In the abstract, you type "dimorpshism" and in your table 1, you type "Shangai-China". These should be corrected.

Piffer Wrote:A note of caution regarding this finding is necessary, as phenotypic variation is correlated to genetic variation but environmental noise due to population stratification (SES, ethnicity, etc.) can easily attenuate the genetic signal.

Concerning that problem, in general, behavior geneticists partial out the "race" variable. To this purpose they can use some sort of fixed effect approach. See here for how to do that in Stata. Very straightforward.

http://www.stata.com/support/faqs/statis...d-effects/
As mentioned in the above link, you can use dummy variables. In SPSS, here's the procedure:

https://www.youtube.com/watch?v=R0qc4rzr9ik
In your case, for instance, you can create dummy for european, arab, asian, african countries. You can end up with perhaps 4 or more dummies, (e.g., race1, race2, race3, race4, etc., all coded 1 if the country corresponds to the race category label and 0 for otherwise), and you should include all of them in the regression equation. (If you do that, your program will probably drop one variable due to collinearity or redundancy, and the variable dropped is treated as the reference category.)

See below (I assume here that "race" variable has been created already and is a numeric variable for which the countries have been categorized into the race category to which they should belong, e.g. white countries=1, arab countries=2, asian countries=3 etc...; I assume here you have 6 race categories, but there may be more, or less, probably... so that if you have 4 races, you should remove all the numbers '5' and '6' in the series shown below) :

RECODE race (1=1) INTO race_cat0.

RECODE race (2=1) (3,4,5,6,1=0) INTO race_cat1.

RECODE race (3=1) (2,4,5,6,1=0) INTO race_cat2.

RECODE race (4=1) (2,3,5,6,1=0) INTO race_cat3.

RECODE race (5=1) (2,3,4,6,1=0) INTO race_cat4.

RECODE race (6=1) (2,3,4,5,1=0) INTO race_cat5.

REGRESSION

/DESCRIPTIVES MEAN STDDEV CORR SIG N

/MISSING LISTWISE

/STATISTICS COEFF OUTS R ANOVA CHANGE ZPP

/CRITERIA=PIN(.05) POUT(.10)

/NOORIGIN

/DEPENDENT PISA_Score_Total

/METHOD=ENTER Difference GDP

/METHOD=ENTER race_cat1 to race_cat5

/PARTIALPLOT ALL

/SCATTERPLOT=(*ZRESID ,*ZPRED)

/RESIDUALS DURBIN HISTOGRAM(ZRESID) NORMPROB(ZRESID)

/CASEWISE PLOT(ZRESID) OUTLIERS(3)

/SAVE MAHAL COOK LEVER PRED RESID.

REGRESSION

/DESCRIPTIVES MEAN STDDEV CORR SIG N

/MISSING LISTWISE

/STATISTICS COEFF OUTS R ANOVA CHANGE ZPP

/CRITERIA=PIN(.05) POUT(.10)

/NOORIGIN

/DEPENDENT PISA_Score_Total

/METHOD=ENTER Difference GDP

/METHOD=ENTER race2 race3 race4 race5 race6 race1

/PARTIALPLOT ALL

/SCATTERPLOT=(*ZRESID ,*ZPRED)

/RESIDUALS DURBIN HISTOGRAM(ZRESID) NORMPROB(ZRESID)

/CASEWISE PLOT(ZRESID) OUTLIERS(3)

/SAVE MAHAL COOK LEVER PRED RESID.

(edit: I forget to precise, those are two different ways to perform this kind of fixed-effect regressions but they produce identical results, at least for the coefficients of variables other than the dummies which have coefficients differing only in function to which category is the reference variable.) Normally in logistic regression, what happens is that race2 to race6 have coefficients expressed in relation to race1 (reference category because entered last) but the point of interest here should be how they affect the other indep variables. It seems as far as I see that it works differently in linear regression, I don't know why but if the other independent var are unchanged, it's not necessarily a problem.

Can you do this analysis ? If you think it's useful of course (personally I think it's useful otherwise i woudn't bother).

Besides, do you have more information about PISA (CPS) ? You say it's like fluid intelligence, but it still is an achievement test. Does it really looks like the fluid subtests in most IQ tests ? (specifically, I'm talking about the items/questionnaires.)