Hispanics, the NLSY 97

[Update: I found NLSY 79 and NLSY 97 AFQT scores broken down by generation in Hansen et al. (2010). The scores from the two studies were not directly comparable, so Hansen et al. transformed them in the manner discussed by Altonji et al. (2008). Hansen et al. did not present standard deviations so I used those for the full survey sample presented in Altonji et al. (2008). Also, Hansen et al. limited the analysis to males. This might have affected the score distribution but the 3rd generation ‘97 scores reported were only slightly different from those which I found using both sexes, so any effect, if existent, is likely small. The category “Children of immigrants” includes all individuals who had at least one foreign born parent. Based on the demographic distributions in the ‘60s and ‘80s, we can deduce that one and two thirds, respectively, of the Hispanic children of immigrants were of the 1st and the 2nd generation. Children of natives, of course, are of the 3rd + generation.

For the IQ estimates I used the 97’ native White score as a norm. It can be seen that 1st and 2nd generation Hispanics narrowed the H/W gap by 2 points, while 3rd and 4th + Hispanics narrowed it by 4 points.

A commentator suggested that there could have been a cohort effect such to render the ’79 and ‘97 Hispanic samples incomparable – e.g., due to changes in immigration patterns, etc. If such an effect occurred, it was probably small, at least for the ‘children of natives,’ as the ‘97 ‘children of the natives’ would represent — though were not — the offspring of 2nd + generation Hispanics in the ’79 cohort. The latter comprised, approximately, 90% of the ’79 Hispanic sample. So there is not much room for an immigrant effect, again, at least for the children of natives. To investigate this issue further, we can look at the CNSLY which was based on the actual children of the NLSY ‘79 cohort. The scores, to note, are not readily comparable because completely different assessments were used. The data were taken from Fryer (2010) and Winship (2003). Averaging the math, reading, and PPVT scores, the magnitude the W/H gap is 8 points, identical to that between native Hispanics and Whites in the 97’ survey. As noted, though, these scores are not directly comparable. Specifically the assessments used, particularly the PIAT tests, had significantly lower g-loadings than the AFQT and so, as with academic achievement scores, do not capture the full magnitude of the difference. Based on the discussion in Murray (2006) concerning the relation between PIAT and IQ scores with respect to the Black-White gap, I estimated that the H/W score differences on these tests is characteristic of a full scale H/W IQ gap of 10 points; this is the same magnitude of difference found based on the more g loaded PPVT in this same sample. Based on this estimate, the differential between 3rd generation Whites and Hispanics narrowed 2 points.

Taken together, we can conclude that there was at least a 2 point narrowing between 3rd + generation Whites and Hispanic between 1979 and 1997. And that the magnitude of this narrowing could have been as high as 4 points.

Now, there is one factor that has yet to be taken into account. As has been noted, there seems to be an intermarriage effect, such that the offsprings of one Hispanic and one non-Hispanic White perform notably superior to the offsprings of two Hispanics from the same generational cohort. This effect, by whatever mechanism it works (e.g., cultural integration, genetic assimilation, etc.) could account for a portion of the found narrowing. This topic will have to be further explored latter.]

Following up with my last post, I looked at the performance of third generation Hispanics in the NLSY 97. Below, ASVAB is the AFQT and PIAT is the Peabody Individual Achievement Test, which was given to different waves in 1999, 2000, 2001, and 2002. I aggregated the PIAT scores across years. The race/ethnic and generation scores are normed on those of Whites (in italics). For both tests, the White-Hispanic gap dropped to 0.5 SD by the third generation. It appears that Ron Unz’s mysterious “prominent rightwing blogger” was not in error, at least if discussion is restricted to the relative performance of 3rd generation Hispanics. (I defined “3rd generation” as noted below.) Search variables and sources are noted. (Lest you ask, the large reduction in numbers going from the full sample (“ALL”) to the 3rd generation sample was largely due to missing information.)

(Interestingly, the NLSY contains detailed information on parental ethnic heritage, in the classic sense of “Italian” and “Dutch,” not merely the euphemistic sense of “race.” Interested investigators can refer themselves to page 92 and 112 of “Screener, Household Roster, and Non-Resident Roster Questionnaire Parent Questionnaire” for the relevant variables.)

This entry was posted in Uncategorized. Bookmark the permalink.

14 Responses to Hispanics, the NLSY 97

  1. Can you post the relative performance of 3rd generation Hispanics.

  2. Pingback: Unz on Race/IQ: The Boston Globe Takes Notice | Ron Unz – Writings and Perspectives

  3. Pingback: Unz on Race/IQ – The Boston Globe Takes Notice | The American Conservative

  4. Peter Frost says:

    This is an apples and oranges comparison. In New York City, third-generation hispanics are overwhelmingly Puerto Rican with some Cuban exiles. First-generation hispanics are much more Dominican and Mexican.

