[N.B.: I originally published this post on 2012.09.22. I revised it for clarity and to replace broken links. Although there are new developments on this topic, this post will not address them; however, I may address them in future posts.]
Consider the following snippets from a recent article in the New Scientist:
[New Scientist’s] analyses of US Census Bureau data reveal large and stubbornly persistent disparities in wealth and educational achievement…the nation may yet avoid a cycle of decline – if it improves educational opportunities for young Hispanics.
If US Hispanics enjoyed similar success to their counterparts of European extraction, there would be little to fear from this shift. But they don’t.
If more young Hispanics are to make it in the US, education will be key. Again, the numbers don’t look good…since the mid-1990s, there has been no narrowing of the gap between whites and Hispanics.
Demographic trends mean that the academies’ recommended benchmarks are unlikely to be reached without an unprecedented rise in Hispanic students getting degrees in science, mathematics and engineering…but that depends on elementary and high schools preparing Hispanic students for college in the first place.
You’ll find the article here. Immediately obvious are concepts like “educational attainment” and the “achievement gap.” Reading between the lines exposes some more – the need for more STEM grads (STEM = Science, Technology, Engineering, and Mathematics) and more effective schooling. The article (implicitly) suggests that certain remedies – namely, closing the achievement gap, increased education spending, and (possibly) expanding affirmative action such that colleges admit more non-White Hispanics – will narrow the wealth and achievement gaps; indeed, that’s what some cats are saying. Are they right?
As I exhaustively explained a few weeks back, policies designed with the achievement gap in mind not only fail (i.e. aren’t as effective as intended), but also encourage adverse behaviors like administrative cheating that further compound the education problem.
Make no mistake – I believe poverty and sub-par education are problems in need of solutions; indeed, I favor programs that strengthen and diversify curricula in schools, improve teacher quality, and combat poverty in disadvantaged areas. However, once you “temper” the environment such that Hispanics truly get equal opportunities (i.e. same educational opportunities, programs, teacher quality, etc. as Whites), then biological and genetic factors begin explaining greater proportions of variance in performance between the groups. As such, I believe we can’t ignore these factors.
(Full disclosure: I am Hispanic (part Dominican, part Puerto Rican) and a STEM grad.)
Now, when I hear politicos and other public figures use the word “Hispanic,” they usually aren’t referring to all Hispanics – they usually mean Mexicans. While Mexicans make up 63% of America’s Hispanic population (according to this Census report), methinks it’s disingenuous to ignore the rest (like me!); to show this (as well as analyze the Hispanic education problem), I conducted some research and ran some numbers.
Using the wealth of data one finds in the General Social Survey (GSS), I compared the educational attainment (i.e. proportion of cats with at least a bachelor’s degree) of the 20 Hispanic subgroups used in the 2010 Census (excluding all instances of “Other”); I used GSS data from the 1% sample of the 2000 Census and the American Community Surveys (ACS) dated 2000-2003, 2005-2007, and 2006-2008. Then, I averaged the proportions. Before I discuss the results, I’ll define some variables:
N = sample size
μ = population mean (or average)
σ = population standard deviation (a measure of how spread apart data points are with respect to the mean)
r2 = coefficient of determination (the percentage of the variation of the dependent variable explained by the relationship between it and one or more independent variables; in layman terms, this tells us how good of a predictor a given model is.)
r = correlation coefficient (a measure of the strength of the relationship between a dependent variable and one or more independent variables)
p = probability value (or p-value; in linear regression analysis, this is the probability that we erroneously reject the null hypothesis that no significant linear relationship between the dependent variable and one or more independent variables exists. Put another way, the smaller this value is, the stronger the evidence that a linear relationship exists between the variables).
Now that we’ve established that, here are my results (click to enlarge):
[Fig. 1] Data table showing the 20 Hispanic nations, average national IQs (see below), and the percentage of each subgroup in the U.S. with at least a bachelor’s degree.
You’ll notice some other data too, so allow me to explain. I obtained average national IQ values from Lynn and Vanhanen’s controversial book Intelligence: A Unifying Construct for the Social Sciences (N.B.: The book is listed as out-of-print on Amazon, but is still available for purchase here); I listed these under the heading IQ (for these, μ = 85.3 and σ = 4.7). By the Central Limit Theorem, the true population deviation for a set of averages approaches σ times the square root of N; if σ is robust in this set, then the population σ value is roughly 20.9. Next is the IQ (adj) column; based on research that Hispanics in America score about a deviation higher on cognitive tests than those outside America, I added 4.7 to each IQ score. The IQ/Achievement Data Quality column is the “IQ and Scholastic Achievement Data Quality” metric from L & V’s book, which limits the scope of my analysis in ways I’ll explain later. N-W means non-White and diff means the difference between the percentage of all Hispanics with at least a bachelor’s degree and the percentage of non-White Hispanics with the same (hereafter overall – non-White difference).
