Using Scanty and Fragmentary Data

The data available may be too few, nonrepresentative, censored, or available only in 

1 Some Statistical Consequences oe Small Sample Sizes 

When the data for a situation of specihc interest are inadequate, a common approach is to make some use of more generic information, including possibly information less representative of the situation of specific interest. For example, if the information for a specihc pesticide is inadequate, then some features of an analysis may be based on information from a set of other pesticides, considered to be comparable. This can be appropriate particularly for an early-tier assessment, in which case the criteria may be designed to be protective, so far as we can judge. 

With regard to relevant statistical methodologies, it is possible to dehne 2 situations, which can be termed a meta-analysis context and a shrinkage estimation context. Similar statistical models, in particular random-effects models, may be applicable in both situations. However, the results of such a model will be used somewhat differently. 

In the hrst situation we hope to dehne a generic distribution based on information from multiple studies, and no study is treated as more representative than another, for the situations where the distribution will be used. Generic assumptions may relate to type of distribution or to distribution parameters (e.g., coefficient of variation, skewness, or kurtosis). An important case is the determination of multiplicative safety factor based on a generic coefficient of variation, and assuming log-normality. 

---