The above seven meta-analyses represent a cross-section of the meta-analyses that have been performed on parapsychological research. They were not chosen to illustrate the greatest effects or to ``paint the rosiest picture'', but rather to provide a window into the range of effects and variety of methodologies found in psi experimentation. The effect sizes in these studies tend to be very small (RNG--PK) to moderate (i.e., DMILS) in size. However, even the smaller effect sizes appear to be reliably found in the databases. Furthermore, the size of an effect does not provide a good indication of its potential meaningfulness or applicability. For example, a recent medical study investigating whether aspirin could help prevent heart attacks was ended prematurely because the effectiveness of the treatment was so clearly demonstrated after six months of trials that the investigators thought it would be unethical to withhold the treatment further from the control group. Indeed, the findings from the study were heralded as a major medical breakthrough. While the findings from this study were highly significant ( = 25.01, p = 0.00001), the effect size is .068, considerably smaller than some of the effect sizes found in the psi literature [12, 13]. It should be noted that small effects have low statistical power [8, 12, 13]. For example, the aspirin study involved over 22,000 subjects. If there had only been 3,000 subjects, the investigators would have had less than a 50 percent chance of finding a conventionally significant effect . Given the small effect sizes which are typical in psi experiments, low replicability is to be expected. Rosenthal  notes that ``even though controversial research areas are characterised by small effects, that does not mean that the effects are of no practical importance.'' (p. 324). Indeed, in an article addressing behavioural research in general, Rosenthal  warned: ``Given the levels of statistical power at which we normally operate, we have no right to expect the proportion of significant results that we typically do expect, even if in nature there is a very real and very important effect'' (p. 16).
What can these findings tell us about the functioning of apparent psi abilities? The conceptualisation of ESP as the anomalous input of information into consciousness and PK as the anomalous output of influence are ``working'' models, which help convey possible interpretations of the obtained phenomena. However, the distinction between ESP and PK is often blurred. If one accepts the precognition database as suggesting that information about an event can be obtained before the occurrence of the event, many of the psi results could be interpreted as representing acts of precognition. For example, while most ganzfeld and DMILS studies are ``real-time'' and involve an ``agent'', it is possible that the actual mechanism at work may be the subject obtaining information about the target by ``looking'' into the future to gain relevant target information and then generating appropriate impressions in the case of ganzfeld studies, or producing the appropriate self-regulatory responses in the case of the DMILS studies. In this context, it should be mentioned that the role of the agent is unclear. No DMILS studies have yet been reported which have not used an agent, but in the case of ganzfeld studies, the recent Edinburgh study  found equally significant outcomes in sender and no sender conditions. Similarly, other ESP studies have obtained significant, positive outcomes without using an agent (for a review of this work see Palmer ). Findings such as these indicate that a sender appears not to be a necessary component in anomalous information transfer studies, although they may still have a beneficial psychological impact upon the study outcomes . The RNG--PK work has been traditionally conceptualised as representing ``influencing'' effects, as has the DMILS work which was initially known as ``bio-PK''. However, alternative interpretations of these apparent effects may involve ESP. May et al.  have proposed that apparent RNG--PK effects could be the result of the observer, via precognition, knowing what would be the right moment to initiate a sequence of random event (i.e., when to push the button) to get the desired outcome, thereby making use of the random fluctuations found in RNG systems to create a non-random outcome. Others have questioned the validity of a precognition interpretation of psi data. For example, Morris  discusses models based on ``real-time'' psi effects as possible alternative explanations of precognition. For example, using PK a subject or investigator could influence the random source used to choose the target in precognition studies to obtain a selection consistent with the subject's response.
As the above comments make apparent, the mechanisms which may be involved in the producing the effects found in these databases are still unknown. Process-oriented research is ongoing in parapsychology. In future studies, correlations such as those found in the precognition database may help us better differentiate between the differing theoretical interpretations of these anomalous effects. While this paper has focused upon presenting summaries of experimental data, there are a variety of theoretical models which address these findings. Although it is outside the scope of this paper to review these models, a thorough presentation of theoretical parapsychology is provided by Stokes .
In conclusion, the findings from these meta-analyses suggest that consistent trends and patterns are to be found in the database. The consistency of outcomes found in the ganzfeld research, the robust PK effects, the modifying variables revealed by the precognition database, the variety of target systems displaying DMILS effects and the correlations found with personality traits are all indicative of lawful relationships. Given these relationships it is difficult to dismiss the findings as ``merely an unexplained departure from a theoretical chance baseline'' p. 301 . Whether these effects will prove to represent some combination of currently unrecognised statistical problems, undetected methodological artefacts, or, as seems increasingly likely, a genuinely new, hitherto unrecognised characteristic of mind or consciousness remains to be seen.
Acknowledgements: I would like to thank Robert L. Morris and Jessica Utts for their helpful comments on earlier drafts of this paper.