Understanding Trends Over Time

One way of examining how much progress your community has made on its journey to better health is to look at the various individual measures in the County Health Rankings.

  • You can get an overall sense of your progress toward becoming a healthier community by looking at the individual health outcome measures for your county.
  • And, to see the impact of your work to address the many health factors that influence your health, you can look at individual health factor measures.

Rankings measures with trend graphs

There are currently fifteen ranked measures (premature death, adult obesity, physical inactivity, alcohol-impaired driving deaths, sexually transmitted infections, uninsured, primary care physicians, dentists, preventable hospital stays, mammography screening, flu vaccinations, unemployment, children in poverty, violent crime, and air pollution) and two additional measures (uninsured adults and uninsured children) with trend data available. There at least three types of questions you will want to consider for each of these measures:

  1. Is your county's estimate increasing, decreasing, or staying the same over time?
  2. Is your county’s trend better or worse than the state trend, or following a similar trend?
  3. Is your county’s trend better or worse than the national trend, or following a similar trend?

For each of the measures where sufficient trend data is available, you can view a detailed graph of the trend data by clicking the graph icons. We have color-coded each to help you answer Question 1:

  •  Your county is getting worse for this measure
  •  Your county is staying the same for this measure
  •  Your county is getting better for this measure

Additional guidance about how to interpret the trend graphs to understand progress in your community can be found here.

Measures with no trend graphs

What about other measures, where there aren’t any trend graphs? Data within the Rankings have many different forms. Our data come from about 25 different sources that use different rules or mechanisms for constructing the estimates we provide. Those differences prove very important when it comes to estimating trends.

Types of data

The major types of indicators the Rankings provide are outlined below. For each type, there is additional guidance available.

A. Single-year census estimates: These indicators are based on a single year of data, and use the entire population as their sample.

  1. Mapping Medicare Disparities (health care quality): trend graphs available
  2. Sexually transmitted infections: trend graphs available
  3. High school graduation
  4. Primary care physicians: trend graphs available
  5. Dentists: trend graphs available
  6. Mental health providers
  7. Access to exercise opportunities
  8. Food environment index
  9. Social associations

These are the simplest measures with regards to estimating trends. However, some of our measures have changed over time, or are too new to have data collected. Access to exercise opportunities and the food environment index are newly developed measures that do not have trend data available. Similarly, other measures on this list do not have enough years of consistently collected data for trends to be analyzed. However, in some cases, such as for the mental health providers measure, state professional associations may have more data available, and should be consulted by communities with a particular interest in this topic.

B. Multi-year census estimates: These indicators are based on more than one year of data, and use the entire population as their sample.

  1. Premature death: trend graphs available
  2. Violent crime rates: trend graphs available
  3. Alcohol-impaired driving deaths: trend graphs available
  4. Low birthweight
  5. Teen birth rate
  6. Injury deaths
  7. Drinking water violations

Estimates that are based on multiple years are more difficult to use for examining trends over time. Data providers sometimes combine years of data in order to protect privacy or provide stability to the measures. However, this substantially reduces the ability to detect trends because estimates are based on overlapping time periods. For example, in 2014, our measure of teen births is based on data from 2005-2011 whereas in 2013, we reported on teen births for 2004-2010. The federal government and most states have privacy protections in place regarding data about births and deaths (known as vital statistics).

Premature death trend graphs use 3 year rolling averages with the middle year displayed on the x-axis. Whereas for Violent Crime and Alcohol-impaired driving deaths we are able to provide single year estimates on the trend graphs.

If your county has sufficient population, you may be able to get access to single-year estimates. We have provided links to state-specific vital statistics on the state page. Using this link you should be able to determine the smallest number of combined years that you can examine without violating privacy protections. Once you have downloaded that data, you can conduct your own trend analysis.

C. Multi-year estimates, sampled data, not modeled.

  1. Children in single-parent households (ACS)
  2. Some college (ACS)
  3. Income inequality (ACS)
  4. Driving alone (ACS)
  5. Long commute (ACS)
  6. Severe housing problems (CHAS)
  7. High school completion (ACS)

Indicators calculated from surveys that are drawn from a sample of the population offer additional challenges for assessing trends. In the best case, groups can access the individual raw data for a single year, and estimate trends using regression estimates accounting for sample design, and weighting. However, these data are often hard to access, and their analysis requires substantial statistical experience.

The data source used in the County Health Rankings that falls into this category is the American Community Survey. In order to ensure the comparability of data across counties, the Rankings use five-year combined ACS estimates produced for all US counties. However, for many counties, there are more precise one- or three-year estimates that they can use for assessing trends.

Counties with populations of 65,000+ can access single year estimates. Counties who can access single-year estimates can use either regression analysis or percent change for these data.

D. Single-year estimates, sampled, modeled data.

  1. Unemployment: trend graphs available
  2. Air pollution- daily fine particulate matter: trend graphs available
  3. Children in poverty: trend graphs available
  4. Uninsured: trend graphs available
  5. Poor or fair health (BRFSS)
  6. Poor physical health days (BRFSS)
  7. Poor mental health days (BRFSS)
  8. Adult smoking (BRFSS)
  9. Excessive drinking (BRFSS)

Several of the measures included in the Rankings are based on single-year estimates of sampled data, but the data providers modeled these data to come up with county-level estimates. Consequently, it is not possible to acquire and analyze the raw data. In these cases, we have conducted tests to examine trends in these estimates -- the findings can be seen by looking at the trend graph icon and viewing additional notes on the trend graphs. These findings come with certain caveats because the statistical power of our tests is substantially decreased without access to the underlying raw data, and the trends may be overestimated due to smoothing undertaken in the modelling process.

E. Sampled, modeled data smoothing estimates over multiple years.

  1. Adult obesity: trend graphs available
  2. Physical inactivity: trend graphs available

These two measures are based on sampled, modeled data based on multiple years: physical inactivity and adult obesity. We have conducted tests to examine trends in these estimates -- the findings can be found by examining the trend graph icon and looking for additional notes on the trend graphs. We used constructed estimates with non-overlapping years of data for trend analysis, but this method has caveats since a great deal of statistical power is lost using this method. An alternative option for examining trends in physical inactivity and adult obesity is to use the raw BRFSS data in the same manner as outlined above under item C.