Figure 2.

Comparing accuracies of network estimation of the MVAR and s-MVAR modeling approaches. (A) Box plot of estimation accuracies (RMSE) with s-MVAR (left) and MVAR (right) approaches across all factors studies. The red line indicates the median of the sample, while the lower and upper bounds of the blue rectangle indicate the 25^th and 75^th percentile levels respectively. Red dots outside blue rectangle indicate sample outliers. p-values are the result of one-tailed sign-tests of the hypotheses that the median of the difference between two samples is zero. Low p-values indicate that the accuracy of the s-MVAR estimation is higher than that of MVAR estimation. (B) Box plots and results of one-tailed paired sign-tests comparing s-MVAR and MVAR at different levels of the samples/parameters ratio. The s-MVAR performs better than the MVAR at all levels, especially when the ratio of samples/parameters is small. (C) Box plots and results of one-tailed paired sign-tests, comparing s-MVAR and MVAR at different levels of network density. s-MVAR performs better than MVAR at all levels, especially when network density is low. (D) Box plots and results of one-tailed paired sign-tests, comparing s-MVAR and MVAR at different levels of signal-to-noise ratio (SNR). s-MVAR performs better than MVAR at all levels. This difference is highest at high values of SNR, but most reliable at low values of SNR.