betaMC: Monte Carlo for Regression Effect Sizes

Ivan Jacob Agaloos Pesigan

Description

Generates Monte Carlo confidence intervals for standardized regression coefficients (beta) and other effect sizes, including multiple correlation, semipartial correlations, improvement in R-squared, squared partial correlations, and differences in standardized regression coefficients, for models fitted by lm(). betaMC combines ideas from Monte Carlo confidence intervals for the indirect effect (Pesigan and Cheung, 2023: http://doi.org/10.3758/s13428-023-02114-4) and the sampling covariance matrix of regression coefficients (Dudgeon, 2017: http://doi.org/10.1007/s11336-017-9563-z) to generate confidence intervals effect sizes in regression.

Installation

You can install the CRAN release of betaMC with:

install.packages("betaMC")

You can install the development version of betaMC from GitHub with:

if (!require("remotes")) install.packages("remotes")
remotes::install_github("jeksterslab/betaMC")

Example

In this example, a multiple regression model is fitted using program quality ratings (QUALITY) as the regressand/outcome variable and number of published articles attributed to the program faculty members (NARTIC), percent of faculty members holding research grants (PCTGRT), and percentage of program graduates who received support (PCTSUPP) as regressor/predictor variables using a data set from 1982 ratings of 46 doctoral programs in psychology in the USA (National Research Council, 1982). Confidence intervals for the standardized regression coefficients are generated using the BetaMC() function from the betaMC package.

library(betaMC)

df <- betaMC::nas1982

Regression

Fit the regression model using the lm() function.

object <- lm(QUALITY ~ NARTIC + PCTGRT + PCTSUPP, data = df)

Monte Carlo Sampling Distribution of Parameters

Normal-Theory Approach

mvn <- MC(object, type = "mvn")

Asymptotic distribution-free Approach

adf <- MC(object, type = "adf")

Heteroskedasticity Consistent Approach (HC3)

hc3 <- MC(object, type = "hc3")

Standardized Regression Slopes

Normal-Theory Approach

BetaMC(mvn, alpha = 0.05)
#> Call:
#> BetaMC(object = mvn, alpha = 0.05)
#> 
#> Standardized regression slopes
#> type = "mvn"
#>            est     se     R   2.5%  97.5%
#> NARTIC  0.4951 0.0759 20000 0.3381 0.6340
#> PCTGRT  0.3915 0.0772 20000 0.2390 0.5411
#> PCTSUPP 0.2632 0.0743 20000 0.1174 0.4099

Asymptotic distribution-free Approach

BetaMC(adf, alpha = 0.05)
#> Call:
#> BetaMC(object = adf, alpha = 0.05)
#> 
#> Standardized regression slopes
#> type = "adf"
#>            est     se     R   2.5%  97.5%
#> NARTIC  0.4951 0.0676 20000 0.3513 0.6142
#> PCTGRT  0.3915 0.0707 20000 0.2428 0.5208
#> PCTSUPP 0.2632 0.0772 20000 0.1041 0.4102

Heteroskedasticity Consistent Approach (HC3)

BetaMC(hc3, alpha = 0.05)
#> Call:
#> BetaMC(object = hc3, alpha = 0.05)
#> 
#> Standardized regression slopes
#> type = "hc3"
#>            est     se     R   2.5%  97.5%
#> NARTIC  0.4951 0.0796 20000 0.3251 0.6375
#> PCTGRT  0.3915 0.0820 20000 0.2191 0.5388
#> PCTSUPP 0.2632 0.0856 20000 0.0904 0.4277

Other Effect Sizes

The betaMC package also has functions to generate Monte Carlo confidence intervals for other effect sizes such as RSqMC() for multiple correlation coefficients (R-squared and adjusted R-squared), DeltaRSqMC() for improvement in R-squared, SCorMC() for semipartial correlation coefficients, PCorMC() for squared partial correlation coefficients, and DiffBetaMC() for differences of standardized regression coefficients.

Multiple Correlation Coefficients (R-squared and adjusted R-squared)

RSqMC(hc3, alpha = 0.05)
#> Call:
#> RSqMC(object = hc3, alpha = 0.05)
#> 
#> R-squared and adjusted R-squared
#> type = "hc3"
#>        est     se     R   2.5%  97.5%
#> rsq 0.8045 0.0620 20000 0.6466 0.8883
#> adj 0.7906 0.0664 20000 0.6214 0.8803

Improvement in R-squared

DeltaRSqMC(hc3, alpha = 0.05)
#> Call:
#> DeltaRSqMC(object = hc3, alpha = 0.05)
#> 
#> Improvement in R-squared
#> type = "hc3"
#>            est     se     R   2.5%  97.5%
#> NARTIC  0.1859 0.0694 20000 0.0499 0.3235
#> PCTGRT  0.1177 0.0543 20000 0.0250 0.2349
#> PCTSUPP 0.0569 0.0378 20000 0.0062 0.1509

Semipartial Correlation Coefficients

SCorMC(hc3, alpha = 0.05)
#> Call:
#> SCorMC(object = hc3, alpha = 0.05)
#> 
#> Semipartial correlations
#> type = "hc3"
#>            est     se     R   2.5%  97.5%
#> NARTIC  0.4312 0.0872 20000 0.2235 0.5688
#> PCTGRT  0.3430 0.0830 20000 0.1583 0.4846
#> PCTSUPP 0.2385 0.0784 20000 0.0784 0.3885

Squared Partial Correlation Coefficients

PCorMC(hc3, alpha = 0.05)
#> Call:
#> PCorMC(object = hc3, alpha = 0.05)
#> 
#> Squared partial correlations
#> type = "hc3"
#>            est     se     R   2.5%  97.5%
#> NARTIC  0.4874 0.1197 20000 0.1751 0.6493
#> PCTGRT  0.3757 0.1155 20000 0.1043 0.5530
#> PCTSUPP 0.2254 0.1126 20000 0.0254 0.4569

Differences of Standardized Regression Coefficients

DiffBetaMC(hc3, alpha = 0.05)
#> Call:
#> DiffBetaMC(object = hc3, alpha = 0.05)
#> 
#> Differences of standardized regression slopes
#> type = "hc3"
#>                   est     se     R    2.5%  97.5%
#> NARTIC-PCTGRT  0.1037 0.1418 20000 -0.1736 0.3791
#> NARTIC-PCTSUPP 0.2319 0.1331 20000 -0.0378 0.4826
#> PCTGRT-PCTSUPP 0.1282 0.1372 20000 -0.1464 0.3871

Documentation

See GitHub Pages for package documentation.

Citation

To cite betaMC in publications, please cite Pesigan & Cheung (2023).

References

Dudgeon, P. (2017). Some improvements in confidence intervals for standardized regression coefficients. Psychometrika, 82(4), 928–951. https://doi.org/10.1007/s11336-017-9563-z

National Research Council. (1982). An assessment of research-doctorate programs in the United States: Social and behavioral sciences. National Academies Press. https://doi.org/10.17226/9781

Pesigan, I. J. A., & Cheung, S. F. (2023). Monte Carlo confidence intervals for the indirect effect with missing data. Behavior Research Methods, 56(3), 1678–1696. https://doi.org/10.3758/s13428-023-02114-4