Statistics 541: Checking assumptions

Admistrivia

• Data from Myers is on web page. Like most data, it needs editing before use.

Checking assumptions

• some fun moving points around
• VO2 max (Orginal version, ascii, CSV)
  • parameters and standard errors
  • Which should be X and which should be Y?

• Tukey: peeling the onion
  • Can look at Y vs X plot
  • Creating residuals: Y - Y-hat
  • residuals provide better microscope
  • Easier to see curvature with residuals
  • Easier to see hetroskadasticity with residuals
  • Saving residuals allows checking for normality
  • After removing a problem--look again for even more detailed problems

• Normal probability/quantile plot (M: p 60 - 66)
  • Sorted guess of true value plotted vs sorted data
  • Can work for any distribution
  • straightness easier to detect than correct curvedness
  • Shapes to consider:
    • curved = skewedness
    • s-shaped = tail thickness problems (mostlikely too many outliers)
    • stairsteps = binomial/integer data

• Kolmogorov-Smirinov test
  • Differences in CDF
  • Sensitive for skewedness and binomial data :-)
  • Arbitarilly large outlier is permitted. Hence it is resistant to change. Is this a good thing?
  • Other tests could be done

• Revisit data
  • Other distributions

  ---

  Last modified: Thu Feb 1 08:56:13 2001