On 2014 Feb 02, John Armour commented:

Thanks Dorothy - I agree that could be a more incisive test of a weak right-shift model, but I also agree that power is a real problem, because the target population for such an analysis (right-handers with discordant MZ twins) is likely to constitute about 18-20% of monozygotic pairs (in our study, we looked at 862 MZ twins, of which 157 were discordant). In our data set we would therefore be looking at fewer “case” individuals than in a more direct sampling of left-handers as the “case” phenotype, so the extra power gained by specifying the phenotype of interest more precisely could be offset in any data set by the loss of numbers. My guess (though I haven’t done any further exploration to back up this instinct) is therefore that any study that would have enough discordant MZ twins to do this job well would probably also be in a good position to demonstrate the same shift even as a simple additive determinant of handedness, treating each twin as a random member of a wider population. As I’m sure you appreciate, the main aim of our study was to clarify unambiguously that the simplest and strongest formulations of single-gene models are simply untenable in the face of the data; there are ways we could have squeezed more power out the data if we wanted to really give ourselves the best chance of finding even a quite feeble locus, but the absence of a strong determinant was really the main outcome from our point of view.

On 2014 Jan 24, Dorothy V M Bishop commented:

I wondered if the authors could do more with their data, given that the sample consists of twin pairs. The most plausible genetic models are ones in which there is a genotype with no bias to left or right, vs one with a bias to right-handedness. A right-hander could come from either group. However, a right-hander with a left-handed MZ twin is likely to come from the no-bias group. Thus by focusing on MZ twins and reclassifying the phenotype on the basis of the twin pair (RR, RL and LL) one would be able to conduct a better test of association with a realistic phenotype. I suspect this study does not have big enough sample size to do this with adequate power, but I think that, in principle, it ought to work, and so might be a way forward for future studies.

On 2014 Jan 24, Dorothy V M Bishop commented:

I wondered if the authors could do more with their data, given that the sample consists of twin pairs. The most plausible genetic models are ones in which there is a genotype with no bias to left or right, vs one with a bias to right-handedness. A right-hander could come from either group. However, a right-hander with a left-handed MZ twin is likely to come from the no-bias group. Thus by focusing on MZ twins and reclassifying the phenotype on the basis of the twin pair (RR, RL and LL) one would be able to conduct a better test of association with a realistic phenotype. I suspect this study does not have big enough sample size to do this with adequate power, but I think that, in principle, it ought to work, and so might be a way forward for future studies.

On 2014 Feb 02, John Armour commented:

Thanks Dorothy - I agree that could be a more incisive test of a weak right-shift model, but I also agree that power is a real problem, because the target population for such an analysis (right-handers with discordant MZ twins) is likely to constitute about 18-20% of monozygotic pairs (in our study, we looked at 862 MZ twins, of which 157 were discordant). In our data set we would therefore be looking at fewer “case” individuals than in a more direct sampling of left-handers as the “case” phenotype, so the extra power gained by specifying the phenotype of interest more precisely could be offset in any data set by the loss of numbers. My guess (though I haven’t done any further exploration to back up this instinct) is therefore that any study that would have enough discordant MZ twins to do this job well would probably also be in a good position to demonstrate the same shift even as a simple additive determinant of handedness, treating each twin as a random member of a wider population. As I’m sure you appreciate, the main aim of our study was to clarify unambiguously that the simplest and strongest formulations of single-gene models are simply untenable in the face of the data; there are ways we could have squeezed more power out the data if we wanted to really give ourselves the best chance of finding even a quite feeble locus, but the absence of a strong determinant was really the main outcome from our point of view.