Covariate formats

SOLAR encodes covariartes in a special format, which is described in manual page for covariate command. Special rules are concerned with interaction terms, quadratic/cubic/etc variables and trait-specific covariates.

Some examples of covariate commands are given in the following table (copied from covariate manual page).

`SOLAR` command	Comment
`covariate age`	covariate age
`covariate age*sex`	age and sex interaction (only)
`covariate agediabetdiameds`	3-way interaction
`covariate age^2`	age squared
`covariate age^1,2`	shorthand for: age age^2
`covariate age#diabet`	shorthand for: age diabet age*diabet
`covariate age^1,2,3#sex`	shorthand for: sex age agesex age^2 age^2sex age^3 age^3*sex
`covariate sex age(q1) age*sex(q3)`	trait-specific covariates

solarius package provides two alternative interfaces to introduce covariates: formula argument or two arguments traits and covlist.

Interaction terms

The next three models are equvalent, and all introduce covariates age, sex and their interaction effect. By looking at the slot cf of the model object, one can ensure that the three covariate terms are presented and their p-values are identical.

mod1 <- solarPolygenic(trait1 ~ age + sex + age*sex, data = dat30, covtest = TRUE)
mod1$cf

##   covariate Estimate      SE    Chi   pval
## 1       age -0.02009 0.02357 0.7238 0.3949
## 2       sex -0.45079 0.31118 2.0887 0.1484
## 3   age*sex  0.03469 0.03104 1.2471 0.2641

mod2 <- solarPolygenic(traits = "trait1", covlist = c("age", "sex", "age*sex"), 
  data = dat30, covtest = TRUE)
mod2$cf

##   covariate Estimate      SE    Chi   pval
## 1       age -0.02009 0.02357 0.7238 0.3949
## 2       sex -0.45079 0.31118 2.0887 0.1484
## 3   age*sex  0.03469 0.03104 1.2471 0.2641

mod3 <- solarPolygenic(traits = "trait1", covlist = c("age#sex"), data = dat30, 
  covtest = TRUE)
mod3$cf

##   covariate Estimate      SE    Chi   pval
## 1       age -0.02009 0.02357 0.7238 0.3949
## 2       sex -0.45079 0.31118 2.0887 0.1484
## 3   age*sex  0.03469 0.03104 1.2471 0.2641

If one encodes covariates in a form age*sex, SOLAR will put only the interaction term (this behaviour is different from, for example, lm model in R).

mod4 <- solarPolygenic(trait1 ~ age*sex, data = dat30, covtest = TRUE)
mod4$cf

##   covariate Estimate      SE    Chi   pval
## 1   age*sex  0.01253 0.01952 0.4136 0.5202

The resulted model mod4 does not make sense, as the interaction term is presented without age and sex covariates.

Once can see the actual command passed to SOLAR.

mod1$solar$cmd

## [1] "trait trait1"                      
## [2] "covariate age sex age*sex"         
## [3] ""                                  
## [4] "polygenic  -prob 0.05 -screen -all"