Confidence levels using noral approxiation to modified t -Statistics for dependent Variables
- 1 August 1981
- journal article
- research article
- Published by Taylor & Francis in Journal of Statistical Computation and Simulation
- Vol. 13 (2) , 131-148
- https://doi.org/10.1080/00949658108810484
Abstract
Dependence between the variables can seriously affect the confidence level when the traditional t-statistic is used to cornstruct confidence limits for the expectation of the variables. The objective of the present paper is to Investigate the normal approximations fo possible modifications of the t-statistic with respect to their use for the construction of confidence intervals. The methods consist in the replacement of the usual variance estimator by a linear combination of estimated covariances, though with different weights and with different rationales. The methods are shown to give morc correct confidence levels than the unmodified t-statistic. Comparing the methods. merits for either one are found. The choice is to some extent dependent upon the underlying model. Some indications about the sample size necessary for the approximation to work at certain levels of satisfaction are given.Keywords
This publication has 6 references indexed in Scilit:
- Testing the Mean of a Normal Population Under DependenceThe Annals of Statistics, 1978
- The robustness of the two—samplet—test over the Pearson systemJournal of Statistical Computation and Simulation, 1978
- The Comparison of Means When Samples Consist of Spatially Autocorrelated ObservationsEnvironment and Planning A: Economy and Space, 1975
- Relation between the shape of population distribution and the robustness of four simple test statisticsBiometrika, 1975
- Effect of Dependence on the Level of Some One-Sample TestsJournal of the American Statistical Association, 1971
- The Statistical Analysis of Series of EventsPublished by Springer Nature ,1966