Skip to content

Latest commit

 

History

History
450 lines (364 loc) · 22.9 KB

NEWS.md

File metadata and controls

450 lines (364 loc) · 22.9 KB

ModernDive 0.5.0.9000

Major refactoring of inference chapters of book

Old Chapter Structure:

  • Chapter 9 - Confidence Intervals
    1. Bootstrapping a) Data explanation b) Exploratory data analysis c) The Bootstrapping Process
    2. The infer package for statistical inference a) Specify variables b) Generate replicates c) Calculate summary statistics d) Visualize the results
    3. Now to confidence intervals a) The percentile method b) The standard error method
    4. Comparing bootstrap and sampling distributions
    5. Interpreting the confidence interval
    6. Example: One proportion a) Observed Statistic b) Bootstrap distribution c) Theory-based confidence intervals
    7. Example: Comparing two proportions a) Compute the point estimate b) Bootstrap distribution
    8. Conclusion a) What’s to come? b) Script of R code
  • Chapter 10 - Hypothesis Testing
    1. When inference is not needed
    2. Basics of hypothesis testing
    3. Criminal trial analogy a) Two possible conclusions
    4. Types of errors in hypothesis testing a) Logic of hypothesis testing
    5. Statistical significance
    6. Hypothesis testing with infer
    7. Example: Comparing two means a) Randomization/permutation b) Comparing action and romance movies c) Sampling -> randomization d) Data e) Model of H0 f) Test statistic delta g) Observed effect delta* h) Simulated data i) Distribution of delta under H0 j) The p-value k) Corresponding confidence interval l) Summary
    8. Building theory-based methods using computation a) Example: t-test for two independent samples b) Conditions for t-test
    9. Conclusion a) Script of R code
  • Chapter 11 - Inference for Regression
    1. Simulation-based Inference for Regression
    2. Bootstrapping for the regression slope
    3. Inference for multiple regression a) Refresher: Professor evaluations data b) Refresher: Visualizations c) Refresher: Regression tables d) Script of R code

New Chapter Structure:

  • Chapter 9 - Confidence Intervals
  1. Activity: Working with a sample of pennies from the bank. Are they representative of all pennies in the US?
  • a) Question: What do I do when I only have one sample?
  • b) Resampling once (paper slips)
  • c) Resampling 33 times
  • d) Diagrams in Keynote
  1. Computer simulation:
  • a) What is resampling?
  • b) Resampling once
  • c) Resampling 33 times
  • d) Resampling 1000 times
  1. Goal: Generate an estimate that accounts for sampling variation
  • a) Constructing a confidence interval: hide code to shade ci region and to get the actual values.
  • b) Constructing a CI using percentile method
  • c) Constructing a CI using SE method
  1. Framework: Boostrap resampling with replacement
  • a) What dplyr verbs did we use?
  • b) There is only one test framework
  • c) the infer package: make sure to draw parallels between dplyr code and infer verbs
  1. Interpretation:
  • a) 95% speaks to reliability of the process, not about an particular interval. "We are 95% confident"
  • b) What determines the width? Sample size, confidence levels (only int at population variance)
  1. Case study: Comparing two proportions with Mythbusters data
  2. Big picture:
  • a) Does this even work? Comparing sampling and bootstrap distribution. Do this using balls.
  • b) Table of inferential scenarios: Add pennies (mu) and Mythbusters (p1 - p2)
  • c) Why does this work? Theoretical result: Efron. The empirical CDF converges to the population CDF. Bootstrap works for any point estimate
  • d) There's a formula for that! Margin of error using critical values z. Talk about normal distributions.
  • Chapter 10 - Hypothesis Testing
  1. Activity: Shuffling resumes between male and female job applicants
  • a) Question: Are men and women rated for jobs differently?
  • b) Alternate universe: No difference
  • c) What about sampling variation?
  • d) What did we actually observe?
  • e) How likely is this result?
  • f) Diagrams in Keynote
  1. Extension of previous framework/infer
  • a) Revisit verb framework
  • b) Permutation test resampling w/o replacement
  • c) There is only one test framework
  • d) Do activity via infer package
  1. Goal: Choose between two possible truths while accounting for sampling variation
  • a) Conducting a hypothesis test
  • b) Null hypothesis that's assumed
  • c) Null distribution of test statistics: A "alternate universe" distribution
  • d) Observed test statistics
  • e) Definition of p-value
  1. Interpretation:
  • a) Analogy of criminal justice system
  • b) Types of errors: 2x2 table
  • c) A yes/no-type decision: statistical significance via alpha
  1. Case study: Comparing two means with action vs romance movie data
  • Use the "There is Only One Test" framework here
  1. Conclusion
  • a) When is inference not needed: EDA can solve the problem.
  • b) Problems with p-values: p-hacking, hard to understand, ASA statement
  • c) Comparison with confidence intervals. HT yields binary decision, but CI's yield plausible range of estimates. This is statistical vs practical significance
  • d) Table of inferential scenarios: Add action vs romance (mu1 - mu2)
  • e) Why does this work? Theoretical result: Neyman-Pearson lemma (maybe)
  • f) There's a formula for that! t-test. Draw a null distribution with t-distribution superimposed.
  • Chapter 11 - Inference for Regression
    1. Activity: Revisit simple linear regression a) Question: Is there a significant relationship between teaching score and bty score above and beyond any evidence due to sampling variation. b) Review exercise/re-run all code c) Regression table
    2. Computer simulation: a) Permuting the relationship: to do a hypothesis test assuming independence of y & x. a) Bootstraping the rows: Having done HT, generate confidence interval.
    3. Goal: Inferring about the population regression slope
    4. Framework:
    5. Interpretation: a) "You don't have to do any of this! Values in table are given!" No simulations necessary! b) Conditions for inference: residual and partial residual plots, assumption of indepdence.
    6. Case study: Multiple regression example from Ch 7.
    7. Big picture: a) ANOVA = Regression with categorical variables b) Table of inferential scenarios: Add (beta1) c) Why does this work? d) There's a formula for that! Fitted intercept and slope. SE of fitted intercept and slope: observe there is a sqrt(n) in denominator.

