John Stossel was on Fox and Friends this morning to promote an upcoming show about public schools. Remember, this is the guy who gave us Stupid in America — his ABC documentary from a few years back about our allegedly failing schools. During his segment, he claimed that “America has tripled spending, but test scores haven’t improved.” The culprits? Teachers unions, school boards and other unnamed bureaucrats. Viewers were then shown a graph that indeed featured a flat line representing test scores over 40 years (improvement 1 point) with a second line escalating to $149,000 over the same period. The source was given as NCES. This got my fact-checking synapses sparking.

While I could not find the exact graph they showed on TV, Stossel did post this rather snazzy display on his blog with the same data:

Go ahead and take a moment to admire the work of the Fox News graphics department. Ok, now let’s talk data. This chart shows scores for three subjects (math, reading and science) and dollar figures (the “cost of education”) from 1970 to 2010. While not noted, I’m assuming the data source is still NCES.

This may get a little wonky, but stay with me. NCES reports trend data over four decades for only two tests: the National Assessment of Educational Progress (NAEP) Long-Term Trends (LTT) and the SAT. NCES also has international test scores, but that data only goes back to the 1990s so that couldn’t be what Stossel used. The SAT does not assess science, which leaves NAEP LTT as the only possibility. It’s not a perfect match. The last NAEP LTT administration was in 2008 although Stossel’s chart shows data to 2010. But I’m going to assume that he fudged a little on the timeframe because nothing else qualifies.

NAEP LTT is given to a representative sample of students age nine, 13 and 17. I’m also going to assume that his analysis is based on 17-year-olds because the data matches his in reading and comes closest in math (more on this later). Between 1971 and 2008, LTT reading scores for 17-year-olds have been relatively flat, posting an increase of just 1 point (not 1% as shown on Stossel’s chart, but we’ll blame the designer for that common mistake). Here’s what it looks like:

Now let’s have some fun. Let’s look at the same test scores disaggregated by race and ethnicity:

Note that every group improved more than the overall score did: White 17-year-olds by 2 points with their Black and Hispanic classmates gaining a whopping 25 and 17 points respectively. This gives me a chance to talk about Simpson’s paradox. The paradox occurs when “a trend that appears in different groups of data disappears when these groups are combined, and the reverse trend appears for the aggregate data.” In this case, the overall trend for 17-year-olds is flat while each group gained, some groups by a lot. The reason is that the distribution of racial/ethnic groups has changed significantly between 1975 and 2008. Here is the distribution of the NAEP samples for the two years:

The proportion of Black and Hispanic 17-year-olds is larger while the proportion of White students in 2008 is 25 percentage points lower than it was in 1975. Even though Black and Hispanic performance also increased by a lot, they were still lower-performing than their White peers in 2008. Thus, all groups gain, but when their performance is combined the overall trend is flat.

Clearly, no one would argue that an achievement gap, though improving, is acceptable and we can move on to other things. But it’s just as absurd to look at these gains and find evidence of failing schools, as Stossel does. And the absurdity doesn’t end there. Stossel, in turns out, is a master cherry picker of data. Let’s look at the rest of NAEP Long Term Trends:

- Reading, 13-year-olds, 1971-2008: Overall scores +12; Black students +23; Hispanic +24.
- Reading, 9-year-olds, 1971-2008: Overall +5; Black +21, Hispanic +10.
- Mathematics, 17-year-olds, 1978 (first year tested)-2008: Overall +6, Black +19, Hispanic +17.
- Mathematics, 13-year-olds, 1978-2008: Overall +17, Black +32, Hispanic +17.
- Mathematics, 9-year-olds, 1978-2008: Overall +24, Black +32, Hispanic +30.

Notice a pattern? If one were to apply Stossel’s grossly oversimplified analysis of education cost to scores — and I’m not saying you should — but if you did, you would have to say our public schools are producing a return on our investment. Then again, how he got those cost figures is another topic for another day.