I looked at the chapter on bullying and I found the methodology weak, given the huge inherent issues with passive observation.
It is really really hard to “control” for other factors and their efforts were quite lame. Several particular problems appear. First they correct for other factors one factor at a time. This is a failure mode when multiple factors are relevent at the same time e.g. IQ and poor parenting. Second they make no allowance for errors in measurement of factors. As one example they correct for childhood IQ to exclude IQ as a factor that may both lead to being bullied and that may be harmful independently of bullying. But they do not correct for measurement error. Any measurement of factor X will have error and thus tests based on the measurement will understate the effect of the actual factor X. In the particular case of IQ, childhood IQ is not very highly correlated to IQ in adulthood and thus is a poor proxy for IQ in adulthood. It is also poorly correlated to parental IQ and thus heavily fails to capture effects operating via shared IQ genes in the parents.
See Judea Pearl’s book “Causality” for a detailed discussion of these issues and what a proper causal model looks like.
I would not base any serious decison on the findings in the book. They are as likely to be anti-knowledge as real knowledge IMHO.
I looked at the chapter on bullying and I found the methodology weak, given the huge inherent issues with passive observation.
It is really really hard to “control” for other factors and their efforts were quite lame. Several particular problems appear. First they correct for other factors one factor at a time. This is a failure mode when multiple factors are relevent at the same time e.g. IQ and poor parenting. Second they make no allowance for errors in measurement of factors. As one example they correct for childhood IQ to exclude IQ as a factor that may both lead to being bullied and that may be harmful independently of bullying. But they do not correct for measurement error. Any measurement of factor X will have error and thus tests based on the measurement will understate the effect of the actual factor X. In the particular case of IQ, childhood IQ is not very highly correlated to IQ in adulthood and thus is a poor proxy for IQ in adulthood. It is also poorly correlated to parental IQ and thus heavily fails to capture effects operating via shared IQ genes in the parents.
See Judea Pearl’s book “Causality” for a detailed discussion of these issues and what a proper causal model looks like.
I would not base any serious decison on the findings in the book. They are as likely to be anti-knowledge as real knowledge IMHO.