All content changes

  • Chapter 6 - Basic regression:
    • Changed skimr::skim() outputs to be of type console.
    • Shortened simple linear regression EDA, in particular geom_jitter() and geom_smooth(se = FALSE)
    • Expanded on "least squares" criteria for "best" fitting line in 6.3.3

ModernDive 0.5.0

Highlights

  • "Data wrangling" chapter now comes after "Tidy data" chapter.
  • Improved explanations and examples of geom_histogram(), geom_boxplot(), and "tidy" data
  • Moving residual analysis from regression Chapters 6 & 7 to Chap 11: Inference for regression
  • Reorganized Chap 8 on Sampling
  • All learning check solutions now in Appendix D
  • PDF build re-added (still a work-in-progress)

All content changes

  • Changed title
    • From: "Statistical Inference via Data Science in R"
    • To: "Statistical Inference via Data Science: A moderndive into R and the tidyverse"
  • Chapter 2 - Getting Started
    • Added subsection 2.2.3 "Errors, warnings, and messages" by @andrewheiss
  • Chapter 3 - Data visualization:
    • Added simpler introductory geom_histogram() and geom_boxplot() examples
    • Started downweighting the amount of data wrangling previews included in this chapter, in particular join.
    • Cleaned up conclusion section
    • Added cheatsheet
  • Switched order of "Chap 4 Tidy Data" and "Chap 5 Data Wrangling": Data Wrangling now comes first
  • Chapter 4 - Data wrangling:
    • Added cheatsheet
  • Chapter 5 - Renamed to "Importing and tidy data"
    • Reordered sections: importing then tidying
    • Added fivethirtyeight::drinks example of "hitting the non-tidy wall", then using tidyr::gather()
    • Made Guatemala democracy score a case study.
    • Added discussion on what tidyverse package is.
    • Moved discussion on normal forms to Ch4: Data Wrangling - joins.
    • Moved discussion on identification vs measurement variables to Ch2: Getting started with data.
  • Chapter 6 - Basic regression:
    • Moved residual analysis to Chapter 11
  • Chapter 7 - Multiple regression:
    • Moved residual analysis to Chapter 11
  • Chapter 8 - Sampling: Major refactoring of presentation/exposition; see below
  • Chapter 11 - Inference for regression:
    • Moved residual analysis from Chapter 6 & 7 here
  • Moved all Learning Check solutions to Appendix D

Chapter 8 Sampling Refactoring

Old chapter structure:

  1. Introduction to sampling a) Concepts related to sampling b) Inference via sampling
  2. Tactile sampling simulation a) Using the shovel once b) Using the shovel 33 times
  3. Virtual sampling simulation a) Using the shovel once b) Using shovel 33 times c) Using shovel 1000 times d) Using different shovels
  4. In real-life sampling: Polls
  5. Conclusion a) Central Limit Theorem b) What’s to come? c) Script of R code

New chapter structure:

