Research

Limitations

These constraints apply to the programme as a whole. Each study report adds the limitations specific to its own design.

A positive outcome does not identify a mechanism

Our designs can measure whether blinded judges match readings to intended targets above a preregistered baseline. They cannot tell you why. Any mechanism, conventional or otherwise, remains an open question after a positive result.

Small samples limit precision

Longitudinal blinded sessions are expensive and slow. Early studies will be small, so confidence intervals will be wide and single-study effect estimates unreliable. Replication, not a first result, is what carries weight.

Self-selected participants

Mediums and sitters volunteer. Findings describe the people who took part under our conditions, not mediumship in general, and cannot be generalised to the wider population of claimants.

Information leakage can never be fully excluded

Blinding, exposure logs and sealed answers reduce leakage risk; they do not reduce it to zero. Every study reports the leakage checks performed and any pathway that could not be closed.

High-base-rate material inflates apparent accuracy

Statements true of many people can look specific to a reader. We code base rates in advance, but the coding scheme is imperfect and remains a source of measurement error.

Judging is a human instrument

Judge agreement is measured and reported, but reliability below a preregistered threshold weakens any conclusion drawn from that round, positive or negative.

Absence of an effect is bounded, not absolute

A negative result rules out effects larger than the smallest size the study could detect. It does not demonstrate that no effect exists at any size.

Publication and analysis are constrained by consent

Some material cannot be released without risking re-identification. Where data cannot be published, the analysis code and summary statistics are, so the result remains checkable.

What would change our minds? sets out the observations that would count against each hypothesis before data are collected.