SOME NON-PARAMETRIC RESULTS FOR EXPERIMENTAL DESIGNS,
Abstract
In experimental designs, the quantities investigaed are often grouped into blocks as a method of obtaining a higher precision for the experiment. This grouping may result in high correlation among observations within the same block. Also there may be substantial variance differences between blocks. Then the t-statistic is not necessarily applicable for comparing the effects of the treatments under investigation. This paper presents some nonparametric results which are usually valid for a well known type of experimental design if there is statistical independence among blocks (number of blocks > or =4). These non-parametric results are reasonably efficient, compared to those based on the t-statistic, for the case where the totality of observations are independent, normally distributed, and have the same variance. High precision can sometimes be obtained by designing the experiment to yield large positive correlation within blocks and then using the non-parametric results. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- Oct 16, 1950
- Accession Number
- AD0603894
Entities
People
- John E. Walsh
Organizations
- RAND Corporation