Concepts (plain language)¶
These ideas help you pick the right statistic without heavy theory.
Mean vs median vs Huber¶
Think of 6 measurements and one big error:
[10, 12, 11, 15, 10, 1000]
- Mean: rises a lot because of 1000.
- Median: stays at the real center.
- Huber: behaves like the mean when data is clean, but "brakes" outliers.
Simple rule: - Use the mean if data is clean. - Use median or Huber if there are outliers.
MAD and IQR¶
- MAD (Median Absolute Deviation) measures dispersion around the median.
- IQR (Interquartile Range) measures the spread between 25% and 75%.
If MAD or IQR are high, there is lots of noise or long tails.
R quantiles¶
R defines 9 ways to compute quantiles. StatGuard implements all of them.
- Type 7: R default. Good balance.
- Types 1-3: more discrete (less interpolation).
- Types 8-9: bias adjustments for certain distributions.
If you are unsure, start with type 7.