
Methodology-Adjusted Substantive Meta-Regression (MASMR)
masmr.RdFits a methodology-only baseline and a joint meta-regression of the original outcome on methodology and substantive moderators. Substantive associations are estimated in the joint model, conditional on all included predictors.
Usage
masmr(
data,
methodology_mods,
substantive_mods,
formula = NULL,
studyID,
effectID = NULL,
sample_size = NULL,
effectsize_type = NULL,
effectsize_name = NULL,
estimation_method = "REML",
variance = NULL,
varcov_type,
weights = NULL,
structure = "univariate",
intercept = FALSE,
missing = "remove",
optim_method = "L-BFGS-B",
robustID = NULL,
robust_args = list(),
lasso = FALSE,
lasso_args = list(lambda_grid = 10^seq(1, -3, length.out = 5), K = 5, all_lasso_metrics
= FALSE, lambda_tolerance = 0),
tol = 1e+07,
adjusted_effect_name = "yi_masmr",
interactions = NULL,
outcome = NULL,
measure = NULL,
units = NULL,
sampling_covariance = NULL,
...
)
# S3 method for class 'masmr'
summary(object, fit_measures = TRUE, ...)
# S3 method for class 'summary.masmr'
print(x, ...)Arguments
- data
Data frame used for analysis.
- methodology_mods
Character vector of methodology moderator terms.
- substantive_mods
Character vector of substantive moderator terms.
- formula
Optional formula input for univariate/multilevel analysis.
- studyID
Study identifier column name.
- effectID
Effect identifier column name (used for SMD-style workflow).
- sample_size
Sample-size column name (used for SMD-style workflow).
- effectsize_type
The non-formula workflow supports "smd". For other measures use formula or outcome-aware precomputed inputs.
- effectsize_name
Effect-size column name.
- estimation_method
Estimation method passed to
mars.- variance
Variance column name.
- varcov_type
Variance-covariance type passed to
mars.- weights
Optional user-defined weights.
- structure
Model structure passed to
mars.- intercept
Intercept flag passed to
mars.- missing
Missing-data mode:
"remove","keep", or"em". EM completes continuous source predictors once across both moderator blocks (and scale predictors), then uses that same completed data in both models. Interactions and transformations are rebuilt afterward. Standard errors condition on the single completion; imputation uncertainty is not propagated. See Details for restrictions and diagnostics.- optim_method
Optimization method passed to
mars.- robustID
Optional robust clustering ID passed to
mars.- robust_args
Named list of robust covariance and inference controls passed to both MARS stages.
- lasso
Must be FALSE; penalized inference is not supported.
- lasso_args
Unused because LASSO is not supported in MASMR.
- tol
Optimization tolerance passed to
mars.- adjusted_effect_name
Unused; MASMR fits the original outcome jointly. Supplying this argument produces a warning.
- interactions
Optional explicit substantive-by-methodology interaction terms, for example
"weeks:tester". These belong only to the joint model.- outcome
Optional column naming outcomes for outcome-aware multivariate fits. Uses outcome-specific intercepts and moderator coefficients.
- measure
Analysis measure code, or a named code per outcome. OR, RR and HR denote already log-transformed estimates; ZCOR denotes Fisher z. No transformation is automatic. See
masmr_sensitivity().- units
Optional units, named per outcome; required for MD.
- sampling_covariance
Positive-definite sampling covariance in original input row order. Mutually exclusive with weights; heterogeneity is separate.
- ...
Additional arguments passed to printing methods for the stage summaries.
- object
An object of class
"masmr".- fit_measures
Logical; if
TRUE, request fit measures from stage summaries when available.- x
An object of class
"summary.masmr".
Value
An object of class "masmr" containing:
- baseline, joint
Methodology-only and joint fitted MARS models.
- stage1, stage2
Additional names for baseline and joint.
- analysis_data
Original-outcome analysis data shared by both models.
- missing_data
Missing-data method and EM diagnostics, including original and completed predictors and the missingness mask in original input order. With EM, both fitted models carry the same diagnostics.
- substantive_coefficients
Joint substantive coefficients with model-based normal 95 percent intervals and two-sided p-values.
- substantive_test
Joint model-based Wald chi-square test of all substantive design columns, conditional on methodology.
- workflow
One of
"formula","multivariate", or"smd".
Details
Use masmr_contrast() to compare substantive scenarios under an explicit
methodological profile or shared distribution. Target weights define the
estimand and are distinct from fitting weights. masmr_support() reports
support diagnostics. Existing coefficient summaries retain model-based normal
inference; contrast inference is selected explicitly in the contrast call.
Both models use the
original outcome and sampling covariance; no stage-1 residuals are refitted.
The moderator lists define the fixed predictors; formula supplies the outcome,
intercept convention and random terms. Terms may not overlap between lists.
The substantive table and test use model-based covariance, not robust
covariance. Robust results, when requested, remain available in the joint fit.
Inference is asymptotic and inherits the assumptions of MARS. Use ML, not REML
likelihood comparisons, when comparing different fixed predictor sets.
EM uses the shared MARS multivariate-normal completion helper once before
fitting. Missing predictors must be continuous; incomplete categorical
predictors are rejected. Repeated studies require study-level invariant
predictors. Missing outcomes and variances are not imputed. Complete scale
predictors and moderator variables condition the completion; outcome values
and additional auxiliary variables are not automatically included.
Completed source predictors are held fixed during sensitivity refits.
Support diagnostics describe completed-data geometry, not purely observed
support. EM convergence does not establish the missingness assumptions or
account for imputation uncertainty. Use external proper multiple imputation
and appropriate pooling for uncertainty-aware missing-data inference.
