Dear Conor, Dear Dirk

We know that *"rmz > 2*rdz suggests an ADE model and rmz < 2*rdz suggests an ACE model"*. How sensitive is this measure, is there a point in choosing and ADE based on e.g. .502 > 2*.250? Or in case of many variables, if some point to an ADE, some to an ACE, do we follow the majority? Or in extreme cases, when two of four variables suggest an ADE, the other two an ACE, do we flip a coin then?

Thanks, Jens

good question ! you touch upon a problem. 1) choosing an ADE model in a univariate analysis based on rmz>2*rdz is ok, but just a rule of thumb which does not take into account statistical significance (e.g., rmz=.4, rdz=.19; 2*rdz=.38 … yes 2*rdz< rmz, but .4 vs .38 …). 2) the rule of thumb can break down in multivariate analysis, for reason you mention. let’s consider some scenario’s. a) all correlations indicate ADE (or ACE)… no problem. b) some indicate ACE and some indicate ADE, but the 2*rdz - rmz are small (say between -,05 and +.05)… not a big problem. it is likely that you end up this AE model. c) some indicate clearly ACE others indicate clearly ADE. in that case you can specify a multivariate twin model that includes ADE for some and ACE for other phenotypes. in this scenario (c), the covariance between phenotype X (ACE model) and phenotype Y (ADE model) cannot be a function of C or D (because these sources of variance are not shared by X and Y. See also