Statistics and Statistical Programming (Winter 2017)/R lecture outline: Week 7

• correlations
  • cor(): works with two variables, or with more!
  • cor(method="spearman") is useful if you have non-normally distributed data because it is simply rank correlations)
• fitting a linear model with one variable: lm()
  • module formulae, which we've already seen!
  • looking at model objects: summary(); m$<tab> or names(m)
    • m$fitted.values; m$residuals
    • also functions: coefficients(m) (or coef), predict(m), residuals(m) (or resid); confint(m)
    • we can also do these by hand:
      • residuals: mtcars$mpg - m$fitted.values
      • confint: est + 1.96 * c(-1, 1) * se
• plotting residuals:
  • hist(residuals(m))
  • plot against our x: plot(mtcars$hp, residuals(m)
  • QQ-plots with qqnorm(residuals(m))
  • doing a plot with ggplot just involves making a dataset: d.fig <- data.frame(hp=mtcars$hp, resids=residuals(m))
• adding controls: just make our formula more complex
  • update.formula()
  • or just a write a new one
  • adding logical variables: no problem!
  • adding categorical variables: no problem! (I'll explain interpretation later, but i want you to see that this works!)
• generating nice regression plots:
  • one of many options: stargazer(m1, m2, type="text") or type="html"
• interpreting linear models with anova() — i'm not going to walk through the details but the important thing to keep in mind is that although the statistics are different, the p-values are identical!