Does having self and other report measures as predictors require dyad analysis? I have three different measures of child anxiety (IVs). For each measure, I have a  self-rating by the child and an other-rating by a parent. 
I want to use these measures to predict a particular diagnosis (DV). 


*

*How do I take into account the dyadic nature of the parent-child relationship? 


I plan to use hierarchical regression or path analysis  to analyse the predictive relationship between the IVs and DV.
 A: Short answer: You don't need to do dyadic analysis
You can include all six predictors (i.e., 2 raters by 3 constructs) in a standard multiple regression (or perhaps logistic regression if diagnosis is binary). 
In fact, this will allow you to examine questions such as whether child or parent ratings are better predictors of diagnosis.
Also, there are other things you may want to look at.
In order to better understand your data, you may wish to explore the intercorrelations between the six predictors. E.g.,


*

*What is the correlation between child and parent ratings?

*Are the scale intercorrelations for parents and children similar?


You might want to explore multimethod-multitrait approaches.
A: As long as each child and each parent is unique (i.e., there are no mothers who rate more than one of their childrens, and no children who are rated by both parents), you are safe with a conventional analysis - there is no hierarchical structure.
Dyadic dependencies arise as soon as you have multiple raters per target (a one-with-many design), or raters that judge multiple targets.
For more details, the definite reference is the book "Dyadic Data Analysis".
Kenny, D. A., Kashy, D. A., & Cook, W. L. (2006). Dyadic data analysis. New York: Guilford.
