Modeling Multivariate Count Time Series Data with a Vector Poisson Log-Normal Additive Model: Applications to Testing Treatment Effects in Single-Case Designs

Sun-Joo Cho,Matthew Naveiras,Erin Barton

DOI: https://doi.org/10.1080/00273171.2020.1860732

2021-01-21

Multivariate Behavioral Research

Abstract:In education and psychology, single-case designs (SCDs) have been used to detect treatment effects using time series data in the presence or absence of intervention. One popular design variant of SCDs is a multiple-baseline design for multiple outcomes, which often collects outcomes with some form of a count. A Poisson model is a natural choice for the count outcome. However, the assumption of the Poisson model that the outcome variable's mean is equal to its variance is often violated in SCDs, as the variance is often larger than the mean (called overdispersion). In addition, when multiple outcomes are from the same participant, it is likely that they are correlated. In this paper, we present a vector Poisson log-normal additive (V-PLN-A) model to deal with (a) change processes (auto- and cross-correlations and data-driven trend) and (b) correlation and overdispersion in multivariate count time series. A multivariate normal distribution was adapted to account for correlation among multiple outcomes as well as possible overdispersion. The V-PLN-A model was applied to an educational intervention study to test treatment effects. Simulation study results showed that parameter recovery of the V-PLN-A model was satisfactory in a large number of timepoints using Bayesian analysis, and that ignoring change processes and overdispersion led to biased estimates of the treatment effects.

psychology, experimental,mathematics, interdisciplinary applications,statistics & probability,social sciences, mathematical methods

What problem does this paper attempt to address?

The problem that this paper attempts to solve is how to effectively detect treatment effects in single - case designs (SCDs), especially under the multiple - baseline design. Specifically, the paper focuses on how to accurately test treatment effects when the data has the following complexities:

1. **Over - dispersion**: Count data in SCDs usually shows a variance larger than the mean, that is, the over - dispersion phenomenon. The traditional Poisson model assumes that the mean is equal to the variance, which is often not valid in practical applications.

2. **Autocorrelation and cross - correlation**: In time - series data, there may be autocorrelation between observations of the same variable at different time points, and there may be cross - correlation between observations of different variables at the same time point. These correlations need to be considered to avoid estimation bias.

3. **Non - linear trend**: Trends in SCDs are usually non - linear, and linear trend models may lead to estimation bias of treatment effects.

4. **Within - participant correlation**: When multiple outcomes come from the same participant, they may be correlated. Ignoring this correlation will lead to estimation bias of treatment effects.

To deal with the above problems, the paper proposes a Vector Poisson Log - Normal Additive Model (V - PLN - A). This model can simultaneously handle over - dispersion, autocorrelation, cross - correlation, non - linear trends, and within - participant correlation, thereby providing more accurate estimates of treatment effects. The paper illustrates the application of the V - PLN - A model through a data set of an educational intervention study, and verifies the accuracy of the model parameter estimates through a simulation study.