INTRODUCTION This chapter reviews issues related to the analysis of psychophysiological data. It focuses on general questions relevant to a variety of techniques, rather than on specific issues related to individual methods. In this new edition we have modified this chapter to account for the fact that new techniques and research domains are rapidly being added to the psychophysiology toolkit. For instance, the last 20 years have seen the growth of a number of integrated methods for analyzing brain and bodily function. The integration of physiological and anatomical data is now becoming standard, as are a series of multivariate methods for the parallel investigation of physiological activity from multiple locations and features (such as network analysis, multivoxel pattern analysis [MVPA], source analysis, and time-frequency analysis). Similarly, it is ever more common to integrate different imaging modalities (e.g., by combining electrophysiological and magnetic resonance data), or to combine manipulative and observational techniques (e.g., transcranial magnetic stimulation [TMS] and electrophysiological or optical recordings). Although the specific methodological details introduced by such combinations are beyond the scope of this chapter, they bring to the fore important statistical and inferential issues (such as the interpretation of causal links) that are central to current trends in biosignal processing and will therefore be discussed here. Stages of Data Processing In general, one of the main objectives of signal processing is maximizing the signal-to-noise ratio, which is reflected in the different stages of analysis of psychophysiological data. The first stage is signal enhancement and involves signal-to-noise enhancing methods such as filtering, as well as the elimination of artifacts and outliers. This stage, sometimes called “signal conditioning,” involves, at least in part, techniques that are specific to each type of physiological measure. The second stage is data reduction (sometimes called “quantification” or “parameter extraction”). Most psychophysiological experiments include a large number of observations per subject (i.e., dependent variables). Whereas this provides datasets that are rich in information content, it also increases the probability of spurious or noisy measurements. A large number of observations may contain many different signals to study, but may also contain a large amount of noise.
ASJC Scopus subject areas