def compute_stats(self):
''' Compute several statistics and metrics. '''
dataset = self.dataset
# Inter-scale correlation matrix
self.y_corrs = np.corrcoef(dataset.y, rowvar=0)
if self.key is not None:
# Cronbach's alpha
self.alpha = stats.cronbach_alpha(dataset.X.values, self.key)
# Predicted scores
self.predicted_y = np.dot(dataset.X, self.key)
# R-squared
self.r_squared = (np.corrcoef(dataset.y,
self.predicted_y,
rowvar=0)[0:self.n_y,
self.n_y::]**2).diagonal()
# Number of items per scale
self.n_items_per_scale = np.sum(np.abs(self.key), 0)
# Correlation matrix for predicted scores
self.predicted_y_corrs = np.corrcoef(self.predicted_y, rowvar=0)
def test_cronbach_alpha(self):
scores = np.array([[3, 4, 3, 3], [5, 4, 4, 3], [1, 3, 4, 5],
[4, 5, 4, 2], [1, 3, 4, 1], [3, 3, 3, 1],
[3, 4, 5, 1], [1, 2, 1, 3]])
# Test three scales: one is sum of all items; second has reverse keyed items;
# third omits an item. Correct alphas from the psych package in R. Values for
# second and third examples are nonsensical, but that's okay for testing.
alphas = gs.cronbach_alpha(
scores[:, :4],
np.array([[1, 1, 1, 1], [1, -1, -1, 1], [0, -1, 1, 0]]).T)
npt.assert_array_almost_equal(alphas, np.array([0.409, -0.461,
-3.333]))
def test_cronbach_alpha(self):
scores = np.array([[ 3,4,3,3],
[ 5,4,4,3],
[ 1,3,4,5],
[ 4,5,4,2],
[ 1,3,4,1],
[ 3,3,3,1],
[ 3,4,5,1],
[ 1,2,1,3]])
# Test three scales: one is sum of all items; second has reverse keyed items;
# third omits an item. Correct alphas from the psych package in R. Values for
# second and third examples are nonsensical, but that's okay for testing.
alphas = gs.cronbach_alpha(scores[:,:4], np.array([
[1,1,1,1],
[1,-1,-1,1],
[0,-1,1,0]]).T)
npt.assert_array_almost_equal(alphas, np.array([0.409, -0.461, -3.333]))
def compute_stats(self):
''' Compute several statistics and metrics. '''
dataset = self.dataset
# Inter-scale correlation matrix
self.y_corrs = np.corrcoef(dataset.y, rowvar=0)
if self.key is not None:
# Cronbach's alpha
self.alpha = stats.cronbach_alpha(dataset.X.values, self.key)
# Predicted scores
self.predicted_y = np.dot(dataset.X, self.key)
# R-squared
self.r_squared = (np.corrcoef(dataset.y, self.predicted_y, rowvar=0)[0:self.n_y, self.n_y::] ** 2).diagonal()
# Number of items per scale
self.n_items_per_scale = np.sum(np.abs(self.key), 0)
# Correlation matrix for predicted scores
self.predicted_y_corrs = np.corrcoef(self.predicted_y, rowvar=0)