2a:null 2b:["$","title",null,{"children":"Coverage of confidence intervals for a proportion - Wald vs Wilson | Mattias Villani | Observable"}] 31:T425,Plot.plot({ caption: plotValidProb ? html` The light gray area marks out the p where np<10 or n(1-p)<10 ` : html``, color: { legend: true, domain: plotValidProb ? ["wald", "wilson", "region where normal approximation is not accurate"] : ["wald", "wilson"], range: plotValidProb ? [colors[0], colors[1], "lightgray"] : [colors[0], colors[1]] }, y: { label: "coverage", domain: [lower_y,1] }, x: { label: "p", domain: [0,1] }, marks: [ Plot.areaY(coveragedata, {filter: plotValidProb, x: "p", y: "invalidp", fill: "lightgray", opacity: 0.3}), Plot.ruleY([confidence/100], {stroke: "gray", strokeWidth: 1}), Plot.ruleY([0], {stroke: "black", strokeWidth: 1}), Plot.ruleX([0], {stroke: "black", strokeWidth: 1}), Plot.lineY(coveragedata, {filter: plotselector.includes("wald"), x: "p", y: "wald", stroke: colors[0]}), Plot.lineY(coveragedata, {filter: plotselector.includes("wilson"), x: "p", y: "wilson", stroke: colors[1]}), ] })32:[] 2e:{"id":"2a76d329c4f2ba09","slug":"ci-proportion-wald-vs-wilson","title":"Coverage of confidence intervals for a proportion - Wald vs Wilson","thumbnail":"972af1ca43674fc31343ef85c30f6b31e347d945c7be21ca563b0fd420d5f48b","access_level":"public","create_time":"$D2025-01-17T11:23:34.161Z","update_time":"$D2025-02-21T10:51:04.018Z","view_time":null,"first_public_time":"$D2025-01-17T11:23:34.161Z","fork":true,"stars":0,"starred":false,"owner":{"type":"team","id":"7592a7ab121f74a0","login":"mattiasvillani","name":"Mattias Villani","avatar_url":"https://avatars-next.observableusercontent.com/a440aaf39a3abefb86cbcc758fd7a5ab42555ae68d6825b52ec006f385a94d1d","tier":"pro_2024"},"version":924,"publish_version":null,"role":null,"authored":false,"authors":[{"type":"individual","id":"56499be8cf43f0ef","login":"mattiasvillani","name":"Mattias Villani","avatar_url":"https://avatars-next.observableusercontent.com/a440aaf39a3abefb86cbcc758fd7a5ab42555ae68d6825b52ec006f385a94d1d","tier":null}],"members":[{"type":"individual","id":"56499be8cf43f0ef","login":"mattiasvillani","name":"Mattias Villani","avatar_url":"https://avatars-next.observableusercontent.com/a440aaf39a3abefb86cbcc758fd7a5ab42555ae68d6825b52ec006f385a94d1d","tier":null,"role":"editor"},{"type":"team","id":"7592a7ab121f74a0","login":"mattiasvillani","name":"Mattias Villani","avatar_url":"https://avatars-next.observableusercontent.com/a440aaf39a3abefb86cbcc758fd7a5ab42555ae68d6825b52ec006f385a94d1d","tier":"pro_2024","role":"editor"}],"comments":[],"custom_thumbnail":false,"trashed":false,"subscribed":null,"forks":0,"collection":{"id":"3247547e8812c274","slug":"statistics","title":"Basic Statistics","owner":{"id":"7592a7ab121f74a0","login":"mattiasvillani","name":"Mattias Villani"}},"type":"notebook","body":{"title":"Coverage of confidence intervals for a proportion - Wald vs Wilson","cells":[{"id":0,"value":"# Coverage of confidence intervals for a proportion - Wald vs Wilson\n","pinned":false,"hidden":false,"mode":"md"},{"id":652,"value":"This notebook investigates the coverage probabilities of two widely used confidence intervals