Intensive longitudinal modelling predicts diurnal activity of salivary alpha-amylase

Abstract

Salivary alpha-amylase (sAA) activity has been widely used in psychological and medical research as a surrogate marker of sympathetic nervous system activation, though its utility remains controversial. The aim of this work was to compare alternative intensive longitudinal models of sAA data: (a) a traditional model, where sAA is a function of hour (hr) and hr squared (sAAj,t = f(hr, hr2), and (b) an autoregressive model, where values of sAA are a function of previous values (sAAj,t = f(sAA j,t-1, sAA j,t-2, …, sAA j,t-p). Nineteen normal subjects (9 males and 10 females) participated in the experiments and measurements were performed every hr between 9:00 and 21:00 hr. Thus, a total of 13 measurements were obtained per participant. The Napierian logarithm of the enzymatic activity of sAA was analysed. Data showed that a second-order autoregressive (AR(2)) model was more parsimonious and fitted better than the traditional multilevel quadratic model. Therefore, sAA follows a process whereby, to forecast its value at any given time, sAA values one and two hr prior to that time (sAA j,t = f(SAAj,t-1, SAAj,t-2) are most predictive, thus indicating that sAA has its own inertia, with a “memory” of the two previous hr. These novel findings highlight the relevance of intensive longitudinal models in physiological data analysis and have considerable implications for physiological and biobehavioural research involving sAA measurements and other stress-related biomarkers.

Pooled Time Series Modeling Reveals Smoking Habit Memory Pattern

“…Descriptive analyses were also performed in the same program. Once the data had been introduced with a "person-level data set" system, they were transferred to a "longitudinal time record" system (42), each record corresponding to a day of smoking, and were organized for analysis with a pooled time series system (17,43,44). After organizing the data, they were analyzed using the statistics software package Mplus (39).…”

Estudio de la memoria conductual de sustancias adictivas mediante análisis de series temporales agrupadas

“…Y es que los métodos de análisis transversal, como pueden ser los contrastes de medias mediante pruebas t o F, las regresiones, …, resultan incorrectos (Hartmann, 1974). Este tipo de análisis transversales asumen que los datos de las variables no están relacionados entre ellos, lo que se conoce como independencia serial (Rosel et al, 2019).…”

“…En conclusión, si se usan métodos de análisis transversales con datos longitudinales, los valores de las pruebas t, z, R 2 o F van a ser más elevados si se da la condición de que los residuos de las variables están correlacionados entre ellos, cosa que no puede acaecer con los datos en panel (Box et al, 2015;Rosel et al, 2019Rosel et al, , 2020. Si se diera este hecho, aumentaría la probabilidad de que ocurrieran errores de tipo α, es decir, se rechazaría la hipótesis nula cuando se tendría que aceptar.…”

“…Esta ausencia de correlación se conoce como ausencia de correlación serial o "ruido blanco" (White Noise). Si los errores muestran correlación entre ellos, las varianzas residuales están infraestimadas, lo que se también traduce en una infraestimación de los errores típicos de los coeficientes β del modelo, produciendo errores de tipo α (Box & Pierce, 1970;Rosel et al, 2019).…”

“…Además, se asume que los valores de los coeficientes de los retardos tendrán valores positivos, conforme a los sistemas de modelos libres (O'Doherty et al, 2017;Wood & Rünger, 2016) o a investigaciones similares llevadas a cabo previamente a esta tesis doctoral (Rosel et al, 2019(Rosel et al, , 2020.…”