Skip to main content

Is This Why Democrats Support Education Funding?

This post started off simply enough: I found some cool data on changes in educational attainment over time. I was going to take a look at how far we'd come as a nation in the last 40 years (even though I had already published this), and show where the biggest gains were.

It wasn't very compelling, at least at first.

Then, I decided to get ambitious (my wife was in her evening class, so I had the night free), and wondered if there were any interesting connections between changes in educational attainment and voting patterns in 2016.  I found a data set with election results by county, (a handful of counties are missing) and merged it in. And thus, this.

Before I start, there are a few points to make about the data.  First, the definitions changed slightly over time.  For instance, in 1970, the field is labeled "College degree," while in 2010, it's labeled "Four or more years of college."  Not the same thing, but we'll have to go with it for now.  Also, 2010 is not really 2010; it's the data from the five year American Community Survey of the Census Bureau, but there's no reason to believe it's not as accurate as the census itself.  In fact, the ACS is used to test the accuracy of the census.

So, onward.

Using the tabs across the top of the visualization:

First, a scattergram, plotting attainment in 1970 and 2010.  The regression line suggests that attainment has essentially doubled in 40 years; those bubbles (counties) above it have done better; those below worse.  Bubbles are sized by votes in the county in the 2016 presidential election.  If you want to look at just one candidate, use the highlighter function.

Note that the counties that went for Clinton tend to be larger (more urban), with higher levels of attainment (moving toward the top right) and more above the line.  Counties that went for Trump tend to be the opposite.  And of course, there are many exceptions (Johnson County, Kansas; Apache County, Arizona), and clearly the binary blue/red can be misleading; some victories are by a point, some by 15 points or more.  Finally, it's almost certain that much of the change in attainment is due to people moving in and out; not everyone lives where they were born. But it's still interesting.

Second, the bar chart (I know from experience that many people won't click on the second tab. Please. You will be glad you did).  The x-axis is broken into college degree attainment. For instance, the long bars in the center show counties where 30--34.99% of adults have a college degree.  You can see how many votes these counties cast for Trump, and how many for Clinton.  I double checked this; it is perhaps the best story I've ever told with one chart.  And although the left end changes as you select single states (using the filter at the top), the right end is fairly stable.

Finally, the last chart just shows three variables: 1970 and 2010 college-degree attainment, and the change over time.  See the box to the right of the chart if you want to sort the data.

Admittedly, this election was a strange one, so perhaps there are no lessons to be learned.  But over the past few decades, Republicans have been fairly staunch opponents of increased educational funding, and you have to wonder if this doesn't explain why; people who lived in areas with higher levels of education voted for Democrats in the last election.

Fifty years ago, the Republicans were the party of the college-educated, white collar classes; the Democrats the blue collar, working-class, high school educated citizens.  That's all changed, if 2016 is any indication.

Agree? Disagree? Let me know in the comments below.


Comments

Popular posts from this blog

The Highly Rejective Colleges

If you're not following Akil Bello on Twitter, you should be.  His timeline is filled with great insights about standardized testing, and he takes great effort to point out racism (both subtle and not-so-subtle) in higher education, all while throwing in references to the Knicks and his daughter Enid, making the experience interesting, compelling, and sometimes, fun. Recently, he created the term " highly rejective colleges " as a more apt description for what are otherwise called "highly selective colleges."  As I've said before, a college that admits 15% of applicants really has a rejections office, not an admissions office.  The term appears to have taken off on Twitter, and I hope it will stick. So I took a look at the highly rejectives (really, that's all I'm going to call them from now on) and found some interesting patterns in the data. Take a look:  The 1,132 four-year, private colleges and universities with admissions data in IPEDS are incl

Freshman Migration, 1986 to 2020

(Note: I discovered that in IPEDS, Penn State Main Campus now reports with "The Pennsylvania State University" as one system.  So when you'd look at things over time, Penn State would have data until 2018, and then The Penn....etc would show up in 2020.  I found out Penn State main campus still reports its own data on the website, so I went there, and edited the IPEDS data by hand.  So if you noticed that error, it should be corrected now, but I'm not sure what I'll do in years going forward.) Freshman migration to and from the states is always a favorite visualization of mine, both because I find it a compelling and interesting topic, and because I had a few breakthroughs with calculated variables the first time I tried to do it. If you're a loyal reader, you know what this shows: The number of freshman and their movement between the states.  And if you're a loyal viewer and you use this for your work in your business, please consider supporting the costs

Yes, your yield rate is still falling, v 2020

I started doing this post on a regular basis several years ago, in response (if I recall) to a colleague talking about their Board of Trustees Chair insisting that "all we need to do" to bring enrollment back to its former level is to get the yield rate up.   That's the equivalent of saying all you need to do is straighten your drives and cut ten putts from each round, and you'll be a great golfer.  Moreover, it's based on the assumption that a falling yield rate is based on something you're doing or not doing.  The challenge is much larger, and a lot harder to address.  It's not a switch you flip. So we've got this: A look at applications, admits, and enrolls over the last twenty years, and three key ratios that are based on those numbers: Admit rate, or the percentage of applicants offered admission; yield rate, or the percentage of those offered admission who enroll; and the lesser-known draw rate, which is calculated by dividing the yield rate by t