### Table 4 Single-primitive targets: base amplitude, location, pose, and radius statistics.

"... In PAGE 10: ... The fraction of runs in which this occurs is listed in the fraction spurious column. Table4 ~described later! presents the results of the continuous parameter estimation, conditional on correct type identification and nonspurious parameters. For each of the eight experiments ~four primitives, two AVSDs!, Monte Carlo runs were continued until we had obtained 500 trials in which the primitive was detected and estimated to be the correct type.... In PAGE 11: ... Table4 presents the error statistics for the continuous parameters. Base amplitude figures are quoted in dBsm, location and radius in inches, and pose in degrees.... In PAGE 11: ... The azimuth/elevation root-mean-squared error ~RMSE! statistics correspond to angular separation in degrees be- tween two points on a sphere. The amplitude bias exhibited in Table4 is largely attrib- utable to two factors. First is the frequency windowing in- herent in the SAR imaging process.... In PAGE 11: ...ominal bias is roughly 20.5 dBsm. ~It is possible to cor- rect for this factor simply by adding 0.5 to the amplitude estimates produced by the algorithm, although we have not done this in the presentation of the results in Table4 .! The second factor influencing amplitude bias is the slight mis- 20 deg cylinder 0.... In PAGE 11: ...s biased by these lower-bounce reflections. In Sec. 5.2 we discuss an approach to removing this bias, if desired. Ad- ditionally, although it is not illustrated in Table4 , the radius errors are correlated with the location errors due to layover effects. Table 3 Single-primitive targets: model order and type confusion statistics.... In PAGE 12: ...000 0.004 The pose results of Table4 demonstrate that primitive pose can in general be accurately estimated to a finer granu- larity than provided by the AVSD. The dihedral pose errors, which are larger than those observed for the other primitive types, are attributable to the near-invariance of the dihedral response to certain changes in the Euler-angle pose.... In PAGE 12: ... The results of this postprocessing on the trials of the previous section are presented in Table 5. The trihedral lo- cation estimation refinement achieves a greater fractional reduction in bias than the dihedral location estimate refine- ment due to the greater accuracy of the trihedral pose esti- mates ~see Table4 !. For similar reasons, the refinement is more successful for the 10-deg AVSD than for the 20-deg AVSD.... ..."

### Table 1. Anticipated Treatment Effects by Measurement Location

2004

"... In PAGE 2: ... To study these effects, we made measurements at the 12 locations indicated (Table 1). Statistical Methods The experiment was conducted according to a split-plot statistical design, studying material loss resulting from 12 exposure levels ( Table1 ) under 2 conditions (defined by pH). We modeled the effects of exposure levels, pH, and their interaction (1) on initial beam dimensions, to estimate uniformity of beam dimensions prior to intervention, and (2) on dimensional changes following intervention, to estimate means and 95% confidence intervals for effects on material loss at pH = 7, at pH = 6, and the difference.... ..."

### Table 1. Results of Location Choice Models for Three Household Types Households with Worker(s) amp; Moved for Job-related Reasons (Model 1)

2007

"... In PAGE 7: ...downloaded from http://elsa.berkeley.edu/~train) was used to estimate model parameters for the three household segments. All parameters found to lack statistical significance were removed in a stepwise fashion, to determine the final model specifications; and Table1 provides final model results. For working households who indicated having moved for job- or school-access reasons, the price-to-income variable is estimated to have a strongly negative set of (random) coefficients, as expected.... In PAGE 7: ... For working households who indicated having moved for job- or school-access reasons, the price-to-income variable is estimated to have a strongly negative set of (random) coefficients, as expected. (Please refer to Model 1 in Table1 .) Thanks to a relatively small estimated variance, only 2.... ..."

### Table 1: Statistical summary of residual analyses

"... In PAGE 12: ...are :13. Results of all three series of experiments are summarized in Table1 . Standard deviations of the residuals ^ are shown, along with the prior estimates ^ est of these quantities.... In PAGE 13: ... Comparison of Figures 4b and 5b with the corresponding panels in Figures 2 and 3 shows that the residuals in the run with the average model are distinctly smaller than those from the original MC model; note the di erence in scale from Figures 2b and 3b to Figures 4b and 5b. Examination of Table1 shows that this is the case at ve of the six assimilation stations, the exception being Nauru. Since the Average model represents a change in the measurement model from MC, rather than a change in the observation or system error models, this indicates that the ltering scheme is over tted.... In PAGE 14: ...This is re ected in the fact that the estimated RMS errors in the ltered model, ^ est in Table1 , are less than 3cm. Table 1 also shows that the standard deviations, ^ , of the post-assimilation residuals, Yt?H ^ Wtjt, are in general much smaller than we expect them to be; in other words, our analysis is closer to the actual observations than it should be, and may contain too much of the observation noise.... In PAGE 14: ...This is re ected in the fact that the estimated RMS errors in the ltered model, ^ est in Table 1, are less than 3cm. Table1 also shows that the standard deviations, ^ , of the post-assimilation residuals, Yt?H ^ Wtjt, are in general much smaller than we expect them to be; in other words, our analysis is closer to the actual observations than it should be, and may contain too much of the observation noise. Since the lter output is a weighted sum of model output and observation, it is natural to suspect that the observations are overweighted.... In PAGE 15: ... Figures 4 and 5. As shown in Table1 , the residuals of the average model have markedly reduced serial correlation at every station but Rabaul, even though they still seem to be serially correlated. The improvement over the MC model is shown by the Partial ACF plots shown in Figures 4d and 5d and in Table 1.... In PAGE 18: ... Analogously to the Average model, 95% con dence bands for the ACF and PACF are :24. From Table1 , the residual standard deviations of the AR(1) approach are smaller than the residual standard deviations of the two other methods at most of the stations. The greatest di er- ences between the AR(1) approach and the Average approach are at Santa Cruz in the experiments with 6 and 10 assimilation stations, and at Rabaul and Kapingamarangi in the experiments with 6 assimilation stations.... In PAGE 19: ...Table1 shows that the posterior estimates of the RMS errors at Kapingamarangi, Tarawa, Canton and Fanning, where no data are assimilated are reasonably reliable. Figures 8, 9, 10 and 11.... In PAGE 20: ... This approach may well be fruitful for more recent, more dense data sets. Our nding that the expected analysis error variances are not highly sensitive to serial correla- tion of the innovation sequences (compare the values of ^ est in Table1 for corresponding experiments with the Average and AR(1) models) is consistent with the theory presented by Daley (1992a). In that paper, Daley argued that when the advection speed U exceeds the quotient x= t, where x is the spatial interval between observations and t is the assimilation interval, the expected anal- ysis error is not sensistive to serial correlation in the observation errors.... ..."

