library(pacman); p_load(VGAM)In a paper that is popular beyond all sense, Torrey & Yolken (2010) argued that the Nazis’ genocide of schizophrenics and the subsequent lack of long-term effect on rates of schizophrenia in Germany was evidence against “erroneous genetic theories… of schizophrenia”, and others have argued along the same lines that the heritability of schizophrenia may well be zero, given the apparent lack of effect of the Nazis’ efforts. The most major problems with this paper can be summarized in one breath: it’s not even wrong. Torrey & Yolken made no effort to understand “eugenics” or genetics. As a result, it was probably hard for them to disqualify the views they wished to contest. Gwern detailed this very well when this came up a few years ago, but since his post is now lost due to Reddit banning the subreddit he made it on, I will do my best to recapitulate it from memory, for posterity.
Firstly, the method Torrey & Yolken used ran into a major problem in the form of countervailing trends. The Nazis sought to sterilize schizophrenics, but the combination of several other factors running the other direction makes it hard to assess any of the effects of that initiative. These factors are genetic, sociological, and psychometric, and they can all affect problems with detecting a change in the amount of schizophrenia in the German population that may have resulted from Nazi eugenics policies.
For what it’s worth, some of these are discussed by Torrey & Yolken. But, the question is not whether schizophrenia rates increased, since they were likely to increase anyway, but whether they increased less than they would have in a counterfactual Germany without their eugenics program. The claim of temporal invariance in schizophrenia is self-evidently wrong because there were constant increases during the postwar period, while a single selection event changes the mean of a trait only for the succeeding generation, providing all subsequent generations with the same mean. Showing postwar increases at multiple points following the war proves one cannot meaningfully compare because of those or other confounds.
Because of these confounds, a simple before and after comparison is not the correct framing for this question. The correct hypothesis is, instead, one that represents what eugenicists may have thought. In this case, that is simply the prediction of the Breeder’s Equation, \(\Delta Z = h^2S\), where \(\Delta Z\) is the change in the trait, \(h^2\) is the heritability, and \(S\) is the selection differential. The objective would be to assess how much of a decrease in the schizophrenia rate should be expected and then to calculate the power to conduct such a test if it is even possible. If the decrease is too small, there will not be enough data to overcome the random error, let alone the effects of the confounds mentioned above or other potential explanations.
Schizophrenia is a highly polygenic trait, with the genetic variance representing thousands of small effects that are - critically - within a liability-threshold framework. Since the effects are well mixed in the population, killing a large fraction of schizophrenics will likely not have a large effect in a single generation, since only those with the greatest number of deleterious alleles will be eliminated, but the level in the general population will be largely unaltered. The claim that almost all of the schizophrenics were killed is nowhere near plausible, since even if all diagnosed schizophrenics were killed, many cases would never be diagnosed or diagnosed as schizophrenic.
The Breeder’s Equation works trivially for normally-distributed traits, as it’s just the narrow-sense heritability multiplied by the selection differential, and the heritability of schizophrenia is very high, around 0.80. However, because of the binary nature of schizophrenia diagnoses, you either have it or you do not, so you must use a modified form of the equation for a threshold. The most plausible provided estimate of the number of schizophrenics killed was 73% (which is still almost certainly too high), but for simplicity, we can set the number to 100% to favor their case and the power of their proposition for showing a low or nil level of genetic influence on schizophrenia. Code follows, derived by Gwern from Walsh & Lynch (2018) and posted by me from when I saved the code, years ago. Code comments are Gwern’s.
ThresholdSelection <- function(FractionCulled, Heritability, pow = 0.80){
ProbitFraction = probitlink(FractionCulled)
#Threshold for failing to manifest schizophrenia:
Selection = dnorm(ProbitFraction) / FractionCulled
print(Selection) #Selection intensity is so low because almost the entire population reproduces, no matter how close to the threshold
#New rate of schizophrenia after one selection where 100% of schizophrenics never reproduce:
SczFraction = pnorm(ProbitFraction + Heritability * Selection); print(SczFraction); print(FractionCulled - SczFraction)
#How much is schizophrenia reduced in percentage terms?
print(((1 - FractionCulled) / (SczFraction/FractionCulled)) * 100)
#How much data do we require to detect before/after differences?
print(power.prop.test(p1 = FractionCulled, p2 = SczFraction, power = pow))
return(SczFraction)}
ThresholdSelection(0.99, 0.80)## [1] 0.02692136
## [1] 0.9905598
## [1] -0.0005598242
## [1] 0.9994348
##
## Two-sample comparison of proportions power calculation
##
## n = 482127.1
## p1 = 0.99
## p2 = 0.9905598
## sig.level = 0.05
## power = 0.8
## alternative = two.sided
##
## NOTE: n is number in *each* group## [1] 0.9905598Under a set of extremely optimistic assumptions about the comprehensiveness of selection without any regard to potential confounding factors, in a regime of simplistic, maximally optimistic effects for Torrey & Yolken’s hypothesis, the possible reduction that can be achieved is about 1%, and we need to sample more than 950,000 Germans to detect that 1% reduction (and this is being generous with the precision!). The data discussed by Torrey & Yolken are nowhere near that. There was a post-war study that covered “424 000 people in 3 counties in Bavaria, including Rosenheim”, but the pre-war studies only covered 46,189. Given the small size of the expected effect and that we need more than 950,000 Germans to detect that change coupled with the lack of sufficient data, we may reasonably conclude that it is practically impossible to detect an effect of the Nazi eugenics program on schizophrenia rates.
