Upload
irina-koksharova
View
209
Download
0
Embed Size (px)
Citation preview
professor of Psychology Rand R.Wilcox, University of Southern California
Written by Marina Haldna
To introduce and compare different statistical methods
Introduction Methods Results Conclusion
Typical situatsions : before-after two places with parallel measurements
Every pair of sampled data has a important information to make a correct conclusion
Data are missing at random. The reason why a data point is missing is
not related to its actual value
Simple strategy is to compute a confidence interval using a normal or t-distribution, this approach may be unsatisfactory (Liang et al. 2008)
Even if there are non-missing data, low power arise when sampling from a heavy-tiled distribution (Wilcox, 2005)
Median 20% trimmed mean Difference between the marginal trimmed
means
M1(means) M2(trimmed means) M3 –bootstrap method with trimmed means M4-medians
In terms of efficiency, the median generally performs better than 20% trimmed mean (Wilcox,2006)
To check the properties of the methods, one sample of correlated and an other non-correlated data from bivariate distribution were used
Simulations were repeated 3000 times, sample size was taken 30.
Two different cases for missing dates: 5 from one and 5 from an other group10 from one and 0 from an other group
In terms of Type I errors The method for comparing means can be
unsatisfactory Percentile bootstrap method with 20%
trimmed means performed well
In terms of power the M2 and M3 are more satisfactory
Statistics means never having to say you are certain
Statistics is the art of never having to say you are wrong
(http://www.btinternet.com/~se16/hgb/statjoke.htm)