Who is using faster internet?
As the federal government, along with states, gets ready to make a once-in-a-lifetime investment in broadband infrastructure, the concept of the digital divide remains somewhat the same as it was back in the mid-1990s, when the term was coined. Namely, who has access to the internet and who does not?
Like anything, the devil is in the details. For decades, a yes/no definition has been used to assert whether somebody has access to the internet. This definition has typically been associated with speeds. All internet connections consist of a download and upload speed and measured in megabits per second or Mbps. The official broadband definition is 25 Mbps download and 3 Mbps upload, or 25/3. However, some federal and state programs are increasingly using 100 Mbps download and 20 Mbps upload. There is also latency and jitter, but we won’t go that technical rabbit hole in this article.
Increasingly, being on the “right” side of the divide is moving beyond a yes/no access at home and more toward determining whether a user’s connection satisfies their needs. As we all saw during the Covid-19 pandemic, online activities became critical. And these activities were not possible without an adequate, reliable, and affordable connection. In fact, before the pandemic most internet users consumed or downloaded. However, increasingly due to e-learning, remote work, and running online micro businesses, users are producing or uploading more.
While there are areas with no connectivity whatsoever, increasingly the question is becoming: Who is using the internet at faster speeds? Are connections symmetrical—where download and upload speeds are similar? Or are most internet users driving down a six-lane paved road (faster download speeds) and driving back on a dirt road (slower upload speeds)?
+++BREAK+++
To address these questions, we looked at speed test results from the Speedtest by Ookla Global Fixed Network Performance Maps and the latest American Community Survey (ACS) 2017-2021 from the U.S Census Bureau. The Ookla results are based on customer-initiated speed tests. These are tests that anyone with an internet connection can initiate to see how fast their connection is. The data from those tests are published quarterly. We averaged by the number of speed tests for 2021 to match the most recent release of ACS data. Results are published in “quads” or 600 x 600-meter geographic tiles and for this study, were over-layed to Census blocks and aggregated accordingly. These speed tests include only those conducted over fixed broadband networks (not cellular or mobile networks) from web browsers as well as mobile and desktop apps.
This is how we crunched the numbers. First, based on the 2017-2021 ACS, there were 84,414 tracts available (not including Puerto Rico). For this analysis, only census tracts with population and at least 50 speed tests (to avoid outliers due to a low number of speed tests) were utilized resulting in 83,107 or 98% of all tracts available. Second, specific socioeconomic variables (e.g., percent rural, percent White non-Hispanic, median household income, etc.) were calculated for each census tract and then divided into deciles or 10 equal groups (each group has roughly the same number of tracts). Lastly, weighted averages—by number of speed tests—for download and upload speeds were calculated for each decile.
Figure 1 shows the average download and upload speed (vertical axis) by deciles (horizontal axis) based on the percentage of population considered rural. A census block was considered rural if it had a housing unit density of less than 425 per square mile per the 2020 Census.
As expected, as the share of a tract’s population is more rural, the average download and upload speeds are slower. Consider that for decile one, where there was no rural population or 0.0%, the average download speed was 202 Mbps, while for decile 10, where the share of rural population was at least greater than 85.3% and up to 100%, was 119 Mbps—almost 100 Mbps slower. Same trend is seen with average upload speeds where in the most urban tracts the average speed was 64 Mbps compared to 39 Mbps in the most rural tracts.
Notice that average download and upload speeds seem to not change much until the share of rural population reaches close to one-third or 31.6% (decile seven). On the other hand, the download/upload asymmetry does not change regardless of if the share of population is urban or rural. Average download speeds were 3x faster for both the most urban (decile one) and the most rural tracts (decile 10).
