General introduction to data viz principles and tools

• lectures
• in-class work
• homework
• student projects & presentations

• Git: distributed version control system
• GitHub: centralized version control server
  • alternatives: BitBucket, GitLab, …
• Git clients: software for working with Git on your computer
  • command-line (e.g. git add foo.rmd)
  • RStudio
  • others (GitHub desktop etc.)

• create repository on Github
• copy repository to local machine
  • git clone
  • RStudio: File > New Project > Version Control > Git > fill in name from “Clone” button on GH

• repeat:
  • pull (fetch and integrate changes from GH, git pull)
    • RStudio: Git panel > click blue down-arrow
  • do stuff (create, edit files, etc.)
  • stage (git add)
    • RStudio: Git panel > click “Staged” button
  • commit (git commit)
    • RStudio: Git panel > click “Commit” icon >
      enter commit message > click “Commit” button (ignore “amend previous commit” button!)
  • push (git push)
    • RStudio: Git panel > click green up-arrow

• set of R packages: https://www.tidyverse.org/
• advantages
  • expressiveness
  • speed
  • new hotness
• disadvantages
  • minor incompatibilities with base R
  • rapid evolution
  • non-standard evaluation

• new verbs
• piping
• tibbles

• filter(x,condition): choose rows equivalent to subset(x,condition) or x[condition,] (with non-standard evaluation)
• select(x,condition): choose columns
  • equivalent to subset(x,select=condition) or x[,condition]
  • helper functions such as starts_with(), matches()
• mutate(x,var=...): change or add variables (equivalent to x$var = ... or transform(x,var=...)

• group_by(): adds grouping information
• summarise(): collapses variables to a single value
• e.g.

x <- group_by(x,course)
summarise(x,mean_score=mean(score),sd_score=sd(score))

• equivalent to

d_split <- split(d,d$var)       ## split
d_proc <- lapply(d_split, ...)  ## apply
d_res <- do.call(rbind,d_proc)  ## combine

• new %>% operator (orig. from magrittr package)
• directs result of previous operation to next function, as first argument
• e.g.

(d_input
    %>% select(row1,row2)
    %>% filter(cond1,cond2)
    %>% mutate(...)
) -> d_output

• extension of data frames
• differences
  • printing
    • only prints first few rows/columns
    • labels columns by type
  • no rownames
  • never drops dimensions (tib[,"column1"] is still a tibble)

• pivot_longer(data,cols, names_to, values_to)
  • see here and here for more info
  • wide to long
  • reshape2::melt()
  • gather() in tidyr pre v 1.0
• pivot_wider(data,names_from, values_from)
  • long to wide
  • reshape2::cast()
  • spread() in tidyr pre v 1.0

• find patterns in data, explore hypotheses
• emphasize robust approaches
• minimize (parametric) assumptions
• John Tukey, William Cleveland (1993)

• evaluate assumptions of a model
  • unbiasedness/goodness of fit
  • homoscedasticity
  • normality
• easily spot deviations
• identify outliers and influential points

• coefficient plots (e.g. dotwhisker package)
• replacement for tables (Gelman et al. 2002)
• also: tests of inference (Wickham et al. 2010)
• Andrew Gelman

• tell an accurate story
• high information density
• Cleveland, Edward Tufte

• grab attention/engage/sell/entertain
• “puzzle” graphics

• present a quick overview of a data set
• user control
• business-oriented

• time dimension
• engage
• allow viewer to drill down
• Dianne Cook