Mean Center and Standardize Selected Variable by std_selected()

In stdmod: Standardized Moderation Effect and Its Confidence Interval

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>",
  fig.width  = 6,
  fig.height = 4,
  fig.align = "center"
)

Purpose

Instead of standardizing all variables, even variables that (a) are categorical and should not be standardized, or (b) measured on meaningful unites and do not need to be standardized, std_selected() from the package stdmod allows users to have more control on how standardization is to be conducted.

A moderated regression model is used as an example but it can also be used for regression models without interaction terms.

More about this package can be found in vignette("stdmod", package = "stdmod") or at https://sfcheung.github.io/stdmod/.

Setup the Environment

library(stdmod)

Load the Dataset

data(sleep_emo_con)
head(sleep_emo_con, 3)

This data set has 500 cases of data. The variables are sleep duration, age, gender, and the scores from two personality scales, emotional stability and conscientiousness of the IPIP Big Five markers. Please refer to (citation to be included) for the detail of the data set.

The names of some variables are shortened for readability:

colnames(sleep_emo_con)[3:4] <- c("cons", "emot")
head(sleep_emo_con, 3)

Moderated Regression

Suppose we are interested in predicting sleep duration by emotional stability, after controlling for gender and age. However, we suspect that the effect of emotional stability, if any, may be moderated by conscientiousness. Therefore, we conduct a moderated regression as follow:

lm_raw <- lm(sleep_duration ~ age + gender + emot * cons,
             data = sleep_emo_con)
summary(lm_raw)

The results show that conscientiousness significantly moderates the effect of emotional stability on sleep duration.

This package has a simple function, plotmod(), for generating a typical plot of the moderation effect:

plotmod(lm_raw,
        x = "emot",
        w = "cons",
        x_label = "Emotional Stability",
        w_label = "Conscientiousness",
        y_label = "Sleep Duration")

The function plotmod() also prints the conditional effects of the predictor, emotional stability in this example.

Mean Center the Moderator

To know the effect of emotional stability when conscientiousness is equal to its mean, we can center conscientiousness by its mean in the data and redo the moderated regression. Instead of creating the new variable and rerun the regression, we can pass the lm() output to std_selected() and specify the variables to be mean centered:

lm_w_centered <- std_selected(lm_raw,
                              to_center = ~ cons)
printCoefmat(summary(lm_w_centered)$coefficients, digits = 3)

The argument for meaning centering is to_center. The variable is specified in the formula form, placing them on the right hand side of the formula.

In this example, when conscientiousness is at mean level, the effect of emotional stability is r formatC(coef(lm_w_centered)["emot"], 4, format = "f").

Mean Center The Moderator and the Focal Variable

This example demonstrates centering more than one variable. In the following model, both emotional stability and conscientiousness are centered. They are placed after ~ and joined by +.

lm_xw_centered <- std_selected(lm_raw,
                               to_center = ~ emot + cons)
printCoefmat(summary(lm_xw_centered)$coefficients, digits = 3)

Standardize The Moderator and The Focal Variable

To standardize a variable we first mean center it and then scale it by its standard deviation. Scaling is done by listing the variable on to_scale. The input format is identical to that of to_center.

lm_xw_std <- std_selected(lm_raw,
                          to_center = ~ emot + cons,
                          to_scale  = ~ emot + cons)

Since 0.2.6.3 of stdmod, to_standardize can be used as a shortcut. Listing a variable on to_standardize is equivalent to listing it on both to_center and to_scale. Therefore, the following call can also be used:

lm_xw_std <- std_selected(lm_raw,
                          to_standardize = ~ emot + cons)

printCoefmat(summary(lm_xw_std)$coefficients, digits = 3)

In this example, when conscientiousness is at mean level, for each one standard deviation increase of emotional stability, the predicted sleep duration increases by r formatC(coef(lm_xw_std)["emot"], 4, format = "f") hour.

plotmod(lm_xw_std,
        x = "emot",
        w = "cons",
        x_label = "Emotional Stability",
        w_label = "Conscientiousness",
        y_label = "Sleep Duration")

The function plotmod() automatically checks whether a variable is standardized. If yes, it will report this in the plot as table note on the bottom.

The pattern of the plot does not change. However, the conditional effects reported in the graph are now based on the model with emotional stability and conscientiousness standardized.

