To-do list

[TengMCing]

Apr 21 - 27

1. Summarize the use of residual plot and hypothesis testing for regression diagnostics in books and articles, especially for detecting nonlinearity and heteroscedasticity.

• [x] Applied regression analysis (3rd), 2014 by Draper and Smith
• [x] Introduction to linear regression analysis (5th), 2012 by Montgomery, Peck and Vining
• [x] Applied regression including computing and graphics, 1999 by Cook and Weisberg
• [x] The examination of residual plots, 1998 by Tsai, Cai and Wu
• [x] Exploring partial residual plots, 1993 by Cook
• [x] Diagnostic Value of Residual and Partial Residual Plots, 1987 by Mansfield and Conerly
• [x] Residuals and influence in regression, 1982 by Cook and Weisberg
• [x] Regression diagnostics: Identifying influential data and sources of collinearity, 1980 by Belsley, Kuh and Welsch
• [x] The examination and analysis of residuals, 1963 by Anscombe and Tukey

2. Write the importance and usefulness of plots and tests in regression diagnostics. Discuss the opinions provided by the above literature.

[TengMCing]

Apr 28 - May 4

1. Work on power comparison for the upcoming week.
2. Computer vision model. Detection rate vs. effect size. Check the agreement and disagreement between CV and humans. Attention map of the model.

• [x] Clean up data (Process survey data & pilot data)
• [x] Double-check simulated power of conventional tests
• [x] Power comparision (visual test vs. conventional test)
• [x] Fix issues in effect size calculation?
• [ ] Investigate the difference (gap), give examples (includes plots and descriptions)
• [ ] Explain and argue why the conventional test is too sensitive.
• [ ] Try to make conventional less sensitive (vary alpha). Do they match the visual test result?

[TengMCing]

May 12 - May 19

Override previous tasks. Now focus on:

• [ ] Produce target plots for investigating power of visual tests. Make sure major factors are being considered.
• [ ] Produce plots for change of parameters using the same random noise realisations.
• [ ] Adjust the main argument of the paper -> Conventional test is useful, but its result may not be as important as we think.