pingouin.cronbach_alpha#

pingouin.cronbach_alpha(data=None, items=None, scores=None, subject=None, nan_policy='pairwise', ci=0.95)[source]#

Cronbach’s alpha reliability measure.

Parameters:

datapandas.DataFrame: Wide or long-format dataframe.
itemsstr: Column in data with the items names (long-format only).
scoresstr: Column in data with the scores (long-format only).
subjectstr: Column in data with the subject identifier (long-format only).
nan_policybool: If ‘listwise’, remove the entire rows that contain missing values (= listwise deletion). If ‘pairwise’ (default), only pairwise missing values are removed when computing the covariance matrix. For more details, please refer to the pandas.DataFrame.cov() method.
cifloat: Confidence interval (.95 = 95%)

Returns:

alphafloat: Cronbach’s alpha

Notes

This function works with both wide and long format dataframe. If you pass a long-format dataframe, you must also pass the items, scores and subj columns (in which case the data will be converted into wide format using the pandas.DataFrame.pivot() method).

Internal consistency is usually measured with Cronbach’s alpha [1], a statistic calculated from the pairwise correlations between items. Internal consistency ranges between negative infinity and one. Coefficient alpha will be negative whenever there is greater within-subject variability than between-subject variability.

Cronbach’s \(\alpha\) is defined as

\[\alpha ={k \over k-1}\left(1-{\sum_{{i=1}}^{k}\sigma_{{y_{i}}}^{2} \over\sigma_{x}^{2}}\right)\]

where \(k\) refers to the number of items, \(\sigma_{x}^{2}\) is the variance of the observed total scores, and \(\sigma_{{y_{i}}}^{2}\) the variance of component \(i\) for the current sample of subjects.

Another formula for Cronbach’s \(\alpha\) is

\[\alpha = \frac{k \times \bar c}{\bar v + (k - 1) \times \bar c}\]

where \(\bar c\) refers to the average of all covariances between items and \(\bar v\) to the average variance of each item.

95% confidence intervals are calculated using Feldt’s method [2]:

\[ \begin{align}\begin{aligned}c_L = 1 - (1 - \alpha) \cdot F_{(0.025, n-1, (n-1)(k-1))}\\c_U = 1 - (1 - \alpha) \cdot F_{(0.975, n-1, (n-1)(k-1))}\end{aligned}\end{align} \]

where \(n\) is the number of subjects and \(k\) the number of items.

Results have been tested against the psych R package.

References

[1]

http://www.real-statistics.com/reliability/cronbachs-alpha/

[2]

Feldt, Leonard S., Woodruff, David J., & Salih, Fathi A. (1987). Statistical inference for coefficient alpha. Applied Psychological Measurement, 11(1):93-103.

Examples

Binary wide-format dataframe (with missing values)

>>> import pingouin as pg
>>> data = pg.read_dataset('cronbach_wide_missing')
>>> # In R: psych:alpha(data, use="pairwise")
>>> pg.cronbach_alpha(data=data)
(0.732660835214447, array([0.435, 0.909]))

After listwise deletion of missing values (remove the entire rows)

>>> # In R: psych:alpha(data, use="complete.obs")
>>> pg.cronbach_alpha(data=data, nan_policy='listwise')
(0.8016949152542373, array([0.581, 0.933]))

After imputing the missing values with the median of each column

>>> pg.cronbach_alpha(data=data.fillna(data.median()))
(0.7380191693290734, array([0.447, 0.911]))

Likert-type long-format dataframe

>>> data = pg.read_dataset('cronbach_alpha')
>>> pg.cronbach_alpha(data=data, items='Items', scores='Scores',
...                   subject='Subj')
(0.5917188485995826, array([0.195, 0.84 ]))