Ich bin auf diese Arbeit in Psycho (MDPI Journal) gestoßen, in der die Beziehung zwischen europäischer Abstammung und kognitiven Fähigkeiten (G -Faktor) untersucht wurde. Link zum Papier.
https://www.mdpi.com/2624-8611/1/1/34
Hier sind einige der Regressionsdiagramme:
Vollständige Probe (n = 10.370): r ≈ 0,36
Hispanic American Sub Probe (n = 2,021): R ≈ 0,23
Afroamerikaner gegen den europäischen amerikanischen Vergleich zeigt einen ähnlichen Trend
Meine Fragen:
-
Wie stark ist eine Korrelation von R ≈ 0,36?
-
Wie viel Varianz erklärt das tatsächlich (R²)?
-
Wie trennen die Forscher beim Betrachten von Streuplots wie diesen die statistische Assoziation von der kausalen Erklärung?
Ich versuche nicht, hier einen politischen Punkt zu machen, nur um zu verstehen, wie man Korrelationen in solchen Datensätzen interpretiert.
Von Trick_Ad_2852
8 Kommentare
R2=0.36 is a poor correlation.
A correlation this low is generally meaningless. In practical terms, you cannot predict with any accuracy the percentage of European ancestry by a person’s g factor. This is basically no correlation.
This is r/dataisbeautiful. Your data is disgusting.
A single scatter plot wouldn’t account for other factors, that’s why people write whole papers.
You’d then plot for socioeconomic factors like income, education etc. and see if they are more significant.
Don’t get caught up on this. It leads down a bad path. Humans are complex, and outcomes aren’t defined by race.
Low r, but massive N likely makes it significant.
R2 is the amount of variance explained by the model, in this case it looks like a simple likely binary variable of ancestry against general intelligence with some control variables. So if that is true then this model explains 36% of the variance in general intelligence.
However I’d strongly caution any casual interpretation here, it’s confusing however in this context ‚explained‘ just means how much of the variance does the correlation explain. Not how much of the variance is caused by ancestry.
How good is a R2 of 0.36, depends in what context. I’d assume a model with many more economic, and personal variables could explain much more variance in general intelligence. While a model of financial markets with an r2 of that out of sample I’d be very rich indeed. In economic literature where models tend to have lower r2 values, since getting good variable and data is hard, a 36% is reasonable in some cases. However they usually have far more variables and are trying to do something far more complicated than this. Which I believe is a classic case of don’t let the endogeneity get in the way of a good story.
I believe I read similar research that was better controlled and only found a very negligible difference in IQ between races when using real world controls and natural experiments. Ie a black kid growing up in a wealthy home from birth etc.
You haven’t accounted for
– education level
– income
– health
– diet
and others.
The data is worthless without those.
When you have a test of intelligence constructed by white europeans, it biases towards white europeans