I first calculated the average proportion of all Hispanics with at least a bachelor’s degree (listed under % college+); for comparison’s sake the U.S. Hispanic graduate rate (i.e. possessing at least a bachelor’s degree) according to the 2010 Census is 12.3%. As is clear, though Mexico’s average IQ is 6th highest of the 20 Hispanic nations, the Mexican graduate rate ranks last (ranks listed under % college+ rank).
To get a clearer picture of the relationship between average IQ and grad rates, I regressed IQ (adj) and % college+ (click to enlarge):
[Fig. 2] Scatterplot and regression line for the relationship between IQ (adj) and the percentage of all Hispanics with at least a bachelor’s degree.
The regression line suggests that a positive relationship between average IQ and graduate rates exist. The positive and statistically significant value of r=0.45 (p<0.05) supports this claim. Moreover, r2 = 0.2024, meaning that this relationship explains up to 20.24% of the variation in graduate rates.
There’s a problem though – this regression includes those Hispanics who identified themselves as White. What would we see if we examined non-White Hispanics exclusively? For my next analysis, I filtered the GSS variables such that they only included data from non-White Hispanics; the racial categories I used for non-Whites were Black, American Indian/Alaskan Native, and Other.
Re-running the regression yields this chart (click to enlarge):
[Fig. 3] Scatterplot and regression line for the relationship between IQ (adj) and the percentage of non-White Hispanics with at least a bachelor’s degree.
The correlation between adjusted average IQ and graduate rate for non-White Hispanics is still positive (r=0.174) but no longer statistically significant. Further, with r2=0.0301, this relationship explains only up to 3.01% of the variance. At first this finding surprised me since, in the intelligence literature, researchers consistently find that Blacks/Hispanics have lower average IQs than Whites and East Asians; one would expect a significant relationship involving IQ in this case. Does this mean that IQ actually has no bearing on the Black/Hispanic graduate rate? To find out, I conducted one more analysis. I regressed the adjusted average IQ of each Hispanic subgroup with the overall – non-White difference; here is the resulting chart (click to enlarge):
[Fig. 4] Scatterplot and regression line for the relationship between IQ (adj) and the difference between the overall Hispanic graduate rate and the non-White Hispanic graduate rate.
Clearly this is a much different result compared to the preceding chart. The positive and extremely significant value of r=0.822 (p<0.001) provides strong evidence of a linear relationship between adjusted IQ and the overall – non-White difference; with r2=0.6752, this relationship explains up to 67.52% of the variance. But what exactly is this model telling us?
The answer – as IQ increases, the overall – non-White difference also increases!
Unfortunately, it gets worse. I obtained estimates of average college grad IQ by major from Steve Sailer’s blog and, while they range from 103 to 133, the STEM major averages are generally on the higher end of this interval (the lowest I saw for any science was 110). Plugging x=110 into the regression equation in Fig. 4 gives an expected value of 14.9% for the overall – non-White difference; for x=130 (i.e. the estimated average IQ of mathematical sciences grads and just above the average for most engineering disciplines), the expected value jumps to roughly 26%!
There are significant caveats in my analysis, however. First, recall from earlier the “IQ/Achievement Data Quality” column from L & V’s book. Here, you’ll see two numbers – the first is the IQ data quality (based on IQ test data) and the second is scholastic achievement data quality (based on other standardized test data). Higher numbers indicate better data quality (highest quality rating is 25 for IQ and 16 for scholastic achievement); dashes indicate insufficient/missing data. Nicaragua’s IQ estimate is an extrapolation based on biologically similar neighbors; insufficient data on Nicaragua exists. The average IQ data quality of this set is 5.94/25 (excluding countries with missing data); the average achievement data quality is 5.5/16 (again, excluding countries with missing data). The generally low data quality for most Hispanic nations means, in my view, that we need more research before drawing definitive conclusions.
Second, I used U.S. Census and GSS data (i.e. survey data) in my analysis. It is possible that Hispanics identifying as White may not actually be White or even part-White, genetically speaking. That is, the analysis doesn’t account for possible confounding effects; hence, why I stated the r2 values found explain up to x% of the variance.
While the data are alarming, I won’t go as far as stating “there’s no hope for Hispanics” (which obviously isn’t true). However, if we keep kidding ourselves into thinking narrowing “achievement gaps” and expanding affirmative action policies (yes, I’m against expanding affirmative action due to the implied perception that we can’t make it without a “push”) will solve the problems prevalent in education and in the field, then we run the risk of having reality slap us in the face (oh, and no lasting academic improvement).
Remember, folks – when you hear the term Hispanic, DON’T assume that “Hispanic” = “Mexican.” 37% of Hispanics are not Mexican – hardly an insignificant figure. Broaden the search and one just might find more Hispanic “STEMs” – one of them is the author of this very blog!