Presentiment research examines whether physiological measures taken before randomly selected stimuli can predict future events. James E Kennedy argues that confirmatory studies should minimise retrospective data-processing flexibility by preregistering software and using pre-stimulus physiological data to predict random stimuli directly.
- Presentiment studies measure physiological activity before randomly selected stimuli are presented.
- Retrospective processing of physiological data can introduce subtle bias unless fully prespecified.
- Confirmatory designs should use pre-stimulus data to predict random stimuli with preregistered classification criteria.
Contents
Presentiment Research
Presentiment research involves physiological measures of precognitive anticipation. Some type of physiological measurement is made before a stimulus is randomly selected and presented to a participant. The research investigates whether the physiological measures indicate that the participant is correctly anticipating which of two possible stimuli will occur. The two stimuli can be for example an emotional picture and a calming picture, or a stimulus that requires a fast physical action and another stimulus that requires no response. The measured anticipation may occur without conscious awareness by the participant. Different physiological measures have been investigated including skin conductance, electrical activity in the brain, muscle activity, heart rate, and pupil dilation.
Presentiment research has certain characteristics that make it unusually susceptible to bias.1Kennedy (2013) For most other types of precognition research, a response from the participant is irreversibly recorded before the precognitive target is displayed. However, in presentiment studies, the response by the participant is typically derived from extensive processing of the physiological data after data collection is complete and after the stimuli have been presented on all the trials. The data processing is often described with terms like baseline adjustment, normalization, filtering, and artifact rejection. Unless the programming is fully prespecified or completed in advance, this retrospective data processing can provide flexibility to obtain the desired results through post hoc adaptations.
The possibility of bias is compounded by the fact that the physiological data usually include reactions to the stimuli that indicate which stimulus was given for a trial. If the data processing is not done very carefully, the data adjustments can introduce bias by incorporating information about the stimuli that is in the physiological record. The bias may be subtle and counterintuitive without the researchers being consciously aware of it.
The basic analysis strategy for most presentiment studies has also been different and more controversial than for most other types of precognition experiments. Most precognition experiments are evaluated by using the participant’s response to predict a randomly selected target. This ability to predict a future event is the definition of precognition for experimental research. A binomial analysis to evaluate whether the number of correct predictions is evidence for precognition is one of the simplest and most theoretically correct (fewest assumptions and approximations) statistical analyses in all of science.
However, in most presentiment studies the physiological measures were not used to predict the random stimulus. Rather, the analysis evaluated whether the physiological measures were different for the two stimuli.
This analysis reverses the traditional analysis for precognition experiments and uses the random targets (stimuli) to predict the previous responses (physiological measures). The traditional analysis that uses the participants’ responses to predict the random targets (response→target) requires a statistical model for the randomly generated targets, which is a simple uncontroversial model. Reversing the analysis (target→response) requires a statistical model for the responses, which is more difficult and controversial because physiological measures and human responses in general cannot be assumed to be random or independent. Attempts to adjust for the nonindependence usually involve debatable assumptions, which is a major disadvantage for controversial research like precognition. Biases can be counterintuitive and controversial to model and understand.2Kennedy (2013) In effect, this reversed analysis stimulates controversy about the statistical analysis whereas the traditional analysis for precognition minimizes such controversy.
As a practical example of the problems that can arise with this reversed analysis, common anticipation tendencies or strategies can produce the same pattern of results as precognition.3Kennedy (2013) When the same stimulus happens to occur on sequential trials, participants tend to increasingly expect the next trial to be the other stimulus. For such a sequence of trials, the last trial is correct and has the highest anticipation, which can produce differences in the physiological measures for the different stimuli. However, these differences are not due to precognition and this expectancy effect cannot be used to successfully predict a random stimulus.
The best strategy for demonstrating that expectancy and other biases did not affect the results is to predict the random stimuli from the physiological measures – which directly evaluates precognition and can be assessed with a simple noncontroversial statistical analysis. If the effects reported in previous research are meaningful precognition effects, the physiological measures should predict the random stimuli.
A study design for confirmatory research
An experimental design that does not have the methodological uncertainties noted above would use the physiological measures to predict the random stimuli. This general analysis strategy is known as classification analysis. Many different classification methods are available, but the specific method does not matter for this overview of the basic strategy. An internet search provides abundant information about different methods for the continually evolving topic of classification analysis. AI may be used as a sophisticated form of classification analysis.
The first step is called the learning phase in which some data is collected and processed to develop the classification criteria. Once the criteria are developed, they are applied to make predictions for the primary experimental data. The criteria may be simple or complex, and may incorporate the outcome of the previous trial in the prediction for the next trial. The criteria could apply to a wide range of participants or could be developed uniquely for each person.
A key point is that the data used for developing the criteria is excluded from the primary analysis. Researchers who are using classification methods for the first time often do not appreciate the need to apply the classification criteria to entirely new data for psi research. The classification criteria applied to the data used to develop the criteria usually produces results that are startlingly more accurate than can be expected when applied to different data. Efforts to statistically adjust for the inflated accuracy with the learning data can be considered exploratory analyses that need to be followed by confirmatory research that applies the criteria to make predictions for new data.
Once the classification criteria have been developed for a participant, the optimal design would have the data collection software apply the criteria as new data is collected. The data collection software would make a prediction about the stimulus on a trial in real time. The prediction would be recorded before the stimulus was randomly generated for the trial. Any trials that are compromised by movement or other artifacts would be rejected by the data collection system before the stimulus is generated.
This would require that all decisions and programming for data processing are developed before data collection begins. This is an appropriate expectation for confirmatory research that is based on adequate previous exploratory research. The programming should be included in the preregistration for the study. If the researchers require flexibility to adapt the data processing and analysis after looking at the data, the research remains at the exploratory stage and has not yet reached the point that the researchers can convincingly demonstrate the effect.
This design also ensures that the prediction on a trial is not influenced by any data or information after the stimulus starts for the trial. And, the traditional simple theoretically correct statistical analysis for predicting a random event before it occurs would be applicable.
Data processing after data collection
If incorporating the data processing into the data collection system is not feasible, the same strategy could be used after the data has been collected. The data processing program would go through the data in chronological order and make a prediction on a trial using only data from before the stimulus for the trial. Data processing would not include any sweeps through all the data to find means, standard deviations, maximum values, frequencies, artifacts, etc. Such sweeps through the data have the potential for information about the stimulus on a trial or about random biases in the data to be incorporated into the predictions.
However, the prediction on a trial can include any type of analysis of data before the stimulus on the trial, including data from previous trials. The programming could be self-adapting as more trials are evaluated. For good confirmatory research, the programming would be expected to be included in the preregistration.
General methodological practices
Good research should incorporate the lessons from the replication crisis including distinguishing between exploratory and confirmatory research, preregistering planned studies, software validation and other quality control measures that prevent and detect unintentional errors and intentional errors (researcher fraud), making the data publicly available, and conducting formal confirmatory research with adequate sample sizes and without exploratory flexibility.
In evaluating research with physiological measures of precognitive anticipation, studies that were preregistered as confirmatory and used the physiological measures to predict the random events based only on data before each random event should be distinguished from all other studies and provide the primary evidence about an effect.
James E Kennedy
Works Cited
Kennedy, J.E. (2013). Methodology for confirmatory experiments on physiological measures of precognitive anticipation. [Full text.] Journal of Parapsychology 77, 237-48.