Discovery Lab Answers

Milton & Wiseman (1999): what the meta-analysis found

The paper — “Does Psi Exist? Lack of Replication of an Anomalous Process of Information Transfer,” published in Psychological Bulletin — pooled 30 ganzfeld ESP studies conducted between 1987 and 1997, i.e. the “new-generation” studies that followed the original PRL autoganzfeld series described by Bem & Honorton (1994). The database and the term to use here is sensory attenuation, not the older label.

The headline result was null: across those 30 studies (N = 1198), Milton and Wiseman (1999) reported a small, non-significant effect (an effect of z/√n ≈ 0.013, z = 0.7, p = .24). This was the field’s inflection point — a well-known positive database (Bem & Honorton) followed by a same-method replication set that did not reach significance.

Coverage caveat, up front: the ESP-Nexus library holds 57 of the 78 studies in the standard ganzfeld meta-analytic database, plus the published abstract of one more. The 16 it is still missing skew toward null results, so any pattern summarized from these holdings — including what follows — runs somewhat more positive than the full published database. So the numbers below describe the library’s audited subset, not a settled account of the whole field.

How ganzfeld researchers responded

The response is best understood as three overlapping moves, and the audited evidence lets you see each.

1. An updated cumulation, including Dalton (1997). One immediate line of argument was that the 30-study cutoff excluded a large, strongly positive study (Dalton 1997a). When later studies through March 1999 were folded in, the same database moved from null to significant — Milton (1999) itself reports an updated recent-ganzfeld cumulation of 39 studies (N = 1460) at z = 2.28, p = .011. Note this is Milton’s own extended figure, not a critic’s; the significance of the post-1987 set was sensitive to a small number of high-yield studies.

2. Later meta-analyses that reached positive pooled results. A series of subsequent pooled analyses — held in the audited evidence for this question — reported significant overall effects. The pattern across the library’s holdings:

Because these rows report different, non-comparable metrics (standardized effect sizes, raw hit rates, a correlation), do not read them as a single pooled figure — they are separate summaries of overlapping but differently-defined databases. The consistent qualitative point is that the wider, later databases pool to small significant positive effects, whereas the specific 30-study 1987–1997 set that Milton & Wiseman analyzed did not.

3. Methodological disputes over which studies belong. Much of the exchange was about study inclusion, standardization, and whether the post-PRL studies were genuine method replications — the argument that a null in one bounded set does not settle the question, and the counter-argument that expanding the set to recover significance risks selection.

What is NOT settled (you should weigh these against the positive pooling)

• Directions disagree. Milton & Wiseman (1999) is a genuine null in the audited set; not every pooled result points the same way.
• Heterogeneity is high. At least two of the pooled results show high between-study heterogeneity (I² ≥ 50%), meaning the studies are not estimating one common effect — a pooled figure obscures real variation.
• Sensitivity to individual studies. The move from Milton & Wiseman’s null to a significant updated cumulation turned substantially on including a small number of large positive studies, which is exactly the kind of dependence that keeps the debate open.

For the full narrative — including the Wiseman-Schlitz experimenter-effects work and the post-1999 meta-analyses — see Ganzfeld Debate and Replication and Meta-Analysis Debates; Julie Milton covers the Milton-Wiseman collaboration directly.