There will never be data to answer this question because the unrealistically detailed and long time series required for it were never collected and are now impossible to collect. If there were a natural experiment to exploit, where various regions were randomly spared the effects of the Nazis’ eugenics programs and everyone was recorded in a way that would allow comprehensive pre-post comparisons of the 1930s and ’40s with the 1950s and ’60s. A longitudinal polygenic score analysis with that cohort would not be theoretically good enough as negative evidence, but it would as positive evidence provided a good enough schizophrenia PGS. This could be confounded with the symptomality of schizophorenia though, making it unable to deliver a meaningful null result.
Furthermore, it should have been known even in 1940 that removing cases of schizophrenia from society would have no impact on the incidence of the disease because the vast majority of individuals with schizophrenia do not have a family history of the disease and do not reproduce. Current research suggests that the cause of schizophrenia involves dozens, and perhaps hundreds, of genes and includes common variants such as single nucleotide polymorphisms or less common variants such as copy number variations. Such variants may be carried by large numbers of people, most of whom never develop schizophrenia. It is possible that such genetic variations may cause disease only if they are activated by life experiences such as perinatal hypoxia, nutritional deficiency, infections, or other environmental factors.
This paragraph, from Torrey & Yolken shows very clearly that they did not understand the eugenicist position from the time. The general position among eugenicists was not that a singular purge would yield large differences immediately, except in cases of certain varieties of Mendelian disorder. It does not seem that the Nazis expected otherwise, and only Rudin & Brugger are cited as believers in schizophrenia as a singular recessive condition, which may or may not be an accurate description of their beliefs, but seems extraordinarily unlikely because family pedigrees for schizophrenia do not look thes slightest bit Mendelian. Some people may have pragmatically believed it because of the immediate cost-savings in asylums, but it’s doubtful this was scientifically serious. Most eugenicists talked about the results accruing over many generations because they appreciated the Breeder’s Equation perfectly well. Over the long run, selection still works, even for highly polygenic traits - you select on the variance, not the variants, as the saying geos. Fisher pioneered the infinitesimal model and you can observe it working out with the Breeder’s Equation. Gwern ran this model for 20 generations to show that it would halve the schizophrenia rate.
Fractions = 0.99
for(i in 2:20) { Fractions[i] <- ThresholdSelection(Fractions[(i - 1)], 0.80); }; Fractions## [1] 0.9900000 0.9905598 0.9910663 0.9915263 0.9919458 0.9923296 0.9926820
## [8] 0.9930064 0.9933060 0.9935834 0.9938409 0.9940804 0.9943038 0.9945124
## [15] 0.9947077 0.9948909 0.9950630 0.9952249 0.9953775 0.9955216The study ended with a particularly childish view:
In addition, perhaps, the most appropriate response the profession of psychiatry can have to the Nazi eugenics and psychiatric genocide program is to focus additional resources on examining more complex forms of genetic and gene-environmental interactions in order to understand the true genetic contribution to schizophrenia. This knowledge should then be used to develop methods for disease prevention and treatment that can be used ethically in all populations.
This little paragraph is a good proof that the authors were not caught up on the literature, which included GWAS and GREML studies, and abundant proof of small, additive effects and high heritability. Quoting Gwern: “Isn’t it amazing how everyone jumps to ‘gene-environmental interactions’ and ‘epigenetics’ these days?”
Gwern’s note on this study was a very good way to understand the Breeder’s Equation, its consequences, its extension to liability-threshold traits, and how implausible and likely ahistorical scientific speculation - clearly - possesses the ability to get past peer review, despite being so full of holes that it should never have moved anyone’s meter. This page is saved for posterity because speculation to the effect that schizophrenia has no genetic basis because a naive eugenics program failed is basically useless for reasons of statistical power and theoretical coherence, both of which were utterly lacking.
Torrey, E. F., & Yolken, R. H. (2010). Psychiatric Genocide: Nazi Attempts to Eradicate Schizophrenia. Schizophrenia Bulletin, 36(1), 26–32. https://doi.org/10.1093/schbul/sbp097
Arvidsson, O., Gillberg, C., Lichtenstein, P., & Lundström, S. (2018). Secular changes in the symptom level of clinically diagnosed autism. Journal of Child Psychology and Psychiatry, 59(7), 744–751. https://doi.org/10.1111/jcpp.12864
Walsh, B., & Lynch, M. (2018). Short-term Changes in the Mean: 2. Truncation and Threshold Selection. In Evolution and Selection of Quantitative Traits. Oxford University Press. https://doi.org/10.1093/oso/9780198830870.003.0014
sessionInfo()## R version 4.1.2 (2021-11-01)
## Platform: x86_64-w64-mingw32/x64 (64-bit)
## Running under: Windows 10 x64 (build 19042)
##
## Matrix products: default
##
## locale:
## [1] LC_COLLATE=English_United States.1252
## [2] LC_CTYPE=English_United States.1252
## [3] LC_MONETARY=English_United States.1252
## [4] LC_NUMERIC=C
## [5] LC_TIME=English_United States.1252
##
## attached base packages:
## [1] splines stats4 stats graphics grDevices utils datasets
## [8] methods base
##
## other attached packages:
## [1] VGAM_1.1-6 pacman_0.5.1
##
## loaded via a namespace (and not attached):
## [1] digest_0.6.28 R6_2.5.1 jsonlite_1.7.2 magrittr_2.0.1
## [5] evaluate_0.14 rlang_0.4.12 stringi_1.7.5 jquerylib_0.1.4
## [9] bslib_0.3.1 rmarkdown_2.11 tools_4.1.2 stringr_1.4.0
## [13] xfun_0.27 yaml_2.2.1 fastmap_1.1.0 compiler_4.1.2
## [17] htmltools_0.5.2 knitr_1.36 sass_0.4.0