Scholarlink

<p>Two-stage individual participant data meta-analysis can be used for synthesizing intervention effects from single-case experimental design (SCED) studies. However, challenges arise when estimating standardized intervention effects in presence of autocorrelation. The current Monte Carlo simulation study examined statistical properties (i.e., relative bias, mean squared error, relative standard error&#xa0;bias, coverage proportion of the 95% confidence/credible intervals, statistical power, and Type Ⅰ error rate) of estimating the standardization factors (i.e., white noise standard deviation or residual standard deviation), the autocorrelation parameter, and the standardized intervention effect size at Stage 1. In addition, we evaluated whether Bayesian estimation procedures with different sets of priors resulted in improved statistical properties compared to the traditional restricted maximum likelihood (REML) estimation. Results indicate that using white noise standard deviation is preferable compared to the residual standard deviation when autocorrelation is present, regardless of the estimation procedure. In terms of estimating autocorrelation and the standardized intervention effect, neither Bayesian nor REML estimation consistently outperformed across all examined statistical properties. Additionally, neither estimation method consistently demonstrated acceptable performance across all statistical properties. Specific recommendations for applied researchers and implications for future research are discussed.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Standardized Effect Sizes for Autocorrelated Single-Case Experimental Design Data: A Monte Carlo Simulation Study

  • 期刊论文

  • 发表日期：

    2025年10月3日

查看全文
  • Yukang Xue,
  • Mariola Moeyaert,
  • Eunkyeng Baek,
  • Benjamin G. Solomon,
  • Wim Van den Noortgate

摘要

Two-stage individual participant data meta-analysis can be used for synthesizing intervention effects from single-case experimental design (SCED) studies. However, challenges arise when estimating standardized intervention effects in presence of autocorrelation. The current Monte Carlo simulation study examined statistical properties (i.e., relative bias, mean squared error, relative standard error bias, coverage proportion of the 95% confidence/credible intervals, statistical power, and Type Ⅰ error rate) of estimating the standardization factors (i.e., white noise standard deviation or residual standard deviation), the autocorrelation parameter, and the standardized intervention effect size at Stage 1. In addition, we evaluated whether Bayesian estimation procedures with different sets of priors resulted in improved statistical properties compared to the traditional restricted maximum likelihood (REML) estimation. Results indicate that using white noise standard deviation is preferable compared to the residual standard deviation when autocorrelation is present, regardless of the estimation procedure. In terms of estimating autocorrelation and the standardized intervention effect, neither Bayesian nor REML estimation consistently outperformed across all examined statistical properties. Additionally, neither estimation method consistently demonstrated acceptable performance across all statistical properties. Specific recommendations for applied researchers and implications for future research are discussed.