Methods, Variance, and Error in Psychoneuroimmunology Research: The Good, the Bad, and the Ugly

Abstract

Every researcher deals with error at some level. In psychoneuroimmunology (PNI) research, there may be error due to substantive fluctuations in immune parameters (e.g., as related to stress, time of day, or activity). This error is significant for some parameters, but it can and should be minimized by taking multiple measurements or converted into “good,” substantive variance by measuring variables that can predict the fluctuations. Type I and Type II “bad” errors are of more concern; many PNI studies have far too few subjects for the number of effects they test. Of studies included in a recent meta-analysis of stress and human immunity, several studies actually had fewer subjects than they had statistical tests. Finally, variance due to assay or supply variability contributes to “ugly” error, and it should be addressed by analysis of covariance or partial variance. However, too often, important variance due to factors such as age is designated as “ugly” rather than incorporated into the model. We suggest solutions for addressing “good,” “bad,” and “ugly” error and look into the future of physiometrics.

Variability and reliability of diurnal cortisol in younger and older adults: Implications for design decisions

“…Compared with other outcomes, the proportions of person variance in diurnal cortisol parameters are similar to several dimensions of daily mood such as anxiety and anger, although lower than others such as fatigue and vigor, as well as positive and negative affect measured broadly (Cranford et al, 2006; Röcke, Li, & Smith, 2009). They are substantially lower than those of many immunological parameters, some of which reflect over 50% variance due to person effects (i.e., stable individual differences; Segerstrom & Smith, 2012, Table A). They also differ in some respects from the proportions of person variance in salivary alpha amylase (sAA) measured in the present Study 1, with the sAA estimate for the diurnal slope only slightly higher, but those for mean and AUC substantially higher (Out, Granger, Sephton, & Segerstrom, 2013).…”

“…In the case of a Type II error caused by insufficient reliability, all of the financial cost has gone to naught. Costs of unreliability therefore include both postponement to scientific advancement and expenditures of researcher and participant resources (Halpern et al, 2012; Segerstrom & Smith, 2012). A related issue concerns replicability, a concern that has re-emerged recently in psychology and other fields.…”

“…Variance due to people, occasions, and days was partitioned and used to calculate four kinds of reliability estimates (see Table 1; Cranford et al, 2006). Finally, we tested whether this variance could be predicted by daily assessment of mood and health behavior, effectively treating “error” variance as variance of interest (Segerstrom & Smith, 2012). …”

“…The present paper generates “physiometric” (Segerstrom & Smith, 2012) data to inform design decisions about diurnal cortisol in such studies. Measures taken on multiple people at multiple times contain three sources of variability: between-person, within-person, and measurement error.…”

The application of protein microarray assays in psychoneuroimmunology

“…This high throughput is an especially attractive feature, as the resources (time and money) required to perform more traditional immunological assays is considerable and can be prohibitive to analyzing larger numbers of participant samples. This is particularly salient in psychoneuroimmunology, which has historically received criticism for an over-reliance on small, often under-powered, studies (Segerstrom & Smith, 2012).…”