Mindfulness RCT

Paper:

Scarabusci, R., & Moneta, G. B. (submitted). Mindfulness and affect regulation: A pilot randomized trial in financially anxious individuals.

Abstract:

Financial anxiety has emerged as a significant psychological burden, yet no randomized controlled trials (RCTs) to date have evaluated mindfulness-based interventions in this population. This pilot RCT is the first to explore the feasibility and preliminary effects of mindfulness training for financially anxious individuals. Thirty-one participants were randomly assigned to either a mindfulness group or a waitlist control. Positive and negative affect were assessed pre- and post-intervention using the Positive and Negative Affect Schedule (PANAS). To account for emotional interdependence, data were analyzed with a multivariate Bayesian model, and simulation-based power analysis was conducted to estimate the sample size needed for a fully powered trial. Results revealed modest but consistent trends toward improved affect, particularly reductions in negative affect in the mindfulness group. Power analysis indicated that a sample size of N = 233 would be required to detect a group × time interaction on positive affect with 80% power. These findings demonstrate the feasibility of a larger trial and highlight the value of simultaneously modeling both affective dimensions when evaluating mindfulness interventions in populations vulnerable to perseverative thinking. 

Mixed-Effects Modeling of Affect Using Python (statsmodels) 

This section provides a fully reproducible Python script for analyzing affective changes over time using a mixed-effects multivariate model (MMLM). The dataset and script demonstrate how to: 

·        Reshape longitudinal data for MMLM analysis

·        Specify a three-way interaction model (Group × Time × Affect)

·        Include correlated residuals for positive affect (PA) and and negative affect NA using vc_formula

·        Compare Python (statsmodels) with R (lme4) in MMLM modeling 

The full dataset and annotated Python script are available for download below. 

Data set

The dataset (MindVerticalAffect.csv) contains restructured affective data from a mindfulness intervention study, formatted for mixed-effects modeling. 

Key variables: 

·        subnum → Participant ID 

·        Group → Experimental (0) vs. Control (1) 

·        Time → Pre (1) vs. Post (2) 

·        PA → Positive Affect score 

·        NA → Negative Affect score 

Download the data:

·        Dataset: MindVerticalAffect.csv

Python script

The full annotated Python script is available from below and includes: 

·        Data reshaping into long format (melt())

·        Fixed effects for Group, Time, and Affect (Score ~ Group * Time * Affect)
Random slopes for Time (re_formula="1 + Time")

·        Correlated residuals for PA and NA (vc_formula={"Affect": "0 + Affect"})

·        Fitting the model with Powell optimization (method='powell')

# Import necessary libraries

import pandas as pd  # For data manipulation

import statsmodels.formula.api as smf  # For specifying and fitting mixed-effects models

# Load the dataset

file_path = "MindVerticalAffect.csv"  # Update the file path if needed

data_affect = pd.read_csv(file_path)  # Read the dataset into a Pandas DataFrame

# Reshape the dataset into a "super vertical" format

data_affect_long = data_affect.melt(

    id_vars=["subnum", "Group", "Time"],  # Keep these columns unchanged (identifiers)

    var_name="Affect",  # Column name for the new categorical variable (PA, NA, etc.)

    value_name="Score"  # Column name for the values of affect scores

)

# Define the mixed-effects model

model_affect_random_slopes = smf.mixedlm(

"Score ~ Group * Time * Affect",  # Fixed effects formula: interaction of Group, Time, and Affect

    data=data_affect_long,  # Use the reshaped dataset

    groups=data_affect_long["subnum"],  # Random intercepts for each participant (subnum)

    re_formula="1 + Time",  # Random intercept + random slope for Time per participant

    vc_formula={"Affect": "0 + Affect"}  # Correlated residuals for PA and NA (variance components)

)

# Fit the model using the Powell optimizer

result_affect_random_slopes = model_affect_random_slopes.fit(method='powell', maxiter=1000)

# Display model summary

print(result_affect_random_slopes.summary())  # Output the results

Key Takeaways and R vs. Python Comparison

Using vc_formula for correlated residuals:

·        Unlike lme4 in R, which requires custom covariance structures, Python’s statsmodels allows easy modeling of correlated residuals for multiple outcome variables (PA & NA) using: 

·        vc_formula={"Affect": "0 + Affect"} 

·        This is useful when affective states are interdependent (e.g., high PA often coexists with low NA). 

Why Python for MLM? 

·        statsmodels provides greater flexibility for specifying variance components 

·        More transparent model fitting compared to lme4's default optimizers 

·        Fully open-source & integrates well with Pandas and NumPy 

Multivariate Mixed Linear Model Regression Results

==================================================================

Model:                 MixedLM    Dependent Variable:    Score    

No. Observations:      124        Method:                REML     

No. Groups:            31         Scale:                 0.2378   

Min. group size:       4          Log-Likelihood:        -135.6666

Max. group size:       4          Converged:             Yes      

Mean group size:       4.0                                        

------------------------------------------------------------------

                        Coef.  Std.Err.   z    P>|z| [0.025 0.975]

------------------------------------------------------------------

Intercept                4.667    0.726  6.428 0.000  3.244  6.090

Affect[T.PA]            -2.429    1.027 -2.366 0.018 -4.442 -0.417

Group                   -0.933    0.455 -2.052 0.040 -1.825 -0.042

Group:Affect[T.PA]       1.596    0.643  2.481 0.013  0.335  2.856

Time                    -1.410    0.396 -3.565 0.000 -2.186 -0.635

Time:Affect[T.PA]        2.207    0.560  3.945 0.000  1.111  3.304

Group:Time               0.677    0.248  2.732 0.006  0.191  1.163

Group:Time:Affect[T.PA] -1.147    0.350 -3.274 0.001 -1.834 -0.461

Group Var                0.000    0.586                           

Group x Time Cov        -0.000    0.405                           

Time Var                 0.000    0.214                           

Affect Var               0.412    0.235                           

==================================================================

Note: The direction of these coefficients reflects the coding of the control group as 1, meaning that the treatment group experienced a greater reduction in NA and a greater increase in PA.

R Simulation Repository

This repository contains all R scripts used to conduct the multivariate multilevel Bayesian simulation and power analysis for the Mindfulness and Affect RCT pilot study. The scripts allow replication of the full simulation workflow, including:

§           Simulation of synthetic data based on parameter estimates

§           Model fitting using brms and cmdstanr

§           Calculation of power across sample sizes

§           Logistic interpolation and plotting of the power curve

Download ZIP archive including all .R scripts

·        Mindfulness_R_Simulation_Repository.zip

Individual scripts

·        simulate_brms_power.R — Core simulation function

·        Supercode_31.R — Simulation run for N = 31

·        Supercode_90.R — Simulation run for N = 90

·        Supercode_120.R — Simulation run for N = 120

·        Supercode_240.R — Simulation run for N = 240

·        logistic_fit_plot.R — Logistic model and visualization of power curve

Requirements

To run these scripts, ensure the following R packages are installed:

§           brms with cmdstanr backend

§           tidyverse

§           MASS

Note: The cmdstanr interface requires a working CmdStan installation. You can set it up using cmdstanr::install_cmdstan().

Feel free to use, modify, and share this script for your own analyses! If you reference this work, please cite: 

📧 Scarabusci, R., & Moneta, G. B. (submitted). Mindfulness and affect regulation: A pilot randomized trial in financially anxious individuals.
📧 For questions, contact: g.moneta@londonmet.ac.uk