Navegación – Mapa del sitio

InicioRúbricasCartografía, Iconografía, Didáctica2023Using Geographically Weighted Reg...

2023
1047

Using Geographically Weighted Regression to Explore County Subdivision Level Predictors of Drug Overdose Death in Connecticut, U.S.

Le recours à la régression géographiquement pondérée pour explorer les facteurs prédictifs de décès par overdose dans l’état du Connecticut aux États Unis
Utilizando regresión geográficamente ponderada para explorar los factores predictores de muertes por sobredosis en el Estado de Connecticut, Estados Unidos
Yunliang Meng

Resúmenes

Los decesos por sobredosis han aumentado significativamente durante las últimas dos décadas en Estados Unidos, convirtiéndose en un relevante problema de salud pública. Sin embargo, el trabajo empírico que explora lo espacial de las correlaciones entre variables de contexto y las tasas de mortalidad por sobredosis continúan limitados y ambiguos. Además, tal temática requiere un enfoque multifacético y multidisciplinario, y a la fecha se observa un bajo volumen de investigaciones en Estados Unidos referente a los factores de riesgo de estas muertes desde una perspectiva criminal. Este artículo utiliza una regresión geográficamente ponderada para examinar la relación entre las tasas de mortalidad por sobredosis y las variables de contexto en los condados del Estado de Connecticut. Los resultados expresan que las variables explicativas como sexo, educación, pobreza, vivienda y la diversidad racial/étnica están asociadas con las tasas de mortalidad por sobredosis. Pero más importante aún, la asociación entre las tasas de mortalidad por sobredosis y todas las variables explicativas en nuestro análisis varía significativamente en el espacio, subrayando la necesidad de programas de intervención y prevención de sobredosis en todo el país, a nivel local y según el contexto. Finalmente, esta investigación contribuye a que médicos, tomadores de decisión y las fuerzas del orden comprendan de mejor manera la geografía de la victimización por sobredosis y asignen recursos de manera efectiva para abordar tales muertes.

Inicio de página

Texto completo

1.Introduction

1Drug overdoses often refer to taking a higher than acceptable amount of a substance, whether it’s prescription, over-the-counter, legal, or illegal. For example, it can happen accidentally or intentionally when someone is taking a medication in higher doses than prescribed for a purpose other than prescribed or opioid-based drugs off the streets for recreational purposes. Drug overdoses can depress breathing rate and cause a loss of consciousness, which may lead to death if not addressed quickly. Deaths caused by drug overdose have increased significantly over the past 2 decades, becoming a public health concern in the U.S. In 2018, 67,367 drug overdose deaths were reported to Centers for Disease Control and Prevention (CDC) in the U.S. While this represents a slight decrease of 4% from 2017 to 2018, the total number of drug overdose deaths in 2018 was still four times higher than in 1999 (CDC, 2020a). Nearly 70% of the drug overdose deaths in 2018 involved an opioid drug and opioid overdose kills about 128 people in the United States daily (CDC, 2020b). The misuse of and addiction to opioids – including prescription pain relievers, heroin, and synthetic opioids, such as fentanyl – is a serious national crisis that affects public health as well as social and economic welfare (National Institutes of Health, 2020a). CDC researchers estimate that the total "economic burden" of prescription opioid misuse in the United States is $78.5 billion per year, including the costs of healthcare, lost productivity, addiction treatment, and criminal justice involvement (Florence et al., 2016).

2The focus of research in the U.S. has been largely on individual‐level risk factors of drug overdose deaths, where predictors such as age, gender, race, and income have often been considered (e.g. Havens et al., 2011; Lippold et al., 2019; Rhee et al., 2019; Scholl et al., 2019; Altekruse et al., 2020). There have been substantially fewer geographic studies that have considered the spatial variation of drug overdose deaths in addition to the above risk factors. However, research demonstrates that drug overdose deaths are not evenly distributed spatially, but instead tends to be geographically patterned (Kerry et al., 2016; Nesoff et al., 2019). Local demographic, socio-economic, and health correlates of drug overdose deaths also differ across geographic space. Given this evidence, more empirical research is needed to investigate the spatial difference in drug overdose deaths and identify underlying contextual characteristics with existing spatial differences in drug overdose deaths. In addition, the geographic studies of drug overdose deaths were almost all carried out at the county level in the U.S., largely because this is the level that CDC compiles and disseminates drug overdose death data at. However, research conducted at the county level provides very limited implications for a state like Connecticut which has no county level government to implement policing, public health, and/or social policies. Furthermore, the growing black-market trade of illicit drugs is a major driver of the number of opioid overdose deaths (U.S. Department of Justice, 2019). Consequently, spatial variations in drug overdose deaths could be explained by a crime theory (i.e. social disorganization theory) which was largely ignored by existing research on drug overdose deaths. This research article fills these gaps by examining the relationship between the drug overdose death rates and demographic, socio-economic, health and social disorganization predictors at the county subdivision level across Connecticut. It should be noted that county subdivisions are better known as cities, towns or municipalities. This research benefits not only the academic community, but also the governmental agencies (i.e. police, public health or social services) that need to understand the geographical variations in demand for their services and the effective interventions that will have the best impact in reducing the deaths caused by drug overdoses. This is especially important in a post COVID-19 pandemic world where resources may be limited or depleted due to the recession brought by the pandemic.

3This article is organized as follows. Section Two provides a literature review of risk factors associated with drug overdose deaths. Section Three presents the methodology outlining a description of the study area, variables used in this study and the local analysis method – Geographically Weighted Regression (GWR). Section Four discusses the results of the local analysis performed using GWR. The final section presents concluding remarks.

2. Literature Review

2.1 Demographic, Socio-economic and Health Factors

4A recent systemic study of demographic and socioeconomic risk factors associated with drug overdose deaths shows that factors such as age, race, gender, marriage status, employment status, education level, home ownership, and poverty can be used to explain the disparities in drug overdose deaths (Altekruse et al., 2020). Drug overdose was an over-represented cause of death among persons aged 20 to 59 years (Altekruse et al., 2020), since 20-39 and 40-59 age groups accounted for 41.1% and 43.8% of total opioid overdose deaths respectively. Compared to other racial/ethnic groups, drug overdose deaths were disproportionately higher among whites and blacks (Rudd et al., 2016; Altekruse et al., 2020). When it comes to gender, men were at greater risk than women to die from drug overdose (Park et al., 2016; Rudd et al., 2016; Altekruse et al., 2020). Drug overdose death risk was higher among widowed than married and unemployed than employed individuals (Johnson et al., 2013; Altekruse et al., 2020). Education level is another important risk factor. People with a high school diploma only were at a greater risk than those with a post-secondary degree (Jenkins et al., 2011; Altekruse et al., 2020). Compared to people who owned homes, those who rented had higher risk (CDC, 2015; Altekruse et al., 2020). People who lived under the poverty line were more likely to die from drug overdose than those who lived above the line (Binswanger et al., 2012; Altekruse et al., 2020).

