Model-based continuous summary tables

library(spicy)

table_continuous_lm() is the model-based companion to table_continuous(). It fits one linear model per selected continuous outcome – lm(outcome ~ by + covariate1 + covariate2 + ...) if covariates are supplied, otherwise lm(outcome ~ by) – and returns a compact reporting table. This is the right choice when you want to stay in a linear-model workflow with heteroskedasticity-consistent or cluster-robust standard errors, bootstrap / jackknife variance, case weights, additive covariate adjustment, or one of four effect-size families (f2, Cohen’s d, Hedges’ g, Hays’ \(\omega^2\)) with noncentral CIs. Formatting follows APA conventions by default; style = "jama" (or "nejm", "lancet", "annals", "aer") switches the whole table to that journal’s published rules — the House styles section of vignette("summary-tables-reporting") shows each style and the guideline sentence behind every rule.

Basic usage

Use select for one or more continuous outcomes and by for the single predictor:

table_continuous_lm(
  sochealth,
  select = c(wellbeing_score, bmi, life_sat_health),
  by = sex
)
#> Continuous outcomes by Sex
#> 
#>  Variable                       │ M (Female)  M (Male)  Δ (Male - Female) 
#> ────────────────────────────────┼─────────────────────────────────────────
#>  WHO-5 wellbeing index (0-100)  │   67.16      71.05          3.89        
#>  Body mass index                │   25.69      26.20          0.51        
#>  Satisfaction with health (1-5) │    3.51       3.59          0.08        
#> 
#>  Variable                       │ 95% CI LL  95% CI UL    p     R²    n   
#> ────────────────────────────────┼─────────────────────────────────────────
#>  WHO-5 wellbeing index (0-100)  │    2.13      5.64     <.001  0.02  1200 
#>  Body mass index                │    0.09      0.93      .018  0.00  1188 
#>  Satisfaction with health (1-5) │   -0.06      0.22      .267  0.00  1192 
#> 
#> Note. Missing values removed: bmi (12), life_sat_health (8).

For categorical predictors, the table reports estimated means by level. When the predictor is dichotomous, it can also show a single mean difference and confidence interval.

Robust standard errors

Use vcov = "HC*" when you want heteroskedasticity-consistent standard errors and tests:

table_continuous_lm(
  sochealth,
  select = c(wellbeing_score, bmi),
  by = sex,
  vcov = "HC3",
  statistic = TRUE
)
#> Continuous outcomes by Sex
#> 
#>  Variable                      │ M (Female)  M (Male)  Δ (Male - Female) 
#> ───────────────────────────────┼─────────────────────────────────────────
#>  WHO-5 wellbeing index (0-100) │   67.16      71.05          3.89        
#>  Body mass index               │   25.69      26.20          0.51        
#> 
#>  Variable                      │ 95% CI LL  95% CI UL   t      p     R²    n   
#> ───────────────────────────────┼───────────────────────────────────────────────
#>  WHO-5 wellbeing index (0-100) │   2.12       5.65     4.32  <.001  0.02  1200 
#>  Body mass index               │   0.09       0.93     2.38   .018  0.00  1188 
#> 
#> Note. Std. errors: heteroskedasticity-robust (HC3).
#> Missing values removed: bmi (12).

The HC* family is computed via sandwich::vcovHC() and includes "HC0", "HC1", "HC2", "HC3" (the sandwich::vcovHC() default, recommended for small to moderate samples by Long and Ervin 2000), "HC4", "HC4m", and "HC5". spicy’s own default remains vcov = "classical".

Cluster-robust standard errors

When observations are not independent (repeated measurements per individual, students nested in classes, panel data), classical and HC* standard errors are typically biased downward — the direction holds under the positive intra-cluster correlation such designs usually induce; with negative intra-cluster correlation the bias reverses. Use the CR* family together with cluster = id_var to get cluster-robust inference, dispatched to clubSandwich:

table_continuous_lm(
  sleep,
  select = extra,
  by = group,
  vcov = "CR2",
  cluster = ID,
  statistic = TRUE
)
#> Continuous outcomes by group
#> 
#>  Variable │ M (1)  M (2)  Δ (2 - 1)  95% CI LL  95% CI UL  t(9)   p     R²   n  
#> ──────────┼─────────────────────────────────────────────────────────────────────
#>  extra    │ 0.75   2.33     1.58       0.70       2.46     4.06  .003  0.16  20 
#> 
#> Note. Std. errors: cluster-robust (CR2), clusters by ID.

