Skip to Main Content
 

Global Search Box

 
 
 

ETD Abstract Container

Abstract Header

Analysis of Removable Interaction

Hanook, Sharoon
http://rave.ohiolink.edu/etdc/view?acc_num=case1413761250

Abstract Details

2014, Doctor of Philosophy, Case Western Reserve University, Epidemiology and Biostatistics.
Researchers have always aspired to estimate the most useful and parsimonious model, which helps to obtain the most precise and efficient estimates. Once we have a large number of rows and columns in a factorial design, we have many parameters; so the question that arises is: can we obtain better estimates of the parameters of interest by modeling, or by transformation, or by both? In the presence of non-additivity in the model, Interpretation of the model is not very straightforward and it is not easy to describe the model in a most meaningful manner. The scale of measurement can cause interaction and it can be efficiently detected by using any auxiliary information in terms of weights (if at all available) regarding the row and column factor levels. We propose a test that uses available appropriate weights for Tukey’s one degree of freedom test known as weighted Tukey’s test for removable interaction to isolate removable interaction effect from the essential interaction effect. This test uses row and column level weights to effectively detect removable interaction effects in the presence of negligible essential interaction effect. We also proposed a modification of the weighted Tukey’s test due to Rasch et al (2009), using non-linear regression and testing overall model against the sub-model (i.e. no interaction model) by a likelihood ratio test, and showed with simulation that proposed modification is better than Tukey’s, weighted Tukey’s and un-weighted modified Tukey’s test. Better estimates can also be obtained by a transformation, in this dissertation we are proposing an empirical monotonic transformation motivated by Fisher (1950) to remove non-additivity from a two-way analysis of variance model, given the condition that in the presence of negligible essential interaction, there exists some non-additivity in the model that is either completely or partially removable. This transformation provides satisfactory results in the case of different interaction structure types by eliminating removable interaction and making the model more parsimonious.
Robert Elston, Dr. (Committee Chair)
Mark Schluchter, Dr. (Committee Member)
Jiayang Sun, Dr. (Committee Member)
Jill Barnholtz-Sloan, Dr. (Committee Member)
123 p.
Biostatistics

Recommended Citations

Citations

  • Hanook, S. (2014). Analysis of Removable Interaction [Doctoral dissertation, Case Western Reserve University]. OhioLINK Electronic Theses and Dissertations Center. http://rave.ohiolink.edu/etdc/view?acc_num=case1413761250

    APA Style (7th edition)

  • Hanook, Sharoon. Analysis of Removable Interaction. 2014. Case Western Reserve University, Doctoral dissertation. OhioLINK Electronic Theses and Dissertations Center, http://rave.ohiolink.edu/etdc/view?acc_num=case1413761250.

    MLA Style (8th edition)

  • Hanook, Sharoon. "Analysis of Removable Interaction." Doctoral dissertation, Case Western Reserve University, 2014. http://rave.ohiolink.edu/etdc/view?acc_num=case1413761250

    Chicago Manual of Style (17th edition)

case1413761250
144
© 2014, all rights reserved.
This open access ETD is published by Case Western Reserve University School of Graduate Studies and OhioLINK.