5When it comes to drug overdose deaths, non-rural residents had higher risk than rural residents (CDC, 2015; Cerdá et al., 2017; Altekruse et al., 2020). Compared to people with health insurance coverage, people without health insurance were more likely to die from drug overdoses (Park et al., 2016; Altekruse et al., 2020). In addition, people who were disabled were at higher risk than those who were not (Kuo et al., 2019; Altekruse et al., 2020). Death rates caused by drug overdose were positively correlated with suicide rates (Howard, Jenson, 1999; Donaldson et al., 2006).

2.2 Risk Factors from A Crime Perspective

6Social disorganization theory has been one of the most widely applied ecological theories of criminal offense (Porter et al., 2015). The theory was proposed by Shaw and McKay (1942) in their study of communities with high levels of crime in Chicago. The core of social disorganization theory is that location matters when it comes to predicting crimes. Shaw and McKay’s study illustrates that crime rates were not randomly distributed throughout the city and that the so-called Zone of Transition that is close to the city center has the highest crime rate (Shaw, McKay, 1942). The areas in the Zone of Transition tend to have low socio-economic status, high levels of ethnic/racial diversity, and high residential mobility (Shaw, McKay, 1942; Wilcox et al., 2003). As a result, Shaw and McKay (1942) suggested that the high crime rates were not a function of the personal attributes of the groups residing in the area, but rather that "the structural factors of poverty, high heterogeneity, and high mobility created ‘social disorganization’, and it was community-level social disorganization that was presumed to cause crime" (Wilcox et al., 2003, p.28).

7However, Shaw and McKay’s theory was criticized because of the lack of testing of the measures they theorized (Sampson, Groves, 1989). Sampson and Groves (1989) were the first to test the causal mechanisms between structural characteristics and crime that had been set out in social disorganization theory. They found that the theory successfully predicted self‐reported crime, using data from the British Crime Survey (Sampson, Groves, 1989; Andresen, 2010). Guided by social disorganization theory, crime was found to increase in an area when there was a lack of collective efficacy and social cohesion. The strongest predictors were found to be three key variables – residential instability, racial/ethnic heterogeneity, and disadvantage status (Sampson, Groves, 1989; Peterson, Krivo, 2009; Hipp, 2011).

8In the U.S., about 70% of drug overdose deaths in 2018 where related to an opioid drug which involved the illegal trafficking, trade, possession, and use of opioid drug (i.e. heroin and fentanyl). Sampson and Bartusch (1998) suggest that in socially disorganized communities, the ‘street culture’ clashes conventionality, where crime, disorder, and drug use are expected and serve as a symbolic embodiment of the precariousness of everyday life. In addition, drug-related crime was successfully explained or predicted using social disorganization theory in numerous empirical studies over the past decades (e.g. Mosher, 2001; Banerjee et al., 2008; Contreras, Hipp, 2019). Therefore, drug overdose death risk could be potentially explained using social disorganization theory.

3. Methodology

3.1 Study Area

9The State of Connecticut is located in the northeastern U.S. It consists of a mix of 169 county subdivisions, including 19 cities, one borough, and 149 incorporated towns (figure 1). It is surrounded by three states – Rhode Island to the east, Massachusetts to the north, New York to the west and south. Connecticut is slightly larger than the country of Montenegro with a land area of 12,559 km2 and a water area of 1,809 km2. Hartford is the capital of Connecticut with a population of 122,591, but its most populous city is Bridgeport with a population of 144,900 according to the 2018 American Community Survey. Connecticut is the third smallest state by area, the 29th most populous, and the fourth most densely populated of the fifty states. However, Connecticut’s rural areas and small towns in the northeast and northwest corners of the state contrast sharply with its industrial cities such as Stamford, Bridgeport, and New Haven, located along the coastal highway Interstate -95 (I-95) running from the New York border to New London.

Figure 1: The Location Map of Study Area: Connecticut, USA

Figure 1: The Location Map of Study Area: Connecticut, USA

Meng, 2022

10Connecticut is an important study area for contextual analysis of drug overdose death rates due to the following reasons: 1) Drug overdose deaths in Connecticut doubled between 2012 and 2018, largely driven by use of multiple drugs together (Rhee et al., 2019). In 2018, the rate of drug overdose death (30.7 per 100,000) in the state was much higher than the national average (20.7 per 100,000). In addition, Connecticut had the eighth highest rate of opioid overdose deaths among all 50 states in the nation in 2017 (Rhee et al., 2019). 2) There has been no scholarly research conducted in Connecticut on drug overdose death patterns and their relationship to contextual characteristics of county subdivisions in the state.

3.2 Data Preparation

11This study is based on drug overdose death data collected and managed by the Connecticut Department of Public Health (CDPH) (https://portal.ct.gov/​). The data consisted of 5,105 reported drug overdose deaths that occurred in 169 county subdivisions in Connecticut from January 1st, 2012 to December 31st, 2018. Then, drug overdose death rates were calculated for each county subdivision in Connecticut as the number of reported drug overdose deaths per 100,000 total population. The descriptive statistics for the dependent variable – drug overdose death rates are shown in table 1. Due to the skewed distribution of the dependent variable, inverse hyperbolic sine (Burbidge et al., 1988) was used to normalize the data due to its ability to approximate the natural logarithm of the variable while retaining zero‐valued observations. Thereafter, the drug overdose death rates table was joined with a shape-file consisting of 169 county subdivisions based on the unique 10-digit county subdivision Federal Information Processing System (FIPS) code assigned by the U.S. Census Bureau using ArcMap 10.6.1 (Esri, 2018). This enabled the drug overdose death rates to be later aligned with contextual variables for further analysis.

Table 1: Descriptive Statistics for the Dependent Variable – Drug Overdose Death Rates

Min

Max

Median

Standard Deviation

Drug Overdose Death Rates: Drug Overdose deaths per 100,000 people

0.0

512.2

74.8

96.4

Meng, 2022

12This study considers 16 contextual variables as potential explanatory variables and the descriptive statistics for each explanatory variable are shown in table 2. The 16 contextual variables, covering demographic, socio-economic, health and crime aspects were chosen to reflect the key dimensions underlying the variation in the risk of drug overdose deaths as suggested by existing empirical research described in Section 2.

