# 1 vs 5 year estimates regarding annual difference in median income

We are trying to calculate the difference in median annual income at the county level between those who are high school graduates vs. those who are less than high school graduates.  There are both 1 year and 5 year estimates from the ACS available.  The goal is to provide a yearly estimate of the difference in income between the two groups (e.g., “in 2013, the annual difference in median income between high school graduates and those with less than a high school education was \$X”).  With 1 Year estimates, we could make a statement like that.  With 5 year estimates, the statement would have to be more like, “in 2013-2017, the annual difference in median income between high school graduates and those with less than a high school education was \$X”.

Understanding that the 5 year estimates are more reliable because the samples are larger and also have smaller margins of error, what would be the better source to calculate the difference by year (i.e., 2013, 2014)?  The 1 year estimates are exactly that, even though the margins of error are larger.  The 5 year estimates share overlapping years, so they are not a true indicator of the difference in a single year.  Since we are looking at county level data, the 1 year estimates include a large sample size, but in comparing the 1 Year and 5 Year data, the actual median values are not wildly different, but the margins of error are very different.  Even though we would ideally like to make a statement about differences by year, would it be better to use the 5 year estimates since they are more reliable?

• Hi Martha,

It is all in how you want to show the data, if you do indeed need annual, then you should use the 1-year estimates. If you're using the 5-year estimates, then you would need compare every 5 years (i.e. 2013-2017 with 2008-2012).

keep in mind that ACS estimates are PERIOD estimates. they are estimates derived from surveys collected throughout the survey period. In other words, 2017 1-year data are based on surveys collected throughout all of 2017 and the 2017 5-year estimates are based on surveys collected throughout all 5 years, 2013-2017.

So a more accurate statement, if using 5-year data, would be something like "For the period 2013-2017, the difference from the prior period (2008-12) in median income between high school graduates and those with less than a high school education was \$xx." Maybe someone else can word that better, as it is a long sentence!

Although, to be honest, in my office we generally refer to the reference year for the 5-year estimates, so we would probably instead say something like "From 2012 to 2017 the different in median income between high school graduates and those with less than a high school education was \$x." Then I'd footnote in the sources to explain that "2012" and "2017" refer to data from the 5-year ACS period estimates.

I would be interested to see how others respond.

A factor to consider here is the county population size. If you are talking about New York County, NY (ie Manhattan) then certainly use the 1 year estimates. If you counties include ones with a population of under 250,000 I'd stick with the 5 year estimates. Even in a city of 100,000, where I work, the 1 year estimates can be quite unstable from year to year. The more uneven your data the more time you will need to spend explaining and defending that unevenness.

It seems to me what is most interesting here is how the trend changes over time - what size is the gap, does it get bigger, smaller, in absolute or percentage terms, etc. Comparing 5 year data will allow you to answer those types of questions.

This is where I miss the 3 year ACS estimates that were discontinued several years ago. For my purposes, they were a good trade off between currency of information and size of margin of error.
• Use the 1 year to show your readers where the difference is trending and the 5 year to give an estimate as to how much that change has been comparing two five year periods, 2008-12 to 2013-17.
• In reply to Richard Taylor:

I too miss the 3 year estimates!

Remember that when you use the median income statistic you're still talking about a one-year income amount. So you can say, e.g., "between 2013 and 2017, the difference in annual median income between those who've graduated high school and those who have not is xxx.

You don't say what specific geography you're working with. In my opinion, the 1 year data are only truly useful for areas with, say, 200K population or more, or 250K as Cliff recommends. The MOEs on the smaller areas are really just too high, even though the bureau publishes for those of 65K ore more. If you have big counties, by all means use the one year data.

(It's all still a lot better than when we had only decennial numbers; around this time of decade the data were so, so stale!)
• Thank you all for the responses! I have some additional follow up information as well as questions:

1. We are using data from Shelby County, TN, which has a population of over 900,000. So it seems that using 1 year data is acceptable.

2. The purpose of our study is not to determine whether the difference in median income between high school graduates and non-graduates is getting bigger or smaller over time, but to determine the cumulative economic impact of having a high school diploma vs. not over time. In other words, what is the median difference in income (in dollars) in 2013, 2014, 2015, 2016, and 2017 between those with a high school diploma vs. not. Then over the entire timeframe from 2013-17, what is the cumulative difference in income between high school graduates and non-graduates (adding the differences for the individual years). So, we’d like to be able to say something to the effect, “The annual difference in median income between high school graduates and non-graduates in 2013 was \$8,046” (actual data, but from the 5 year estimate). Then make a similar statement for each individual year 2014-17, with the ultimate statement something to effect of, “Between 2013 and 2017 the cumulative difference in median income between high school graduates and non-graduates was \$34,663” (actual data, but from the 5 year estimate). The point would be to show how having a high school diploma vs. not affects income, and how that difference accumulates over time. We also want to say how that effect for the county is compounded based on the number of high school graduates each year (e.g., if there were 500 graduates in Shelby County in 2013, that is a cumulative difference for the county of \$4,023,000 (\$8,046 x 500)).

3. Do you typically report a range of values for the median based on the margin of error. And for our example, whether they would report a range of values for the difference between high school graduates and non-graduates based on the margin of error.
Thank you!