Agreement is treated as a record of another original
The claim is that matching memories of a missing visual feature establish that an earlier version of reality contained it. One person could be mistaken; many people making the same confident report seems different. The stronger question is therefore not whether anyone can make a memory error. It is whether ordinary remembering can repeatedly produce one specific error across different people.
The wrong answer was shared in a controlled task
Prasad and Bainbridge tested familiar icons with an original image and altered alternatives. For Fruit of the Loom, the alternatives added a cornucopia or a plate. Their author-deposited response file contains one row per participant. The Fruit column records 32 original choices, 50 choices for the more popular alteration, 13 for the other alteration and 5 missing responses.
The experiment therefore contains 95 valid choices for this icon. More than half selected the same incorrect option. That result is more specific than a general failure to recognise a logo: errors clustered around one feature. The authors also investigated drawing from memory, so the research was not confined to choosing whichever supplied picture looked most plausible.
The research did not identify a single mechanism that explains every icon. An association between fruit and a container, for example, does not by itself explain why the cornucopia defeated the plate. The useful finding is that a shared error is experimentally observable. Explaining its precise form is a further question within memory research.
Measure the agreement that the sample contains
The instrument reproduces the counts for the Fruit of the Loom task. It first adds the three valid-response categories: 32 + 50 + 13 = 95. The preferred wrong image received 50/95, or 52.63 percent, of the valid responses. Missing responses stay out of this denominator because they contain no selected image.
Now draw two different valid responses at random from this recorded sample. The first has a 50/95 chance of being the preferred wrong choice. If it is, 49 such choices remain among 94 responses. Multiplying gives 27.44 percent: more than one in four randomly selected pairs agree on that particular incorrect image.
This calculation draws without replacement. It describes pairs already present in the sample, so it does not need to assume that the participants’ mental processes were independent or assign a universal probability of false memory. Changing the counts explores a different sample; leaving the opening counts in place reproduces the deposited observation.
A useful contrast keeps the same 63 errors but divides them almost evenly between the two wrong images. The majority-wrong pattern weakens even though overall accuracy stays unchanged. Accuracy and error concentration are distinct measurements. The surprise in a visual Mandela Effect is partly that the errors collect in the same place.
Measure shared wrong-image choices
Opening counts reproduce the valid Fruit of the Loom responses in Experiment 1. The pair calculation selects two recorded responses without replacement.
The percentage of valid responses choosing the same altered image.
Exact probability for two different responses randomly selected from these counts.
Working tape
- Total valid responses
32 + 50 + 13 = 95 - Preferred wrong-image fraction
50 ÷ 95 = 0.526316 - Preferred wrong choices after first draw
50 − 1 = 49 - Responses after first draw
95 − 1 = 94 - Second matching choice after first match
49 ÷ 94 = 0.521277 - Both responses select the preferred error
0.526316 × 0.521277 = 0.274356 - Pair probability as a percentage
0.274356 × 100 = 27.43561 - Preferred wrong-image percentage
0.526316 × 100 = 52.631579
The agreement was real; the selected image was altered
Fifty of 95 valid responses selected the same altered image: 52.63 percent. Two different responses drawn from that sample have a 27.44 percent chance of matching on this specific error. The task produces substantial agreement while the original and altered alternatives remain known.
That leaves an interesting, measurable question: what properties of a particular icon make its wrong version so compelling? Counting the errors keeps that question alive. Turning the count into evidence of a changed reality skips the experimental result that made the count possible.