Table 2: Descriptive Statistics for the Explanatory Variables

Variables

Min

Max

Median

Standard Deviation

Demographic and Socioeconomic

Age: % aged between 20 and 59

32.1

60.2

49.6

3.9

Race 1: % Whites

15.8

97.0

91.3

14.7

Race 2: % Blacks

0.1

56.2

1.3

7.4

Gender: % males

42.5

58.1

48.9

2.3

Marriage status: % widowed population

1.7

10.8

5.6

1.6

Population density: the number of people per square kilometer

11.4

2946.3

116.4

468.2

Health

Healthcare coverage: % without health insurance

1.0

19.1

3.8

2.8

Disability: % disabled people

5.1

18.5

10.6

2.7

Suicide rate: the number of suicide deaths per 100,000 people

0.1

49.3

11.0

7.2

Social disorganization

Disadvantage Status

Joblessness: Unemployment rate

1.2

16.0

5.6

2.3

Household type: % lone parent families

2.9

36.6

12.3

5.5

Education: % without a college degree

14.7

77.9

51.0

13.8

Poverty: % living under the poverty line

1.1

30.1

10.4

4.4

Residential instability

Recent movers: % lived elsewhere 12 months ago

1.6

18.6

6.6

3.0

Housing: % renters

2.3

76.2

19.0

13.7

Racial/ethnic diversity

Index: Shannon equitability index

0.04

0.592

0.186

0.133

Meng, 2022

13The demographic, socio-economic, health and crime explanatory variables – residential population, age, race, gender, marriage status, healthcare coverage, disability, joblessness, household type, education, poverty, recent movers, and housing were taken from the 2014-2018 American Community Survey (ACS) five-year estimates (data.census.gov). Age variable was calculated by the percentage of population aged between 20 and 59. Race variables were measured by the percentage of White and Black people in the residential population. Gender variable was quantified by the percentage of males in the population. Marriage status variable was determined as the percentage of widowed population. Healthcare coverage variable was calculated by the percentage of people without health insurance coverage. Disability variable was measured by the percentage of people with a disability. Joblessness variable was quantified by the unemployment rate. Household type was determined by the percent of lone parent families. Education variable was calculated by the percentage of people without a college degree. Poverty variable was measured as the percentage of people living under the poverty line. Recent movers variable was quantified as the percentage of people who lived elsewhere 12 months ago. Housing variable was determined by the percentage of renters in the population.

14Population density was calculated by the number of people per square kilometer. Suicide rates from 2010 to 2014 were compiled at the county subdivision level and acquired through the CDPH (https://portal.ct.gov/​DPH). The dataset is the most recent available and reliable suicide death data in the state, as the CDPH collects and releases such data in 5-year intervals. The most recent 2010-2014 suicide data were released in 2017. The racial/ethnic diversity variable was measured using Shannon equitability index (Shannon, Weaver, 1949) based on eight racial/ethnic groups’ population in county subdivisions in Connecticut, including Hispanic, Non-Hispanic White alone, Non-Hispanic Black alone, Non-Hispanic Asian alone, Non-Hispanic American Indian and Alaska Native alone, Non-Hispanic Native Hawaiian and Other Pacific Islander alone, Non-Hispanic other race alone, and Non-Hispanic two or more races. Population data for these eight racial/ethnic groups were collected from American Community Survey (ACS) five-year estimates (2014-2018) which were released on the U.S. Census Bureau’s website (data.census.gov). Shannon’s index, which originated in ecology research, accounts for both abundance and evenness of the species present, but also has implications for the relative racial/ethnic heterogeneity of human populations (White, 1986). The Shannon diversity index is calculated using the following equation:

15Where S is total number of racial/ethnic groups in the community, pj is the proportion of the jth racial/ethnic group to the total population.

16Then, the Shannon equitability index (EH) can be then generated by dividing H by Hmax, which has a value between 0 (no diversity or a county subdivision is completely dominated by one racial/ethnic group) and 1 (perfectly diverse or all racial/ethnic groups are equally represented in a county subdivision). Due to the skewed distribution of independent variables, they were normalized with the logarithmic transformation. Thereafter, the explanatory datasets were joined with Connecticut county subdivision’s shape-file and stored for further analysis using ArcMap 10.6.1 (ESRI, 2018).

3.3 GWR Model Building

17Early contextual studies on the geographical distribution of disease or crime are often criticized for their use of global regression modeling techniques, such as Ordinary Least Squares (OLS) regression (Moore, Carpenter, 1999; Malczewski, Poetz, 2005), because the regression method violates some basic assumptions when spatial data is used (e.g. independence of observations and spatial stationarity of the relationship between independent and dependent variables). GWR (Brunsdon et al., 1996; Fotheringham et al., 2002) relaxes these assumptions and enables the analysis of spatially clustered data. GWR is often considered as a spatial extension of OLS regression, since it allows local parameters instead of global parameters to be estimated, hence making it possible to model spatial variations within the data (Fotheringham et al., 2002). Where OLS regression produces a single global model across space, GWR simultaneously conducts multiple regressions for different data units so that there is one regression model per spatial data unit (e.g. a county subdivision boundary) (Hipp, Chalise, 2015). In a GWR model, observations near a particular data unit will have more influence in the estimation than observations distance away (Hipp, Chalise, 2015). Given the weaknesses of OLS regression and strength of GWR, this research uses GWR for analyzing the spatial non-stationarity relationship between the dependent variable and explanatory variables in the State of Connecticut.

18The first step is to examine the dependent variable – drug overdose death rates, and explore its spatial autocorrelation. If the dependent variable is not significantly spatially clustered, there is no need to build a spatially explicit model. The Moran’s I Index (Anselin, 1995) based on the queen’s contiguity matrix was used to identify the clustering of drug overdose death rates across county subdivisions in the State of Connecticut. Moran’s I ranges from −1.0, perfectly dispersed (e.g. a checkerboard pattern), to +1.0, perfectly clustered.

19Then, the OLS multivariate model (Aiken, West, 1991) in Statistical Package for the Social Sciences (SPSS) 25 was then used to conduct initial data exploration and model specification. Specifically, the backward stepwise selection procedure was used when the model starts with all 16 explanatory variables in the model. At each step, the variable that is the least significant is removed. This process continues until no non-significant variables remain. Two factors motivated the decision to first specify the OLS model: 1) to detect explanatory variables that are significantly correlated with drug overdose death rates before specifying the GWR model; and 2) the GWR software used for spatial analysis does not provide a variance inflation factor (VIF) to measure multicollinearity, but SPSS provides such measurement. If the standard regression equation in the investigation of the dependent variable is given by:

where Yi is the drug overdose death rate at county subdivision i, β0 is a constant term (i.e. the intercept), βk measures the relationship between the independent variable xk and Y for the set of i county subdivisions, and εi is the error associated with county subdivision i. It should be noted that i Є C = {1,2,…,n} which is the index set of locations of n observations (i.e. all county subdivisions).

