You are here

some doubts

7 posts / 0 new
Last post
Lola's picture
Offline
Joined: 12/28/2016 - 18:42
some doubts

Hi everyone,
I am really new in OpenMx. I am trying to do a bivariate analysis (dicotomic-quantitative) and I have some questions.
1-I would like to know if I should do a univariate analysis previously I mean in the bivariate analysis The best fitting model could be AE for example, for two variables but in the univariate I could have AE model for one variable and ACE model for the other variable.
2-There is a big difference in the means of one variable and I have problems in the saturated model. I think I should try a sex limitation model, but I cannot find a bivariate sex limitation script.
3- In the sense of that I have read a bivariate sex-limitation model is quite difficult and I would like to know when I have to do cholesky decomposition and how. I am quite confused with that.
PD: I use Boulder´s scripts
Thanks in advance any help or material will help me.

Best regards

AdminRobK's picture
Offline
Joined: 01/24/2014 - 12:15
some answers
1-I would like to know if I should do a univariate analysis previously I mean in the bivariate analysis The best fitting model could be AE for example, for two variables but in the univariate I could have AE model for one variable and ACE model for the other variable.

Others may disagree with me, but I would say not to bother with the single-trait analyses if you're going to do a two-trait analysis. That's because the covariance between the two phenotypes carries additional information about the influence of differing sources of variation that is not present in a monophenotype analysis. Incidentally, that's also why a diphenotype analysis might suggest dropping C whereas monophenotype analyses might suggest retaining it for at least one trait.

2-There is a big difference in the means of one variable and I have problems in the saturated model. I think I should try a sex limitation model, but I cannot find a bivariate sex limitation script.

Could you be more specific here? In particular, what is meant by "big difference in the means of one variable," and what kinds of problems are you encountering with the saturated model?

3- In the sense of that I have read a bivariate sex-limitation model is quite difficult and I would like to know when I have to do cholesky decomposition and how. I am quite confused with that.

I would refer you to this article concerning the problems with using the Cholesky parameterization in polyphenotype sex-limitation analyses. I don't have a diphenotype sex-limitation script offhand, but such an example script probably exists somewhere (perhaps another user reading this post has one?).

Lola's picture
Offline
Joined: 12/28/2016 - 18:42
Doubts

1-Ok! I think I understand it a little better, but if one model fit a clear ACE and in the other a clear AE model, would we losing information? since we would have to chose one of them ¿no?

2-The mean between men and women are quite different. I mean the men´s mean is quite lower than the women´s mean, in one variable so I would like to do a sex-limitation. Moreover, in the saturated model, in the condition " twin order and zygosity" I have significant differences.

3-ok! Since I'm really new to this. Any additional material apart from Neale &Cardon 92?

Thank you so much I really appreciate your help

AdminRobK's picture
Offline
Joined: 01/24/2014 - 12:15
more answers
1-Ok! I think I understand it a little better, but if one model fit a clear ACE and in the other a clear AE model, would we losing information? since we would have to chose one of them ¿no?

In a diphenotype model, you could still retain C for one phenotype and drop it for the other. The point is that an analysis of both phenotypes at the same time provides more information because it biometrically decomposes the covariance between the two phenotypes, as well as the variances of each.

2-The mean between men and women are quite different. I mean the men´s mean is quite lower than the women´s mean, in one variable so I would like to do a sex-limitation. Moreover, in the saturated model, in the condition " twin order and zygosity" I have significant differences.

Do you hypothesize that two sexes differ not just in phenotypic means, but in phenotypic variance-covariance structure as well? Because if not, I don't think that the tricky aspects of sex-limitation will apply, since you would only need to condition the mean of the phenotype on sex (either by estimating separate means by sex or by using sex as a definition variable).

3-ok! Since I'm really new to this. Any additional material apart from Neale &Cardon 92?

Do you lack access to the article I linked in my previous post? I think it directly addresses your question about the Cholesky parameterization in sex limitation models pretty well.

Lola's picture
Offline
Joined: 12/28/2016 - 18:42
Thank you!

1- Perfect, I think I understand it.
2-Sorry if I do not explain it well but my English is not very good.
I think I should do a sex-limitation model since the means are different between males and females also in the condition " twin order and zygosity" I have significant differences (saturated model).
3- I already read the article as you told me but I meant more documentation in general not just sex-limitation
Once again thank you very much I am too new and this is quite difficult

neale's picture
Offline
Joined: 07/31/2009 - 15:14
Twin order & zygosity effects

Hi Lola

The significant differences with respect to twin order and zygosity are a concern, unless they are entirely due to the sex differences. If they aren't they should be examined carefully. Often such significant differences occur if there are outliers or the distribution of the data departs a lot from multivariate normal (especially in the case where the data are being treated as continuous). It may be better to treat the data as ordinal in such cases.

A colleague, Dr. Maes, will be in touch with a website or a script for you to use for sex limitation via the correlated factors model.

RFrank's picture
Offline
Joined: 06/08/2012 - 14:22
treating data as ordinal

I've had this issue before and I think it probably relates to treating data as continuous when it's not really. I've seen scripts for using dichotomous data, but does anyone have any resources for analyzing "count" data in a bivariate cholesky? (Specifically when one of the variables IS continuous?)