Contributions to influence diagnostics in nonlinear regression analysis
2013 (English)Licentiate thesis, comprehensive summary (Other academic)
This thesis contributes to influence diagnostics in nonlinear regression analysis. The topic of influence analysis springs from the presence of influential observations. The detection of influential observations is an important part of the statistical paradigm, since influential observations can compromise the reliability of the conclusions and the inference.
In this thesis a graphical aid, used to visually identify influential observations, is proposed and referred to as the Added Parameter Plot. This plot is a nonlinear analogue to the Added Variable Plot used in linear regression. The Added Parameter Plot displays the effect of the individual observations on a certain parameter estimate in the nonlinear regression model. The plot is proven to be a graphical representation of the score test statistic and therefor observations that influence the score test statistic can also be identified.
Three influence diagnostics are also derived. These diagnostics measures the marginal and joint influence of the individual observations on the parameter estimates and the joint influence of the observations on the score test statistic.
Place, publisher, year, edition, pages
Stockholm University, 2013.
Research subject Statistics
IdentifiersURN: urn:nbn:se:su:diva-104635OAI: oai:DiVA.org:su-104635DiVA: diva2:724318
von Rosen, Tatjana