20Then, the same set of significant variables generated using the OLS model was used to specify a GWR model using the GWR4 software (https://gwrtools.github.io/​). GWR is a modeling technique used to explore spatial non-stationarity (Brunsdon et al., 1996). The GWR extends the standard regression equation by generating a separate regression equation for every county subdivision in the study area. In other words, by estimating a parameter for each county subdivision in the State of Connecticut, the GWR equation would only alter the standard regression equation as follows:

where Yi is the drug overdose death rate at county subdivision i, β0i is the constant term at county subdivision i, xki is the explanatory variable at county subdivision i, βki is the value of the parameter for the corresponding explanatory variable at county subdivision i, and εi is the error term at county subdivision i. It should be noted that i Є C = {1,2,…,n} which is the index set of locations of n observations (i.e. all county subdivisions).

21While conducting GWR, the adaptive kernel was utilized, which was produced using the bi-square weighting function. The adaptive kernel uses varying spatial areas, but a fixed number of observations for each estimation. It is the most appropriate technique when the distribution of observations varies across space. In this case, observations (i.e. county subdivisions) are smaller and closer together in the center of the state than they are at the edge. Finally, a process that minimizes the Akaike Information Criteria (AIC) was used to determine the best kernel bandwidth. The parameter estimates and t values produced by GWR4 were then exported and joined with the Connecticut county subdivision shape-file for mapping using ArcGIS 10.6.1.

4. Results and Discussion

4.1 Moran’s I results

22In this research, Moran’s I scores (0.294) and p value (0.000) were obtained, indicating drug overdose death rates are significantly clustered spatially. Local Moran’s I Cluster Analyses of the dependent variable was conducted, and the dependent variable and LISA results are shown in figures 2a and 2b below. Five different types of spatial clustering are demonstrated in figure 2b: (1) high-high, for county subdivisions with high drug overdose death rates that are in close proximity to county subdivisions with high drug overdose death rates; (2) low-low, for county subdivisions with low drug overdose death rates that are in close proximity to county subdivisions with low rates; (3) high-low (known as spatial outliers), for county subdivisions with high drug overdose death rates, but are proximate to county subdivisions with low rates; (4) low-high (also known as spatial outliers), for county subdivisions with low drug overdose death rates, yet are in close in proximity to county subdivisions with high rates; (5) not significant, for county subdivisions where there is no significant spatial clustering. As shown in figure 2b, the low-low spatial clusters were mainly found in the east and northeast of Connecticut, except a few low-low spatial clusters were found in county subdivisions located in the northwest. The high-high spatial clusters were mainly found in county subdivisions located in and around Hartford metropolitan area in the center of Connecticut. In other words, the spatial pattens illustrate that the high drug overdose death rates are mainly clustered in a few county subdivisions in the Hartford metropolitan areas in Connecticut. However, a few high-high spatial clusters were also found in and around New Haven metropolitan area located in the southwest of the state.

Figure 2: Drug Overdose Death Rates and Local Moran’s I Cluster Analyses of Drug Overdose Death Rates in Connecticut

Figure 2: Drug Overdose Death Rates and Local Moran’s I Cluster Analyses of Drug Overdose Death Rates in Connecticut

Meng, 2022

4.2 OLS results

23The summary of the OLS analysis results is presented in table 3. In the OLS regression, variables were included that are significantly correlated statistically with the drug overdose death rates (p < 0.05). The OLS model is statistically significant (F = 36.5, p < 0.01). The R2 and adjusted R2 values are 0.527 and 0.498 which means that the OLS model explained 49.8% of the variance in county subdivision-level overdose death rates in Connecticut. The VIF values for all variables was less than 5.0, a commonly used cutoff point (Becker et al., 2014; Ringle et al., 2015), suggesting no severe multicollinearity issue was detected among the explanatory variables (see table 3). In other words, the correlations among the 5 included explanatory variables are low.

Table 3: Results from Ordinary Least Square Model of Drug Overdose Death Rates at County Subdivision-Level in Connecticut

Dependent Variable

Independent Variables

β

p value

VIF

Drug Overdose Death Rates

Intercept

-----

< 0.01

-----

Gender: % males

0.412

< 0.05

2.470

Education: % without a college degree

0.642

< 0.01

1.823

Poverty: % living under the poverty line

0.519

< 0.01

2.036

Housing: % renters

0.556

< 0.01

3.025

Racial diversity: Shannon equitability index

0.301

< 0.05

1.925

Meng, 2022

24As shown in table 3, there is a positive and significant relationship between drug overdose death rates and all explanatory variables (i.e. gender, education, poverty, housing, and racial diversity). In other words, the higher the percentage of males, people without a college degree, people living under the poverty line, renters, and the higher the Shannon equitability index in a county subdivision, the higher the drug overdose death rate. The rest of the explanatory variables are insignificantly related to the dependent variable in this study. However, the residuals of the OLS model were spatially auto-correlated (Moran’s I = 0.235, p < 0.05), indicating that the OLS model overestimated drug overdose death rates for some county subdivisions, while it underestimated the results for some others.

4.3 GWR results

25The GWR model of drug overdose death rates generated β coefficients for each county subdivision (see table 4 and figure 3), local R2 value (see figure 3F), and t values for each county subdivision (see figure 4). A local Moran’s I cluster analysis (Anselin, 1995) was conducted for the residuals of the GWR model as a diagnostic for the spatial autocorrelation in GWR residuals. There were no violations of residual independence.

Table 4: Results from GWR Model of Drug Overdose Death Rates in Connecticut

Dependent Variable

Independent Variables

β coefficients

Percentage of County Subdivision by 95% of t Statistics

Min

Max

t│ ≤ 1.96

1.96 < │t│ ≤ 2.58

t│ > 2.58

Drug Overdose Death Rates

Intercept

0.018

2.240

-----

-----

-----

Gender: % males

0.066

0.732

61.6%

25.4%

13.0%

Education: % without a college degree

0.442

0.920

29.0%

17.8%

53.2%

Poverty: % living under the poverty line

0.125

0.905

8.9%

41.0%

50.1%

Housing: % of renters

0.213

0.826

23.1%

30.2%

46.7%

Racial diversity: Shannon equitability index

-0.088

0.708

34.9%

35.5%

29.6%

Note: 1.96 and 2.58 are the cut-off values for t-test. When │t│ > 1.96, the β coefficient estimate for a variable is significant at a level of 0.05. When │t│ > 2.58, the β coefficient estimate for a variable is significant at a level of 0.01.

Meng, 2022

Figure 3: Spatial Variations in β Coefficient Estimates in the State of Connecticut for Explanatory Variables from the GWR model of Drug Overdose Death Rates

Figure 3: Spatial Variations in β Coefficient Estimates in the State of Connecticut for Explanatory Variables from the GWR model of Drug Overdose Death Rates

Meng, 2022

Figure 4: Spatial Variations in t Values for Explanatory Variables in Modeling Drug Overdose Death Rates in Connecticut

Figure 4: Spatial Variations in t Values for Explanatory Variables in Modeling Drug Overdose Death Rates in Connecticut

Note: 1.96 and 2.58 are the cut-off values for t-test. When │t│ > 1.96, the β coefficient estimate for a variable is significant at a level of 0.05. When │t│ > 2.58, the β coefficient estimate for a variable is significant at a level of 0.01.