"CR2" is the recommended default (Bell & McCaffrey 2002; Pustejovsky & Tipton 2018). It generally produces fractional Satterthwaite degrees of freedom, rendered in the displayed test header as e.g. t(8.7) or F(2, 12.4); in the balanced sleep design above (10 subjects, 2 observations each) the Satterthwaite df is exactly 9, hence the t(9) header. "CR1" applies the G/(G − 1) small-sample correction only. Stata’s vce(cluster id) uses the larger G(N − 1)/((G − 1)(N − p)) factor with t(G − 1) inference — clubSandwich’s "CR1S", which spicy does not expose — so no CR* option here reproduces Stata’s cluster-robust output.

Bootstrap and jackknife

For situations where the residual distribution is non-standard or the sample is small, vcov = "bootstrap" and vcov = "jackknife" provide resampling-based variance estimators in pure base R (no dependency added):

set.seed(2026)
table_continuous_lm(
  sochealth,
  select = wellbeing_score,
  by = sex,
  vcov = "bootstrap",
  boot_n = 1000,  # default
  statistic = TRUE
)
#> Continuous outcomes by Sex
#> 
#>  Variable                      │ M (Female)  M (Male)  Δ (Male - Female) 
#> ───────────────────────────────┼─────────────────────────────────────────
#>  WHO-5 wellbeing index (0-100) │   67.16      71.05          3.89        
#> 
#>  Variable                      │ 95% CI LL  95% CI UL   z      p     R²    n   
#> ───────────────────────────────┼───────────────────────────────────────────────
#>  WHO-5 wellbeing index (0-100) │   2.12       5.66     4.30  <.001  0.02  1200 
#> 
#> Note. Std. errors: nonparametric bootstrap (1000 replicates).

When cluster is supplied, bootstrap switches to a cluster bootstrap (Cameron, Gelbach & Miller 2008) and jackknife to leave-one-cluster-out (Quenouille 1956; Tukey 1958). Both estimators use asymptotic inference: z for single-coefficient contrasts (the z header above, shown because statistic = TRUE) and chi^2(q) for the global Wald test of a categorical predictor with k > 2 levels (q = k - 1 restrictions).

Case weights

Use weights when you want weighted estimated means or slopes in the same model-based table:

table_continuous_lm(
  sochealth,
  select = c(wellbeing_score, bmi),
  by = education,
  weights = weight,
  show_weighted_n = TRUE
)
#> Continuous outcomes by Highest education level
#> 
#>  Variable                      │ M (Lower secondary)  M (Upper secondary) 
#> ───────────────────────────────┼──────────────────────────────────────────
#>  WHO-5 wellbeing index (0-100) │        57.71                68.39        
#>  Body mass index               │        27.85                25.79        
#> 
#>  Variable                      │ M (Tertiary)    p     R²    n    Weighted n 
#> ───────────────────────────────┼─────────────────────────────────────────────
#>  WHO-5 wellbeing index (0-100) │    76.55      <.001  0.19  1200   1196.47   
#>  Body mass index               │    24.23      <.001  0.13  1188   1183.32   
#> 
#> Note. Missing values removed: bmi (12).

This is often the most natural summary-table function when your reporting workflow already relies on weighted linear models.

