Model Configurations

Shixiang Wang

Central South University
wangshx@csu.edu.cn

2025-08-19

library(bregr)

Interaction Terms

The bregr package provides comprehensive support for the processing and visualization of interaction terms in regression models, provided that the interaction term is inherently supported by the model.

Below is an illustrative example using a linear model:

rv <- br_pipeline(
  data = mtcars,
  y = "mpg", x = c("cyl", "disp*hp"), x2 = "am",
  method = "lm"
)

Examine the results:

br_get_results(rv, tidy = TRUE)
#> # A tibble: 6 × 8
#>   Focal_variable term    estimate std.error statistic p.value conf.low conf.high
#>   <chr>          <chr>      <dbl>     <dbl>     <dbl>   <dbl>    <dbl>     <dbl>
#> 1 cyl            cyl     -2.50e+0 0.361         -6.93 1.28e-7 -3.24e+0 -1.76    
#> 2 cyl            am       2.57e+0 1.29           1.99 5.64e-2 -7.42e-2  5.21    
#> 3 disp*hp        disp    -5.42e-2 0.0186        -2.92 7.06e-3 -9.23e-2 -0.0160  
#> 4 disp*hp        hp      -9.20e-2 0.0244        -3.77 8.16e-4 -1.42e-1 -0.0419  
#> 5 disp*hp        am       2.29e+0 1.45           1.57 1.27e-1 -6.96e-1  5.27    
#> 6 disp*hp        disp:hp  2.27e-4 0.0000938      2.42 2.26e-2  3.45e-5  0.000419

Visualize the results:

br_show_forest(rv)

Additionally, specified models can be visualized using the functions provided by bregr:

br_show_forest_ggstats(rv, idx = 2)

Or compare them:

br_show_forest_ggstats(rv)

Customized GLM Family

For GLM models, the family object can be further customized with various arguments. By default, bregr directly supports GLM family functions with their default arguments, which can be viewed using:

br_avail_methods()
#>  [1] "lm"               "coxph"            "survreg"          "clogit"          
#>  [5] "cch"              "binomial"         "gaussian"         "Gamma"           
#>  [9] "inverse.gaussian" "poisson"          "quasi"            "quasibinomial"   
#> [13] "quasipoisson"     "nls"              "aov"

However, for alternative GLM configurations, modifications can be made directly. For instance, consider the following example using a quasi-family with specified variance and link functions:

data <- data.frame(
  x = rnorm(100)
)
data$y <- rpois(100, exp(1 + data$x))
head(data)
#>            x y
#> 1  0.5594572 6
#> 2 -0.7838015 0
#> 3 -1.9278205 1
#> 4  1.1311586 8
#> 5 -1.3567963 0
#> 6  1.1693306 8
rv <- br_pipeline(
  data = data,
  y = "y", x = "x",
  method = 'quasi(variance = "mu", link = "log")'
)
#> Warning: nonstandard method `quasi(variance = "mu", link = "log")` passed to
#> `stats::glm()`, double-check if it's correct
br_get_results(rv, tidy = TRUE)
#> # A tibble: 1 × 8
#>   Focal_variable term  estimate std.error statistic  p.value conf.low conf.high
#>   <chr>          <chr>    <dbl>     <dbl>     <dbl>    <dbl>    <dbl>     <dbl>
#> 1 x              x         1.02    0.0517      19.7 7.70e-36    0.916      1.12

This is equivalent to:

glm(y ~ x, data = data, family = quasi(variance = "mu", link = "log")) |>
  summary()
#> 
#> Call:
#> glm(formula = y ~ x, family = quasi(variance = "mu", link = "log"), 
#>     data = data)
#> 
#> Coefficients:
#>             Estimate Std. Error t value Pr(>|t|)    
#> (Intercept)  0.99441    0.06114   16.26   <2e-16 ***
#> x            1.01762    0.05167   19.69   <2e-16 ***
#> ---
#> Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
#> 
#> (Dispersion parameter for quasi family taken to be 0.8259798)
#> 
#>     Null deviance: 424.092  on 99  degrees of freedom
#> Residual deviance:  99.266  on 98  degrees of freedom
#> AIC: NA
#> 
#> Number of Fisher Scoring iterations: 5