Please enable JavaScript.

Coggle requires JavaScript to display documents.

PhD Introduction, Concluding statement This manuscript demonstrates why…

- - - - Why don't these methods work, why is this a stumbling block, why haven't we made progress in this area?
      - Researchers overgeneralise threshold based variable selection to inferential models, due to the conflation between predictive and inferential models
        
        More recent literature demonstrates that inferential and predictive models are definitely not the same.
        
        Concluding statement
        This manuscript demonstrates why researchers should be cautious overgeneralising the threshold
        based variable selection approach presented in Dormann et al. (2013) when their research goals
        are inferential. We quantify and present the potentially severe impacts that collinearity has on
        standardised parameter estimates under a regression framework using simulated and real-world datasets. We show why a variable selection approach is a potentially undesirable technique for
        dealing with collinearity in inferential models. Finally, we propose a potential alternative that builds upon Sewall Wrights path analytic modelling approach by including mediated associations
        between independent variables in parameter estimate calculations.
        
        Need more detail on the ’how’. So in this study we undertake a targeted simulation study
        which allows use to separate inference and prediction so we can assess the impacts of collinearity.
        
        I think don’t lean to heavily on Dorman here. I think there are two keys you are advancing our understanding
        
        (1) you demonstrate that the impact of collinearity on accurate inference is far more perverse and severe than previously believed and
        
        (2) you demonstrate an approach, through the use of ’mediated associations’ whereby the severe impacts of collinearity on inference can besubdued.
        
        I think need to also mention the cases studies here and you can flag the results a bit
        
        We apply our approach to several real world data sets to show how the use of mediated associationto correct/control for collinearity changes are parameter estimate s (..?allows for more accurate inference?)
        
        Confirm Sewall Wrights key objectives
        
        Finally, we propose a potential alternative that builds upon Sewall Wrights path analytic modelling approach by including mediated associations
        between independent variables in parameter estimate calculations.
        
        We need to be specific
        
        We hypothesised that it could be possible to make better inference utilising path models and mediation techniques, that we could improve inference due to...
        
        What is the theoretical grounding?
        
        The focus of inference is to understand the data generation mechanism. It goes against the objective to remove variables.
- - - - Applied researchers have broadly and incorrectly applied the r > |0.7| threshold approach to deal with collinearity for predictive AND inferential research.
        
        Methods like booting out variables undermines the foundations of inferential modelling
        
        Remove important relationships
        Introduces bias to parameter estimates
        
        However
        
        Lack of understanding of the difference between predictive models and inferential models (predictive models and inferential models conflated)
        
        The lack of understanding has the potential to have serious implications for research findings, particularly for inference
        
        1 more item...