{
  "$type": "site.standard.document",
  "canonicalUrl": "https://jacob.blog/notes/type-i-and-type-ii-errors",
  "description": "Precision, recall, and F1",
  "path": "/notes/type-i-and-type-ii-errors",
  "publishedAt": "2025-07-25T00:00:00.000Z",
  "site": "at://did:plc:ckthoyuvsmkp254fyuinyzb2/site.standard.publication/3mndm6tiamb26",
  "tags": [
    "decision-making",
    "statistics"
  ],
  "textContent": "> [!info] In short…\n>\n> Type I errors are false positives. _Higher precision means fewer Type I errors._\n>\n> Type II errors are false negatives. _Higher recall means fewer Type II errors._\n\nType I errors, or false positives, are errors of commission where the status quo is incorrectly rejected in favor of new, misleading information. e.g. convicting an innocent defendant.\n\nType II errors, or false negatives, are errors of omission where a misleading status quo is allowed to remain due to failure in identifying it as incorrect. e.g. failing to convict a guilty defendant.\n\nType I errors are often said to be worse than type II errors, but in practice _it depends_. Mostly depends on the potential consequences of each error case.\n\nPrecision & Recall\n\n_Precision_ measures how many of the things were positive out of all the things you said were positive. _Note that the denominator is the group of things you chose, and other true positives you didn’t select don’t show up here._\n\n> Precision = true positives / (true positives + false positives)\n\n_Recall_ measures how many of the positive things you actually selected. _Note that the denominator here is all the actual positives including those you did not choose._\n\n> Recall = true positives / (true positives + false negatives)\n\n_F1_ is the _harmonic mean_ of precision and recall. It’s a good rule of thumb about whether you should be ok with something or maybe mut more effort in.\n\n> F1 = 2 × (precision × recall) / (precision + recall)",
  "title": "Type I and Type II errors"
}