Error structure

The diad fractions for the low conversion experiments only are reproduced in Table II. The high conversion data cannot be used since the Mayo-Lewis model does not apply. Again diad fractions have been standardized such that only two independent measurements are available. When the error structure is unknown, as in this case, Duever and Reilly (in preparation) show how the parameter distribution can be evaluated. Several attempts were made to use this solution. However with only five data points there is insufficient information present to allow this approach to be used. 

One way to proceed with this example is to estimate the error structure. Then using two different estimates, the sensitivity of the results to the assumed error structure can be examined. The first estimate of the covariance matrix used here is... 

The result of analyzing the data under this assumption for the error structure is shown in Figure 4. 

Therefore to make meaningful inferences from experiments such as those reported by Yamashita et al. either the error structure must be known or sufficient data must be provided, preferably in the form of optimally designed replicates. This analysis confirms that it is generally insufficient to evaluate only point estimates. In fact these are secondary to evaluating and reporting joint probability regions. 

The data analyzed in this work were reported by Hill et al. ( ) for the copolymerization of styrene with acrylonitrile. They are shown in Table III in the form of triad fractions measured by C-NMR for copolymers produced at various feed compositions. One reason for choosing this particular dataset is that the authors did indicate the error structure of their measurement. 

Applications of the method to the estimation of reactivity ratios from diad sequence data obtained by NMR indicates that sequence distribution is more informative than composition data. The analysis of the data reported by Yamashita et al. shows that the joint 95% probability region is dependent upon the error structure. Hence this information should be reported and integrated into the analysis of the data. Furthermore reporting only point estimates is generally insufficient and joint probability regions are required. 

If basic assumptions concerning the error structure are incorrect (e.g., non-Gaussian distribution) or cannot be specified, more robust estimation techniques may be necessary. In addition to the above considerations, it is often important to introduce constraints on the estimated parameters (e.g., the parameters can only be positive). Such constraints are included in the simulation and parameter estimation package SIMUSOLV. Beeause of numerical inaccuracy, scaling of parameters and data may be necessary if the numerical values are of greatly differing order. Plots of the residuals, difference between model and measurement value, are very useful in identifying systematic or model errors. 

To gain some perspective on the problem, it is well to realize that assessment of the detection limit is subject to all of the assumptions and restrictions of the estimation process. That is, the functional and error structure of the calibration curve must be known or assumed, and the respective parameters and their uncertainties must be estimated. Although large... 

Schepers, U., Ermer, J., Preu, L., and Waetzig, H. (2004). Wide concentration range investigation of recovery, precision and error structure in liquid chromatography.. Chromatogr. B 810, 111—118. 

NP2. In particular, Tyr82 flips from outside the protein in NP1/NP4, where it contacts bulk solvent, to inside the protein in NP2, where it hydrogen bonds to Glu55 (110). Thus, modeling the NP2 structure based on those of NPl and NP4, and, as is normally done, keeping the invariant residues structurally conserved, would lead to errors. Structures of protein families such as the nitrophorins therefore provide key information for the future improvement of structure prediction and structural genomics. 

Fig. 6. Error structure of HTS results. This figure shows how the reproducibility of activity data varies with differences in activity. The circles represent compounds designated as inactive and crosses designate active compounds. From this figure it is obvious that assay variability is closest at the cutoff between active and inactive compounds.

The posterior probabilities on the Monte Carlo realizations of the model are determined by the error structure in the data. 

Equations 14 through 16 indicate that careful consideration must be given to the analysis of the error structure and to the confidence intervals of the correlation variables, otherwise trends due to errors will appear as real when, in fact, they may be contained within the error bands of either y or z. 

In the previous sections it has been stipulated that there are several response variables which can be modeled. The success of the optimization procedure depends on the selection of the response variable(s). There are several criteria which can be used to select a response variable [12,17]. The response variable should have a homoscedastical error structure and have to change continuously and smoothly. Both experimental data and chromatographic theory can be used to check these properties. 

From chromatographic theory [2] it is clear that the R value should result in simple models. For this reason it is preferred over, the k or the Rj. These latter response values can be calculated from predicted R values. It is more difficult to determine the error structure of the R . It is believed however that logarithmic transformation of the k values should result in homoscedastical error structures [3]. 

If there is no theory available to determine a suitable transformation, statistical methods can be used to determine a transformation. The Box-Cox transformation [18] is a common approach to determine if a transformation of a response is needed. With the Box-Cox transformation the response, y, is taken to different powers A, (e.g. -2transformed response can be fitted by a predefined (simple) model. Both an optimal value and a confidence interval for A can be estimated. The transformation which results in the lowest value for the residual variance is the optimal value and should give a combination of a homoscedastical error structure and be suitable for the predefined model. When A=0 the trans-... 

