ATProto Browser

Record data

{
  "uri": "at://did:plc:ivv4cfsfdt5d3ozdngj35cbc/app.bsky.feed.post/3llh2lo5euk2z",
  "cid": "bafyreigi3fhrhkmsincwrcfcnlkspq5n6spajtziw5iwa5q25z7f3oujnu",
  "value": {
    "text": "Related to our NCME workshop, but in a more cynical mode: I will be at the International Objective Measurement Workshop discussing clearly non-objective measurement in a talk titled \"How to Lie with Fit Statistics.\" The IOMW committee didn't seem to mind my sarcastic abstract:",
    "$type": "app.bsky.feed.post",
    "embed": {
      "$type": "app.bsky.embed.images",
      "images": [
        {
          "alt": "IOMW abstract: Title: How to Lie with Fit Statistics\nAbstract (397 words):\nStatistical models enable scientists to describe data, explain phenomena, form testable predictions, and (ideally) understand the world. Success in such endeavors is contingent on the model’s usefulness, which – especially in the social sciences – is often gauged by inspecting its goodness-of-fit (GOF) to some observed data set. In many modeling applications, it seems that the primary aim is to satisfy some prescribed benchmark criteria for GOF. This aim is of minimal inferential value [see Roberts and Pashler (2002), who wrote that the use of GOF for theory testing is “rotten to the core” (p. 605)], but researchers who achieve good fit will be able to craft whatever story they wish to tell, bolstering their specious claims with statistical “evidence” and thereby increasing their chances of publication. In this presentation, I will teach you how to use GOF to perpetrate lies of commission and of omission when analyzing your data and disseminating your findings. (Although these lies transcend statistical modeling frameworks, I will illustrate them in the context of structural equation modeling and factor analysis.)",
          "image": {
            "$type": "blob",
            "ref": {
              "$link": "bafkreidsv4ktbqk773j6epjihaomk3z4xalnzxevnnjw23o2ulvtjhkvdq"
            },
            "mimeType": "image/jpeg",
            "size": 291842
          },
          "aspectRatio": {
            "width": 1175,
            "height": 597
          }
        },
        {
          "alt": "During data analysis, lies of commission include deception (e.g., using exploratory methods as the basis for confirmatory modeling) and tampering (e.g., using modification indices); lies of omission include ignorance (e.g., failing to account for model complexity beyond the number of parameters). During dissemination, lies of commission include misconstrual (e.g., writing that a model’s GOF “proves” that the underlying theory is correct) and distortion (e.g., reporting GOF cutoffs that are not supported by the statistical literature); lies of omission include underreporting (e.g., failure to disclose issues known to affect GOF, such as non-normality) and paltering (e.g.,, cherry-picking the fit indices that best support the model).",
          "image": {
            "$type": "blob",
            "ref": {
              "$link": "bafkreih6u5stimpon4idzm2nmjirlxmiggjk66cml6ezbzvwknyp5djcvy"
            },
            "mimeType": "image/jpeg",
            "size": 180900
          },
          "aspectRatio": {
            "width": 1176,
            "height": 319
          }
        },
        {
          "alt": "Each lie above is based on previously established research on GOF. I will expand on this existing knowledge by presenting more subtle forms of statistical hoodwinking that are not widely known to impact GOF (e.g., adjusting the model specification in terms of the number of factors, the factor (im)balance, etc.). Since these tricks have not yet been recognized in the model evaluation literature, their use is not an intentional lie, but more an effect of nescience (though, after my presentation, you will no longer be able to plead ignorance about the relationship between model specification and GOF!).\nWith this array of lies at your disposal, I am confident that you will produce many “publishable” findings that will “confirm” your theories.\nReference: \nRoberts, S. & Pashler, H. (2002). Reply to Rodgers and Rowe (2002). Psychological Review, 109(3), 605-607.",
          "image": {
            "$type": "blob",
            "ref": {
              "$link": "bafkreibt2ih5dy5czdxswcussoztculca42oc6kmz6n46nbk5w46u7cndu"
            },
            "mimeType": "image/jpeg",
            "size": 254163
          },
          "aspectRatio": {
            "width": 1185,
            "height": 536
          }
        }
      ]
    },
    "langs": [
      "en"
    ],
    "createdAt": "2025-03-28T14:46:38.128Z"
  }
}