Meng, 2022

26As shown in table 4 and figure 3, gender, defined by the percentage of males, is positively associated with county subdivision level drug overdose death rates across Connecticut. This supports previous research which suggests that men are more likely than women to use almost all types of illicit drugs, and illicit drug use is more likely to result in emergency department visits or overdose deaths for men than for women (National Institutes of Health, 2020b). The influence of gender variable is stronger in the west of the state (see figure 4A). When holding all other variables constant, a 1% increase of male in the population would see a 0.732% increase in the drug overdose death rate in the areas colored in the darkest red on the maps, compared to a 0.066% increase in the darkest blue areas.

27As illustrated in table 4 and figure 3B, education, defined by the percentage of people without a college degree, is positively associated with county subdivision-level drug overdose death rates across Connecticut. This finding is consistent with previous research which shows people with a high school diploma only were at a greater risk than those with a post-secondary degree (Jenkins et al., 2011; Altekruse et al., 2020). A 1% increase of the population without college would see between a 0.442% and 0.92% increase in the drug overdose death rate, with highest coefficient values in the west of Connecticut. In addition, the influence of the education variable is stronger in the west and north of the state (see figure 4B). Higher levels of drug overdose mortality among the less educated may be driven by numerous factors. One important factor is that low educated individuals are more likely to be unemployed or employed in the low-skill sectors. As a result, they face increased risk of depression or anxiety, workplace injuries, and chronic health conditions (Oh, Shin, 2003; Minkler et al., 2006; Seeman et al., 2008), which lead to greater likelihood of being prescribed and addicted to opioid painkillers or drugs. This study calls policymakers in Connecticut to reduce barriers to higher education so that the percentage of adults who believe they have little to offer the labor market can be decreased and consequently, the drug overdose death rates can be lower in the future.

28As demonstrated in table 4 and figure 3C, poverty, defined by the percentage of people living under the poverty line, is positively associated with county subdivision-level drug overdose death rates across Connecticut. This finding confirms previous research which suggests people who live under the poverty line are more likely to die from drug overdose than those living above the line (Binswanger et al., 2012; Altekruse et al., 2020). The poverty variable is an influential variable in the GWR model, and it is also the variable that shows the greatest range in coefficient values. The influence of the variable is stronger in the east half of the state (see figure 4C). When holding all other variables constant, a 1% increase of poverty in the population would see a 0.905% increase in the drug overdose death rate in the areas colored in the darkest red, compared to a 0.125% increase in the darkest blue areas. The areas with the low coefficient values are some of the most affluent parts of Connecticut, which suggests that the effect of drug overdose death rate is not as pronounced in the more affluent areas and that other variables have more influence in these areas. This finding calls for policies (i.e. job counselling and training, affordable housing programs) to lift people living in low-income areas out of poverty and to increase opportunities for low-income families in Connecticut. In addition, as a recently proposed option, implants that release buprenorphine to treat drug addiction, must be replaced every six months by a trained doctor (Bebinger, 2016). This kind of treatments is very expensive, and cost and access barriers for the people living in poverty are likely to be substantial.

29As shown in table 4 and figure 3D, housing, defined by the percentage of renters, is largely positively associated with county subdivision-level drug overdose death rates across the state, except for a few outliers located in the southeast of the state. This finding supports previous research which shows that drug overdose death risk is higher among renters than homeowners (CDC, 2015; Altekruse et al., 2020). The influence of the housing variable is stronger in the west half of the state (see figure 4D). A 1% increase of renters in the population would see between a 0.213% and 0.826% increase in the drug overdose death, with highest coefficient values in the northwest of Connecticut. Previous research shows that when compared to renters, homeowners were more likely to engage in instrumental civic engagement, such as participating in community organizations or patrols (Manturuk et al., 2012). In doing so, communities with a higher percentage of rental homes tend to build less social cohesion and capital than those with a higher percentage of homeownership. Consequently, drug use and overdose death risk can be higher in communities with a higher percentage of rental homes. According to the latest American Community Survey (ACS) estimates, the homeownership rate in Connecticut is 66.59%, lower than the national average: 67.9%. In addition, in the first seven months of 2020 (in the peak of the COVID-19 pandemic), the median home sale price in Connecticut still increased 8 percent compared to the same period in 2019, to $288,000 from $267,000 (Gosselin, 2020). If Connecticut policy makers fail to provide affordable housing and boost homeownership, drug overdose death rates might increase steadily in the coming decades.

30As shown in table 4 and figure 3E, racial diversity, measured by the Shannon equitability index, is mostly positively associated with county subdivision-level drug overdose death rates across the state. This finding illustrates that drug overdose death risk is higher among county subdivisions with higher racial/ethnic diversity. The influence of racial/ethnic diversity variable is stronger in the mid-to-east of the state (see figure 4E). When holding all other variables constant, a 1% increase of racial/ethnic diversity in the population would see a 0.708% increase in the drug overdose death rate in the areas colored in the darkest red, compared to a 0.088% decrease in the darkest blue areas. Since 1970s, racial/ethnic diversity within cities has been increasing to different extents across the U.S. The trend is expected to continue in the future. Most noteworthy is the increased racial/ethnic diversity in the younger portion of the population. In 2019, for the first time, more than half of U.S. population under age 16 were identified as a racial/ethnic minority and the latest Census Bureau projections indicate that non-Hispanic whites will no longer account for the majority of the U.S. population by 2045 (Vespa et al., 2020). If Connecticut policy makers fail to boost social cohesion and informal social control in highly racially/ethnically diversified county subdivisions, drug overdose death rates might increase gradually in the coming decades.

31The local Moran I and GWR results are potentially useful for the CDPH, social policy makers and police in targeting priority county subdivisions for drug overdose prevention and intervention, since preventive strategies should be informed by an understanding of spatial distribution of drug overdose death rates (see figure 2) and the contextual factors (see figures 3 and 4). In particular, these factors should be examined locally, and different strategies aimed at preventing and reducing drug overdose death rates should be applied in different county subdivisions in Connecticut. For example, a strategy designed to help males avoid drug addiction may be sufficient to reduce drug overdose death rates in county subdivisions located in western Connecticut, since the variable of gender has a greater effect in the area (figure 4A). However, the same approach is unlikely to be effective in the county subdivisions located in eastern Connecticut. In those county subdivisions, drug overdose prevention/intervention programs aimed at helping, for example, people living below the poverty line, should be considered as tools for drug overdose death rates, since the variable of poverty has a greater effect in the area (figure 4C). In addition, programs aimed at increasing social interaction and cohesion should be considered as tools for preventing and reducing drug overdose death in eastern Connecticut, since the racial/ethnic diversity index has a greater effect in the area (figure 4E).

