covariates

[ZhiyiZhu0226]

Hi, I was wondering why is agesex, agesex^2 included as covariates. Is age and sex not enough to account as the covariates?

[robwwilliams]

Dear Zhiyi—more context please. But I think the answer is yes. If sample size permits a more complex model with interaction terms and high order covariates, they can sometimes be useful.

…

On Tue, Feb 21, 2023 at 2:02 PM Zhiyi Zhu ***@***.***> wrote: Hi, I was wondering why is age*sex, age*sex^2 included as covariates. Is age and sex not enough to account as the covariates? — Reply to this email directly, view it on GitHub <#43>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AC55V7ARIS7NPT4E2ZT2VEDWYUNPDANCNFSM6AAAAAAVDO3I7A> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

-- Rob Robert W Williams PhD Chair: Department of Genetics, Genomics and Informatics UT-ORNL Governor's Chair in Computational Genomics 71 S Manassas St, Memphis TN 38163 University of Tennessee Health Science Center, TSRB 405 Cell: 901 604 4752 EMAIL: ***@***.*** Zoom: https://tennesseehipaa.zoom.us/j/9492269658 ----------------

[ZhiyiZhu0226]

Hi, thanks for responding! For the UKBB GWAS, imputed-v3 Association model (https://github.com/Nealelab/UK_Biobank_GWAS), it was stated that the following are included as covariates: 1st 20 PCs + sex + age + age^2 + sex * age + sex * age^2. My question is why are the last two terms, sex * age and sex * age^2, are included as covariates. I have rarely seen other studies that include combinations of age and sex as covariates (I have seen one that uses age-by-sex as one of their covariates: https://www.nature.com/articles/s41586-021-04103-z), and I'm curious about the motivation behind using those two covariates. What variance are those covariates trying to account for?

[robwwilliams]

Ah, not my work but sex* age is a useful interaction term, sex * age^2 will model steeper age related sex interaction effects that become much more important from 40 years up.

On Tue, Feb 21, 2023 at 15:55 Zhiyi Zhu ***@***.***> wrote: Hi, thanks for responding! For the UKBB GWAS, imputed-v3 Association model (https://github.com/Nealelab/UK_Biobank_GWAS), it was stated that the following are included as covariates: 1st 20 PCs + sex + age + age^2 + sex * age + sex * age^2. My question is why are the last two terms, sex * age and sex * age^2, are included as covariates. I have rarely seen other studies that include combinations of age and sex as covariates (I have seen one that uses age-by-sex as one of their covariates: https://www.nature.com/articles/s41586-021-04103-z), and I'm curious about the motivation behind using those two covariates. What variance are those covariates trying to account for? — Reply to this email directly, view it on GitHub <#43 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AC55V7GTY5GMLOXB66JEMQ3WYU2UJANCNFSM6AAAAAAVDO3I7A> . You are receiving this because you commented.Message ID: ***@***.***>

-- Rob Robert W Williams PhD Chair: Department of Genetics, Genomics and Informatics UT-ORNL Governor's Chair in Computational Genomics 71 S Manassas St, Memphis TN 38163 University of Tennessee Health Science Center, TSRB 405 Cell: 901 604 4752 EMAIL: ***@***.*** Zoom: https://tennesseehipaa.zoom.us/j/9492269658 ----------------

[ZhiyiZhu0226]

Thanks for the answer! I have a follow up question: how is the age * sex interaction calculated? Currently I'm encoding females as 0 and males as 1. So is the age * sex interaction for a 56 year old female 0 (or for a 47 year old male, 47), or is the interaction calculated in some more complicated way? It doesn't make much sense to me to multiple a categorical variable (sex) and a quantitative variable (age)

[robwwilliams]

Dear Zhihui: Should be pretty standard LMM. This is one reason fir the word “mixed” in LMM.

On Thu, Feb 23, 2023 at 16:49 Zhiyi Zhu ***@***.***> wrote: Thanks for the answer! I have a follow up question: how is the age * sex interaction calculated? Currently I'm encoding females as 0 and males as 1. So is the age * sex interaction for a 56 year old female 0 (or for a 47 year old male, 47), or is the interaction calculated in some more complicated way? It doesn't make much sense to me to multiple a categorical variable (sex) and a quantitative variable (age) — Reply to this email directly, view it on GitHub <#43 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AC55V7FSYQRPYNJO3E5QHMDWY7SPDANCNFSM6AAAAAAVDO3I7A> . You are receiving this because you commented.Message ID: ***@***.***>

-- Rob Robert W Williams PhD Chair: Department of Genetics, Genomics and Informatics UT-ORNL Governor's Chair in Computational Genomics 71 S Manassas St, Memphis TN 38163 University of Tennessee Health Science Center, TSRB 405 Cell: 901 604 4752 EMAIL: ***@***.*** Zoom: https://tennesseehipaa.zoom.us/j/9492269658 ----------------