estimation.feols_compressed_.FeolsCompressed

estimation.feols_compressed_.FeolsCompressed(
    self,
    FixestFormula,
    data,
    ssc_dict,
    drop_singletons,
    drop_intercept,
    weights,
    weights_type,
    collin_tol,
    fixef_tol,
    lookup_demeaned_data,
    solver,
    demeaner_backend='numba',
    store_data=True,
    copy_data=True,
    lean=False,
    context=0,
    reps=100,
    seed=None,
    sample_split_var=None,
    sample_split_value=None,
)

Non-user-facing class for compressed regression with fixed effects.

See the paper “You only compress once” by Wong et al (https://arxiv.org/abs/2102.11297) for details on regression compression.

Parameters

Name	Type	Description	Default
FixestFormula	FixestFormula	The formula object.	required
data	pd.DataFrame	The data.	required
ssc_dict	dict[str, Union[str, bool]]	The ssc dictionary.	required
drop_singletons	bool	Whether to drop columns with singleton fixed effects.	required
drop_intercept	bool	Whether to include an intercept.	required
weights	Optional[str]	The column name of the weights. None if no weights are used. For this method, weights needs to be None.	required
weights_type	Optional[str]	The type of weights. For this method, weights_type needs to be ‘fweights’.	required
collin_tol	float	The tolerance level for collinearity.	required
fixef_tol	float	The tolerance level for the fixed effects.	required
lookup_demeaned_data	dict[str, pd.DataFrame]	The lookup table for demeaned data.	required
solver	str	The solver to use.	required
store_data	bool	Whether to store the data.	True
copy_data	bool	Whether to copy the data.	True
lean	bool	Whether to keep memory-heavy objects as attributes or not.	False
context	int or Mapping[str, Any]	A dictionary containing additional context variables to be used by formulaic during the creation of the model matrix. This can include custom factorization functions, transformations, or any other variables that need to be available in the formula environment.	0
reps	int	The number of bootstrap repetitions. Default is 100. Only used for CRV1 inference, where a wild cluster bootstrap is used.	100
seed	Optional[int]	The seed for the random number generator. Only relevant for CRV1 inference, where a wild cluster bootstrap is used.	None

Methods

Name	Description
predict	Compute predicted values.
prepare_model_matrix	Prepare model inputs for estimation.
vcov	Compute the variance-covariance matrix for the compressed regression.

predict

estimation.feols_compressed_.FeolsCompressed.predict(
    newdata=None,
    atol=1e-06,
    btol=1e-06,
    type='link',
)

Compute predicted values.

Parameters

Name	Type	Description	Default
newdata	Optional[DataFrameType]	The new data. If None, makes a prediction based on the uncompressed data set.	None
atol	float	The absolute tolerance.	1e-06
btol	float	The relative tolerance.	1e-06
type	str	The type of prediction.	'link'

Returns

Name	Type	Description
	np.ndarray	The predicted values. If newdata is None, the predicted values are based on the uncompressed data set.

prepare_model_matrix

estimation.feols_compressed_.FeolsCompressed.prepare_model_matrix()

Prepare model inputs for estimation.

vcov

estimation.feols_compressed_.FeolsCompressed.vcov(vcov, data=None)

Compute the variance-covariance matrix for the compressed regression.