for a proportion ${tex`p`} in the iid Bernoulli model\n${tex.block`X_1,\\ldots,X_N \\mid p \\sim \\mathrm{Bernoulli}(p)`}","pinned":false,"hidden":false,"mode":"md"},{"id":144,"value":"md`\n#### Wald interval\nThe Wald confidence interval with approximate coverage probability ${tex`1-\\alpha`} for a population proportion *p* is\n\n${tex.block`\\hat p \\pm z_{\\alpha/2} \\sqrt{\\frac{\\hat p (1-\\hat p)}{n}}`}\n\nwhere ${tex`\\hat p`} is the proportion in the sample and ${tex`z_{\\alpha/2}`} is obtained from the standard normal distribution as the value that has ${tex`\\alpha/2`} probability to the right of the value. \n\nThe Wald confidence interval is based on a [normal approximation to the binomial distribution](https://observablehq.com/@mattiasvillani/approximating-the-binomial-distribution) which is known to be accurate if both ${tex`np \\geq 10`} and ${tex`n(1-p) \\geq 10`}. \n\n---\n`","pinned":false,"hidden":false,"mode":"ojs"},{"id":391,"value":"md`\n#### The Wilson interval\n\nThe Wilson interval uses the formula:\n\n${tex.block`\\tilde p \\pm z_{\\alpha/2} \\frac{\\sqrt{\\frac{\\hat p (1-\\hat p)}{n} + \\frac{z_{\\alpha/2}^2}{4 n^2}}}{1+z_{\\alpha/2}^2/n}`}\n\nwhere the midpoint of the interval is ${tex.block`\\tilde p = \\hat p + \\frac{z_{\\alpha/2}^2}{2n} \\frac{1-2\\hat p}{1+z_{\\alpha/2}^2/n}`}\n\nThe Wilson interval is a little more complicated and less intuitive formula than the Wald interval, but it is often recommended since it tends to have a more accurate coverage^{[[1](#footnote1)]}; its actual coverage is often closer than the nominal coverage, compared to the Wald interval.\n\n---\n`","pinned":false,"hidden":false,"mode":"ojs"},{"id":5,"value":"viewof settings = Inputs.form([\n Inputs.range([0, 100], {value: 90, label: \"confidence level\", step: 1}),\n Inputs.range([1, 200], {value: 50, label: \"sample size, n\", step: 1}),\n ])","pinned":false,"hidden":false,"mode":"ojs"},{"id":808,"value":"viewof plotValidProb = Inputs.toggle({label: md`Plot region where normal approx is inaccurate`, value: false})","pinned":false,"hidden":false,"mode":"ojs"},{"id":456,"value":"viewof plotselector = Inputs.checkbox([\"wald\", \"wilson\"], {label: \"plot interval type:\", value: [\"wald\"]})","pinned":false,"hidden":false,"mode":"ojs"},{"id":447,"value":"$31","pinned":false,"hidden":false,"mode":"ojs"},{"id":920,"value":"

\n#### Footnotes","pinned":false,"hidden":false,"mode":"md"},{"id":918,"value":"footnote1 = md`1. [^](#paragraph1) \nAndersson, P. G. (2023). The Wald confidence interval for a binomial p as an illuminating “bad” example, *The American Statistician*, 77:4, 443-448.\n`","pinned":false,"hidden":false,"mode":"ojs"},{"id":12,"value":"
\n
\n
\n
\n
\n
\n
\n
\n
\n
\n
\n
\n
\n
\n
\n
