3.6Markov and Chebyshev inequalities
The idea
Even without the full distribution, Markov and Chebyshev give guaranteed upper limits for how likely a variable is to be far from typical.
Think of it like this: If the average salary is Rs 50,000, at most 1 person in 10 can earn Rs 500,000 or more: you do not need to know anything else.
Why it exists
They give provable bounds when only the mean and variance are known, and they appear in true/false statements.
How it works
- Markov (for X ≥ 0): P(X ≥ a) ≤ E[X]/a.
- Chebyshev: P(|X − μ| ≥ kσ) ≤ 1/k².
- Both hold for every distribution of that kind, so they are usually loose.
- Compare the bound with the real tail in the lab.
- A bound above 1 is true but useless.
Worked example: A non-negative variable with mean 10
P(X ≥ 50) ≤ 10/50 = 0.2. If you also know σ = 5, Chebyshev with k = 2 gives P(|X − 10| ≥ 10) ≤ 1/4. The actual probability is often much smaller; try the exponential distribution in the lab.
Now play with it
- →Move k from 1 to 5 and compare Chebyshev's bound with the actual tail for the normal distribution.
- →Switch to the exponential and look at the Markov bound for a = 3 × mean.
Both bounds hold for every distribution (Markov needs X ≥ 0), so they are usually loose: the real tail is smaller. They are useful because they only need the mean (and the variance for Chebyshev). Chebyshev with k = 2 always gives at most 25%, even though a normal curve has only about 4.6% beyond 2σ.
Check yourself
E[X] = 4 and X ≥ 0. What does Markov say about P(X ≥ 20)?
Exam notes (the dense version)
Markov (X ≥ 0): P(X ≥ a) ≤ E[X]/a. Chebyshev: P(|X − μ| ≥ kσ) ≤ 1/k², for any distribution with finite variance. They give guaranteed (usually loose) bounds when you know only the mean (and variance).
Previous year questions: Random Variables, Expectation & Variance (11)
PYQs are grouped by topic, so this is the full set for Random Variables, Expectation & Variance; try the ones that match this subtopic. Open all 11 on the topic page.