Disparities in Poverty, Life Expectancy and Healthcare Services

From Cmsc734_f12
Jump to: navigation, search

Contents

INTRODUCTION

Although the United States is home to some of the wealthiest people and counties in the world, there is an alarming percent of the population that is both below the poverty level and without insurance. The purpose of this analysis is gain insights into the disparities in poverty, insurance, health care services, and health outcomes in the United States. First, this project investigates the median and variability of the uninsured poverty population within states. Next, the relationships between a county's median income, level of the uninsured poverty population, physically unhealthy days, and availability of healthcare services are assessed. Finally, how physically unhealthy days and being uninsured and in poverty relates to life expectancy is examined.

TEAM

  • Allan Fong, PhD Student, Department of Computer Science
af2012 at gmail dot com
  • Brendan Fruin, Master's Student, Department of Computer Science
brendan at cs dot umd dot edu

DATA USED

1. 2009 Estimates of Uninsured Populations at or below 138% of Poverty[1] in counties with a population of 65,000 or more

  • Population for all incomes by county
  • Uninsured population for all incomes by county
  • Population living at or below 138% of poverty by county
  • Population uninsured living at or below 138% of poverty by county

2. 2010 Businesses by County Census Data[2]

  • Number of a particular type of business by county classified by the North American Industry Classification System
    • Health care and social assistance
      • Ambulatory health care services
      • Hospitals
      • Nursing and residential care facilities
      • Social assistance
  • Number of overall employees in a particular business by county

3. 2010 County Health Rankings[3] 'contains rankings and data details for each county compared to other counties including:

  • Geographic Identifiers
  • Number of deaths under age 75 (2004-2006)
  • Aggregate Population under age 75 (2004-2006)
  • Age-adjusted years of potential life lost (YPLL) rate per 100,000 people
  • Percent of adults that report fair or poor health
  • Average number of reported physically unhealthy days per month

4. 2009 County Life Expectancy[4]

5. 2010 Median Income Level per County[5]

6. 2008 Presidential Election Results By State[6]

  • State voting outcomes in 2008 Presidential Election
  • Battleground/Swing States


The data sources above were combined in order to make a "Master Table" which was used in Spotfire[7] to make the insights and headlines outlined below. The "Master Table" contains 797 rows and 97 columns.

Notes:

  • Many of the datasets listed contain more information than described. The above is a complete list of the information that was extracted in order to present the visualizations.
  • For simplicity, the ratio of a particular group will refer to the number of people in that particular group over the population of the county.
  • People living at or below 138% of poverty will be noted as people living in or below poverty.

HEADLINES

Headline 1: The ten states with the lowest median levels of poverty uninsured all voted democrat in the 2008 Presidential Election

Updated Boxplot for Census.jpg

Fong fruin box plot.png

The box plot shows, for each county, the percent of people in poverty who are uninsured. The counties are binned by state and the states are ordered by descending median percent. Visualizing the distributions, variances, and outliers of the insurance rate of people in poverty highlights the large disparities in insuring the poor both between and within states. Furthermore, the ten states with the lowest percent of people living in poverty without insurance are all “blue” states (states that voted Democrat in the 2008 presidential election). All of these states were considered strongly Democrat states except Pennsylvania which could be considered a “swing” state. Out of the ten states with the highest percentage of uninsured in poverty, five were "red" states and three (Nevada, Florida, Colorado) were "swing" states.

Note: Vermont and Wyoming did not have enough counties with populations over 65,000 to draw their figures in the box plot. However, their descriptions are shown in the table below the graph.

Headline 2: In states with large variances in the poverty and uninsured population, less wealthy counties generally report more sick days despite the ratio of overall health care employees, except in counties with larger ratios of social assistance employees

Fong fruin treemap.png Fong fruin parallel.png

In this treemap, states are numbered by their average median income (1-Maryland has the highest average median income), sized by the variance in the percent of the population in poverty and uninsured (the larger the rectangle, the larger the variance), and colored by the average of physically unhealthy days (the more red a state, the more physically unhealthy days were reported). States with some of the lowest median incomes, such as Oklahoma, Alabama, Kentucky, and West Virgina, have the largest number of physically unhealthy days. Physically unhealthy days was chosen as a metric because other health metrics, such as life expectancy and premature deaths, are effected by many more factors that are outside the scope of this project. Nevertheless, the number of physically unhealthy days is correlated with the life expectancy in a county. Furthermore, most of the states with the largest variances in the uninsured poverty population, also widely vary in their care for these populations. This is especially true for Arizona, Georgia, Louisiana, New Mexico, and Texas.


