Hispanic “Flynn Effect” in the NAEP

Below are the NAEP Main assessment, Grade 12, math and reading standardized differences between Whites and Hispanics broken down variously. Ron Unz has argued that we should restrict our focus to US born Hispanics, on the grounds that the performance of first generation Hispanics is probably artificially depressed. This is reasonable. Unfortunately, the NAEP online explorer only provides two nativity variables which jointly cover only 4 years. And these variables are not found for the math test but only the reading. There were, though, variables constructed from a student answered family language question, which asked how frequently a language other than English was used at home. The answers ranged from “always” to “never.” According to Pew Hispanic Research, unsurprisingly, less than 5% of Hispanic 1st gens come from homes in which only English is spoken. As such, the “never” group can reasonably be used as a proxy for Hispanics excluding those of the first generation. It’s noted, though, that a substantial percent of Hispanic 2nd + gens do hail from homes in which Spanish is spoken at least some of the time; as such, the “never” group is more exclusive. This group probably contains Hispanics whose families are more assimilated, than average, into the dominant culture and gene pool.

As can be seen below, with regards to math performance, the 2005 and 2009 differences between non-Hispanic Whites (“Whites,” from now on) and English only Hispanics were little different from those between Whites and all Hispanics. Restricting our consideration to English only Mexicans alters little. With regards to reading performance, there is little difference between US born Hispanics, English only Hispanics, and English only Mexicans; this confirms our suspicion that the White-English only gap closely approximates the White- Hispanic 2nd + gen one. These groups, though, perform approximately 0.15 SD better than all Hispanics in reading. Based on both the math and reading scores, 0.5-0.6 SD is a reasonable estimate of the contemporaries White-Hispanic 2nd+ gen national achievement gap. This can be compared to a difference of about 0.65 SD between Whites and all Hispanic. For international tests, which are more g-loaded, and for which a larger gap is expected granting Spearman’s hypothesis, the respective gaps were 0.8 SD and 0.7 SD. Between Whites and Hispanic 3rd + gens, the gap, averaged across international tests, was 0.6 SD. If the NAEP gap follows the same generational pattern as does the international test gap, we might infer that the White-Hispanics 3rd + gen gap is slightly less than 0.5 SD. (This, of course, is consistent with a general intelligence gap of 0.7, as the correlation between these tests and g is 0.7.)

Now, Ron has claimed that there has been a large narrowing of the White-Hispanic 2nd + gen IQ gap. As evidence of this he has pointed to Wordsum, NLSY97, and, more recently, SAT results. Ron’s clever SAT argument has been shown to be problematic by Unsilenced Science. The NLSY studies — comparing the 1979 to the 1997 results — indicate some narrowing for all Hispanics (from 14 points to 11.5 points); moreover, the 97 study showed a White-Hispanic 3rd + gen gap of 7.5 points. But no estimate of the NLSY 79 White-Hispanic gap, decomposed by generation, is available. As such, one can only infer a narrowing (between Whites and Hispanic 2nd + gens). As for Wordsum, the correlation between this measure and general intelligence is less than that between general intelligence and the NAEP tests (0.63 versus 0.7). So we can look to the NAEP, age 17, long term trends (LTT) to see to what extent we can verify Ron’s claim. (For those not familiar with the NAEP setup, the LTT and Main are comparable but not identical.)

The overall grade 12 gap as calculated by Eric Hanushek:


(Hanushek, 2010. How well do we understand achievement gaps?)

The grade 12 gaps by family language:

It can be seen that there was a secular narrowing in the overall White-Hispanics NAEP math and reading gaps. If one compares the earliest points to the latest this represents a 30-40% narrowing. But this large decrease is partially illusionary. A portion of it is attributable to the 2004 NAEP format change. In 2004, a linking study was conducted in which both formats were presented to random samples. It can be seen that relative to Whites, Hispanics perform 0.2 SD better on the newer format. Taking the effect of the format change into account, the White-overall Hispanics gap narrowed only 15-30%. Similar results can be seen in the case of the White-English only Hispanic gaps, which, we said above, approximates the White-Hispanic 2nd + gen ones.

Reply to Ron.

Ron Unz commented on this post, thusly:

There’s actually a major problem in relying upon the English-Only Hispanic segment. First, it’s a very small slice of all Hispanics past childhood, only about 15-20% I think. Similarly, the 3rd+ Hispanic generation is very small. The problem is that Hispanics traditionally tended to have a pretty high intermarriage rate, so I’d think a good majority of Americans having 3rd or 4th generation Hispanic ancestry are mostly non-Hispanic, and probably don’t usually identify with the “Hispanic” category on the survey. The minority who are overwhelmingly Hispanic and do so identity generally tend to be concentrated in impoverished, often rural areas, and their low performance numbers throw off the results. That’s the reason for the “shocking” fact that rapid socio-economic improvement trends for Hispanics tend to sharply reverse after about the 3-4th generation.