32As shown in table 4 and figures 3 and 4, the change in magnitude and direction of the coefficients suggests spatial non-stationarity of the relationship between the dependent variable (drug overdose death rates) and the explanatory variables. The variation in parameter estimates from GWR suggests the necessity of applying this spatial statistical tool to future drug overdose modeling studies that would be restricted by using global OLS models, since GWR provides insights on how a particular explanatory variable influences the drug overdose death rates across the study area. The importance of using spatial statistical tools such as GWR in future drug overdose death rates can also be confirmed by the local R2 value (see figure 3F). The adjusted R2 value for the GWR model of drug overdose death rates ranged from 0.502 to 0.841, with an average of 0.695, while the adjusted R2 in the OLS model was 0.498. The OLS R2 of 0.498 masks a wide distribution of local associations between the explanatory variables and drug overdose death rates. In other words, without GWR, it would be unable to estimate the variance of local associations. For example, as shown in figure 3F, in southeastern Connecticut, the GWR model explained up to 84.1% of the variance in drug overdose death rates. However, among county subdivisions clustered in the northwest of the state, the model only explains 50.2% to 56.9% of the variance, a spatial variation that would have been neglected with the OLS model alone.

33This study contains limitations. The first group of limitations is related to spatial boundary impact, such as Modifiable Areal Unit Problem (MAUP) and edge effect. It should be noted that the statistical relationships generated by the GWR model must be carefully interpreted and applied to other areas. In this study, county subdivision boundaries in Connecticut were used as the areal units and the relationships between drug overdose death rates and contextual characteristics at the county subdivision-level thus cannot be interpreted as and/or applied to individual-level relationships. The distribution of drug overdose death rates within the county subdivisions may not be spatially homogenous. Therefore, making assumptions that everyone shares the same risk in an areal unit could create an ecological fallacy problem (Robinson, 2006). In addition, the GWR model is restricted by the edge effect, whereby county subdivisions located on the edges of Connecticut do not have the 360° influence of county subdivisions in the state’s interior. Finally, census tracts are created by U.S. Census Bureau to provide a stable set of geographic units for the presentation of statistical data, and they are relatively uniform in terms of population size, which is targeted from 1,200 to 8,000 persons to avoid data suppression. Unlike census tracts, the size of the population varies greatly across the 169 county subdivisions including 19 cities, one borough and 149 incorporated cities. As a result, the demographic and socio-economic statuses of people could be greatly different even within a county subdivision. It would be ideal to conduct the analysis at census tract level if drug overdose death data are disseminated at the census tract level in the future.

34The second group of limitations is related to drug overdose death and explanatory variable data. It should be noted that the dataset prepared by the CDPH may underestimate the drug overdose death rates to some extent because of incomplete cause-of-death reporting on death certificates, indicating that the mortality from drug overdose may be higher than it appears (Buchanich et al., 2018). In addition, encountered is a time alignment problem. Drug overdose death data were collected between the year of 2012 and 2018. However, the most recent and complete demographic and socio-economic data are based on American Community Survey collected between the year of 2014 and 2018. The most recent and complete suicide rate data were collected by CDPH between the year of 2010 and 2014. In addition, the local R2 values accounted for 50.2% to 84.1% of the drug overdose death rates, which means that other risk factors (i.e. pre-existing conditions, drug dosage, access to urgent care, etc.) need to be added into the GWR models in future studies.

5. Conclusions

35While there are some limitations, this research has some important findings and implications. In a post-COVID-19 recession world, this research can help government agencies to allocate resources more efficiently, since it reveals the spatial distribution and more importantly the hotspots of drug overdose death rates. The spatial relationships between the drug overdose death rates and the explanatory variables are still under investigation and little research has been done to investigate the spatial heterogeneity of the relationships. This study fills the research gap by illustrating there is a significant spatial correlation between the drug overdose death rates and the explanatory contextual variables (i.e. gender, education, poverty, housing, and racial/ethnic diversity). In addition, this relationship has a spatial but nonstationary association which highlights the need for designing local and context-specific drug overdose prevention-intervention programs rather than using blanket national approaches designed by CDC.

36This research also contributes to extant literature by taking the crime aspect of drug overdose into consideration when modeling the overdose death rates. Given the fact that demographic (i.e. gender), socio-economic (i.e. education and poverty), and social disorganization indicators (i.e. racial heterogeneity and residential mobility measures) are correlated with drug overdose death rates, this research disclosed and confirmed the multi-dimensional nature of drug overdose prevention and intervention. Researchers, health practitioners, policy makers, and law enforcement should work closely together in the future to design effective drug overdose prevention-intervention programs.

37This research also illustrates that it has been possible to periodically predict the rate of drug overdose death in a county subdivision in Connecticut with a moderate-to-high degree of accuracy using data covering demographic, socio-economic, health, and crime aspects, since the data are widely available online and updated regularly. This study presents an initial and exploratory step towards better understanding of drug overdose death rates in Connecticut, but much more in-depth work remains before researchers understand why explanatory factors, such as gender, education, poverty, housing, and racial diversity, have relatively low explanatory effects in some cities and towns, but explain up to 84.1% of the drug overdose death rates in others.

Inicio de página

Bibliografía

Aiken L.S., West S.G., 1991, Multiple regression: Testing and interpreting interactions, Newbury Park, Sage.

Altekruse S.F., Cosgrove C.M., Altekruse W.C., Jenkins R.A., Blanco C., 2020, "Socioeconomic risk factors for fatal opioid overdoses in the United States: Findings from the Mortality Disparities in American Communities Study", PLoS One, vol.15, No.1, e0227966.

Andresen M.A., 2010, "The place of environmental criminology within criminological thought", In: Andresen M.A., Brantingham M.J., Kinney J.B. (Eds.), Classics in environmental criminology, Burnaby, Simon Fraser University Publications, 29-75.

Anselin L., 1995, "Local indicators of spatial association – LISA", Geographical Analysis, Vol.27, No.2, 93-115.

Banerjee A., LaScala E., Gruenewald P.J., Freisthler B., Treno A., Remer L.G., 2008, "Social disorganization, alcohol and other drug markets and violence: A space-time model of community structure", In: Thomas Y.F., Richardson D., Cheung I. (Eds), Geography and Drug Addiction, Netherlands, Springer, 117-130.

