measure_reliability
How much of a measured rate is the provider and how much is the sample it was measured on. Route here whenever a rate is about to be compared, ranked, penalised or rewarded, because a rate computed on a small denominator moves on its own and a comparison that does not know by how much is reading noise as performance. Declare `numerator` and `denominator`, the counts the rate was computed from, and `between_provider_variance`, the variance of the true rates across the providers you are comparing. All three are required and this world ships no default for any of them. The variance is the input that decides the answer and it is the operator's to declare: it is a property of the group being compared, not of any one provider, and a defaulted value would be this world choosing the conclusion. Send it as a bare number to declare it, which is what most callers do and what keeps that input an assumption. Where this world has measured the spread for you, send it instead as an object naming hospital_variance and the measure_id you read it under, and that input alone is stamped measured with the tool and measure in the receipt. Both shapes are on the schema. THE OBJECT FORM IS CHECKED RATHER THAN TAKEN. hospital_variance serves several fields for one measure, on different scales, and a value carried across from the wrong one is a plausible answer that is wrong by orders of magnitude. So this tool re-runs that tool for the measure_id you named and compares the value you sent against every field it serves. That comparison is arithmetic over a figure this world produced rather than a reading of what you meant, which is why no scale argument is asked for: this world could verify a declared scale no better than it could verify the value. Every outcome names the FIELD your value matched. A value matching the field on the measure's own published scale is refused, naming it. A value matching the other proportion-scale conversion, on a measure whose units this world can state, does not answer at the value you sent: what comes back is a payload that declines it, carries no reliability where an answer would carry one, and serves under a key of its own what the field that measure calls for would give, so you can see the figure you were reaching for. A value matching none of the served fields did not come from that tool for that measure and is refused, with no field and no measure guessed for it. Where this world holds no specification for a measure's units, it names the field you matched, serves the other conversion beside it, and picks neither. Serves the reliability itself, the within-provider and between-provider variance it was computed from, the smallest denominator that would reach each of several stated reliability levels, and the between-provider variance at which each of those levels would be reached at your own counts, which is the range over which a conclusion survives rather than the single point you asked about. It serves no verdict. Whether a reliability is high enough to act on is a decision about consequences this world cannot see, and a key asking for one is refused rather than answered and ignored. Nothing here is ordered, costed or recommended.
Use this form for new arguments. Follow concrete links for known values. Fetch-only readers may be unable to submit new URLs.
Exact input contract
{
"additionalProperties": false,
"properties": {
"between_provider_variance": {
"description": "Variance of the true rates across the providers being compared, ON THE PROPORTION SCALE. A property of the group, declared by whoever chose the group. Two shapes, and which one you send decides this input's basis and nothing else.",
"oneOf": [
{
"description": "A bare number is your own declaration. This input's basis stays assumption, and it names no source because a declaration has nothing to name.",
"type": "number"
},
{
"additionalProperties": false,
"description": "An object is a figure this world measured, passed back in. This input's basis becomes measured and the receipt names the tool and the measure, so the claim is checkable by re-running that tool with that measure_id and comparing. The top-level basis does not move: the numerator and denominator are still yours, and a figure carries the weakest basis among its inputs.",
"properties": {
"from": {
"description": "The tool that measured it. Only this world's own hospital_variance earns a measured basis here, because only it names a measure this world can re-read.",
"enum": [
"hospital_variance"
],
"type": "string"
},
"measure_id": {
"description": "The measure_id you passed to hospital_variance, spelled the same way. Required rather than optional: a measured basis that names no measure is checkable against nothing, and a variance in one measure is not a variance in any other.",
"type": "string"
},
"value": {
"description": "The proportion-scale figure from that tool's answer, which is the field named on_the_proportion_scale_if_score_is_a_percentage or the one named on_the_proportion_scale_if_score_is_a_rate_per_thousand, whichever matches the units of that measure's own score column. NOT the field named between_provider_variance: that one is in the square of the measure's own units, and this tool computes the sampling variance of a proportion. Which of them you sent is CHECKED here rather than trusted: that tool is re-run for the measure_id you named and this value is compared against every field it serves, so a value from the wrong field or from another measure is named as such instead of being turned into a reliability. The comparison allows for a figure carried across at the precision your tooling printed it at; it does not allow for a figure from somewhere else.",
"type": "number"
}
},
"required": [
"from",
"measure_id",
"value"
],
"type": "object"
}
]
},
"denominator": {
"description": "Cases eligible for the measure, as counted.",
"type": "number"
},
"numerator": {
"description": "Cases meeting the measure, as counted.",
"type": "number"
}
},
"required": [
"numerator",
"denominator",
"between_provider_variance"
],
"type": "object"
}