LASSO is not supported. SMD user weights and transformed calls
inside random terms are rejected explicitly. Create transformed random-slope
columns in the data before fitting.
Examples
out <- masmr(
data = teacher_expectancy,
methodology_mods = c("factor(tester)"),
substantive_mods = c("year", "weeks", "factor(setting)"),
formula = yi ~ 1 + year + weeks + factor(setting) + factor(tester),
studyID = "study",
variance = "vi",
varcov_type = "univariate",
structure = "univariate"
)
summary(out)
#> Methodology-Adjusted Substantive Meta-Regression (MASMR)
#> Workflow: formula
#> Methodology moderators: factor(tester)
#> Substantive moderators: year, weeks, factor(setting)
#>
#> Baseline (Methodology Only)
#> Results generated with MARS:v 0.6.0.8
#> Thursday, November 5, 2026
#>
#> Model Type:
#> univariate
#>
#> Estimation Method:
#> Restricted Maximum Likelihood
#>
#> Coefficient covariance: gls
#> Optimizer convergence code: 0
#>
#>
#> Model Formula:
#> yi ~ 1 + factor(tester)
#> <environment: 0x653fddd02148>
#>
#> Data Summary:
#> Number of Effect Sizes: 19
#> Number of Fixed Effects: 2
#> Number of Random Effects: 1
#>
#> Random Components:
#> term var SD
#> intercept 0.02476 0.1574
#>
#> Fixed Effects Estimates:
#> attribute estimate SE z_test p_value lower upper
#> (Intercept) 0.04635 0.07187 0.6448 0.5190 -0.09452 0.1872
#> blind 0.10248 0.11209 0.9142 0.3606 -0.11722 0.3222
#>
#> Model Fit Statistics:
#> logLik Dev AIC BIC AICc
#> -3.838 7.675 13.68 16.17 23.86
#>
#> Q Error: 35.157 (17), p = 0.0059
#>
#> I2 (General): 47.52336
#>
#> Residual Diagnostics:
#> Diagnostic Value
#> Observations 19
#> Finite raw residuals 19
#> Mean raw residual 0.06879
#> SD of raw residuals 0.3547
#> Root mean squared residual 0.352
#> Mean absolute residual 0.2418
#> Sum of squared Pearson residuals 22.86
#> Mean |studentized residual| 0.907
#> Maximum |studentized residual| 2.846
#> Proportion |studentized residual| > 2 0.1053
#> Proportion |studentized residual| > 3 0
#>
#> Normality screen (Shapiro-Wilk, whitened residuals):
#> N W p-value
#> 19 0.9141 0.08801
#>
#> Heteroscedasticity screen (|raw residual| vs fitted):
#> N Correlation Slope p-value
#> 19 -0.1174 -0.5868 0.6322
#>
#> Joint Model (Methodology and Substantive Predictors)
#> Results generated with MARS:v 0.6.0.8
#> Thursday, November 5, 2026
#>
#> Model Type:
#> univariate
#>
#> Estimation Method:
#> Restricted Maximum Likelihood
#>
#> Coefficient covariance: gls
#> Optimizer convergence code: 0
#>
#>
#> Model Formula:
#> yi ~ 1 + factor(tester) + year + weeks + factor(setting)
#> <environment: 0x653fddd02148>
#>
#> Data Summary:
#> Number of Effect Sizes: 19
#> Number of Fixed Effects: 5
#> Number of Random Effects: 1
#>
#> Random Components:
#> term var SD
#> intercept 0.01109 0.1053
#>
#> Fixed Effects Estimates:
#> attribute estimate SE z_test p_value lower upper
#> (Intercept) 35.888447 51.292888 0.69968 0.4841 -64.64377 136.420660
#> blind 0.001689 0.101523 0.01664 0.9867 -0.19729 0.200670
#> year -0.018133 0.026036 -0.69645 0.4861 -0.06916 0.032897
#> weeks -0.014011 0.005933 -2.36151 0.0182 -0.02564 -0.002382
#> indiv 0.226317 0.190726 1.18661 0.2354 -0.14750 0.600132
#>
#> Model Fit Statistics:
#> logLik Dev AIC BIC AICc
#> -1.499 2.998 15 18.83 87.8
#>
#> Q Error: 24.227 (14), p = 0.043
#>
#> I2 (General): 27.96069
#>
#> Residual Diagnostics:
#> Diagnostic Value
#> Observations 19
#> Finite raw residuals 19
#> Mean raw residual 0.05269
#> SD of raw residuals 0.3123
#> Root mean squared residual 0.3085
#> Mean absolute residual 0.2049
#> Sum of squared Pearson residuals 19.12
#> Mean |studentized residual| 0.9164
#> Maximum |studentized residual| 2.794
#> Proportion |studentized residual| > 2 0.05263
#> Proportion |studentized residual| > 3 0
#>
#> Normality screen (Shapiro-Wilk, whitened residuals):
#> N W p-value
#> 19 0.9416 0.2817
#>
#> Heteroscedasticity screen (|raw residual| vs fitted):
#> N Correlation Slope p-value
#> 19 0.2794 0.398 0.2467
#>
#> Substantive associations (model-based 95% intervals)
#> term estimate std_error conf_low conf_high
#> year -0.01813305 0.026036384 -0.06916343 0.032897323
#> weeks -0.01401058 0.005932892 -0.02563883 -0.002382321
#> factor(setting)indiv 0.22631690 0.190725607 -0.14749842 0.600132217
#> p_value
#> 0.48614676
#> 0.01820075
#> 0.23538152
#> $statistic
#> [1] 8.784025
#>
#> $df
#> [1] 3
#>
#> $p_value
#> [1] 0.03230458
#>
#> $method
#> [1] "Model-based joint Wald chi-square test of substantive coefficients"
#>