08-01-2016 04:54 PM
I am trying to look at the correlation between mom's weight at time of birth and her kids' weight later in life. I'm doing this by comparing a given mom's weight at the time of each of her kids' births, and their weight later on in life. I currently have my data constructed so that all siblings (same mom and dad) are paired off (so if there are four siblings in a family, there will be 6 observations of different sibling pairs). Within the data, each observation has the following variables: the ID of two siblings (ID_1 and ID_2), the mom's weight at each birth (mom_wt_1 and mom_wt_2), and the kids' weight later on (kid_wt_1 and kid_wt_2). There are several thousand observations. I am looking for a significant increase in kids' weight compared to their siblings if the mom was heavier when giving birth.
What procedure should I use to look for this correlation?
Thanks for your help!
08-01-2016 05:24 PM
Are all of the childrens later weight measures done at the same age? At approximately what ages?
You may want to consider using a standardized height/weight measure for the age such as from CDC or WHO growth charts as if one set of children are measured at 11 months and another at 12 months they aren't quite the same thing.
08-08-2016 01:04 PM
Before attacking the sibling issue, how have you addressed the effect of maternal weight on subsequent weight of individuals?
That would be an interesting question to come at first. You are probably going to have to spline the response variable as a function of time, as you undoubtedly have offspring weights at different intervals. A good example to start with would be Example 45.15 Comparing Multiple B-Splines in the PROC GLIMMIX documentation. I am starting there as the addition of the sibling effect is going to require random effects in the model.
If you have the age at subsequent weights, consider the following:
proc glimmix data=yourdata; class maternal_id; effect spl = spline(age); model offspring_wt = spl*maternal_wt / s noint; random intercept / subject=maternal_id; run;
If I have completely missed the design here, let me know, and I can try to fix it.
08-09-2016 05:16 PM
Thanks for your help, I really appreciate it! I'm still a bit confused about how I would use this to look between sibling pairs. I see that you're looking at the correlation of offspring weight from maternal weight in the model statement, but how does this then allow for a comparison between different sibling pairs?
08-15-2016 11:08 AM
Well, the approach I suggested was to not look at the sibling effect yet, but just see if the model as proposed will give reasonable results. The more I read this, the less I think of sibling pairs as units, but rather birth order as a categorical factor in the model. The differences between siblings could then be obtained by looking at the differences in the least squares means. However, there is a fair chance I have misinterpreted the design.