I’m not going to elaborate on the figures as they should be self explanatory. My reply:

(1) Do third generation H comprise only a “very small” segment of the relevant US population? No.

TIMSS 2007 distribution, grade 8

(Over 1/3rd of H adolescent students are 3rd gen Hs. Since we are comparing student scores, this is the relevant population.)

(2) Do third generation Hispanics mostly not identify as Hispanics? And does Hispanic identity attrition throw off the 3rd gen results?

Depends on how Hispanic is defined and No.

Add Health Wave I, VIQ:

(Third generation students who identify as H have slightly lower VIQs than those who are inclusively identified as H by parental identity (based on having one or more H parents) but slightly higher VIQs than those who are genealogically defined, allowing for one-half Hs. 92% of students with 2 H parents identify as H; for students with 1 H parents, the % is 61.)

(3) Are English only Hispanics concentrated in rural areas? No.

Frequency distributions in the HSTS 2009:

(Rather, they are concentrated in the city.)

(4) Can the rural/urban divide explain the White-Hispanic gap? No. (So much for busing Hispanics into the ghetto to close the gap.)

HSTS 2009, SAT/ACT

(No & the rural gap is smaller than the city gap.)

(5) Is the gap driven by low SES 3rd gen/language only Hispanics? No.

TIMSS 2007, Math

(No, at least not in the TIMSS 2007.)

Here were the composite SAT/ACT/PSAT Hispanic scores in NELS 88′, which can be analyzed online at DAS. These were from 1992-1994. SES on Left. Standard deviations in pic below.

(Third gen Hispanics have much higher SES but littler higher SAT/ACT/PSAT scores relative to their first and second generation peers.)

About these ads

5 thoughts on “Hispanic “Flynn Effect” in the NAEP

  1. There’s actually a major problem in relying upon the English-Only Hispanic segment. First, it’s a very small slice of all Hispanics past childhood, only about 15-20% I think. Similarly, the 3rd+ Hispanic generation is very small. The problem is that Hispanics traditionally tended to have a pretty high intermarriage rate, so I’d think a good majority of Americans having 3rd or 4th generation Hispanic ancestry are mostly non-Hispanic, and probably don’t usually identify with the “Hispanic” category on the survey. The minority who are overwhelmingly Hispanic and do so identity generally tend to be concentrated in impoverished, often rural areas, and their low performance numbers throw off the results. That’s the reason for the “shocking” fact that rapid socio-economic improvement trends for Hispanics tend to sharply reverse after about the 3-4th generation.

    Therefore, I think we need to mostly discount the English-Only NAEP results, and (unfortunately) are forced to rely on the overall NAEP trends. It seems to me that if you compare your NAEP trends with the Wordsum-IQ trends for all Mex-Ams and also just American-born Mex-Ams, you’ll see that they fall somewhere between those latter two categories. Since Mex-Ams are only about 60% of all Hispanics, you wouldn’t anyway expect a great fit, but if it’s a rough fit, wouldn’t that tend to support the plausibility of the Wordsum-IQ trends I noted?

    In fact, in that post you unfortunately took down, I think you’d found only slight improvement trends in overall Mex-Am Wordsum-IQ, so wouldn’t these NAEP Hispanic trends actually be *more* encouraging?

    • “Since Mex-Ams are only about 60% of all Hispanics, you wouldn’t anyway expect a great fit”

      And yet, you find a good one, because, as noted, the Mexican and Hispanic scores closely track each other.

      “but if it’s a rough fit, wouldn’t that tend to support the plausibility of the Wordsum-IQ trends I noted?”

      What do you mean by “support the plausibility of”? The wordsum trends are the wordsum trends. If you mean “support the generalizability of” then, obviously, “no,” As noted, if we restrict our consideration to the overall NAEP trends, then we see only a 15-30% narrowing, not whatever outlandish amount you claimed.

      “The minority who are overwhelmingly Hispanic and do so identity generally tend to be concentrated in impoverished, often rural areas, and their low performance numbers throw off the results.”

      As usual, you don’t have a shred of evidence to support this claim.

  2. Pingback: Umm…no. « Occidentalist

comments do not require an email

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s