\n\n---\n## 🙈 Appendix 🙈\n\nHere is just code and details that you probably do not want to read.","pinned":false,"hidden":false,"mode":"md"},{"id":246,"value":"#### Imports","pinned":false,"hidden":false,"mode":"md"},{"id":15,"value":"jstat = require('jstat')","pinned":false,"hidden":false,"mode":"ojs"},{"id":206,"value":"import {Scrubber} from \"@mbostock/scrubber\"","pinned":false,"hidden":false,"mode":"ojs"},{"id":248,"value":"#### Functions","pinned":false,"hidden":false,"mode":"md"},{"id":329,"value":"wald = function(s, n, zcrit){\n const p̂ = s/n\n const se = Math.sqrt( p̂*(1 - p̂)/n)\n return [p̂ - zcrit*se, p̂ + zcrit*se]\n} ","pinned":false,"hidden":false,"mode":"ojs"},{"id":339,"value":"wilson = function(s, n, zcrit){\n const p̂ = s/n\n const ptilde = p̂ + (zcrit**2/(2*n))*((1-2*p̂)/(1+zcrit**2/n))\n const se = Math.sqrt( p̂*(1 - p̂)/n + (zcrit**2/(4*n**2)) )/(1+zcrit**2/n) \n return [ptilde - zcrit*se, ptilde + zcrit*se]\n} ","pinned":false,"hidden":false,"mode":"ojs"},{"id":422,"value":"bayesian = function(s, n, confidence, alpha, beta){\n const lower = jstat.beta.inv((1 - confidence/100)/2, alpha + s, beta + n - s)\n const upper = jstat.beta.inv(1 - (1 - confidence/100)/2, alpha + s, beta + n - s)\n return [lower, upper]\n}","pinned":false,"hidden":false,"mode":"ojs"},{"id":250,"value":"#### Data","pinned":false,"hidden":false,"mode":"md"},{"id":462,"value":"colors = [\"#6C8EBF\", \"#c0a34d\", \"#780000\", \"#007878\", \"#bf9d6c\", \"#3A6B35\", \"#b5c6df\",\"#eadaaa\", \"#bb989a\", \"#98bbb9\", \"#bf8d6c\", \"#CBD18F\"]","pinned":false,"hidden":false,"mode":"ojs"},{"id":28,"value":"confidence = settings[0]","pinned":false,"hidden":false,"mode":"ojs"},{"id":26,"value":"n = settings[1]","pinned":false,"hidden":false,"mode":"ojs"},{"id":488,"value":"lower_y = 0","pinned":false,"hidden":false,"mode":"ojs"},{"id":336,"value":"zcrit = jstat.normal.inv(1 - (1 - confidence/100)/2, 0 , 1)","pinned":false,"hidden":false,"mode":"ojs"},{"id":683,"value":"lower = 0.001// onlyValid_p ? 10/n : 0.001","pinned":false,"hidden":false,"mode":"ojs"},{"id":686,"value":"upper = 0.999 // onlyValid_p ? 1-10/n : 0.999","pinned":false,"hidden":false,"mode":"ojs"},{"id":688,"value":"pgrid = d3.range(lower, upper, (upper-lower)/1000)","pinned":false,"hidden":false,"mode":"ojs"},{"id":438,"value":"coveragedata = {\n let coverages = []\n const sgrid = d3.range(0, n+1)\n for (let i = 0; i < pgrid.length; ++i) {\n let p = pgrid[i]\n let invalidp = (n*p < 10) || (n*(1-p) < 10) ? 1 : 0;\n let prValid = (n >= 20) ? jstat.binomial.cdf( n-10, n, p ) - jstat.binomial.cdf( 9, n, p ) : 0\n let binomprob = sgrid.map(s => jstat.binomial.pdf(s, n, p))\n let coverageWald = 0;\n let coverageWilson = 0;\n for (let s = 0; s <= n; ++s) {\n let waldinterval = wald(s, n, zcrit)\n if (waldinterval[0] <= p && waldinterval[1] >= p){coverageWald += binomprob[s]}\n let wilsoninterval = wilson(s, n, zcrit)\n if (wilsoninterval[0] <= p && wilsoninterval[1] >= p){coverageWilson += binomprob[s]}\n }\n coverages.push(({p: p, invalidp: invalidp, prValid: prValid, wald: coverageWald, wilson: coverageWilson, conflevel: confidence/100}))\n }\n return coverages\n}","pinned":false,"hidden":false,"mode":"ojs"},{"id":550,"value":"abserrors = [d3.mean(coveragedata.map(d => Math.abs(d.wald - confidence/100))), \n d3.mean(coveragedata.map(d => Math.abs(d.wilson - confidence/100)))\n]","pinned":false,"hidden":false,"mode":"ojs"},{"id":671,"value":"priorcoverage = d3.sum(coveragedata.map(d => d.priorpdf*d.bayes*(pgrid[1]-pgrid[0])))","pinned":false,"hidden":false,"mode":"ojs"}],"files":"$W32","stdlib":"1","titling":"auto"}} 2f:"$2e:body"