← Bank of America Interview Insights
I went straight to sample standard deviation and felt good about it, then they asked the unbiased part and I fumbled a bit.
Start by stating the maximum likelihood estimator (MLE) for the standard deviation, which is the square root of the average squared deviations from the known mean of zero. Then explain that this estimator is biased for sigma, and discuss the bias correction using the sample standard deviation with n-1 in the denominator, noting that even that is not perfectly unbiased for sigma.
Pro tip: Mention that while the sample variance with n-1 is unbiased for the variance, its square root is not unbiased for the standard deviation due to Jensen's inequality. This shows depth and avoids a common pitfall.
Since the mean is known to be zero, the MLE for variance is the average of the squared observations: σ̂² = (1/n) Σ x_i². The MLE for σ is the square root of that.
The estimator σ̂² is unbiased for the true variance because E[σ̂²] = σ². This is a key property to mention.
The square root of an unbiased estimator is not unbiased for the square root of the parameter. By Jensen's inequality, E[√σ̂²] < √E[σ̂²] = σ, so the MLE for σ is biased downward.
A common bias-corrected estimator is the sample standard deviation with denominator n-1: s = √(Σ x_i² / (n-1)). However, s is still not perfectly unbiased for σ; it is unbiased for σ only asymptotically. For small samples, a correction factor using the gamma function can be applied.
State that no simple closed-form unbiased estimator exists for σ in general, but the MLE is asymptotically unbiased and consistent. For practical purposes, the MLE or the n-1 corrected version is often used.
AI-generated suggestions, not part of the candidate's original notes. May be inaccurate — verify before relying on them.