The parallel coordinate plot shows counties in Arizona, Georgia, Louisiana, New Mexico, and Texas. The visualization highlights the percent of the population that is in poverty and without insurance, the median household income, the ratio of health care and social assistance employees for the population, and the number of physically unhealthy days for a given county. A surprising insight is that counties with high median income typically do not have high health care and social assistance employees ratios. Another insight is that counties with less wealth generally report more sick days despite the ratio of health care employees. However, counties that have a larger ratio of social assistance employees despite lower overall health care employees tend to have less physically unhealthy days.

Headline 3: Compared to males, females have a tighter clustered relationship between poverty, lack of insurance, physically unhealthy days and life expectancy; though the life expectancy for males tend to be overall more negative trending.

Update Female Life Expectancy for Census.jpg Update Male Life Expectancy for Census.jpg

Fong fruin LE unhealthy scatter.png Fong fruin LE Uninsured Scatter.png

The left scatter plot shows the life expectancy in years for females (in pink) and males (in blue) for each county in relation to how many physically unhealthy days are reported in the span of a month. Their respective linear regression lines show how the life expectancy of each gender is negatively correlated with the number of reported physically unhealthy days. This visualization shows a surprising insight that a male’s life expectancy decreases by almost two years for each additional physically unhealthy day reported per month while a female’s life expectancy decreases by less than 1.5 years.


The right scatter plot shows the life expectancy for females and males for each county in relation to the percent of a county’s population that is both uninsured and in poverty. Note that the data points have been “jittered” or slightly adjusted along the horizontal axis to better display overlapping values. A quadratic best fit curve was drawn for each gender to show how life expectancy is affected by the percent of the population that is both uninsured and in poverty. A quadratic curve was used as opposed to a straight line in order to show the non-linear relationship between the uninsured poverty and life expectancy variables. A “leveling” out effect appears for both males and females when the percent of uninsured poverty exceeds 8%. The steeper slope of the male curve shows that a male’s life expectancy is more affected by the percent of a county’s population uninsured and poverty than the female’s life expectancy.

CONCLUSION

The visualizations presented highlight several insights on the disparities in poverty, life expectancy and health services in the United States. There are large disparities in insuring the poor both between and within states. In addition, the 10 states with the lowest median levels of poverty uninsured all voted Democrat in the 2008 presidential election. Furthermore, in states with large variances in the poverty and uninsured population, less wealthy counties generally report more sick days despite the ratio of overall health care employees. However, counties that have a larger ratio of social assistance employees despite lower overall health care employees tend to have less physically unhealthy days. Lastly, poverty, lack of insurance, and physically unhealthy days correlate stronger to lower life expectancy of males as opposed to females.

This analysis is limited by several factors such as the accuracy of the data, county health estimations, how facilities self-classified with the NAICS codes, and missing or incomplete data. There are also several factors in addition to the ones explored in this project that can effect healthcare in a specific county. Such factors include proximity to a major academic teaching hospitals, climate, pollution levels, and the education level of the residents. The effects of these factors warrant additional exploration and research.

The team is especially grateful to Dr. Seth Powsner and Mr. Eric Newburger for their feedback and insights concerning the visualizations and for their domain knowledge expertise.

CRITIQUE OF SPOTFIRE

Advantages:

  • Spotfire accepts a wide variety of data file formats.
  • Spotfire was a great tool for quickly visualizing and filtering data. Filtering was possible for both numeric and string values.
  • Since everything is in memory, the response/update time in Spotfire is incredibly fast.
  • There was a wide selection of visualization options that are both informative and visually aesthetic.
  • Spotfire provided tutorials to help users quickly get started.
  • Using the same filter to link different visualizations on the same page is a powerful exploratory tool for users especially to understand change relationships.


Limitations:

  • Using the same filter to link visualizations on different pages is very inconvenient and does not allow comparative explorations of data. Perhaps a temporary locking feature for visualizations could alleviate this problem.
  • Modifying the text for the axis and legend in Spotfire is not intuitive and involves several steps. Such features are important especially when generating graphics for presentations and reports.
  • Spotfire has very limited online documentation.