Geoscience Reference
In-Depth Information
LAG SEQUENTIAL AND NESTED ANALYSIS
Sequential analysis facilitates study of the temporal structure of sequences of
events. A common variant is lag sequential analysis, which involves calculating
frequencies of transitions between pairs of events within a certain lag in a time
series. The first event of each pair is called the criterion event (also called
antecedent, X-event, or given event) and the second event is the target event
(also called the consequent or Y-event). Lag sequential analysis allows you to
answer questions such as “How many times is the (criterion) event Head Up
followed by the (target) event Stare?” The time window in which the transition
from criterion to target event occurs is either a state lag or a time lag. In a state-
lag sequential analysis, a transition is counted from a criterion event to the first
target event following this criterion event. In time-lag sequential analysis, a
transition is counted from a criterion event to the target event occurring
within a specified time window following the criterion event. All other events
that are not defined as either criterion or target events are ignored in the analy-
sis. Lag-sequential analysis differs profoundly from other approaches (such as
log-linear models), which compare estimated expected frequencies with
observed frequencies because these do not have a sequential property. The only
way to analyze changes of behavior through time is in terms of transitions
between the antecedent and the consequent behavior. Lag sequential analysis
is currently the only method that applies a contextual approach: Instead of
analyzing the frequencies of different behavior over a certain period of time,
lag sequential analysis uses the frequencies of behavioral transitions (Roberts
1992).
However, the results of lag sequential analysis are rarely expressed in terms
of the absolute frequencies of transition from one event to the others; instead,
they are often written in probabilistic terms as transition rates. The transition
rates, written in transition matrices, can be calculated from the frequencies of
transitions between preceding and consequent behavior using equation 1.6 in
Haccou and Meelis (1992), which is formally written as a maximum likeli-
hood estimator (
MLE
)—equation 10.1 or 10.2—when duration of an act is
constant:
N
N
1
}
a
AB
=
}
´
x
(10.1)
}
A
B
A
or
