further research is needed to establish the interdependence of livability domains, along with how livability is related across different geographies (e.g., how livability of high-income countries and middle- and lower-income countries are related), and discriminant validity of objective and subjective measures of livability. As livability measurement continues to grow in popularity, especially with the development of multilevel, multi-method indices, scholars and practitioners have an ethical obligation to guide measurement efforts to center community and life-affirming structures and systems.

Data availability

The data that has been used is confidential.