### Table 3. Median of the true error and estimation error (in meters) for the ZB75 and ZB90 algorithm at each of the four poster locations.

"... In PAGE 16: ... To test the accuracy of our error estimation algorithms, we supplied these locations into our MR, ZB75 and ZB90 algorithms. Table3 shows the median of the true error and estimation error for each poster. MR performs well for posters 1 and 2, but over estimates the error for poster 3 and under estimates for poster 4.... ..."

### Table 3 Posterior distributions of response parameters Model for store incidence Model for ln expenditures

2005

"... In PAGE 17: ... We verified both restrictions, and found that display does not affect store incidence significantly, and feature does not affect spending significantly. E RESULTS Store incidence We present the store incidence results in the left-hand part of Table3 . Except for perceived Produce Quality - the impact of which is not significant - the benefit variables (StoreSurface, Feature and LagExpend) have positive and significant effects on store incidence.... In PAGE 18: ...079), plausibly because consumers want to shop for holiday meals, and the longer opening hours (relative to Christmas) allow them to do so. [Insert Table3 about here] Focusing on the impact of the price war variables, several interesting findings emerge. First, Albert Heijn did not manage to increase its store incidence propensity, as the coefficient of the PW*AlbertHeijn variable is insignificant.... In PAGE 18: ....355), and part of which is temporary (-.121). This empirical finding corroborates the prediction of Busse (2002) and Heil and Helsen (2001) that price wars increase the price sensitivity of consumers. Expenditures The estimates for the log (ln) of expenditures equation are given in right-hand part of Table3 . All the benefit variables have the expected positive and significant effects.... ..."

### Table 3. 2 analysis for ux density distribution of galaxies, nll = 25.

"... In PAGE 9: ... = 2 is very small. Following Yahil et al. (1991), we calculate the 2 statistic of the di erence between the predicted and observed curves, using Poisson error bars and summing only over bins with ve or more galaxies. The results are tabulated in Table3 . The t with p = 1 is acceptable in all bins.... In PAGE 9: ... For p = 3 (not shown in the gure) the t in most redshift bins is unacceptable. The quantity in Table3 is the number of bins of ux density in which the comparison is done. One might argue that this number should be reduced by the number of free parameters in the luminosity function (25, in these cases!).... ..."

### Table 3. Per-condition Coefficients of Correlation Between Subjective Scores and Objective Estimators

1998

"... In PAGE 23: ... The resulting coefficients of correlation are shown in Table 3. The correlation values in Table3 were calculated after averaging all available subjective scores for each condition to a single score for that condition. Similarly, for each condition, all available objective... In PAGE 27: ... A more advanced analysis technique, described in [37], recognizes the importance of the distributions of the objective estimates and the subjective scores for each condition, how they influence confidence intervals, and in turn, the final conclusions that one draws from objective and subjective tests. Table3 demonstrates the limitations of SNR, SNRseg, and PWSNRseg as estimators of perceived speech quality. CD and BSD tend to show higher correlations for tests 5, 6, and 7, which contain only conditions that tend to preserve waveforms.... In PAGE 27: ... Table 4 shows per-condition correlation values for L(AD) as calculated by the two MNB structures. Since L(ND) is used as a reference, that column from Table3 is repeated as column 2 of Table 4 to allow for easy comparisons. Two versions of the estimators were evaluated.... ..."

### Table 16: Estimated values using acceptance sampling based on the simulation with Weak prior as sampling density. Marginal output density from the Real prior used as target density.

"... In PAGE 26: ... 1a the simulation results with the Weak prior were reused based on the density from the Real simulations. As shown in Table16 the resulting distribution of the in the resampled simulation runs were close to the results in the original Real simulation runs in Table 6 with mean 139.9 compared to 139.... In PAGE 29: ...The presented algorithm seems a good choice for generating samples from the post- model distribution. The very low success rate (less than 3%) shown in Table16 may serve as an argument for not including the parameter uncertainty when using simulation models. However, as the example illustrates the presented algorithm very rapidly ensures that the succes rate of costly simulation runs will be much higher.... ..."