Figure 1. Average Download & Upload Speeds in Mbps by Rural Decile
How about age groups? Research has found that age groups play a significant role in internet access and use. Figure 2 shows census tracts divided into 10 equal groups based on their share of those age 65 or older and their average download and upload speeds. As expected, as the share of this group increases, the average download and upload speeds decreases. The average download speeds in tracts where less than 8% of the population were age 65 or older (decile one) was 205 Mbps compared to 163 Mbps in tracts where more than one-quarter and up to100% of the population were this age. A similar trend is observed with average upload speeds where decile one had an average upload speed of 72 Mbps compared to 44 Mbps for decile 10.
The download/upload asymmetrical issue is worse in areas with a higher share of those ages 65 or older. Consider that in tracts with the lowest share of this age group (decile one), average download speeds were 2.8x faster compared to upload speeds. However, in tracts where at least a little more than one-quarter to up to 100% of the population were this age group (decile 10), download speeds were 3.7x faster compared to upload speeds. A possible explanation for this may be that senior folks may afford or choose slower plans overall and these in turn, have the largest difference between download and upload speeds.
Figure 2. Average Download & Upload Speeds in Mbps by Age 65 or Older Decile
Figure 3 looks at race/ethnicity, specifically the share of White, non-Hispanics across 10 groups and average download/upload speeds. Surprisingly, as the share of White, non-Hispanics increases, the average download and upload speeds decrease. For decile one, where the share of White, non-Hispanics is 11.4% or less, average download and upload speeds were 204 and 58 Mbps, respectively. These numbers go down to 138 and 37 Mbps download and upload respectively for decile ten, where the share of White, non-Hispanics is greater than 93%.
We are not sure what is going on. One explanation may be that this variable is capturing a “double whammy”. We know that rural areas have slower speeds on average and have a higher share of White, non-Hispanic population. Two, a higher share of White, non-Hispanic is older (e.g., a higher share of those age 65 or older), who also tend to use the internet at slower speeds. In a future analysis, we will look at only rural tracts and see if these patterns hold.
Figure 3. Average Download & Upload Speeds in Mbps by White, non-Hispanic Deciles
Figure 4 shows average download and upload speeds based on median household income deciles. As expected, areas with higher median household incomes also reported higher download and upload speed tests. For decile one, where the median household income was $37,000 or lower, the average download and upload speeds were 178 and 48 Mbps, respectively. These numbers increased to 199 and 71 Mbps for download and upload speeds respectively for decile ten where the median household income was at least $250,000.
A similar download/upload asymmetry issue plays out here as well. Download speeds were 3.7x faster than upload speeds in decile one (less wealthy) compared to 2.8x faster in wealthier Census tracts (decile 10).
Figure 4. Average Download & Upload Speeds in Mbps by Median Household Income Deciles
Lastly, Figure 5 looks at the share of workers working from home and average download and upload speeds. As expected, Census tracts with a higher share of workers from home had higher average download and upload speeds. Areas where less than 2% of workers worked from home clocked 176 Mbps download and 49 Mbps upload compared to 195 Mbps download and 70 Mbps upload in areas where at least one-fifth of workers worked from home (decile 10).
Here too, the asymmetrical download/upload issue is seen. Download speeds were 3.8x faster compared to upload speeds in decile one versus 2.7x faster in decile ten (a larger share of workers from home). The lower asymmetry may be explained by the fact that workers from home require faster upload speeds as well.
Figure 5. Average Download & Upload Speeds in Mbps by Working from Home Deciles
In a future analysis, we will look at rural tracts only (defined as those where at least 50% of their population is rural) and see if these patterns hold. Most of what we uncovered here was expected. However, the race/ethnicity finding is surprising. Again, additional number crunching is needed to peel the multiple layers involved but while we do that, what do you think is happening Daily Yonder readers? Why are White, non-Hispanic using the internet at slower speeds? This seems to go contrary to the assumption that minorities are on the wrong side of the divide. Look forward to reading your comments.
Roberto Gallardo is vice president for engagement at Purdue University, director of the Purdue Center for Regional Development, and an associate professor in the Agricultural Economics Department.
Republish our articles for free, online or in print, under a Creative Commons license.
