Madole & Harden (M&H) describe how genetics can be used in a causal framework. We agree with the authors that viewing genetic causality as probabilistic, instead of deterministic, is a fruitful way to view the effects of genetics on behavior. Moreover, we agree that using within-family designs will strengthen our understanding of the causal effects of genetics on development (Hart, Taylor, & Schatschneider, Reference Hart, Taylor and Schatschneider2013; van Dijk, Norris, & Hart, Reference van Dijk, Norris and Hartin press). However, there are two points we wish to discuss from Madole & Harden (M&H). First, we believe we do not need to equate within-family designs to experiments, including randomized control trials (RCTs) and natural experiments. Second, we believe that the effects of genetics on behavior are better described as what they are, inus conditions, rather than describing what they are not, non-uniform, non-unitary, and non-explanatory.
Philosophers have been debating the meaning of causality for centuries, including John Stewart Mill, who formalized three conditions for establishing causality. First, the cause must precede the effect, second, the cause must be related to the effect, and third, we can find no plausible explanation for the effect other than the cause (see Shadish, Cook, & Campbell, Reference Shadish, Cook and Campbell2002). The first two conditions are easily established by many designs. The third condition is the hardest to meet. Experiments, defined by the use of random assignment with a manipulation, are common and powerful tools we use to meet the third condition, as they allow us to rule out other plausible explanations. However, ultimately causality is determined by meeting those three conditions, whether you have experimentation or not. M&H suggest that genetic transmission from parent to child can be interpreted the same as an average treatment effect from an RCT. We agree that a within-family design and an RCT carry a lot of strong causal information, but we disagree with describing the two designs as similar. Within-family designs are different than RCTs in an important way. An RCT is a specific type of experiment that introduces a manipulable cause to examine whether that cause increases or decreases a given behavior. Within-family designs investigate non-manipulable causes to examine whether genetic influences have effects on variability across the range of behaviors. We believe there is no need to make the comparison between within-family designs and RCTs as on their own within-family designs do meet the conditions laid out by John Stewart Mill for causal conclusions. We also believe that we do not need to equate within-family designs to natural experiments, which is often done, as natural experiments also have a manipulation. By forcing the language of experiments, no matter the type, on within-family designs, we leave ourselves open to criticisms that within-family designs do not have a manipulation and therefore can never establish causality. However, establishing causality does not need an experimental manipulation, and as a field we do not need to borrow the language of experimental designs, whether it is a natural experiment or RCTs, to show that our results can still inform us about causality. With a within-family design that is estimating the genetic transmission from parent to child, if the assumption of the equal environments is met, there are no other plausible genetic or environmental explanations for the effects of the specific genes on behavior. We can instead use the language of what we do have, which is a unique design with no formal name that we know of, which has randomization of genes because of miosis and non-manipulated causes, and looks at the impact of variation, as opposed to mean differences. We believe this unique and powerful design can meet John Stewart Mill's three conditions for establishing causality.
Second, M&H assert that genetic effects should most likely be viewed as non-uniform, non-unitary, and non-explanatory. We agree with this position. It is highly likely that most of the effects of genes on behavior are not causally deterministic but instead only impact the chances of a behavior occurring. However, this position describes these effects in terms of what they are not. The reason that these effects may be non-uniform, non-unitary, and non-explanatory is because they most likely operate as inus conditions (Mackie, Reference Mackie1974; Shadish et al., Reference Shadish, Cook and Campbell2002). Inus stands for an insufficient but nonredundant part of an unnecessary but sufficient condition. Insufficient means that the existence of this factor by itself is not enough to cause the effect. Nonredundant means that in the constellation of factors that come together to produce an effect, a particular factor provides something unique that the other factors do not. Unnecessary means that the effect could be produced by other factors even in the absence of a particular factor. Sufficient means that in concert with other factors, it is enough to produce an effect. For example, having a particular polymorphism alone is insufficient to produce a particular behavior. Every behavior, for example, needs an environment that is capable of allowing that effect. It is nonredundant in that a particular polymorphism is unique within a person. It is unnecessary in that the behavior could be observed without the polymorphism, but it is sufficient in that in combination with a constellation of other factors it either promotes or inhibits a behavior. It is because of these conditions that the effects of genes may be non-uniform, non-unitary, and non-explanatory. Most behaviors of interest to scientists fall under the category of inus conditions because most behaviors that we study have multiple causes in the sense that there are many factors that either promote or inhibit a behavior. It is likely that genetic effects operate on behaviors in the same way.
We applaud M&H for reminding us that within-family designs can inform causality and extending this discussion by laying out how genetic influences might operate within a causal framework. We believe our commentary will help sharpen the concepts and language around this causal framework.