  1. Activity: Sampling from a bowl a) Question: What proportion of this bowl is red? b) Using shovel once c) Using shovel 33 times
  2. Computer simulation: a) What is a simulation? We just did a "tactile" one by hand, now let's do one using the the computer b) Using shovel once c) Using shovel 33 times d) Using shovel 1000 times e) Using different shovels
  3. Goal: Study fluctuations due to sampling variation a) You probably already knew: Bigger sample size means "better" guess. b) Comparing shovels: Role of sample size
  4. Framework: Sampling a) Terminology for sampling (population, sample, point estimate, etc) b) Statistical concepts: sampling distribution and standard error c) Computer's random number generator
  5. Interpretation: a) Visual display of differences
  6. Case study: Obama poll
  7. Big picture: a) Table of inferential scenarios: Add bowl and obama poll (both p) b) Why does this work? Theoretial result: CLT c) There's a formula for that: SE formula that has sqrt(n) at the bottom d) Appendix: Normal distribution discuss

ModernDive 0.4.0

Highlights

  1. The infer package is ready for prime-time! Thus we made a first pass at incorporating it into the book in Chapters 9 and 10 on confidence intervals and hypothesis testing!
  2. Chapter 12 on "Thinking with Data" now includes a case study using the Seattle house prices dataset on Kaggle.com. Chapters 3 and 4 from new "Modeling with Data in the Tidyverse" DataCamp course by Albert Y. Kim are based on this analysis!
  3. Speaking of DataCamp, we point readers to various DataCamp courses that directly align with various chapters in the book!
  4. We significantly cleaned up Chapter 8 on sampling! In particular: adding a 2013 Obama approval rating poll example to tie in with our sampling bowl tactile and virtual simulations and making it very clear that ultimately we are performing statistical inference via sampling.

All content changes

  • Introduction: Added section on correspondence of chapters to various DataCamp courses. Furthermore, links to relevant DataCamp course are included at the outset of each chapter.
  • Chapter 3 - Data visualization:
    • Added simplified geom_jitter() example
    • More explanations for how whiskers and outliers are constructed in geom_boxplots
    • Added summary of table of all 5 named graphs
  • Chapter 4 - Tidy data:
    • Added section on importing Excel data via RStudio
    • Added example of tidy vs non-tidy: fivethirtyeight::drinks
  • Chapter 5 - Data wrangling:
    • Added computing available seat miles data wrangling case study
    • Abandoned "5 Main Verbs" 5MV notion
    • Added _join() and group_by() multiple variables
  • Chapter 6 - Basic regression:
    • Clarified explanations of indicator/dummy variables when using categorical variable in regression.
    • Expanded "Correlation is not necessarily causation" subsection with example of "does sleeping with shoes on cause headaches?" including causal diagram
    • Introduced concept of a "wrapper function" when introducing moderndive::get_regression_table() function
    • Replaced all base::summary() with skimr::skim() for quick numerical summaries
  • Chapter 7 - Multiple regression:
    • Changed all "everything else being equal" interpretation statements with "taking into account/controlling for all other variables in our model"
  • Chapter 8 - Sampling:
    • Significantly cleaned up sampling terminology and definitions and made more clear that we are sampling for inference
    • Cleaned up section and subsection structure to be much cleaner:
      1. Tactile sampling simulation
      2. Virtual sampling simulation
      3. In real-life sampling: Introduced example of 2013 Obama approval rating poll and then tie everything with sampling bowl.
  • Major overhaul: Chapter 9 - Confidence intervals
    • infer package now being ready for prime-time, we made first pass at incorporation into book.
  • Major overhaul: Chapter 10 - Hypothesis testing
  • Chapter 11 - Inference for Regression
    • Added a simple linear regression example using the infer package
  • Major overhaul: Chapter 12 - Thinking with data
    • Added case study of Seattle house prices dataset from Kaggle, which is now available in house_prices dataframe in moderndive package.
      1. Chapters 3 and 4 from new "Modeling with Data in the Tidyverse" DataCamp course are based on this analysis
      2. Includes a discussion on the importance of log10-transformations
      3. Introduces modeling/regression for prediction: predicting house prices
    • Laid outline for "effective data storytelling" using fivethirtyeight data and added one small example using US births data
    • At the beginning of chapter, we now come full circle and revisit the discussion on the ModernDive flowchart in the introduction.