Bebinger M., 2016, FDA Considering Pricey Implant As Treatment For Opioid Addiction, [online]. URL: http://www.npr.org/sections/health-shots/2016/05/20/478577515/fda-considering-pricey-implant-as-treatment-for-opioid-addiction

Becker J.M., Ringle C.M., Sarstedt M., Völckner M., 2014, "How collinearity affects mixture regression results", Marketing Letters, Vol.26, No.4, 643-659.

Binswanger I.A., Blatchford P.J., Mueller S.R., Stern M.F., 2013, "Mortality after prison release: opioid overdose and other causes of death, risk factors, and time trends from 1999 to 2009", Annals of Internal Medicine, Vol.159, No.9, 592-600.

Brunsdon C., Fotheringham A.S., Charlton M.E., 1996, "Geographically weighted regression: A method for exploring spatial nonstationarity", Geographical Analysis, Vol.28, No.4, 281-298.

Buchanich J.M., Balmert L.C., Williams K.E., Burke D.S., 2018, "The effect of incomplete death certificates on estimates of unintentional opioid-related overdose deaths in the United States, 1999-2015", Public Health Reports, Vol.133, No.4, 423-431.

Burbidge J.B., Magee L., Robb A.L., 1988, "Alternative transformations to handle extreme values of the dependent variable", Journal of the American Statistical Association, Vol.83, No.401, 123-127.

Centers for Disease Control and Prevention, 2015, "Today’s heroin epidemic", Vital Signs, July 7, 2015, [online]. URL: https://www.cdc.gov/vitalsigns/heroin/

Centers for Disease Control and Prevention, 2020a, Understanding the Epidemic, [online]. URL: https://www.cdc.gov/drugoverdose/epidemic/index.html

Centers for Disease Control and Prevention, 2020b, Drug Overdose Deaths, [online]. URL: https://www.cdc.gov/drugoverdose/data/statedeaths.html

Cerdá M., Gaidus A., Keyes K.M., Ponicki W., Martins S., Galea S., Gruenewald P., 2017, "Prescription opioid poisoning across urban and rural areas: identifying vulnerable groups and geographic areas", Addiction, Vol.112, No.1, 103-112.

Contreras C., Hipp J.R., 2019, "Drugs, Crime, Space, and Time: A Spatiotemporal Examination of Drug Activity and Crime Rates", Justice Quarterly, Vol.37, No.2, 1-23.

Donaldson A.E., Larsen G.Y., Fullerton-Gleason L., Olson L.M., 2006, "Classifying undetermined poisoning deaths", Injury Prevention, Vol.12, No.5, 338-343.

Environmental System Research Institute (ESRI), 2018, ArcGIS Desktop: Release 10.6.1, Redlands, CA.

Florence C., Luo F., Xu L., Zhou C., 2016, "The economic burden of prescription opioid overdose, abuse, and dependence in the United States, 2013", Med Care, Vol.54, No.10, 901-906.

Fotheringham A.S., Brunsdon C., Charlton M., 2002, Geographically weighted regression: The analysis of spatially varying relationships, Chichester, Wiley.

Gosselin K.R., 2020, Connecticut home sales heat up as New Yorkers flee and at-home workers look for more space, [online]. URL: https://www.courant.com/business/hc-biz-connecticut-home-sale-boom-20200904-gdwejd3eafhotfajaj74pdydku-story.html

Havens J.R., Oscer C.B., Knudsen H.K., Lofwall M., Stoops W.W., Walsh S.L., Leukefeld C.G., Kral A., 2011, "Individual and network factors associated with non-fatal overdose among rural Appalachian drug users", Drug & Alcohol Dependence, Vol.115, No.1-2, 107-112.

Hipp J.P., 2010, "Spreading the wealth: The effect of the distribution of income and race/ethnicity across households and neighborhoods on city crime trajectories", Criminology, Vol.49, No.3, 631-665.

Hipp J.A., Chalise N., 2015, "Spatial analysis and correlates of county-level diabetes prevalence, 2009-2010", Preventing Chronic Disease, Vol.12, [online]. URL: https://www.cdc.gov/pcd/issues/2015/14_0404.htm

Howard M.O., Jenson J.M., 1999, "Inhalant use among antisocial youth: Prevalence and correlates", Addictive Behaviors, Vol.24, No.1, 59-74.

Jenkins L.M., Banta-Green C.J., Maynard C., Kingston S., Hanrahan M., Merrill J.O., Coffin P.O., 2011, "Risk factors for nonfatal overdose at Seattle-area syringe exchanges", Journal of Urban Health, Vol.88, No.1, 118-128.

Johnson E.M., Lanier W.A., Merrill R.M., Crook J., Porucznik C.A., Rolfs R.T., Sauer B., 2013, "Unintentional prescription opioid-related overdose deaths: description of decedents by next of kin or best contact, Utah, 2008-2009", Journal of General Internal Medicine, Vol.28, No.4, 522-529.

Kerry R., Goovaerts P., Vowles M., Ingram B., 2016, "Spatial analysis of drug poisoning deaths in the American West, particularly Utah", International Journal of Drug Policy, Vol.33, 44-55.

Kuo Y.F., Raji M.A., Goodwin J.S., 2019, "Association of disability with mortality from opioid overdose among US Medicare adults", JAMA Netw Open, Vol.2, No.11, e1915638.

Lippold K.M, Jones C.M., Olsen E.O., Giroir B.P., 2019, "Racial/ethnic and age group differences in opioid and synthetic opioid–involved overdose deaths among adults aged ≥18 years in metropolitan areas - United States, 2015-2017", Morbidity and Mortality Weekly Report, Vol.68, No.43, 967-973.

Malczewski J., Poetz A., 2005, "Residential burglaries and neighborhood socioeconomic context in London, Ontario: Global and Local Regression Analysis", The Professional Geographer, Vol.57, No.4, 516-529.

Manturuk K., Lindblad M., Quercia R., 2012, "Homeownership and civic engagement in low-Income urban neighborhoods a longitudinal analysis", Urban Affairs Review, Vol.48, No.5, 731-760.

Minkler M., Fuller-Thomson E., Guralnik J.M., 2006, "Gradient of disability across the socioeconomic spectrum in the United States", New England Journal of Medicine, Vol.355, No.7, 695-703.

Mosher C., 2001, "Predicting Drug Arrest Rates: Conflict and Social Disorganization Perspectives", Crime & Delinquency, Vol.47, No.1, 84-104.

Moore D.A., Carpenter T.E., 1999, "Spatial analytical methods and geographic information systems: Use in health research and epidemiology", Epidemiologic Reviews, Vol.21, No.2, 143-161.