While you will use the least squares method in most cases, do not forget that selecting an estimation criterion you make assumptions on the error structure, even without a real desire to be involved with this problem. Therefore, it is better to be explicit on this issue, for the sake of consistency in tlie further steps of the estimation. 

Once we have the molecular formula, it can provide information that limits the amount of trial-and-error structure writing we have to do. Consider, for example, heptane and its molecular formula of C7H16. We know immediately that the molecular formula belongs to an alkane because it corresponds to C H2 +2. 

As an example, 80 batches with four observations per batch were each simulated for the following random variation forms normal, exponential, and lognormal, (see Fig. 5 A-C). x and R charts were constructed for each set as if the true random variation were normal. The charts appear in Figs. 2-4. The results appear in Table 2. This table shows that roughly the same number of points falls outside the x control limits, regardless of the form of the random variation. However, the lognormal distribution has many more R values outside the control limits than the other four distributions. The operator of the process would mistakenly think this process was frequently out of control. The R chart shows greater susceptibility to nonnormality in the random error structure. 

Another claim is that the nature of the errors is different between chemical and physical measurements. It is claimed [2] that for physical systems, systematic errors predominate and that these are corrected out of the result whereas for chemical systems random errors predominate. I do not know the basis for these claims but they do not align with my own experience, e.g. the systematic error associated with recovery can easily be equal or greater than the random error. An important point about the error structures is that the ability to detect and correct for systematic errors is limited by the size of the random error, but this is true for all types of measurement. It requires 13 replicate measurements to have the sensitivity to detect an effect equal to the size of the standard deviation on 1 measurement. There is also the view that the uncertainties on physical measurements are of the order of one part per million, but again this is not true. Standards of radioactivity and neutron dose uncertainties are often in the range of 1-30%. 

The differences between physical and chemical measurements should not be over emphasised. Many of the basic problems are very similar, e.g. sample effects are important in all types of measurement. Also similar error structures occur in both types of measurements and not... 

Of importance for any experimental technique which is to be used to fit some complex reaction model is the way in which experimental errors influence the result [93]. The error structure for the EHD method utilising the RDE has been analysed in detail by Orazem et al. [94]. These authors showed that information could reliably and accurately be extracted even at high modulation frequencies (up to 20 Hz). In principle the determination of Sc should only require data at low modulation frequency. They demonstrated that extraction of accurate values for Sc required data which had been recorded over a relatively wide frequency range and had been weighted according to a reliable model for the errors. They also showed that the EHD (7 - Cl) response could be fitted empirically to the form ... 

However, as in univariate calibration, the coefficients obtained using both approaches may not be exactly equal, as each method makes different assumptions about error structure. 

A quarter of a century ago Behnken [224] as well as Tidwell and Mortimer [225] pointed out that the linearization transforms the error structure in the observed copolymer composition with the result that such errors after transformation have no longer zero mean and constant variances. It means that such transformed variables do not meet the requirements for the least-squares procedure. The only statistically accurate means of estimation of the reactivity ratios from the experimental data is based on the non-linear least-squares procedure. An effective computing program for this purpose has been published by Tidwell and Mortimer (TM) [225]. Their method is considered to be such a modification of the curve-fitting procedure where the sum of the squares of the difference between the observed and computed polymer compositions is minimized. 

If the number of samples with characteristics presenting a normal distribution is not significant, then we can have an error structure. This situation can also be due to outliers, i.e. samples that are atypical of the population or that might have been incorrectly labeled or grouped. 

In addition, the variance of impedance measurements depends strongly on frequency, and this variation needs to be addressed by the regression strategies employed. An assumed dependence of the variance of the impedance measurement on impedance values was employed in early stages of regression analysis, and this gave rise to some controversy over what assumed error structure was most appropriate. An experimental approach using measurement models, described in Chapter 21, was later developed, which eliminated the need for assumed error structures. 

Mathematical analysis Heaviside theory Capacitance vs. frequency Vf Fitting Kramers-Kronig analysis, assumed error structure Measurement model, measured error structure... 

Carson et al. °° showed that phase-sensitive detection measurements with a single reference signal biases the error structure of the impedance data due to errors introduced when the square-wave reference signal is in pheise with the measured signal. Modem phase-sensitive detection instruments employ more than one reference signal and may thereby avoid this imdesired correlation. 

Phase-sensitive detection is accurate and can be relatively inexpensive. Modem instrumentation uses more than one reference signal and can mitigate the bias in the error structure. 

---