- Bootstrap analysis, both as a tool for nonparametric statistical inference and as a tool for describing sample results stability and replicability,
- using bootstrap for nonparametric statistical inference (Efron, 1985)
- Bootstrapping it has also been advocated as a descriptive tool and an internal replication mechanism for assessing the stability and replicability of sample results of an individual study (Thompson, 1993). This descriptive use of bootstrap is meaningful when our interest may not be about statistical inference but rather about understanding how stable the results may be across repeated sampling.
- distinct differences between the inferential and the descriptive applications of the bootstrap (cf. Thompson, 1993). Larger numbers of bootstrapped resamples are required in inferential (i.e., statistical significance) bootstrap applications as against descriptive
applications of the bootstrap. when invoke the bootstrap to conduct statistical significance tests, extremely large numbers of resamples are required (e.g., 2,000, 5,000).
However, when our application is descriptive, we are primarily interested in the
mean (or median) statistic and the SD/SE from the sampling distribution. These values are less dependent on large numbers of resamples. (Thompson, 1999, p. 50) Of course, in an era of powerful microcomputers, we may elect to use large numbers of resamples even for descriptive applications of the bootstrap because the cost of additional resampling is effectively zero
- bootstrap analysis can be accomplished by using a standard statistical analysis package (e.g., SAS) with only a reasonable amount of effort.
- In Amos, "Perform Bootstrap” needs to be checked. One may choose the percentile confidence interval or the bias-corrected confidence interval and the desired confidence interval level (default is 90%).
- For doing regression analysis in Amos, if we are also interested in the replicability of the sample R2 of the regression model, under the “Output” tab of AMOS “Analysis
Properties,” the option of “Squared Multiple Correlations” was checked, if the model R2 is very high (e.g., R2 = .93), and it is likely to be highly replicable across comparable samples.
- Efron,B., and Tibshirani,R.J. (1993) An introduction to the bootstrop. Chapman&Hall, Inc.
- Jun Shao and Dongsheng Tu (1995) The jackknife and bootstrap. Springer.
- Xitao Fan (2003) Using Commonly Available Software For Bootstrapping In Both Substantive And Measurement Analyses. Educational and Psychological Measurement, v63: 24-50.