Question
A text's total error count follows . One motivating approximation uses 100 pages, 100 lines on each page, and 100 character positions per line: one million independent, identically distributed Bernoulli error opportunities. For a portion A with k character positions, the approximate mean error count is . Two proofreaders independently inspect the text. Each error is detected independently by reader i with probability , i=1,2. Define as errors found only by reader 1, as errors found only by reader 2, as errors found by both, and as errors found by neither. Estimate the distribution of and the benefit of an independent third reader who detects each error with probability .
(a) Specify the joint law of , , , .
(b) Verify and .
(c) Estimate , , by replacing with the observed X_i.
(d) Estimate the mean number of errors missed by both readers.
(e) With , , , estimate the distribution of .
(f) For these counts, an independent third reader has =0.9. By how much does the expected missed-error count decrease?
(a) Joint distribution
Let be the total count. Each error enters exactly one of four categories, with probabilities
Conditional on , the four counts have a multinomial distribution and sum to . Without conditioning on , Poisson splitting gives mutually independent counts . Their joint probability is
To see the independence, multiply the Poisson probability for by the multinomial probability for those category counts. The factorial for cancels, and factors the exponential into four terms. Independence here is unconditional; fixed-total multinomial counts are generally dependent.
(b) Ratios of means
Writing and assuming ,
(c) Estimates of the detection probabilities and total mean
For observed counts with , replacing means by counts gives
Reader 1's probability is estimated using the proportion of reader 2's discoveries also found by reader 1. The formula for follows from . These are model-based plug-in estimates, not assertions about a known total number of typos. The zero-overlap case makes this estimation formula unavailable; do not divide by zero or infer that there are no unseen errors.
(d) Mean number missed by both
The resulting estimated law is .
(e) Estimated distribution for the supplied counts
For ,
Thus the plug-in distribution for is Poisson with mean 45:
The estimate does not say that exactly 45 errors remain. It is a distribution with estimated mean and variance 45.
(f) Benefit of a third independent proofreader
Let be the number still missed after the third reader. Given , . Thinning therefore gives . With , the estimated remaining mean is
The expected missed-error count decreases by errors, a 90% reduction. These are expected counts, so they need not be integers. This conclusion assumes independent detection with the same stated probability for every remaining error.
Evidence boundary
This is the six-part March 10, 2011 Columbia IEOR 4106 version. The final part is labeled (f) here to distinguish the source’s two successive (e) labels. Counts 60, 30, 40 and detection probability 0.9 are supplied by the exercise. The Poisson laws and plug-in estimates rely on independent, homogeneous error detection; estimated means are not observed remaining counts. The formulas for parameter estimation require positive observed overlap.
Sources
Эти источники подтверждают изложенные выше концепции и методы.