6.6.4.2.1. statsmodels.sandbox.panel.random_panel.PanelSample¶
-
class
statsmodels.sandbox.panel.random_panel.
PanelSample
(nobs, k_vars, n_groups, exog=None, within=True, corr_structure=<function eye>, corr_args=(), scale=1, seed=None)[source]¶ data generating process for panel with within correlation
allows various within correlation structures, but no random intercept yet
Parameters: nobs : int
total number of observations
k_vars : int
number of explanatory variables to create in exog, including constant
n_groups int
number of groups in balanced sample
exog : None or ndarray
default is None, in which case a exog is created
within : bool
If True (default), then the exog vary within a group. If False, then only variation across groups is used. TODO: this option needs more work
corr_structure : ndarray or ??
Default is np.eye.
corr_args : tuple
arguments for the corr_structure
scale : float
scale of noise, standard deviation of normal distribution
seed : None or int
If seed is given, then this is used to create the random numbers for the sample.
Notes
The behavior for panel robust covariance estimators seems to differ by a large amount by whether exog have mostly within group or across group variation. I do not understand why this should be the case from the theory, and this would warrant more investigation.
This is just used in one example so far and needs more usage to see what will be useful to add.
6.6.4.2.1.1. Methods¶
__init__ (nobs, k_vars, n_groups[, exog, ...]) |
|
generate_panel () |
generate endog for a random panel dataset with within correlation |
get_y_true () |