Numeric predictors

If by is numeric, table_continuous_lm() reports the slope rather than group means:

table_continuous_lm(
  sochealth,
  select = c(wellbeing_score, bmi),
  by = age,
  vcov = "HC3"
)
#> Continuous outcomes by Age (years)
#> 
#>  Variable                      │  B    95% CI LL  95% CI UL    p     R²    n   
#> ───────────────────────────────┼───────────────────────────────────────────────
#>  WHO-5 wellbeing index (0-100) │ 0.04    -0.02      0.10      .176  0.00  1200 
#>  Body mass index               │ 0.04     0.02      0.05     <.001  0.02  1188 
#> 
#> Note. Std. errors: heteroskedasticity-robust (HC3).
#> Missing values removed: bmi (12).

When you need the underlying returned data for further processing, use output = "data.frame" for the wide raw table or output = "long" for the analytic long table.

Covariate adjustment

When the comparison of by-group means is potentially confounded by other variables, supply them through covariates. Each per-outcome linear model becomes lm(outcome ~ by + covariate1 + covariate2 + ...), and the reported Δ (categorical by) or slope (numeric by), its CI, p, R² and effect size are all adjusted for the covariates. The default ASCII print emits an APA-style footer naming the adjustment so the reader knows which estimand they are reading.

table_continuous_lm(
  sochealth,
  select = c(wellbeing_score, bmi),
  by = sex,
  covariates = c(age, education),
  vcov = "HC3"
)
#> Continuous outcomes by Sex
#> 
#>  Variable                      │ M (Female)  M (Male)  Δ (Male - Female) 
#> ───────────────────────────────┼─────────────────────────────────────────
#>  WHO-5 wellbeing index (0-100) │   67.27      70.93          3.65        
#>  Body mass index               │   25.65      26.23          0.58        
#> 
#>  Variable                      │ 95% CI LL  95% CI UL    p     R²    n   
#> ───────────────────────────────┼─────────────────────────────────────────
#>  WHO-5 wellbeing index (0-100) │   2.08       5.23     <.001  0.22  1200 
#>  Body mass index               │   0.19       0.97      .004  0.16  1188 
#> 
#> Note. Adjusted for age, education (proportional).
#> Std. errors: heteroskedasticity-robust (HC3).
#> Missing values removed: bmi (12).

covariates accepts the full tidyselect vocabulary (c(age, sex), tidyselect::starts_with("life_sat_"), tidyselect::all_of(cov_names)) or a literal character vector. A covariate that also appears in select is silently auto-excluded from the outcome list (mirroring how by / weights are excluded from select); a covariate equal to by is rejected up front.

Two estimands for the adjusted marginal means

Two methodologically distinct conventions exist for computing the covariate-adjusted per-group means displayed in the table (the M (level) columns of the printed output, and the emmean / emmean_se / emmean_ci_lower / emmean_ci_upper columns of the output = "long" data frame) when by is categorical. table_continuous_lm() exposes the choice via adjustment; both are valid but produce different numerical answers when at least one categorical covariate (factor, character, or logical) has non-uniform observed proportions.

• adjustment = "proportional" (the default) — G-computation / population-weighted standardisation. For each focal level of by, predictions are evaluated at every observation in the data with by replaced by that level (covariates kept at their observed values), then averaged. This is the convention of Stata’s margins and marginaleffects::avg_predictions(variables). Interpretation: “What is the predicted mean in this population if everyone had by = lvl, holding their actual covariate profile?”. Best for causal / counterfactual reporting.

• adjustment = "balanced" — equal-weight averaging on a synthetic grid. Categorical-covariate level combinations are crossed and weighted uniformly (1 / k); numeric covariates are fixed at their sample mean. Logical covariates count as categorical here: lm() encodes a logical as a treatment-contrast TRUE dummy, so the grid crosses FALSE / TRUE with equal weight. This is the convention of emmeans::emmeans() (default), SPSS UNIANOVA EMMEANS, and SAS LSMEANS. Interpretation: “What is the predicted mean assuming a balanced covariate design?”. Best for descriptive / ANOVA-style APA reporting.

The two methods coincide trivially when covariates is NULL, and also when all covariates are numeric: in a linear model, averaging predictions over the observed values of a numeric covariate equals predicting at its mean. They diverge whenever a categorical covariate — including a logical — has non-uniform observed proportions, because "proportional" weights its levels by their empirical frequencies while "balanced" weights them equally.

