total_effect | R Documentation |
Calculate a data frame of total effects, resulting from a pulse experiment
(i.e., an exogenous and temporary change in a single variable in time t=0
) or
a press experiment (i.e., an exogenous and permanent change in a single variable
starting in time t=0
and continuing for n_lags
times), representing the
estimated effect of a change in any variable on every other variable and any time-lag
from 0 (simultaneous effects) to a user-specified maximum lag.
total_effect(object, n_lags = 4, type = c("pulse", "press"))
object |
Output from |
n_lags |
Number of lags over which to calculate total effects |
type |
Whether a pulse or press experiment is intended. A pulse experiment
answers the question: |
Total effects are taken from the Leontief matrix \mathbf{(I-P)^{-1}}
,
where \mathbf{P}
is the path matrix across variables and times.
\mathbf{(I-P)}^{-1} \mathbf{\delta}
calculates the effect of a perturbation represented by vector \mathbf{\delta}
with length n_{\mathrm{lags}} \times n_{\mathrm{J}}
where n_{\mathrm{J}}
is the number of variables.
\mathbf{(I-P)}^{-1} \mathbf{\delta}
calculates the total effect of
a given variable (from)
upon any other variable (to) either in the same time (t=0
), or subsequent times
(t \geq 1
), where \mathbf{\delta} = \mathbf{i}_{\mathrm{T}} \otimes \mathbf{i}_{\mathrm{J}}
,
where \mathbf{i}_{\mathrm{J}}
is one for the from
variable and zero otherwise.
For a pulse experiment, \mathbf{i}_{\mathrm{T}}
is one at t=0
and zero for other times.
For a press experiment, \mathbf{i}_{\mathrm{T}}
is one for all times.
We compute and list the total effect at each time from time t=0
to t=n_lags-1
. For press experiments, this includes transient values as the the total effect
approaches its asymptotic value (if this exists) as t
approaches infinity.
If the analyst wants an asymptotic effect from a press experiment, we recommend
using a high lag (e.g., n_lags = 100
) and then confirming that it has
reached it's asymptote (i.e., the total effect is almost identical for the last
and next-to-last lag), and then reporting the value for that last lag.
A data frame listing the time-lag (lag), variable that is undergoing some exogenous change (from), and the variable being impacted (to), along with the total effect (total_effect) including direct and indirect pathways, and the partial "direct" effect (direct_effect)
### EXAMPLE 1
# Define linear model with slope of 0.5
sem = "
# from, to, lag, name, starting_value
x -> y, 0, slope, 0.5
"
# Build DSEM with specified value for path coefficients
mod = dsem(
sem = sem,
tsdata = ts(data.frame(x=rep(0,20),y=rep(0,20))),
control = dsem_control( run_model = FALSE )
)
# Show that total effect of X on Y from pulse experiment is 0.5 but does not propagate over time
pulse = total_effect(mod, n_lags = 2, type = "pulse")
subset( pulse, from=="x" & to=="y")
### EXAMPLE 2
# Define linear model with slope of 0.5 and autocorrelated response
sem = "
x -> y, 0, slope, 0.5
y -> y, 1, ar_y, 0.8
"
mod = dsem(
sem = sem,
tsdata = ts(data.frame(x=rep(0,20),y=rep(0,20))),
control = dsem_control( run_model = FALSE )
)
# Show that total effect of X on Y from pulse experiment is 0.5 with decay of 0.8 for each time
pulse = total_effect(mod, n_lags = 4, type = "pulse")
subset( pulse, from=="x" & to=="y")
# Show that total effect of X on Y from press experiment asymptotes at 2.5
press = total_effect(mod, n_lags = 50, type = "press")
subset( press, from=="x" & to=="y")
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.