AI summary by TwinMind
Maximum Likelihood, clearly explained!!!
StatQuest with Josh Starmer
6m 12s
Introduction and Setup
- StatQuest educational video from UNC Chapel Hill genetics department
- Topic: Maximum likelihood estimation concept
- Example dataset: mouse weights
- Goal: find optimal way to fit distribution to data
Distribution Types Overview
- Multiple distribution types for different data types:
- Normal distribution
- Exponential distribution
- Gamma distribution
- Many others available
- Benefits of fitting distributions:
- Easier to work with than raw data
- More general - applies to all experiments of same type
Normal Distribution Characteristics
- Assumption: mouse weights normally distributed
- Key properties:
- Most measurements close to mean/average
- Relatively symmetrical around mean (not perfectly symmetrical but not heavily skewed)
- Come in various shapes: skinny, medium, large
- Data validation: observed mouse weights show expected normal distribution patterns
Maximum Likelihood Process for Mean
- Started with arbitrary normal distribution placement
- Initial poor fit: distribution average (black dotted line) far from actual data average
- Low probability/likelihood of observing actual weights with poorly positioned distribution
- Shifted distribution so mean matched average weight
- Result: higher probability/likelihood of observing actual measurements
- Continued shifting distribution across all possible locations
- Plotted likelihood vs distribution center location
- Maximum likelihood estimate = location maximizing likelihood of observed weights
- For normal distribution: distribution mean = data mean
Maximum Likelihood Process for Standard Deviation
- After determining optimal mean, calculated maximum likelihood estimate for standard deviation
- Plotted likelihood vs different standard deviation values
- Found standard deviation value maximizing likelihood of observed weights
- Final result: normal distribution fitted using maximum likelihood estimates for both mean and standard deviation
Terminology Clarification
- Everyday usage: probability and likelihood mean same thing
- Statistics context: likelihood specifically refers to finding optimal parameter values (mean, standard deviation, etc.) for distribution given observed measurements
- This process = fitting distribution to data
