Getting started with baselinr

The problem

Every quasi-experimental impact study in education has to answer the same question before anyone looks at outcomes: were the treatment and comparison groups similar enough at baseline? The What Works Clearinghouse (WWC) sets the de facto standard for this in education research:

• a covariate with a standardized mean difference (Hedges’ g) of 0.05 or less satisfies baseline equivalence on its own;
• between 0.05 and 0.25, equivalence holds only if the covariate is statistically adjusted for in the impact model;
• above 0.25, the covariate cannot establish equivalence.

baselinr computes those effect sizes and categories so the baseline table is not something you assemble by hand for every report.

A worked example

study <- data.frame(
  treat = c(1, 1, 1, 0, 0, 0),
  pretest = c(5, 6, 7, 4, 5, 6), # continuous -> Hedges' g
  female = c(1, 0, 1, 0, 0, 1) # binary     -> Cox index
)

baseline_equivalence(study, treatment = "treat")
#>   covariate       type n_treatment n_comparison mean_treatment mean_comparison
#> 1   pretest continuous           3            3      6.0000000       5.0000000
#> 2    female     binary           3            3      0.6666667       0.3333333
#>   sd_treatment sd_comparison effect_size  wwc_category
#> 1    1.0000000     1.0000000   0.8000000 not_satisfied
#> 2    0.5773503     0.5773503   0.8401784 not_satisfied

By default, every numeric, logical, and factor column other than the treatment indicator is treated as a covariate. A covariate with exactly two unique values is treated as binary and summarized with the Cox index; other numeric covariates use Hedges’ g. Pass covariates = to control the set explicitly.

The building blocks

baseline_equivalence() is built from exported helpers you can also call directly.

# Standardized mean difference (Hedges' g) for a continuous covariate
hedges_g(study$pretest, study$treat)
#> [1] 0.8

# Cox index for a binary covariate
cox_index(study$female, study$treat)
#> [1] 0.8401784

# Classify any effect size(s) into the WWC categories
wwc_classify(c(0.03, 0.12, 0.80))
#> [1] "satisfied"                 "satisfied_with_adjustment"
#> [3] "not_satisfied"

Visualise and format

A Love plot shows the standardized effect size of each covariate against the WWC thresholds (0.05 and 0.25), coloured by category:

love_plot(baseline_equivalence(study, treatment = "treat"))

For a report-ready table, gt_baseline() returns a formatted gt table:

gt_baseline(baseline_equivalence(study, treatment = "treat"))

Scope

Continuous covariates use Hedges’ g (with the WWC small-sample correction); binary covariates use the WWC Cox index. Collapse the table into an overall verdict with wwc_summary(), assess sample loss with attrition(), visualise with love_plot(), and format with gt_baseline(). See NEWS.md for the roadmap.