# Same model, balanced (emmeans-style) marginal means:
table_continuous_lm(
  sochealth,
  select = c(wellbeing_score, bmi),
  by = sex,
  covariates = c(age, education),
  adjustment = "balanced",
  vcov = "HC3"
)
#> Continuous outcomes by Sex
#> 
#>  Variable                      │ M (Female)  M (Male)  Δ (Male - Female) 
#> ───────────────────────────────┼─────────────────────────────────────────
#>  WHO-5 wellbeing index (0-100) │   65.90      69.56          3.65        
#>  Body mass index               │   25.89      26.46          0.58        
#> 
#>  Variable                      │ 95% CI LL  95% CI UL    p     R²    n   
#> ───────────────────────────────┼─────────────────────────────────────────
#>  WHO-5 wellbeing index (0-100) │   2.08       5.23     <.001  0.22  1200 
#>  Body mass index               │   0.19       0.97      .004  0.16  1188 
#> 
#> Note. Adjusted for age, education (balanced).
#> Std. errors: heteroskedasticity-robust (HC3).
#> Missing values removed: bmi (12).

Effect sizes under adjustment

Adding covariates changes which effect sizes remain meaningful:

• effect_size = "f2" and "omega2" are reported as partial f² / partial ω² via the partial F restricted to the focal term (the Type-II test of by after all covariates, equal to stats::drop1() in this additive setting) — the correct generalisation of the bivariate effect size to a covariate-adjusted model.
• effect_size = "d" and "g" raise spicy_unsupported: Cohen’s d and Hedges’ g have no defined extension under adjustment (the pooled SD has no canonical analogue). The error message points to f2 / omega2 as the partial-F generalisations.

Both effect-size variants and their noncentral CIs (when effect_size_ci = TRUE) are computed from the same fitted model as the displayed coefficients and remain invariant to vcov (HC* / CR* / bootstrap / jackknife only affect the contrast SE).

Scope: additive only in v1

covariates accepts additive terms only. Formula syntax with interactions or transforms (covariates = ~ age * sex, covariates = ~ I(age^2)) raises spicy_unsupported with a migration hint. Interactions and polynomials are reserved for a future spicy release. To fit a fully-specified model right now, do so with lm() directly and inspect via summary() or table_regression() (which accepts interaction / polynomial / I() terms natively).

Effect sizes

table_continuous_lm() supports four effect sizes via the effect_size argument. All are derived from the same fitted model as the displayed coefficients and R², so the table stays internally consistent — and all are invariant to vcov (the choice of classical or HC* standard errors changes the contrast SE, CI, and test statistic but not the standardized effect size itself).

Cohen’s d and Hedges’ g are the conventions for two-group comparisons (Cohen 1988; Hedges and Olkin 1985; APA 2020) and require by to have exactly two non-empty levels:

table_continuous_lm(
  sochealth,
  select = c(wellbeing_score, bmi),
  by = smoking,
  effect_size = "d"
)
#> Continuous outcomes by Current smoker
#> 
#>  Variable                      │ M (No)  M (Yes)  Δ (Yes - No)  95% CI LL 
#> ───────────────────────────────┼──────────────────────────────────────────
#>  WHO-5 wellbeing index (0-100) │ 69.36    67.65      -1.72        -3.91   
#>  Body mass index               │ 25.96    25.93      -0.03        -0.55   
#> 
#>  Variable                      │ 95% CI UL   p     R²     d     n   
#> ───────────────────────────────┼────────────────────────────────────
#>  WHO-5 wellbeing index (0-100) │   0.47     .124  0.00  -0.11  1175 
#>  Body mass index               │   0.49     .905  0.00  -0.01  1163 
#> 
#> Note. Missing values removed: bmi (12). Rows with missing smoking removed: 25.