National Institutes of Health, 2020a, Opioid Overdose Crisis, [online]. URL: https://www.drugabuse.gov/drug-topics/opioids/opioid-overdose-crisis

National Institutes of Health, 2020b, Substance use in women research report: Sex and gender differences in substance use, [online]. URL: https://www.drugabuse.gov/publications/research-reports/substance-use-in-women/sex-gender-differences-in-substance-use

Nesoff E.D., Branas C.C., Martins S.S., 2019, "The geographic distribution of fentanyl-involved overdose deaths in Cook County, Illinois", American Journal of Public Health, Vol.110, No.1, 98-105.

Oh J., Shin E.H., 2003, "Inequalities in nonfatal work injury: the significance of race, human capital, and occupations", Social Science & Medicine, Vol.57, No.11, 2173-2182.

Park T.W., Lin L.A., Hosanagar A., Kogowski A., Paige K., Bohnert A.S., 2016, "Understanding risk factors for opioid overdose in clinical populations to inform treatment and policy", Journal of Addiction Medicine, Vol.10, No.6, 369-381.

Peterson R.D., Krivo L.J., 2009, "Segregated Spatial Locations, Race-Ethnic Composition, and Neighborhood Violent Crime", The Annals of the American Academy of Political and Social Science, Vol.623, No.1, 93-107.

Porter J.R., Capellan J., Chintakrindi S., 2015, "Social Disorganization Theory", In: Jennings W.G. (Ed.), The Encyclopedia of Crime and Punishment, New York, Wiley and Sons, 1-7.

Rhee T.G., Ross J.S., Rosenheck R.A., Grau L.E., Fiellin D.A., Becker W.C., 2019, "Accidental Drug Overdose Deaths in Connecticut, 2012-2018: The Rise of Polysubstance Detection?", Drug and Alcohol Dependence, Vol.205, 107671.

Ringle C.M., Wende S., Becker J.M., 2015, SmartPLS 3, Bönningstedt, SmartPLS.

Robinson A.L., 2006, "Reducing repeat victimisation among high‐risk victims of domestic violence: The benefits of a coordinated community response in Cardiff, Wales", Violence Against Women, Vol.12, No.8, 761-788.

Rudd R., Aleshire N., Zibbell J.E., Gladden M., 2016, "Increases in drug and opioid overdose deaths - United States, 2000-2014", Morbidity and Mortality Weekly Report, Vol.64, No.50, 1378-1382.

Sampson R.J., Groves W.B., 1989, "Community structure and crime: Testing social‐disorganization theory", American Journal of Sociology, Vol.94, 774-802.

Sampson R.J., Bartusch D.J., 1998, "Legal cynicism and (subcultural?) tolerance of deviance: The neighborhood context of racial differences", Law & Society Review, Vol.32, No.4, 777-804.

Scholl L., Seth P., Kariisa M., Wilson N., Baldwin G., 2019, "Drug and opioid-involved overdose deaths-United States, 2013-2017", Morbidity and Mortality Weekly Report, Vol.67, No.51-52, 1419-1427.

Seeman T., Merkin S.S., Crimmins E., Koretz B., Charette S., Karlamangla A., 2008, "Education, income and ethnic differences in cumulative biological risk profiles in a national sample of US adults: NHANES III (1988-1994) ", Social Science & Medicine, Vol.66, 72-87.

Shannon C.E., Weaver W., 1949, The mathematical theory of communication, Urbana, The University of Illinois Press.

Shaw C.R., McKay H.D., 1942, Juvenile delinquency and urban areas, Chicago, IL, University of Chicago Press.

U.S. Department of Justice, 2019, 2019 national drug threat assessment, [online]. URL: https://www.dea.gov/sites/default/files/2020-01/2019-NDTA-final-01-14-2020_Low_Web-DIR-007-20_2019.pdf

Vespa J., Medina L., Armstrong D.M., 2020, Demographic turning points for the United States: Population projections for 2020 to 2060, [online]. URL: https://www.census.gov/content/dam/Census/library/publications/2020/demo/p25-1144.pdf

Wilcox P., Land K.C., Hunt S.A., 2003, Criminal Circumstance: A Dynamic Multicontextual Criminal Opportunity Theory, New York, Aldine de Gruyter.

Inicio de página

Índice de ilustraciones

Título Figure 1: The Location Map of Study Area: Connecticut, USA
Créditos Meng, 2022
URL http://journals.openedition.org/cybergeo/docannexe/image/40286/img-1.jpg
Ficheros image/jpeg, 772k
URL http://journals.openedition.org/cybergeo/docannexe/image/40286/img-2.png
Ficheros image/png, 2,8k
URL http://journals.openedition.org/cybergeo/docannexe/image/40286/img-3.png
Ficheros image/png, 3,2k
URL http://journals.openedition.org/cybergeo/docannexe/image/40286/img-4.png
Ficheros image/png, 3,4k
Título Figure 2: Drug Overdose Death Rates and Local Moran’s I Cluster Analyses of Drug Overdose Death Rates in Connecticut
Créditos Meng, 2022
URL http://journals.openedition.org/cybergeo/docannexe/image/40286/img-5.png
Ficheros image/png, 345k
Título Figure 3: Spatial Variations in β Coefficient Estimates in the State of Connecticut for Explanatory Variables from the GWR model of Drug Overdose Death Rates
Créditos Meng, 2022
URL http://journals.openedition.org/cybergeo/docannexe/image/40286/img-6.png
Ficheros image/png, 553k
Título Figure 4: Spatial Variations in t Values for Explanatory Variables in Modeling Drug Overdose Death Rates in Connecticut
Leyenda Note: 1.96 and 2.58 are the cut-off values for t-test. When │t│ > 1.96, the β coefficient estimate for a variable is significant at a level of 0.05. When │t│ > 2.58, the β coefficient estimate for a variable is significant at a level of 0.01.
URL http://journals.openedition.org/cybergeo/docannexe/image/40286/img-7.png
Ficheros image/png, 378k
Inicio de página

Para citar este artículo

Referencia electrónica

Yunliang Meng, « Using Geographically Weighted Regression to Explore County Subdivision Level Predictors of Drug Overdose Death in Connecticut, U.S. », Cybergeo: European Journal of Geography [En línea], Cartografía, Iconografía, Didáctica, documento 1047, Publicado el 05 abril 2023, consultado el 18 abril 2024. URL : http://journals.openedition.org/cybergeo/40286 ; DOI : https://doi.org/10.4000/cybergeo.40286

Inicio de página

Autor

Yunliang Meng

Dr., Department of Geography, Central Connecticut State University, New Britain, Connecticut, 06053, United States of America. Email: mengy@ccsu.edu

Artículos del mismo autor

Inicio de página

Derechos de autor

CC-BY-4.0

Únicamente el texto se puede utilizar bajo licencia CC BY 4.0. Salvo indicación contraria, los demás elementos (ilustraciones, archivos adicionales importados) son "Todos los derechos reservados".

Inicio de página
Buscar en OpenEdition Search

Se le redirigirá a OpenEdition Search