June 4, 2025
The authors discussed a weighted repeated measures correlation coefficient which could work even in the presence of missing data. The Pearson correlation coefficient cannot be used in these data due to violations of independent data. Also, some measures were created to handle repeated measures by they failed to take into account differing numbers of measures. In their study they considered 6 different correlation coefficients with 3 based on Pearson correlation, one based on ANCOVA, and a correlation based on linear mixed effects model (LME), and all compared to theirs. The one based on ANCOVA used sums of squares. The one based on LME was calculated from the covariance matrix estimated using the maximum likelihood method. Their measures was a modification of the correlation coefficient based on ANCOVA. They bootstrapped to obtain the confidence intervals.
In simulations overall their method proved better in terms of bias and coverage probability than the other five methods to which it was compared, especially in the presence of missing data. However, in the presence of missing data their method had a wider confidence interval than the ANCOVA method. In a real dataset application, the Pearson correlations were higher than theirs and the methods based on ANCOVA or LME. Per their discussion, the Pearson correlation coefficient only worked in certain situations but overall was not appropriate for the repeated measures data. Their method worked well in general. The correlation based on LME did not provide a convergent solution when applied to overall data and did not converge 20.2% of the time. One issue is that their proposed method assumed independence across subject and they did not check that issue. Also they still want to evaluate calculation of the standard deviation through a theoretical way rather than bootstrapping. Overall in conclusion their proposed weighted repeated measures correlation coefficient performed well compared to other methods and also in the presence of missing data.
Written by,
Usha Govindarajulu
Keywords: weighted repeated measures, correlation coefficient, missing data, Pearson correlation, ANCOVA, linear mixed effect model
References:
Kondo M, Nagashima K, Isono S, and Sato Y (2025) “Weighted Repeated Measures Correlation Coefficient: A New Correlation Coefficient for Handling Missing Data With Repeated Measures” Statistics in Medicine.
https://doi.org/10.1002/sim.70046