Where does coverage collide?

In frequentist statistics, coverage is the long-run proportion of interval estimates that contain the true value of the parameter being estimated: an interval procedure with 95 percent coverage produces intervals that trap the true value in 95 of every 100 repeated samples. Neyman's 1937 Outline of a Theory of Statistical Estimation Based on the Classical Theory of Probability introduced the confidence interval and defined its confidence coefficient as exactly this probability of enclosing the true value. Here coverage is a probability attached to a procedure across hypothetical repetitions, not the money a policy will pay out or the number of times a genome base is read.[1]

Unit

A probability, the fraction of repeated intervals that contain the true parameter value

Recovery means

Not applicable

Measured by

Derived analytically as the interval procedure's confidence coefficient, or estimated by repeated-sampling simulation

In insurance, coverage is the scope of losses that an indemnity policy will pay for: it is fixed by the perils named, the limits, the deductible, and the exclusions written into the contract, and it determines how much of a realised loss the insurer reimburses. Vernon Smith's 1968 Optimal Insurance Coverage analysed how much coverage a risk-averse buyer should rationally purchase given its price, treating the amount of loss indemnified as the choice variable; the underlying industry and legal meaning of the term is much older than this economic treatment. Here coverage is the reach of a contractual promise to pay, not a repeated-sampling probability or a count of sequencing reads.[2]

Unit

The set of losses a policy reimburses, bounded by limits, deductibles, and exclusions

Recovery means

Not applicable

Measured by

Read from the policy contract and priced actuarially by the expected indemnified loss

In genome sequencing, coverage (also called read depth) is the average number of times each base in a target is independently read: sequencing to 30x coverage means each position is spanned by about 30 reads on average, which raises confidence that true bases are told apart from errors. Lander and Waterman's 1988 Genomic mapping by fingerprinting random clones: A mathematical analysis gave the mathematical model relating the number of random fragments, their length, and the expected coverage and gaps of an assembled genome. Here coverage is a physical redundancy of observation over a sequence, not a probability that an interval brackets a parameter or the losses a policy will indemnify.[3]

Unit

A dimensionless multiple (e.g. 30x), the mean number of reads spanning each base

Recovery means

Not applicable

Measured by

Computed as total sequenced bases divided by genome length, or per base from aligned reads