King Et Al. Respond

We appreciate interest in our article and are particularly grateful that Durkin et al. highlighted the importance of considering parameterization in research on parental age and autism. Making sense of parental age has always been complicated because homophily on age plays a significant role in partner selection. The decomposition presented in our article mitigates these statistical problems. As Durkin et al. noted, our decomposition utilizes a continuous parameterization of parental age. It also eliminates problems arising from multicollinearity and provides a means of assessing whether the effects of maternal and paternal age on autism risk are joint or independent. We hope that future work will pay greater attention to choice of parameterization, as parental age appears particularly sensitive to specification. The decomposition results prove more stable than the categorical results. Our analyses demonstrate that the instability of the categorical age results for autism could lead to erroneous conclusions in pooled data. For instance, in 1995 both the estimates from the maternal and paternal cohort analyses differed from the pooled estimate by more than 20%. In the pooled data there appears to be a critical crossover in the effect of maternal and paternal age in that year. In fact, the cohort data reveals a 44% difference in risk. Similarly, there is more than a 10% difference in risk between the pooled and cohort estimate for paternal age in 2000. Here, again, the risk associated with maternal age exceeds that of paternal age in the cohort data. The opposite appears to be the case in the analysis of the pooled data. What should one conclude from this? First, different analyses conducted at different points in time and with different analysis strategies can lead to different conclusions about risk factors. Second, analyses that do not suffer from the statistical fallacies produced by pooling categorical data demonstrate that maternal age, rather than paternal age, may pose greater risk. Are these differences meaningful? Given that individuals may make life choices on the basis of these reports, our sense is that they are. Durkin et al. are absolutely right about the need to be sensitive to model specification. We agree completely. We also think it critical to be attentive to temporality. Although this is true for all research, it appears to be particularly true for autism, because the size and the composition of the population with autism has changed over time.1

Autism · Biology · Categorical variable · Cohort · Developmental psychology · Multicollinearity · Paternal age · Preadolescence · Regression analysis · Sociology · Statistics · Autism Spectrum Disorder Research · Demography · Family and Disability Support Research · Genetics · Mathematics · Psychology · Virology and Viral Diseases