Here, I will use the motivating example of the poorly performing American Empire inner-city school to illustrate that the successful performance of multiple subpopulations (that are not the current one in consideration) in the process, and the existence of a metric to show that the process is being performed, together are roughly (statistically powerful, not infallible) sufficient to show that the non-performance in wider application is due to other factors.
We recall not only the exemplars of individuals of all races/backgrounds, but also specifically examples e.g. from the Escalante example with calculus, or the KIPP/Success Academy type approaches, where a subpopulation is able to deliver as good results when tutored according to a certain kind of approach. There are several confounders:
- Self selection, such as the parents being knowledgeable/motivated enough to sign up for school lotteries
- Survivor bias: the kids actually have to get far enough to be able to benefit from the instruction, which means they are the smarter/more determined subpopulation, or at least the well-behaved (reference a popular attribution for part of charter school success, that they kick out all the punks)
- Just dumb luck, or the existence of exceptional educators that make the demonstrated achievement not generalizable
At some level, the self-selection confounder never can be discounted. Even rigorous metrics don’t cover 10% of the overall stimulus to children. Further, we also recall that some poor kids make it out, some don’t, even from the same family. We also recall people who make a change in mid-life, who never displayed above-average potential prior to that. These data points inherently limit the prediction accuracy of any approach. However, there is one critical point about this: this prediction accuracy, also establishes a “background” or “baseline” level of performance, that even in a poorly performing population, you should expect to see some number of outperformers. That you see none, or occurrences well below the average, indicates that there is active suppression or other measures that are hindering the outperformers.
Survivor bias, as a distinct factor from self selection (free will), therefore only can refer to inherent attributes of the individual such as g (general intelligence), or to the specific home or other environment in which the individual was raised. For example, if only a cut line above a certain g (the purported general intelligence constant) strongly predicts the number of outperformers. This then suggests that there are metrics that could determine the performance of an assigned population to a process.
Dumb luck and exceptionalism also always is an explanation. This only can be reduced in probability with multiple trials: that is, multiple applications of the process and the successful results (though not in the specific case at hand).
Now let us consider the example of the poorly-educating inner city school, and the suggestion that people of means should send their children there. There are several other factors such as general crime level and delinquency outside of school, but to focus the point I only am examining the academically controlled variables. Further to illustrate the real life example and the theoretical edge cases, I directly will address the issues of race/class that tend to come up in the American Empire context.
If there is a history of the white and Asian/immigrant parents sending their children to this school and achieving the similar results as for the best performing schools, of course this is the best indicator for this quality of education at this school. The question then becomes why do the black kids not perform as well (I am ignoring the brown/immigrant because this brings in the obstacle of ESL that lowers academic performance on standard benchmarks). Non-obvious explanations (e.g. excluding gross facility issues and hunger) would include tracking to less rigorous classes, less parental positive involvement in education, and less parental environmental support e.g. how many words do they speak to their children.
The tracking to less rigorous classes easily can be solved by forcing high standards (a metric that can be measured as per the introduction).
The parental involvement factors partially can be measured, for example parents who have to work outside the home most of the time. These are under the control of the parents, so they are not issues with the school.
Hence, we consider that in this case, if there are no obvious issues with the school itself, no disruptions in the classroom, and that there are high standards, then under the further assumption that the races have equal scholastic ability, then we would attribute the outperformance to parental and associated social factors, so there should be no issues with people of means (who can do the necessary parenting) sending their children to this school.
Now let us consider the more challenging example: say that there is no high-performing subpopulation at this school. The conditions for the similar success would exist (because this is what is going on in the school with the successful subpopulation) if:
- The school is in a decent facilities state, textbooks, etc.
- The classes are well-behaved and make efficient use of time
- The high standards easily can be verified by unannounced spot observation and by review of curricula
- Optional but mentioned because it has significance for verification: that the school hours are longer
This is the definition of the education process.
Bringing back a critical point: in a “normal” setting there always are outperformers (and underperformers). So, there should be some number of black outperformers in the school who clearly are doing about as well as the populations from the best educated schools (incidentally that this points out the issue about arbitrariness of metric in defining subpopulation on a not obviously performance-relevant concept of race, when the number of high achieving students might be the same in each of the two schools).
As previously noted, the factors that play into the level of survivor bias are not necessarily measured. If they are, those factors could be assessed relative to other successful educational environments to determine the prospects for the children of the parents of means.
Because the above listed education process (especially with the longer hours) repeatedly has been replicated with improved results, prior dumb luck and exceptional teachers are highly unlikely to cause the results to underperform in this case.
If these preceding attributes of the education process have been verified, there are the outperformers that you would expect, and your children are in a favorable position with respect to the survivor bias factors, then with confidence, you could send your children to this school that has no high-performing subpopulation.