Choice of experimental design

The analytical recovery and the method precision (relative standard deviation of individual results) should be determined from experiments conducted  

These experiments are designated as phases II and III of a method validation or an analyst familiarization for those using the approach recommended in the USDA/FSIS Chemistry Laboratories Guidebook (see QA section of the posted methods)7 The results provide an assessment of the recovery (trueness) and the analyst precision attained with the method under routine conditions of use. In addition, the data generated may be used to calculate statistical estimates of the reliability of the results, including estimates of MU. i  

Obvious questions are how many replicates should be included in the design and also how many analytical runs should be completed. Eurachem guidance suggests a minimum of 10 replicates for recovery (accuracy) and precision. A collaborative smdy design is recommended to include a minimum of five materials (matrices) at three concentrations, in blind duplicate, which means that each participating laboratory produces 10 results for each concentration. However, when one considers the recommendation that at least six different sources of matrix should be used in validation for methods for veterinary drug residues in foods and that the validation should include analyses conducted on multiple days, with inclusion of other variables, such as analyst, equipment, and reagents, it is obvious that 10 replicates will prove insufficient to provide the necessary data for a suitable assessment of 

The underlying concept of the validation experiments is to provide a prediction of the performance to be expected over an extended period in routine use, and a minimal dataset will not satisfy this expectation. Therefore, a typical validation design will include the six different sources of matrix, usually at each of three concentrations bracketing the MRL, repeated as analyst spikes in three or four analytical runs, followed by one or two additional runs where the materials are provided as unknowns (blind) to the analyst. The design is usually repeated for each required matrix (e.g., each species-tissue combination) for the initial target species and may be also be required when the method is applied routinely to other species. However, when there are obvious commonalities (such as tissues from different ruminants), method extension may require only a reduced dataset, based on experience with the method. 

A typical experimental design to assess method performance, particularly recovery and repeatability for an analyte for which a regulatory limit has been established, would therefore include the following experiments  

Many experiments could be carried out either as paired or unpaired studies. For example the rifampicin/theophylline experiment (Table 6.1) was performed on an unpaired basis -15 people received one treatment and a separate group of 15 received the other. This is referred to as a parallel groups trial. We could have used a paired structure, with 15 subjects receiving one treatment on one occasion and the other treatment at some other time (a cross over trial). The paired alternative would almost certainly have been a lot more powerful. However, it does not follow automatically that we should always be looking for a paired experimental design. The following points need to be born in mind  

The use of a paired design produces data that can be analysed by the more powerful paired t-test, whereas data from an unpaired experiment can only be analysed by the less powerful two-sample /-test. 

Greater practical difficulties In a paired design, each subject has to be studied twice. This may be slower to implement, especially if you need to leave a significant period of time between the two stages of the study. With human studies, there is also the problem that people may be less likely to volunteer if they know they will be experimented upon twice instead of just once. 

Greater problems in the case of data loss If we used an unpaired design, we might find ourselves unable to obtain a measurement from one of our subjects. In that case, we would be left with 15 observations in one column, and only 14 in the other. With a two-sample /-test, we would still be able to use all of the data obtained. However, if we were performing a paired study (and presumably a paired f-test), we would not only lose that data point, but additionally its accompanying paired value would become useless and would have to be discarded. 

---

The methods and examples outlined in this chapter show how multivariate statistical tools may assist in this context, both for the exploration of the reaction space and in the experimental space. Proper designs ensure that sufficiently large variations of the experimental systems are covered by the experiments. The number of necessary experiments can be kept low by a proper choice of experimental design. A properly designed experiment permits systematic variations of the experimental results to be modelled so that these variations can be traced to the perturbations of the experimental system. 

As a general rule many of the complexities of analysing transient kinetic measurements may be circumvented by careful choice of experimental design. The objective is to employ conditions, where possible, under which the transient follows an exponential time course (or the sum of a small number of exponentials). In this circumstance the kinetic process can be described by a single parameter, a rate coefficient, which is independent of the amplitude of the transient. 

The choice of model in the analysis of experimental data is closely tied to the choice of experimental design. We have so far used mainly polynomial models and for mixture experiments we have used the canonical forms of the polynomial equations. These models are usually, but not always, the most appropriate for analysing mixtures. In this section we will briefly describe some others and indicate circumstances where they may be useful. 

D. Choice of Experimental Design According to Model Proposed 502... 

D. R. Cox, P/anning of Experiments,]ohxi Wiley Sons, Inc., New York, 1958. This book provides a simple survey of the principles of experimental design and of some of the most usehil experimental schemes. It tries "as far as possible, to avoid statistical and mathematical technicalities and to concentrate on a treatment that will be intuitively acceptable to the experimental worker, for whom the book is primarily intended." As a result, the book emphasizes basic concepts rather than calculations or technical details. Chapters are devoted to such topics as "Some key assumptions," "Randomization," and "Choice of units, treatments, and observations."... 

As will be discussed in the following section, a variety of experimental designs are available and have been described to conduct such transport studies. While these variations are all available, the choice of a system and design of the experiment will be dictated by the information desired from the study. However, as illustrated in the applications section, if the variables present in the experimental design are taken into proper consideration, it will be possible to extract mechanistic information which is essentially independent of the system used. In this way it should be possible to compare results from one system or laboratory with those of another. 

Unfortunately, other experimental factors, such as contact capacitance at the junction of the cell leads and the measurement system, lead capacitance, and capacitance due to the dielectric properties of the thermostatting medium, may contribute substantially to the parallel capacitance. These effects may be minimized by proper choice of cell design and use of oil rather than water in the thermostatting bath. The art of making ac conductance measurements has been refined to a high degree of precision and accuracy, and detailed discussions of the rather elaborate procedures that are often necessary are available [9,10]. 

In the following paragraphs we will discuss some of the important problems and issues that are faced in designing a MESA system. Some of the most important experimental issues include the choice of material, design and fabrication of the objects, forces, and how the objects are assembled. 

In this article, examples are chosen from recent literature to highlight progress in the NMR study of biological materials. Many emphasise the need for a careful choice of experimental procedure and/or instrumental design. The field is now expanding and too large to cover comprehensively, but several review articles on previous work are available e.g. enzymes [9-12], protein structure [13-16], interactions of biological molecules [17-19], and medicinal chemistry [20]. 

---