### Table VI. State-Based Validation: An Example

1990

Cited by 79

### Table VII. State-Based Validation:

1990

Cited by 79

### Table 2. State-Based Navigation Module

2002

Cited by 1

### Table 2. State-Based Navigation Module

2002

Cited by 1

### Table 4. State-Based Reconstructability Analysis Results

"... In PAGE 5: ...Table 4. State-Based Reconstructability Analysis Results A state-based modeling approach can do significantly better in this particular example, as indicated by the results in Table4 . The notation used to describe state-based models is somewhat more complex and requires explanation.... In PAGE 6: ... A second quot;default quot; conditional distribution over the states of Z can be used for any other combination of A and B, without significantly compromising the quality of our prediction. As shown in Table4 , the state-based models AB:Z:a0b0Z and AB:Z:a0b1Z each do much better, in terms of information captured, than the best variable-based model. Because the sample size (n=1247) is fairly large, however, the discrepancies between the observed and predicted frequencies are sufficient to keep p low and justify a statistical rejection of both models.... ..."

### Table 5. Marginal Z Distributions for State-Based Models

"... In PAGE 6: ...6 conditional distributions on Z for all states other than a1b0 are as similar as possible, subject to satisfaction of the AB and Z constraints. This is illustrated in Table5 for four different state-based models, including the AB:Z:a0b1Z model.... ..."

### Table 4. State-Based Reconstructability Analysis Results

"... In PAGE 5: ...Table4... In PAGE 6: ... A second quot;default quot; conditional distribution over the states of Z can be used for any other combination of A and B, without significantly compromising the quality of our prediction. As shown in Table4 , the state-based models AB :Z:a 0 b 0 Z and AB:Z:a 0 b 1 Z each do much better, in terms of information captured, than the best variable-based model. Because the sample size (n=1247) is fairly large, however, the discrepancies between the observed and predicted frequencies are sufficient to keep p low and justify a statistical rejection of both models.... ..."

### Table 5. Marginal Z Distributions for State-Based Models

"... In PAGE 6: ... This relaxation procedure has the effect that, in the model, the conditional distributions on Z for all states other than a 1 b 0 are as similar as possible, subject to satisfaction of the AB and Z constraints. This is illustrated in Table5 for four different state-based models, including the AB :Z:a 0 b 1 Z model. data AB:Z:a0b0Z AB:Z:a0b1Z AB:Z:a1b0Z AB:Z:a1b1Z p(z0) p(z1) p(z0) p(z1) p(z0) p(z1) p(z0) p(z1) p(z0) p(z1) a0b0 0.... ..."