Uncertainty estimates in the selected thermodynamic data

The electrode calibration, especially the conversion between measured pH and - logio[H ] is an important source of systematic error. The reviewers have when possible corrected this error, as seen in many instances in Appendix A. 

The assumption of a normal distribution of the random errors is a choice made in the absence of better alternatives. 

One of the objectives of the NEA Thermochemical Data Base (TDB) project is to provide an idea of the uncertainties associated with the data selected in this review. As a rule, the uncertainties define the range within which the corresponding data can be reproduced with a probability of 95% at any place and by any appropriate method. In many cases, the statistical treatment is limited or impossible due to the availability of only one or few data points. A particular problem has to be solved when significant discrepancies occur between different source data. This appendix outlines the statistical procedures, which were used for fundamentally different problems, and explains the philosophy used in this review when statistics were inapplicable. These rules are followed consistently throughout the series of reviews within the TDB Project. Four fundamentally different cases are considered  

---

Data values were extracted from the SOURCE database by suitable software and then screened to identify the "best" (most accurate) values. The selection was based on a comparison of the estimated uncertainties of values for the same substance. The selection also took into account the distribution of virial coefficients with temperature. An appropriate algorithm has been developed at the Thermodynamics Research Center and has been used for several compilations. Briefly, the uncertainty for each data valne in a set was compared to a weighted mean of the uncertainties of all the other values. The weighting factor was an inverse exponential function of the absolute value of the difference between the temperature of the value being evaluated and the temperatures of the other values. The screeiung level, the size of the data set and range of temperatures it covered determined the parameters used in the comparison. Additional details are given in [96-wil/mar]. Selected values are marked with various symbol in the tables of data. 

In view of the uncertainties in the Debye temperature of Th(cr) discussed in detail above, we have preferred to estimate the data from 4 to 20 K without recourse to these uncertain values and have adopted the values suggested to us by Arblaster [2004ARB], who kindly provided us with his recent (so far impublished) assessment of the thermodynamic properties of Th(cr). For this the selected values are the following ... 

---