Hedges’ g applies the small-sample correction J = 1 - 3/(4·df_resid - 1). It is generally preferred over raw d in published reports (Goulet-Pelletier and Cousineau 2018):

table_continuous_lm(
  sochealth,
  select = c(wellbeing_score, bmi),
  by = smoking,
  effect_size = "g"
)
#> Continuous outcomes by Current smoker
#> 
#>  Variable                      │ M (No)  M (Yes)  Δ (Yes - No)  95% CI LL 
#> ───────────────────────────────┼──────────────────────────────────────────
#>  WHO-5 wellbeing index (0-100) │ 69.36    67.65      -1.72        -3.91   
#>  Body mass index               │ 25.96    25.93      -0.03        -0.55   
#> 
#>  Variable                      │ 95% CI UL   p     R²     g     n   
#> ───────────────────────────────┼────────────────────────────────────
#>  WHO-5 wellbeing index (0-100) │   0.47     .124  0.00  -0.11  1175 
#>  Body mass index               │   0.49     .905  0.00  -0.01  1163 
#> 
#> Note. Missing values removed: bmi (12). Rows with missing smoking removed: 25.

For categorical predictors with three or more levels (or numeric predictors), use Hays’ omega² for a less biased estimate of the population variance explained (Hays 1963; Olejnik and Algina 2003; Lakens 2013):

table_continuous_lm(
  sochealth,
  select = c(wellbeing_score, bmi),
  by = education,
  effect_size = "omega2"
)
#> Continuous outcomes by Highest education level
#> 
#>  Variable                      │ M (Lower secondary)  M (Upper secondary) 
#> ───────────────────────────────┼──────────────────────────────────────────
#>  WHO-5 wellbeing index (0-100) │        57.22                68.97        
#>  Body mass index               │        28.09                26.02        
#> 
#>  Variable                      │ M (Tertiary)    p     R²    ω²    n   
#> ───────────────────────────────┼───────────────────────────────────────
#>  WHO-5 wellbeing index (0-100) │    76.85      <.001  0.21  0.21  1200 
#>  Body mass index               │    24.39      <.001  0.13  0.13  1188 
#> 
#> Note. Missing values removed: bmi (12).

Cohen’s f² (= R² / (1 - R²)) is the effect size familiar from power-analysis frameworks (e.g. G*Power) and is defined for any predictor type:

table_continuous_lm(
  sochealth,
  select = c(wellbeing_score, bmi),
  by = age,
  effect_size = "f2"
)
#> Continuous outcomes by Age (years)
#> 
#>  Variable                      │  B    95% CI LL  95% CI UL    p     R²    f²  
#> ───────────────────────────────┼───────────────────────────────────────────────
#>  WHO-5 wellbeing index (0-100) │ 0.04    -0.02      0.10      .177  0.00  0.00 
#>  Body mass index               │ 0.04     0.02      0.05     <.001  0.02  0.02 
#> 
#>  Variable                      │  n   
#> ───────────────────────────────┼──────
#>  WHO-5 wellbeing index (0-100) │ 1200 
#>  Body mass index               │ 1188 
#> 
#> Note. Missing values removed: bmi (12).

Confidence intervals for effect sizes

Setting effect_size_ci = TRUE adds a confidence interval for the effect size. The implementation uses the modern noncentral-distribution inversion approach — the consensus standard in commercial statistical software (Stata esize / estat esize, SAS PROC TTEST and PROC GLM EFFECTSIZE) and in mainstream R packages (effectsize, MOTE, TOSTER; effsize offers it as an option via noncentral = TRUE, its default being the central-t approximation):

• Noncentral t inversion for "d" and "g" (Steiger and Fouladi 1997; Goulet-Pelletier and Cousineau 2018; Fitts 2021), which has nominal coverage across sample sizes — unlike the older Hedges-Olkin normal approximation that is biased for small samples.
• Noncentral F inversion for "omega2" and "f2" (Steiger 2004; Smithson 2003).

