How accurate are our 50-State population projections?

Image

Whenever a prediction, forecast, or projection for the future is made, the natural question that arises is “how good is it?” After our office prepared population projections for the 50 states and D.C., we’ve been eagerly awaiting the 2020 Census results so we can answer this question for ourselves.

The answer: the accuracy of our projections is very good!

While the newly released 2020 Census count provides a once-in-a-decade snapshot of the current size of the population, it also allows us to evaluate the precision of the population projections our office produced for the nation as a whole and for all 50 states by comparing it to the actual Census head count.

Nationally, we projected the 2020 U.S. population to be 332,527,548, which is slightly over the Census count of 331,449,281, representing a difference of 0.33%. Our projection is also slightly closer to the actual count than the U.S. Census Bureau’s latest¹ projection of 332,639,000.

In the 2010s, as the U.S. Census Bureau stopped producing state-level population projections, our office began to develop projections for 2020, 2030, and 2040 for each state and Washington D.C.  We decided to fill this void because many federal agencies, state governments, businesses, and non-profits rely on a consistent set of state-by-state projections for their planning and research needs. Our projections, for example, were used by the Centers for Disease Control and Prevention, U.S. Environmental Protection Agency, and National Park Service to better plan for the future of cardiovascular disease cases, climate change, and park visitors respectively, to name a few. Figure 3 (at the end of the article) lists a selected compilation of how our projections were used.

With the release of the 2020 Census results in April this year, we evaluated our projections for the 50 states and Washington, D.C. The key findings are summarized below, as well as illustrated in Figure 1.

1. Of all 50 states, the difference between our projected population and the Census count ranges from 0.06% for Kansas to 3.4% for Idaho. In other words, all differences are under 3.5%, a highly compact, narrow range.
2. Furthermore, 30 states were within 1% difference from the Census count, including 16 states being within 0.5%.
3. As a result, the average difference of the 50 states is 1.1%. This is very low, compared to the 3%-5% range based on the Census Bureau’s empirical evaluations of its own projections in previous decades².
4. Washington, D.C. is the only outlier with the highest difference of 6.2%. This happened back in the 2000s as well, when the Census Bureau released its final set of 50 state projections in 2005 for the 2010 population – D.C. was still off by 12%, higher than every other state projected.
    

Image

The map in Figure 2 further breaks down the states into those that were over-projected and under-projected. The 21 states shaded in blue were projected higher than the actual counts, while the 29 states shaded in yellow were projected lower. States with an asterisk indicate a difference within +/-1%.

It is interesting to note that we over-projected the most southern and western states and under- projected New England and the Midwest states. The fast growing sun-belt region did not grow as fast as anticipated, and the slower growth states in the northeast and Midwest were not as slow either.

For our home state of Virginia, the 2020 projection was very close – only 0.27% higher than the Census count.
 

Image

While we take pride in the accuracy of our national and state-level projections, the “crystal ball” we use—methodology, input data, and assumptions—is actually quite simple. Some lessons we learned through our projections development are:

1. The extrapolation method is a highly efficient and effective method for projecting population totals.
2. Input data quality is critical. We used the 2010 Census data and 2017 Census estimates to construct the 2020 state projections in our 2018 vintage. Our projections accuracy is highly dependent on the quality of input data produced by the U.S. Census Bureau.
3. With high quality input data, simple assumptions work better than elaborate assumptions, just as fresh seafood needs minimum seasoning.
4. Projections need to be updated every two or three years to better capture the latest population trends. The projections we produced in 2018, for example, were more accurate than the ones we produced in 2016.
    

Acknowledgements: Rebecca Draughon, our PhD student research assistant, contributed to this post.

Footnotes:

1. U.S. Census Bureau released the 2017 National Population Projections series as their latest vintage.
2. Smith, Tayman, and Swanson, State and Local Population Projections Methodology and Analysis, Kluwer Academic/Plenum Publishers, 2001, page 327