THE VARIANCE OF MEANS OF SYMMETRICALLY TRIMMED SAMPLES FROM NORMAL POPULATIONS, AND ITS ESTIMATION FROM SUCH TRIMMED SAMPLES. (TRIMMING/ WINSORIZATION I)
Abstract
A statistical analysis is made on the use of the mean of a symmetrically trimmed sample as an indicator of location. The use of the total sum of squares of deviations of the same trimmed sample as an indicator of the variability of the trimmed mean is explored. The increase in variance (of the trimmed mean as compared with the untrimmed mean) when trimming samples from an exactly normal distribution is found to be less than 3%, 6%, 9%, and 14%, respectively, when a total of 1/10, 2/10, 3/10/ or 4/10 of the sample is trimmed away. The loss of normaltheory efficiency is given for all symmetric trimmings of samples of size is less than or equal to 20. The appropriate divisor, by which the trimmed sum of squares of deviations is to be divided to obtain an estimate of the variance of the trimmed mean, is given in tabular summary.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jul 01, 1961
- Accession Number
- AD0263101
Entities
People
- Donald H. Mclaughlin
- John W. Tukey
Organizations
- Princeton University