You have
Access
Madole & Harden (M&H) describe how genetics can be used in a causal framework. We agree with the authors that viewing genetic causality as probabilistic, instead of deterministic, is a fruitful way to view the effects of genetics on behavior. Moreover, we agree that using within-family designs will strengthen our understanding of the causal effects of genetics on development (Hart, Taylor, & Schatschneider, Reference Hart, Taylor and Schatschneider2013; van Dijk, Norris, & Hart, Reference van Dijk, Norris and Hartin press). However, there are two points we wish to discuss from Madole & Harden (M&H). First, we believe we do not need to equate within-family designs to experiments, including randomized control trials (RCTs) and natural experiments. Second, we believe that the effects of genetics on behavior are better described as what they are, inus conditions, rather than describing what they are not, non-uniform, non-unitary, and non-explanatory.
Philosophers have been debating the meaning of causality for centuries, including John Stewart Mill, who formalized three conditions for establishing causality. First, the cause must precede the effect, second, the cause must be related to the effect, and third, we can find no plausible explanation for the effect other than the cause (see Shadish, Cook, & Campbell, Reference Shadish, Cook and Campbell2002). The first two conditions are easily established by many designs. The third condition is the hardest to meet. Experiments, defined by the use of random assignment with a manipulation, are common and powerful tools we use to meet the third condition, as they allow us to rule out other plausible explanations. However, ultimately causality is determined by meeting those three conditions, whether you have experimentation or not. M&H suggest that genetic transmission from parent to child can be interpreted the same as an average treatment effect from an RCT. We agree that a within-family design and an RCT carry a lot of strong causal information, but we disagree with describing the two designs as similar. Within-family designs are different than RCTs in an important way. An RCT is a specific type of experiment that introduces a manipulable cause to examine whether that cause increases or decreases a given behavior. Within-family designs investigate non-manipulable causes to examine whether genetic influences have effects on variability across the range of behaviors. We believe there is no need to make the comparison between within-family designs and RCTs as on their own within-family designs do meet the conditions laid out by John Stewart Mill for causal conclusions. We also believe that we do not need to equate within-family designs to natural experiments, which is often done, as natural experiments also have a manipulation. By forcing the language of experiments, no matter the type, on within-family designs, we leave ourselves open to criticisms that within-family designs do not have a manipulation and therefore can never establish causality. However, establishing causality does not need an experimental manipulation, and as a field we do not need to borrow the language of experimental designs, whether it is a natural experiment or RCTs, to show that our results can still inform us about causality. With a within-family design that is estimating the genetic transmission from parent to child, if the assumption of the equal environments is met, there are no other plausible genetic or environmental explanations for the effects of the specific genes on behavior. We can instead use the language of what we do have, which is a unique design with no formal name that we know of, which has randomization of genes because of miosis and non-manipulated causes, and looks at the impact of variation, as opposed to mean differences. We believe this unique and powerful design can meet John Stewart Mill's three conditions for establishing causality.
Second, M&H assert that genetic effects should most likely be viewed as non-uniform, non-unitary, and non-explanatory. We agree with this position. It is highly likely that most of the effects of genes on behavior are not causally deterministic but instead only impact the chances of a behavior occurring. However, this position describes these effects in terms of what they are not. The reason that these effects may be non-uniform, non-unitary, and non-explanatory is because they most likely operate as inus conditions (Mackie, Reference Mackie1974; Shadish et al., Reference Shadish, Cook and Campbell2002). Inus stands for an insufficient but nonredundant part of an unnecessary but sufficient condition. Insufficient means that the existence of this factor by itself is not enough to cause the effect. Nonredundant means that in the constellation of factors that come together to produce an effect, a particular factor provides something unique that the other factors do not. Unnecessary means that the effect could be produced by other factors even in the absence of a particular factor. Sufficient means that in concert with other factors, it is enough to produce an effect. For example, having a particular polymorphism alone is insufficient to produce a particular behavior. Every behavior, for example, needs an environment that is capable of allowing that effect. It is nonredundant in that a particular polymorphism is unique within a person. It is unnecessary in that the behavior could be observed without the polymorphism, but it is sufficient in that in combination with a constellation of other factors it either promotes or inhibits a behavior. It is because of these conditions that the effects of genes may be non-uniform, non-unitary, and non-explanatory. Most behaviors of interest to scientists fall under the category of inus conditions because most behaviors that we study have multiple causes in the sense that there are many factors that either promote or inhibit a behavior. It is likely that genetic effects operate on behaviors in the same way.
We applaud M&H for reminding us that within-family designs can inform causality and extending this discussion by laying out how genetic influences might operate within a causal framework. We believe our commentary will help sharpen the concepts and language around this causal framework.
Financial support
This project was supported by funding from the Eunice Kennedy Shriver National Institute of Child Health and Human Development grants P50HD052120 and R01HD095193. Views expressed herein are those of the authors and have neither been reviewed nor approved by the granting agencies.
Competing interest
None.