Standardize The Moderator, The Focal Variable, and the Outcome Variable

We can also mean center or standardize the outcome variable (dependent variable). We just add the variable to the right hand side of ~ in to_center and to_scale as appropriate.

lm_xwy_std <- std_selected(lm_raw,
                           to_center = ~ emot + cons + sleep_duration,
                           to_scale  = ~ emot + cons + sleep_duration)

Since 0.2.6.3, to_standardize can be used as a shortcut:

lm_xwy_std <- std_selected(lm_raw,
                           to_standardize = ~ emot + cons + sleep_duration)
printCoefmat(summary(lm_xwy_std)$coefficients, digits = 3)

In this example, when conscientiousness is at mean level, the standardized moderation effect of emotional stability on sleep duration is r formatC(coef(lm_xwy_std)["emot"], 4, format = "f").

plotmod(lm_xwy_std,
        x = "emot",
        w = "cons",
        x_label = "Emotional Stability",
        w_label = "Conscientiousness",
        y_label = "Sleep Duration")

Again, the pattern of the plot does not change, but the conditional effects reported in the graph are now based on the model with emotional stability, conscientiousness, and sleep duration standardized.

Standardize All Variables

If we want to standardize all variables except for categorical variables, if any, we can use ~ . as a shortcut. std_selected() will automatically skip categorical variables (i.e., factors or string variables in the regression model of lm()).

lm_all_std <- std_selected(lm_raw,
                           to_center = ~ .,
                           to_scale  = ~ .)

Since 0.2.6.3, to_standardize can be used as a shortcut:

lm_all_std <- std_selected(lm_raw,
                           to_standardize = ~ .)
printCoefmat(summary(lm_all_std)$coefficients, digits = 3)

The Usual Standardized Solution

For comparison, this is the results of standardizing all variables, including the product term and the categorical variable.

library(lm.beta) # For generating the typical standardized solution
packageVersion("lm.beta")
lm_usual_std <- lm.beta(lm_raw)
printCoefmat(summary(lm_usual_std)$coefficients, digits = 3)

In moderated regression, the coefficient of standardized product term, r formatC(coef(lm_usual_std)["emot:cons"], 4, format = "f"), is not interpretable. The coefficient of standardized gender, r formatC(coef(lm_usual_std)["gendermale"], 4, format = "f"), is also difficult to interpret.

Improved Confidence Interval For "Betas"

It has been shown (e.g., Yuan & Chan, 2011) that the standard errors of standardized regression coefficients (betas) computed just by standardizing the variables are biased, and consequently the confidence intervals are also invalid. The function std_selected_boot() is a wrapper of std_selected() that also forms the confidence interval of the regression coefficients when standardization is conducted, using nonparametric bootstrapping as suggested by Cheung, Cheung, Lau, Hui, and Vong (2022).

We use the same example above that standardizes emotional stability, conscientiousness, and sleep duration, to illustrate this function. The argument nboot specifies the number of nonparametric bootstrap samples. The level of confidence set by conf. The default is .95, denoting 95% confidence intervals. If this is the desired level, this argument can be omitted.

if (file.exists("eg_lm_xwy_std_ci.rds")) {
    lm_xwy_std_ci <- readRDS("eg_lm_xwy_std_ci.rds")
  } else {
    set.seed(58702)
    lm_xwy_std_ci <- std_selected_boot(lm_raw,
        to_center = ~ emot + cons + sleep_duration,
        to_scale  = ~ emot + cons + sleep_duration,
        nboot = 2000)
    saveRDS(lm_xwy_std_ci, "eg_lm_xwy_std_ci.rds", compress = "xz")
  }

set.seed(58702)
lm_xwy_std_ci <- std_selected_boot(lm_raw,
                    to_standardize = ~ emot + cons + sleep_duration,
                    nboot = 2000)

summary(lm_xwy_std_ci)

tmp <- summary(lm_xwy_std_ci)$coefficients

The standardized moderation effect is r formatC(tmp["emot:cons", "Estimate"], 4, format = "f") , and the 95% nonparametric bootstrap percentile confidence interval is r formatC(tmp["emot:cons", "CI Lower"], 4, format = "f") to r formatC(tmp["emot:cons", "CI Upper"], 4, format = "f").

Note: As a side product, the nonparametric bootstrap confidence of the other coefficients are also reported. They can be used for other variables that are standardized in the same model, whether they are involved in the moderation or not.

Further Information

Further information on the functions can be found in their help pages (std_selected() and std_selected_boot()). For example, parallel computation can be used when doing bootstrapping, if the number of bootstrapping samples requested is large.

Reference

Cheung, S. F., Cheung, S.-H., Lau, E. Y. Y., Hui, C. H., & Vong, W. N. (2022) Improving an old way to measure moderation effect in standardized units. Health Psychology, 41(7), 502-505. https://doi.org/10.1037/hea0001188.

Yuan, K.-H., & Chan, W. (2011). Biases and standard errors of standardized regression coefficients. Psychometrika, 76(4), 670-690. https://doi.org/10.1007/s11336-011-9224-6

stdmod documentation built on Sept. 30, 2024, 9:42 a.m.