In the printed and rendered outputs, the effect size is displayed with the bracketed CI (article-ready):

table_continuous_lm(
  sochealth,
  select = c(wellbeing_score, bmi),
  by = smoking,
  effect_size = "g",
  effect_size_ci = TRUE
)
#> Continuous outcomes by Current smoker
#> 
#>  Variable                      │ M (No)  M (Yes)  Δ (Yes - No)  95% CI LL 
#> ───────────────────────────────┼──────────────────────────────────────────
#>  WHO-5 wellbeing index (0-100) │ 69.36    67.65      -1.72        -3.91   
#>  Body mass index               │ 25.96    25.93      -0.03        -0.55   
#> 
#>  Variable                      │ 95% CI UL   p     R²            g          
#> ───────────────────────────────┼────────────────────────────────────────────
#>  WHO-5 wellbeing index (0-100) │   0.47     .124  0.00  -0.11 [-0.25, 0.03] 
#>  Body mass index               │   0.49     .905  0.00  -0.01 [-0.15, 0.13] 
#> 
#>  Variable                      │  n   
#> ───────────────────────────────┼──────
#>  WHO-5 wellbeing index (0-100) │ 1175 
#>  Body mass index               │ 1163 
#> 
#> Note. Missing values removed: bmi (12). Rows with missing smoking removed: 25.

The CI level follows ci_level (default 0.95). For programmatic access, the wide raw output exposes separate numeric columns, and the long output always includes them:

table_continuous_lm(
  sochealth,
  select = wellbeing_score,
  by = smoking,
  effect_size = "g",
  effect_size_ci = TRUE,
  output = "data.frame"
)
#>                                      Variable   M (No)  M (Yes) Δ (Yes - No)
#> wellbeing_score WHO-5 wellbeing index (0-100) 69.36317 67.64538    -1.717793
#>                 95% CI LL 95% CI UL         p          R²          g
#> wellbeing_score -3.906437 0.4708504 0.1238546 0.002017481 -0.1098571
#>                 es_ci_lower es_ci_upper    n
#> wellbeing_score  -0.2497291  0.03006164 1175

out <- table_continuous_lm(
  sochealth,
  select = wellbeing_score,
  by = smoking,
  effect_size = "g",
  effect_size_ci = TRUE,
  output = "long"
)
out[, c("variable", "es_type", "es_value", "es_ci_lower", "es_ci_upper")]
#>          variable es_type   es_value es_ci_lower es_ci_upper
#> 1 wellbeing_score       g -0.1098571  -0.2497291  0.03006164
#> 2 wellbeing_score    <NA>         NA          NA          NA

When weights is supplied, all four effect sizes (and their CIs) are computed from the weighted least-squares fit, keeping them consistent with the weighted contrast and its CI (DuMouchel and Duncan 1983). This is the right convention for case-weighted reporting; for propensity-score balance assessment (Austin and Stuart 2015) or complex-survey designs, dedicated packages (cobalt::bal.tab() and survey) are more appropriate.

Article-style polish

Pretty outcome labels and a comma decimal separator (useful for European reporting):

table_continuous_lm(
  sochealth,
  select = c(wellbeing_score, bmi),
  by = sex,
  labels = c(
    wellbeing_score = "WHO-5 wellbeing (0-100)",
    bmi = "Body-mass index (kg/m²)"
  ),
  effect_size = "g",
  effect_size_ci = TRUE,
  decimal_mark = ","
)
#> Continuous outcomes by Sex
#> 
#>  Variable                │ M (Female)  M (Male)  Δ (Male - Female)  95% CI LL 
#> ─────────────────────────┼────────────────────────────────────────────────────
#>  WHO-5 wellbeing (0-100) │   67,16      71,05          3,89           2,13    
#>  Body-mass index (kg/m²) │   25,69      26,20          0,51           0,09    
#> 
#>  Variable                │ 95% CI UL    p     R²           g           n   
#> ─────────────────────────┼─────────────────────────────────────────────────
#>  WHO-5 wellbeing (0-100) │   5,64     <,001  0,02  0,25 [0,14; 0,36]  1200 
#>  Body-mass index (kg/m²) │   0,93      ,018  0,00  0,14 [0,02; 0,25]  1188 
#> 
#> Note. Missing values removed: bmi (12).