### TABLE III Conditional Logit Estimates Probability of Location

### Table 24 Estimated Odds Ratios for being a light-care home care client (versus a

"... In PAGE 94: ... Home Care in Ontario A second logistic regression equation predicating the likelihood of being a light care CCAC home care client (versus a CHA client) is shown in Table24 . The final c statistic for the model is 0.... ..."

### Table 23 Estimated Odds Ratios for being a home care client (versus a Community

"... In PAGE 89: ... Not being self-reliant, having ADL decline, having more falls, self reporting to be in poor health, not reporting being lonely and living with others make one more likely to be receiving CCAC home care support (rather than support through a community support agency). The final c statistic for the first model ( Table23 ) was 0.85.... In PAGE 94: ....70. With whom the client lived is no longer significant and was dropped form the equation. The point estimates are much lower for this equation than for the first equation predicting being a CHA or HC client (see Table23 ); however, the same pattern is observed such that age and gender remain non-significant and self-reliance, ADL decline, having more falls, self reporting to be in poor health and not reporting being lonely make one more likely to be receiving CCAC home care support (LC home support specifically). A third logistic regression model predicting the likelihood of being a light care CCAC client (versus a CHA client who does not access supportive housing services) shows a similar pattern of variables (see Table 25).... ..."

### Table 2. Statistical Parameters of Significant Models model

2005

"... In PAGE 7: ...e., significant models , were included in the lists ( Table2 ). We also tested the top ranked model for each data set against all other significant models for the same set.... In PAGE 7: ... Moreover, one can easily notice that models for logK1 have in general lower MAE compared to log 2, because of the lower quality of the experimental data in the later models, as it is explained in the Discussion. The analysis of models in Table2 indicated complexities with comparison methods using just one data set. A combination of model + descriptors, which provided top and very significant results for one data set, was at the bottom of the list for another data set.... In PAGE 8: ... Cumulative Plots of Descriptors Contributions to Top- Ranked Models. To provide some general estimations of all results, we selected n top-ranked models per data set characterized by the smallest MAE (see Table2 ) and the analyzed contribution of different descriptors and methods types to these models. Figure 4 show which types of descriptors were more frequently used in the top-ranked models.... In PAGE 8: ... Percentage of models (y axis) calculated using the corresponding descriptor system as a function of the number of best models (x axis) per each data type. For each data set we selected the n-best models ( Table2 ) and the calculated percents of models generated using each descriptor system. ISIDA-5 and ISIDA- average models generated using SMF fragments were not included in the analysis.... In PAGE 8: ... Percentage of models (y axis) as a function of the number of n top-ranked significant models (x axis) selected per each data type. For each data set we selected the n-best models ( Table2 ) and counted the percents of models contributed using each method. A, B, and C correspond to models calculated using a single descriptor type.... ..."

### Table 4. Estimation of Population and Household Suburbanization, Using an Alternative Indicator of Telecommuting [Dependent Variable: Population and Household Gradients (N=60)]

"... In PAGE 20: ... that use the Internet at home for job-related tasks ( Table4 ). Table 4 includes the same variables in the estimation, but replaces the telecommuter variable PROPMW, with the alternative indicator of telecommuting, IUSEHOME.... In PAGE 20: ... That is, higher is the proportion of population that uses the Internet from home for office-related work, greater would be the flexibility to locate farther away from the CBD and greater would be the extent of suburbanization. None of the other variables are significant in the population or the household gradient regressions in Table4 , as in the earlier specifica- tions, except population, which continues to have the expected negative ef- fect on population and household suburbanization. The model in all cases is a better explanation of variations occurring in the population gradient rather than the household gradient, as may be seen in the value for the adjusted R2 reported in Tables 3 and 4.... In PAGE 21: ... used in Table4 , IUSEHOME, could be biased especially if full-time conven- tional workers bring home part of their office work. In this case, their use of the Internet from home for office-related work will not result in any reduc- tion in their conventional trips to the workplace.... ..."

### Table 1. SCAG 1994 Traffic Counts on Regional Screenlinesa

in Author 4

"... In PAGE 10: ... Figure 3 gives the location of the 11 screenlines. Table1 gives the 1994 actual count data by screenline. The first panel gives autos, HDTs, total vehicles, and the HDT share.... In PAGE 12: .... US Census Bureau (2001) Statistical Abstract of the United States, 121st edition. Washington, DC. Table1 051. 2.... In PAGE 12: ...ashington, DC. Table 1051. 2. US Department of Transportation (USDOT) (2002) National Transportation Statistics 2001, Table1 -41, Washington, DC. 3.... ..."