Random Control Groups Are Not Attribution

We should be careful in our use of the word “attribution”. That word
does not apply to RCTs.

The dictionary definition of “attribution” is:

      • the action of regarding something as being caused by a person
        or thing.
      • “the electorate was disillusioned with his immediate attribution
        of the bombings to a separatist group”
      • the action of ascribing a work or remark to a particular author,
        artist, or person.
      • “the study of Constable is fraught with problems of attribution”
        the action of regarding a quality or feature as characteristic of or
        possessed by a person or thing.
        “the attribution of human emotions to inanimate objects”

In other words, attribution is suppositional. Suppose X is the cause of
Y. What if X were the cause of Y? Attribution is subjunctive
dictionary defined as “relating to or denoting a mood of verbs
expressing what is imagined or wished or possible.

So, my point is that we should not fall into the sloppy habit of
treating the word “attribution” as if it is the umbrella supercategory
that sits over and includes RCTs as well as MMM, MTA, and
singlesource. RCT is in a class by itself – it is not regarding X as a
cause, it is proving that X is a cause. It is establishing scientific fact.

This is not a trivial epistemological distinction – it is the whole point
of using attribution to generate testable hypotheses, and then using
RCTs to test those hypotheses, before committing millions of dollars
to a specific media/creative combination called a campaign.
If RCT were attribution too, there would be no point in using it to
confirm or disconfirm hypotheses drawn from attribution.

Best to all,