Publication-ready output

The function supports the same output formats as the other summary-table helpers, including tinytable, gt, flextable, excel, word, and clipboard.

pkgdown_dark_gt(
  table_continuous_lm(
    sochealth,
    select = c(wellbeing_score, bmi, life_sat_health),
    by = sex,
    vcov = "HC3",
    statistic = TRUE,
    output = "gt"
  )
)

Continuous outcomes by Sex
Variable	M (Female)	M (Male)	Δ (Male - Female)	95% CI		t	p	R²	n
				LL	UL
WHO-5 wellbeing index (0-100)	67.16	71.05	3.89	2.12	5.65	4.32	<.001	0.02	1200
Body mass index	25.69	26.20	0.51	0.09	0.93	2.38	.018	0.00	1188
Satisfaction with health (1-5)	3.51	3.59	0.08	-0.06	0.22	1.11	.267	0.00	1192

Note. Std. errors: heteroskedasticity-robust (HC3). Missing values removed: bmi (12), life_sat_health (8).

Tidying for downstream pipelines

table_continuous_lm() returns an object that can be coerced to a plain data.frame / tbl_df (stripping the spicy formatting attributes) or piped into broom::tidy() / broom::glance() for any downstream tidyverse-stats workflow:

out <- table_continuous_lm(
  sochealth,
  select = c(wellbeing_score, bmi),
  by = sex,
  effect_size = "g",
  effect_size_ci = TRUE
)
#> Continuous outcomes by Sex
#> 
#>  Variable                      │ M (Female)  M (Male)  Δ (Male - Female) 
#> ───────────────────────────────┼─────────────────────────────────────────
#>  WHO-5 wellbeing index (0-100) │   67.16      71.05          3.89        
#>  Body mass index               │   25.69      26.20          0.51        
#> 
#>  Variable                      │ 95% CI LL  95% CI UL    p     R²  
#> ───────────────────────────────┼───────────────────────────────────
#>  WHO-5 wellbeing index (0-100) │   2.13       5.64     <.001  0.02 
#>  Body mass index               │   0.09       0.93      .018  0.00 
#> 
#>  Variable                      │         g           n   
#> ───────────────────────────────┼─────────────────────────
#>  WHO-5 wellbeing index (0-100) │ 0.25 [0.14, 0.36]  1200 
#>  Body mass index               │ 0.14 [0.02, 0.25]  1188 
#> 
#> Note. Missing values removed: bmi (12).

# One row per estimated parameter: emmean per level, contrast for
# binary predictors, slope for numeric predictors.
broom::tidy(out)
#> # A tibble: 6 × 10
#>   outcome        label term  estimate_type estimate std.error conf.low conf.high
#>   <chr>          <chr> <chr> <chr>            <dbl>     <dbl>    <dbl>     <dbl>
#> 1 wellbeing_sco… WHO-… Fema… emmean          67.2       0.623  65.9       68.4  
#> 2 wellbeing_sco… WHO-… Male  emmean          71.0       0.644  69.8       72.3  
#> 3 wellbeing_sco… WHO-… Male… difference       3.89      0.896   2.13       5.64 
#> 4 bmi            Body… Fema… emmean          25.7       0.150  25.4       26.0  
#> 5 bmi            Body… Male  emmean          26.2       0.155  25.9       26.5  
#> 6 bmi            Body… Male… difference       0.512     0.216   0.0888     0.935
#> # ℹ 2 more variables: statistic <dbl>, p.value <dbl>

# One row per outcome with model-level statistics: r.squared,
# adj.r.squared, F / t, df, p.value, nobs, weighted_n, plus the
# effect-size summary.
broom::glance(out)
#> # A tibble: 2 × 16
#>   outcome     label predictor_type test_type statistic    df df.residual p.value
#>   <chr>       <chr> <chr>          <chr>         <dbl> <int>       <dbl>   <dbl>
#> 1 wellbeing_… WHO-… categorical    F             18.8      1        1198 1.55e-5
#> 2 bmi         Body… categorical    F              5.64     1        1186 1.78e-2
#> # ℹ 8 more variables: r.squared <dbl>, adj.r.squared <dbl>, es_type <chr>,
#> #   es_value <dbl>, es_ci_lower <dbl>, es_ci_upper <dbl>, nobs <int>,
#> #   weighted_n <dbl>

