Golden, R. M. (1994).
Analysis of categorical time-series text recall data using a connectionist model.
Journal of Biological Systems, 2,
283-305.
ABSTRACT
Categorical time-series are generated by discrete-time probabilistic
dynamical systems which can only be in one of a small number of finite
states at any given instant in time. A novel statistical methodology
based upon log-linear modelling is proposed for analyzing categorical
time-series data which allows one to incorporate a considerable
amount of prior knowledge directly into the data analysis.
The statistical model can be shown to be formally equivalent to a
connectionist (i.e., artificial neural network) model. Methods for
model selection and hypothesis testing using the new statistical model
for samples with large numbers of observations are then developed using
asymptotic statistical theory. To illustrate this new method of
categorical time-series data
analysis, the model is applied to the analysis of text free recall data
from children and adults.
These analyses indicated that
the model can successfully use the order of recalled text propositions
to discriminate among alternative theories of prior knowledge and alternative
treatment conditions. The reliability of the large sample statistical tests
were also checked using a boot-strap methodology and found to be acceptable.
Semi-final post script draft (possibly missing figures)
Golden's Text Comprehension and Memory Publications