Other changes

  • Updated moderndive package on CRAN to 0.2.0. See NEWS.md

ModernDive 0.3.0

Content changes

  • Reorganized chapter sequencing according to flowchart at top of Section 1.1
  • Chapter 2 - Getting Started: Added more explanation on R packages, including analogy for install.packages() and library() (akin to downloading apps onto phone)
  • Added "Data Modeling" portion to book
    • Chapter 6 - Basic regression: one numerical explanatory variable, correlation, one categorical explanatory variable)
    • Chapter 7 - Multiple regression: two numerical explanatory variables, one numerical and one categorical, interaction effects, Simpson's Paradox
    • Uses new moderndive package, which includes get_regression_table() and get_regression_points() wrapper functions to simplify outputting of clean regression tables and observed/fitted values + residuals
  • Added "statistical inference" portion to book
    • Added Chapter 8 - Sampling (still under construction) using sampling bowl
    • Chapters 9 and 10 on confidence intervals and hypothesis testing have not yet been updated, as we were awaiting the now launched package: infer: A tidyverse-friendly R package fo statistical inference
    • Added Chapter 11 - Inference for regression (still under construction), where we'll revisit the regression models fit in Chapters 6 & 7

Other changes

ModernDive 0.2.0

Content changes

  • Incorporated feedback from consultations with Prof. Yana Weinstein, cognitive psychological scientist and co-founder of The Learning Scientists.
  • Restructured/revamped chapters
    • Chapter 1: Introduction
    • Chapter 2: Getting Started New chapter added meant for new R users/coders, including
      • Discussions on R vs RStudio and how to install both (with support videos)
      • A "How do I code in R?" section with links to DataCamp.com courses that covers the console, data types, vectors, factors, data frames, boolean operators, functions etc
      • Thorough discussion on R packages
      • An end-to-end starter example analysis of the data frames in the nycflights13 package using the console, View(), glimpse() etc.
    • Chapter 3: Data Visualization via ggplot2 now first non-intro chapter.
      • Replaced Menard's "Napolean's March on Moscow" with Hans Rosling's (RIP) "Gapminder" plots as introductory example to Grammar of Graphics.
      • Added geom_col() for making barcharts when data is pre-tabulated, instead of using geom_bar(stat="identity")
    • Chapter 4: Tidy Data via tidyr bumped back. Added sections on converting from wide to long/tidy format and importing CSV's
    • Chapter 5: Data Manipulation Wrangling via dplyr
    • Chapter 6: Data Modeling using Regression via broom bumped up from end of book to here given its pedagogical importance, added notes on viewing regression in a prediction framework.
    • Chapter 7-9: Sampling, Hypothesis Testing, Confidence Intervals Mostly unchanged for now; see pending changes section below.

Technical changes

  • Book is now hosted on ModernDive.com
  • Development version now on original ModernDive site https://ismayc.github.io/moderndiver-book/
  • Added links to digital copies and source code of all past versions of ModernDive in Chapter 1.
  • Cut build/compilation time of book from ~20 minutes to ~1 minute
  • Disabled gitbook PDF output

Pending changes for next version

  • Chapter 6: Data Modeling using Regression via broom
    • Better treatment of experimental design and its effect on bias/causation than currently exists in chapter.
    • Examples of regression with categorical predictors with 3 or more levels.
    • Multivariate regression, in particular the following predictor scenarios: 2 numerical, 2 categorical, and 1 numerical + 1 categorical
    • Interaction effects
  • Chapter 7-9: Sampling, Hypothesis Testing, Confidence Intervals have largely not been updated, pending developments of infer: A tidyverse-friendly R package fo statistical inference

ModernDive 0.1.3

  • Attempting to fix Shiny app in Figure 6.2 appearing as white box in published site noted here
    • Reverted to using screenshot with link instead
  • Updated link to dplyr cheatsheet and ggplot2 cheatsheet
  • Began adding DataCamp chapters as Review Questions to the end of Chapters 3 and 4 (More to come)
  • Updated link to MailChimp
  • Fixed wording in a few Ch 3 Learning Checks

ModernDive 0.1.2

  • Converted last updated in index.Rmd to inline instead of R chunk
  • Fixed edit link to point to moderndive-book GitHub repo instead of moderndive-source repo
  • Fixed broken links to script files at the end of Chapters 4-9
  • Added purl=FALSE to chunks that do not contain useful code to the reader
  • Attempting to fix Shiny app in Figure 6.2 appearing as white box in published site noted here

ModernDive 0.1.1

  • Fixed the problems of chapter cross-references not working by removing the backticks in chapter names
    • Issue created on bookdown here
  • Looked for typos throughout all chapters
  • Added coggle diagrams to Chapter 4 and Appendix B
  • Followed the same format of having a Conclusion section at the end of each chapter
  • Fixed $T$ distribution plot with histogram in Chapter 7
    • May be weird issue with cache = TRUE that incorrectly plotted values on 1/10^th^ the correct scale
    • Will need to keep an eye on it going forward
  • Fixed typo on Reach for the Stars chapter name

ModernDive 0.1.0

  • Fiat Lux!
  • Basic chapter structure in place
  • First pass at II Inference section (Chapters 6-9) complete
  • First revisions of I Data Exploration (Chapters 3-5) nearly complete