Biometrical Letters vol. 45(2), 2008, pp. 1-8
|
ROBUSTIFICATION OF CONFIDENCE INTERVALS FOR THE MAXIMUM POINT OF A QUADRATIC REGRESSION AGAINST AUTOCORRELATION D. Kozioł1, W. Zieliński2 1Warsaw University of Life Sciences, Department of Agricultural Economics and IER, Nowoursynowska 166, 02-787 Warsaw, Poland, e-mail: dorota.koziol@ekonometria.info 2Warsaw University of Life Sciences, Department of Econometrics and Statistics, Nowoursynowska 166, 02-787 Warsaw, Poland, e-mail: wojtek.zielinski@statystyka.info |
In this paper the problem of interval estimation of a maximum point of a quadratic regression function in the case of correlated random errors is considered. An approximate Student confidence interval for the maximum point is highly unrobust against autocorrelation (Kozioł and Zieliński, 2003). In the paper a method for robustification of this interval is presented. This solution relies on modification of the data after which the minimum confidence level is close to the nominal one.
quadratic regression, autocorrelaction, maximum of quadratic regression, confidence interval, robustness
