### Table 2: Comparison of maximum accuracy (averaged over BDBC iterations) in the small sample size, for using words (Test3) vs. using word-clusters (Test4).

2001

Cited by 38

### Table A1 A numeric guide to sample size for small, medium, and large effects sizes for dif

2005

### Table 9: The key to associate regression techniques with lines on the graphs in Figure 2. This table is laid out in blocks, similar to the graphs in the figure. Each block in the table lists the regression techniques associated with the lines of the corresponding graph for the case of small sample size.

1997

"... In PAGE 23: ... Again, to simplify the graph labels, the MISE has been multiplied by 10,000. Table9 provides a key for associating regression techniques with the lines on the graphs in Figure 2: the regression techniques are ordered in the table for each graph according to their position when sample size is small. For example, the graph lines in the middle graph of Figure 2, from better to worse MISE (i.... ..."

Cited by 1

### Table 10: The key to associate regression techniques with lines on the graphs in Figure 3. This table is laid out in blocks, similar to the graphs in the figure. Each block in the table lists the regression techniques associated with the lines of the corresponding graph for the case of small sample size.

1997

"... In PAGE 23: ... As before, all MISE values are multiplied by 10,000. Table10 provides a key for associating regression techniques with the lines on the graphs in Figure 3: the regression techniques are ordered in the table for each graph according to their position when sample size is small. For example, the graphs lines in the top left of Figure 3, from better to worse MISE (i.... ..."

Cited by 1

### Table 9: The key to associate regression techniques with lines on the graphs in Figure 2. This table is laid out in blocks, similar to the graphs in the figure. Each block in the table lists the regression techniques associated with the lines of the corresponding graph for the case of small sample size.

"... In PAGE 21: ... Again, to simplify the graph labels, the MISE has been multiplied by 10,000. Table9 provides a key for associating regression techniques with the lines on the graphs in Figure 2: the regression techniques are ordered in the table for each graph according to their position when sample size is small. For example, the graph lines in the middle graph of Figure 2, from better to worse MISE (i.... ..."

### Table 10: The key to associate regression techniques with lines on the graphs in Figure 3. This table is laid out in blocks, similar to the graphs in the figure. Each block in the table lists the regression techniques associated with the lines of the corresponding graph for the case of small sample size.

"... In PAGE 21: ... As before, all MISE values are multiplied by 10,000. Table10 provides a key for associating regression techniques with the lines on the graphs in Figure 3: the regression techniques are ordered in the table for each graph according to their position when sample size is small. For example, the graphs lines in the top left of Figure 3, from better to worse MISE (i.... ..."

### Table 8. Error ratio as a function of number of added Boolean irrelevant attributes for small sample sizes (using ensembles of eleven stochastically-learned models; combined with Uniform Voting). The number below each data set identifier indicates the number of training examples.

1996

"... In PAGE 22: ...ata sets. So, to test this hypothesis, we performed 100 trials with training sets of size 20. In particular, we were interested to see if the exceptionally low error ratio obtained on the wine data set could be made to increase with increasing numbers of irrelevant attributes. Table8 shows that for very small training set sizes, adding irrelevant attributes makes no significant difference to error ratios in 4 domains and increases the error for the wine data set thus validating our hypothesis. In summary, the answer to the question posed in this section ( How does increasing the number of irrelevant attributes affect the amount of error reduction? ) is that error ratios initially decrease as irrelevant attributes are added thus providing an opportunity for the multiple models approach.... ..."

Cited by 113

### Table 2. Comparisons with the Previous Studies Reference

### Table 1 Relative efficiency of IFS estimators with different set of maps W1, W2, Q1 and Q2 with respect to the empirical distribution function (i.e. IFS/EDF). Based on 100 Monte Carlo simulation for each distribution. Small sample sizes.

Cited by 1