Large (400 cm2) electrodes of the optimum density were prepared and found to have the same large capacity as their smaller, test-size counterparts.
Recall that a larger training data set size (n train ) causes a smaller test set size (n test ) since the number of objects was kept constant at n train + n test = 80 objects.
The larger deviations for smaller test set sizes in Figure 4b are probably due to this larger variability and would vanish if the number of simulations were increased.
Follow-up ANOVAs in the small target test size condition with study/test relative size condition as a within-subject factor did not approach significance, F 2,126) = 2.06, MSe = 0.02, p = 0.13, η2 = 0.03.
It appears that a small test sample size (78 19) will reduce the power.
Using the same sample source, the expected log2-ratio value was 0. We found that the number of differently expressed genes increased when comparing the smallest test sample size (250 cells) with the reference.
This suggests a warning to trace analysts: very reproducible results at very small test portion sizes may be erroneous.
Figure 5 shows that the prediction error estimates in the outer loop were highly variable for small test data sizes.
However, there was also a drawback since larger training data sets imply smaller test sets: in case of small test data set sizes, the variability of the prediction error estimates increased considerably.
Analogous to the simulation study, small test data set sizes yielded largely varying prediction error estimates in the outer loop owing to highly variable test data especially in case of 10-fold CV (Figure 7).
In case of small test data set sizes, the error estimates in the outer loop scattered largely around their average as compared to the error estimates derived from the 'oracle' data set (Figure 9).