See also

• See vignette("table-continuous", package = "spicy") for descriptive continuous summary tables with classical group-comparison tests.
• See vignette("table-regression", package = "spicy") for the full coefficient table when the analysis moves from a per-outcome one-predictor model to a single multi-predictor lm / glm, including interactions, polynomials, and standardised coefficients.
• See vignette("summary-tables-reporting", package = "spicy") for a cross-function reporting workflow using the summary-table helpers along the APA Table 1 / 2 / 3 sequence.

References

• American Psychological Association (2020). Publication Manual of the American Psychological Association (7th ed.).
• Austin, P. C., & Stuart, E. A. (2015). Moving towards best practice when using inverse probability of treatment weighting (IPTW) using the propensity score to estimate causal treatment effects in observational studies. Statistics in Medicine, 34(28), 3661–3679.
• Bell, R. M., & McCaffrey, D. F. (2002). Bias reduction in standard errors for linear regression with multi-stage samples. Survey Methodology, 28(2), 169–181.
• Cameron, A. C., Gelbach, J. B., & Miller, D. L. (2008). Bootstrap-based improvements for inference with clustered errors. Review of Economics and Statistics, 90(3), 414–427.
• Cohen, J. (1988). Statistical Power Analysis for the Behavioral Sciences (2nd ed.). Lawrence Erlbaum.
• DuMouchel, W. H., & Duncan, G. J. (1983). Using sample survey weights in multiple regression analyses of stratified samples. Journal of the American Statistical Association, 78(383), 535–543.
• Fitts, D. A. (2021). Expected and empirical coverages of different methods for generating noncentral t confidence intervals for a standardized mean difference. Behavior Research Methods, 53(6), 2412–2429.
• Goulet-Pelletier, J.-C., & Cousineau, D. (2018). A review of effect sizes and their confidence intervals, Part I: The Cohen’s d family. The Quantitative Methods for Psychology, 14(4), 242–265.
• Hays, W. L. (1963). Statistics for Psychologists. Holt, Rinehart and Winston.
• Hedges, L. V., & Olkin, I. (1985). Statistical Methods for Meta-Analysis. Academic Press.
• Lakens, D. (2013). Calculating and reporting effect sizes to facilitate cumulative science: A practical primer for t-tests and ANOVAs. Frontiers in Psychology, 4, 863.
• Long, J. S., & Ervin, L. H. (2000). Using heteroscedasticity consistent standard errors in the linear regression model. The American Statistician, 54(3), 217–224.
• Olejnik, S., & Algina, J. (2003). Generalized eta and omega squared statistics: Measures of effect size for some common research designs. Psychological Methods, 8(4), 434–447.
• Pustejovsky, J. E., & Tipton, E. (2018). Small-sample methods for cluster-robust variance estimation and hypothesis testing in fixed effects models. Journal of Business & Economic Statistics, 36(4), 672–683.
• Quenouille, M. H. (1956). Notes on bias in estimation. Biometrika, 43(3/4), 353–360.
• Smithson, M. (2003). Confidence Intervals. Sage.
• Steiger, J. H. (2004). Beyond the F test: Effect size confidence intervals and tests of close fit in the analysis of variance and contrast analysis. Psychological Methods, 9(2), 164–182.
• Steiger, J. H., & Fouladi, R. T. (1997). Noncentrality interval estimation and the evaluation of statistical models. In L. L. Harlow, S. A. Mulaik, & J. H. Steiger (Eds.), What if there were no significance tests? (pp. 221–257). Lawrence Erlbaum.
• Tukey, J. W. (1958). Bias and confidence in not-quite large samples (Abstract). The Annals of Mathematical Statistics, 29(2), 614.