### Table 8. Estimate Breaks in Variance by the ICSS, and Estimate d and LM statistics of the Squared Stock Return, Break Process and Squared Residuals

1999

"... In PAGE 22: ...negative estimates of d in the residuals are obtained, so there is some possibility of overdifference as pointed out in section 5. As an additional analysis, we also examined long memory in the squared stock returns in Table8 . As occasional breaks are incorporated directly into return series, the existence of long memory in volatility is mixed, too.... ..."

Cited by 19

### Table 4. Proportion of nonsigni cant di erences (by z-tests, = :05) between the mean estimates for each estimator and the true density. The number of points at which the density was estimated is given in parentheses next to each estimator. The \Distribution Density quot; is the density estimator derived from ts to the CDF.

"... In PAGE 34: ... However, they are ine cient: both the decile and Vincent estimators are highly biased, and signi cant bias persists up through samples of size N = 10; 000. As shown in Table4 , even when N equals 10,000, most of the estimated points on the density function shown in Figures 4 and 5 deviated signi cantly from the true density function. Overall, the Vincent histograms were less variable than the decile-based histograms and showed... In PAGE 36: ...di erences at the early points on the bias curves correspond to the mode or the points immediately preceding the mode. As shown in Table4 , the number of signi cant deviations between the mean xed-width histogram estimate and the true density function actually increased with sample size. The reason for this was explained above.... In PAGE 36: ... However, the extent of bias attenuates quickly with sample size and seems to have all but disappeared by N = 1000 (except for the di usion model), although signi cant bias remains at N = 10; 000. Table4 shows that the proportion of nonsigni cant di erences is uniformly high across sample sizes, unlike the histogram estimates, making this estimator the most accurate of the density estimators we have examined. CDF estimates The average CDFs are shown in Figure 12, and their bias is presented in Figure 13.... In PAGE 37: ... Variance seems to be largest at the higher deciles (the 90th, in particular), but negligible at the smallest deciles even when N = 50. Table4 gives the proportion of nonsigni cant di erences between the mean quantiles and the true quantiles. Unlike the other estimators, there is little e ect on the number of signi cant di erences with increasing sample size, as most of the di erences are nonsigni cant even at N = 50.... In PAGE 48: ... By N = 100 observations, the variance of the estimates are small in comparison to the nonparametric estimates, except for the portions of the densities that are changing very rapidly. Table4 shows that there are fewer nonsigni cant di erences between the estimate and the true density function for the MLE and CDF techniques, and that the rate at which bias decreases depends strongly on the model being t. When N = 500, the proportion of nonsigni cant di erences for the ex-Gaussian, gamma, and Wald models is equivalent to the proportion observed for the Gaussian kernel estimator, but the kernel estimator was more accurate for the rest.... In PAGE 51: ... Variance around the SSE estimate does not decrease with sample size, nor does bias. Table4 shows the proportion of nonsigni cant deviations between the average SSE estimate and the true density at each sample size. The accuracy of the estimate does not improve with increases in N, and overall accuracy is nowhere near as good as the accuracy of the kernel estimates.... In PAGE 73: ... Table4 continued. Model Estimator N Di usion ex-Gaussian Gamma Race Wald Weibull 50 .... ..."

### Table 3: Error distribution for velocity, ow, density, and headway.

1997

"... In PAGE 6: ... Fig. 8 shows scatter plots of the ow and velocity estimates pro- vided by the loop and vision data, and Table3 summarizes the error distribution for velocity, ow, density, and head- way. As one would expect from a feature based tracker, the measured velocity is very accurate.... ..."

Cited by 71

### Table 3: Error distribution for velocity, ow, density, and headway.

1997

"... In PAGE 6: ... Fig. 8 shows scatter plots of the ow and velocity estimates pro- vided by the loop and vision data, and Table3 summarizes the error distribution for velocity, ow, density, and head- way. As one would expect from a feature based tracker, the measured velocity is very accurate.... ..."

Cited by 71