    So the data could reflect the effects of assimilation. Or it could be an ethnic difference. Impossible to tell.

  5. Chuck11 says:


    I looked specifically at Mexicans in the Add Health data and found a similar generational effect.

    I could redo this analysis, limiting the search to Mexicans, but I’m pretty sure that I will find the similar results.

  6. Peter Frost says:

    Even if you limited your analysis to Mexicans, there would still be a shift in ethnicity and socio-economic status. In the past, Mexican immigration to the US came mainly from the mestizo states of northern Mexico. There was also a higher percentage of middle-class Mexicans (especially in the wake of the Mexican Revolution of the early 20th century). Today, Mexican immigration comes much more from poorer social strata and from the southern, more Amerindian states.

  7. Peter Frost says:


    Yes, a longitudinal study would be the best way to approach this question. Otherwise, any control for national origin would be illusory.

    This is a general problem that applies to all immigrant groups. From the mid-1920s to about 1970, immigration to the U.S. was highly restrictive. Non-European immigrants were subject to more controls, both officially and unofficially, and were more likely to have elite backgrounds. A subsidiary problem is that people who considered emigration to the U.S. were also more likely to be at least partially Westernized and highly educated.

  8. Pingback: Unz on Race/IQ: The Entire Series and Debate | Ron Unz – Writings and Perspectives

  9. Pingback: Race/IQ: The Entire Series and Debate | The American Conservative

  10. 猛虎 says:

    Can someone here tell me how to use the NLSY ? I have an account but I really don’t understand how that works. I’m lost.

  11. Steve Sailer says:

    There was a huge change in the format of the AFQT between the NLSY79 and the NLSY97. When the NLSY79 were given the AFQT in 1980, it was a 105 page paper test. Black males did unprecedentedly badly on it — the overall white-black gap was 18.6 points, I believe, while the black male gap must have been out around 20 points.

    In the mid 1990s, a researcher demonstrated that a lot of low-scorers had given up before completing all 105 pages. Black males were particularly likely to just “bubble in” after they’d gotten discouraged, bored, tired, or whatever. This problem with giving up didn’t hurt high scorers much, just kids who realized partway through that this was hopeless. So, a lot of people with 80 IQs wound up scoring 70, or whatever.

    I believe the AFQT given with the NLSY97 was on computers and was responsive — miss some questions, the computer gives you an easier question next and vice versa. I believe the white-black gap fell to 14.7 points, a 3.9 point improvement. Charles Murray’s best guess is that the causes of that 3.9 point improvement were roughly evenly divided between real gains among blacks and the elimination of problems that particularly plagued some groups more than others.

    What impact all these changes had on the white-Hispanic gap is another question.

    • Chuck says:

      Ya, we discussed this issue before. See here.

      For some reason, the CAT versions of tests show smaller B/W and H/W gaps than the paper version. I assume that this has something to do with memory factor loadings. From my previous reply to you:

      Here’s an excerpt from a NAEP technical paper:

      Perhaps the most comprehensive study of the comparability of delivery modes for population groups is that of Gallagher, Bridgeman, and Cahalan (2000), who addressed the issue with large samples of examinees taking a variety of admissions and licensure tests. The tests were the Graduate Record Examinations (GRE®) General Test, Graduate Management Admission Test (GMAT®), SAT I: Reasoning Test, Praxis: Professional Assessment for Beginning Teachers, and Test of English as a Foreign Language (TOEFL®). These investigators discovered that delivery mode consistently changed the size of the differences between focal- and reference-group performance for some groups on both verbal and mathematical tests, but only by small amounts. Of particular interest to the current study is that for Black students and Hispanic students the difference in mathematical performance relative to White students was smaller on computer-based tests than on paper tests. From one mode to the other, the difference in performance between groups changed by up to .24 standard deviation units, depending upon the test. Also, the difference on mathematical tests between White female students and White male students was smaller on the paper versions than on the online editions. This difference changed as a function of delivery mode by up to .12 standard deviations, again depending upon the particular test.

      Personally, I found the CAT version of the GRE more difficult than the paper version, but then I took the former several years after the latter and, well …

      Regardless, what you see is a sort of Darwinian selection for formats and test items. Imagine n number of formats and m number of test items. Those that show reduced ds, but minimally reduced validity, will tend to be selected for. Of course, Spearman’s fact limits the possible effect that this tinkering can have. But, with it, you can shave off a couple of points. Were you to randomly select formats and tests items, say for the GRE or the AFQT, you would see a somewhat larger difference than currently are found. As for the NLSY surveys, if you wanted, you could compare each subtest score to see on which there was change. You can email Meng Hu for these. See also here.

Leave a Reply

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