AI summary by TwinMind
Quantiles and Percentiles, Clearly Explained!!!
StatQuest with Josh Starmer
6m 30s
Introduction and Context
- StatQuest by Josh Starmer from UNC Chapel Hill Genetics Department
- Topic: quantiles and percentiles clearly explained
- Research revealed complexity: every webpage had different explanations
- Two main problems:
- Difference between quantile definition vs practical usage
- Multiple calculation methods (R's quantile function has 9 different methods)
Technical Definition of Quantiles
- Example dataset: gene expression measurements from 15 genes
- Median definition:
- Median value = 4.5
- 50% of genes have higher expression, 50% have lower
- Median is a quantile because it splits data into equal-sized groups
- 7 data points below line, 7 data points above line
- Labeled as 0.5 quantile or 50% quantile
Quartiles as Quantiles
- Added two more lines to create four equally-sized groups
- 25th percentile (0.25 quantile):
- Value = 2.5
- 25% of points are less than this value
- 75th percentile (0.75 quantile):
- Value = 7.3
- 75% of points are less than this value
- General definition: quantiles are lines dividing data into equally-sized groups
Percentiles vs Quantiles (Technical vs Practical)
- Technical definition: percentiles divide data into 100 equally-sized groups
- Practical usage: terminology much more flexible
- Dataset too small for 100 groups, but still use percentile terminology:
- 50th quantile = 50th percentile
- 75th quantile = 75th percentile
- 25th quantile = 25th percentile
Individual Data Point Quantiles
- Each data point gets its own group classification
- Bottom data point: 0% quantile (zeroth percentile) - no values less than it
- Calculation method: (number of points less than value) ÷ (total points)
- Examples:
- 1÷15 = 7% quantile (seventh percentile)
- 3÷15 = 20% quantile (twentieth percentile)
Calculation Methods and Limitations
- Multiple calculation methods exist (R has 9 different approaches)
- Each method produces slightly different results
- Small datasets: don't rely heavily on quantiles (high variability between methods/samples)
- Large datasets: all methods give fairly similar results
Future Content Preview
- Upcoming StatQuest topics:
- Quantile-quantile plots
- Quantile normalization
- Planned as series due to topic importance and frequency of use
