The jingle-jangle fallacies
In 1904, Edward Thorndike pointed to the phrase “college student.” It covers the full-timer and the part-timer, the freshman and the grad student. Very different people under one name. The jingle.[1]
The jangle came later, from Truman Lee Kelley in 1927. Setting the two side by side in Interpretation of Educational Measurements, he wrote:
Equally contaminating to clear thinking is the use of two separate words or expressions covering in fact the same basic situation, but sounding different, as though they were in truth different. Kelley, 1927, p. 64 [2]
His example was achievement and intelligence. The field kept the two names apart; the tests measured much the same thing.[2] The names jangled; the constructs did not.
Construct proliferation: new constructs that turn out to be old ones under a new name.[8] The usual checks run inside one field. ConstructCollision puts every field side by side, so a construct can be checked against the idea itself, regardless of discipline.
How to use it
- Search a construct to see what it means in each field, and the paper behind each meaning.
- Build your model to see where your constructs land on the map and what already lives nearby.
- Prime your project to get the key terms for your question and the papers that defined them.
- Claim your construct if your work is cited, to confirm it or correct us.
What the map is, and what it is not
Every dot is one meaning, placed near similar meanings. “Close” on the map means close in meaning, not proof that two measures behave the same in data. The map is a compass, not a judge: the real test is reading the papers, and testing on your own sample.
This site is in beta, and it may have errors. If you find one, tell us. We appreciate honest feedback.
The scale of the problem is now measurable
A construct is only as precise as its items. When two scales share a name but not their items, pooling them mixes unlike things. When two scales share their items but not their name, one variable ends up with two literatures.
Applying language-model embeddings to the text of items and scales, Wulff and Mata found that the semantic similarity of items predicted their empirical correlations, and used this to flag hundreds of candidate fallacies across a corpus of several hundred personality scales, the majority jingles and a substantial minority jangles.[3] In a companion review, they argue the approach generalizes beyond personality and propose that the field's construct taxonomy could be compressed dramatically with little loss.[4]
How to check for one
Before you cite, pool, or coin
- Go to the items. A construct's name is a claim; its items are the evidence. Read what the scale actually asks.
- Compare across fields, not just within. Jingles hide at disciplinary borders, where the same name was adopted independently.
- Look for the nearest neighbor. Before you name a new construct, find the existing construct whose items are closest to yours on the map. Then consider how yours differs, item by item, not just in name.
- Put a number on it. Distinctness is something you measure, not something you assert. Two scales that share most of their item wording, or whose scores rise and fall together, are probably one construct with two names.
None of this is new. Checking a construct against its nearest neighbor is discriminant validity; the redundancy it guards against is what organizational research named construct proliferation. The difference here is that the check runs across fields, not within one study.[7][8]